From db8f44d907425fa8514714302e6a3b9b3dc29242 Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Mon, 16 Feb 2026 13:50:37 +0100 Subject: [PATCH 1/7] amd-isp4 Signed-off-by: Peter Jung --- Documentation/admin-guide/media/amdisp4-1.rst | 63 + Documentation/admin-guide/media/amdisp4.dot | 6 + .../admin-guide/media/v4l-drivers.rst | 1 + MAINTAINERS | 25 + drivers/media/platform/Kconfig | 1 + drivers/media/platform/Makefile | 1 + drivers/media/platform/amd/Kconfig | 3 + drivers/media/platform/amd/Makefile | 3 + drivers/media/platform/amd/isp4/Kconfig | 14 + drivers/media/platform/amd/isp4/Makefile | 10 + drivers/media/platform/amd/isp4/isp4.c | 235 ++++ drivers/media/platform/amd/isp4/isp4.h | 20 + drivers/media/platform/amd/isp4/isp4_debug.c | 271 ++++ drivers/media/platform/amd/isp4/isp4_debug.h | 41 + .../platform/amd/isp4/isp4_fw_cmd_resp.h | 314 +++++ drivers/media/platform/amd/isp4/isp4_hw_reg.h | 124 ++ .../media/platform/amd/isp4/isp4_interface.c | 789 +++++++++++ .../media/platform/amd/isp4/isp4_interface.h | 141 ++ drivers/media/platform/amd/isp4/isp4_subdev.c | 1057 +++++++++++++++ drivers/media/platform/amd/isp4/isp4_subdev.h | 131 ++ drivers/media/platform/amd/isp4/isp4_video.c | 1165 +++++++++++++++++ drivers/media/platform/amd/isp4/isp4_video.h | 65 + 22 files changed, 4480 insertions(+) create mode 100644 Documentation/admin-guide/media/amdisp4-1.rst create mode 100644 Documentation/admin-guide/media/amdisp4.dot create mode 100644 drivers/media/platform/amd/Kconfig create mode 100644 drivers/media/platform/amd/Makefile create mode 100644 drivers/media/platform/amd/isp4/Kconfig create mode 100644 drivers/media/platform/amd/isp4/Makefile create mode 100644 drivers/media/platform/amd/isp4/isp4.c create mode 100644 drivers/media/platform/amd/isp4/isp4.h create mode 100644 drivers/media/platform/amd/isp4/isp4_debug.c create mode 100644 drivers/media/platform/amd/isp4/isp4_debug.h create mode 100644 drivers/media/platform/amd/isp4/isp4_fw_cmd_resp.h create mode 100644 drivers/media/platform/amd/isp4/isp4_hw_reg.h create mode 100644 drivers/media/platform/amd/isp4/isp4_interface.c create mode 100644 drivers/media/platform/amd/isp4/isp4_interface.h create mode 100644 drivers/media/platform/amd/isp4/isp4_subdev.c create mode 100644 drivers/media/platform/amd/isp4/isp4_subdev.h create mode 100644 drivers/media/platform/amd/isp4/isp4_video.c create mode 100644 drivers/media/platform/amd/isp4/isp4_video.h diff --git a/Documentation/admin-guide/media/amdisp4-1.rst b/Documentation/admin-guide/media/amdisp4-1.rst new file mode 100644 index 000000000000..878141154f96 --- /dev/null +++ b/Documentation/admin-guide/media/amdisp4-1.rst @@ -0,0 +1,63 @@ +.. SPDX-License-Identifier: GPL-2.0 + +.. include:: + +==================================== +AMD Image Signal Processor (amdisp4) +==================================== + +Introduction +============ + +This file documents the driver for the AMD ISP4 that is part of +AMD Ryzen AI Max 300 Series. + +The driver is located under drivers/media/platform/amd/isp4 and uses +the Media-Controller API. + +The driver exposes one video capture device to userspace and provide +web camera like interface. Internally the video device is connected +to the isp4 sub-device responsible for communication with the CCPU FW. + +Topology +======== + +.. _amdisp4_topology_graph: + +.. kernel-figure:: amdisp4.dot + :alt: Diagram of the media pipeline topology + :align: center + + + +The driver has 1 sub-device: Representing isp4 image signal processor. +The driver has 1 video device: Capture device for retrieving images. + +- ISP4 Image Signal Processing Subdevice Node + +--------------------------------------------- + +The isp4 is represented as a single V4L2 subdev, the sub-device does not +provide interface to the user space. The sub-device is connected to one video node +(isp4_capture) with immutable active link. The sub-device represents ISP with +connected sensor similar to smart cameras (sensors with integrated ISP). +sub-device has only one link to the video device for capturing the frames. +The sub-device communicates with CCPU FW for streaming configuration and +buffer management. + + +- isp4_capture - Frames Capture Video Node + +------------------------------------------ + +Isp4_capture is a capture device to capture frames to memory. +The entity is connected to isp4 sub-device. The video device +provides web camera like interface to userspace. It supports +mmap and dma buf types of memory. + +Capturing Video Frames Example +============================== + +.. code-block:: bash + + v4l2-ctl "-d" "/dev/video0" "--set-fmt-video=width=1920,height=1080,pixelformat=NV12" "--stream-mmap" "--stream-count=10" diff --git a/Documentation/admin-guide/media/amdisp4.dot b/Documentation/admin-guide/media/amdisp4.dot new file mode 100644 index 000000000000..978f30c1a31a --- /dev/null +++ b/Documentation/admin-guide/media/amdisp4.dot @@ -0,0 +1,6 @@ +digraph board { + rankdir=TB + n00000001 [label="{{} | amd isp4\n | { 0}}", shape=Mrecord, style=filled, fillcolor=green] + n00000001:port0 -> n00000003 [style=bold] + n00000003 [label="Preview\n/dev/video0", shape=box, style=filled, fillcolor=yellow] +} diff --git a/Documentation/admin-guide/media/v4l-drivers.rst b/Documentation/admin-guide/media/v4l-drivers.rst index 393f83e8dc4d..0fb88449fffd 100644 --- a/Documentation/admin-guide/media/v4l-drivers.rst +++ b/Documentation/admin-guide/media/v4l-drivers.rst @@ -9,6 +9,7 @@ Video4Linux (V4L) driver-specific documentation .. toctree:: :maxdepth: 2 + amdisp4-1 bttv c3-isp cafe_ccic diff --git a/MAINTAINERS b/MAINTAINERS index e08767323763..2f1439d6fe7e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1153,6 +1153,31 @@ T: git git://git.kernel.org/pub/scm/linux/kernel/git/iommu/linux.git F: drivers/iommu/amd/ F: include/linux/amd-iommu.h +AMD ISP4 DRIVER +M: Bin Du +M: Nirujogi Pratap +L: linux-media@vger.kernel.org +S: Maintained +T: git git://linuxtv.org/media.git +F: Documentation/admin-guide/media/amdisp4-1.rst +F: Documentation/admin-guide/media/amdisp4.dot +F: drivers/media/platform/amd/Kconfig +F: drivers/media/platform/amd/Makefile +F: drivers/media/platform/amd/isp4/Kconfig +F: drivers/media/platform/amd/isp4/Makefile +F: drivers/media/platform/amd/isp4/isp4.c +F: drivers/media/platform/amd/isp4/isp4.h +F: drivers/media/platform/amd/isp4/isp4_debug.c +F: drivers/media/platform/amd/isp4/isp4_debug.h +F: drivers/media/platform/amd/isp4/isp4_fw_cmd_resp.h +F: drivers/media/platform/amd/isp4/isp4_hw_reg.h +F: drivers/media/platform/amd/isp4/isp4_interface.c +F: drivers/media/platform/amd/isp4/isp4_interface.h +F: drivers/media/platform/amd/isp4/isp4_subdev.c +F: drivers/media/platform/amd/isp4/isp4_subdev.h +F: drivers/media/platform/amd/isp4/isp4_video.c +F: drivers/media/platform/amd/isp4/isp4_video.h + AMD KFD M: Felix Kuehling L: amd-gfx@lists.freedesktop.org diff --git a/drivers/media/platform/Kconfig b/drivers/media/platform/Kconfig index 3f0b7bb68cc9..0b33e927bd59 100644 --- a/drivers/media/platform/Kconfig +++ b/drivers/media/platform/Kconfig @@ -63,6 +63,7 @@ config VIDEO_MUX # Platform drivers - Please keep it alphabetically sorted source "drivers/media/platform/allegro-dvt/Kconfig" +source "drivers/media/platform/amd/Kconfig" source "drivers/media/platform/amlogic/Kconfig" source "drivers/media/platform/amphion/Kconfig" source "drivers/media/platform/arm/Kconfig" diff --git a/drivers/media/platform/Makefile b/drivers/media/platform/Makefile index 6d5f79ddfcc3..16c185752474 100644 --- a/drivers/media/platform/Makefile +++ b/drivers/media/platform/Makefile @@ -6,6 +6,7 @@ # Place here, alphabetically sorted by directory # (e. g. LC_ALL=C sort Makefile) obj-y += allegro-dvt/ +obj-y += amd/ obj-y += amlogic/ obj-y += amphion/ obj-y += arm/ diff --git a/drivers/media/platform/amd/Kconfig b/drivers/media/platform/amd/Kconfig new file mode 100644 index 000000000000..25af49f246b2 --- /dev/null +++ b/drivers/media/platform/amd/Kconfig @@ -0,0 +1,3 @@ +# SPDX-License-Identifier: GPL-2.0+ + +source "drivers/media/platform/amd/isp4/Kconfig" diff --git a/drivers/media/platform/amd/Makefile b/drivers/media/platform/amd/Makefile new file mode 100644 index 000000000000..8bfc1955f22e --- /dev/null +++ b/drivers/media/platform/amd/Makefile @@ -0,0 +1,3 @@ +# SPDX-License-Identifier: GPL-2.0+ + +obj-y += isp4/ diff --git a/drivers/media/platform/amd/isp4/Kconfig b/drivers/media/platform/amd/isp4/Kconfig new file mode 100644 index 000000000000..d4e4ad436600 --- /dev/null +++ b/drivers/media/platform/amd/isp4/Kconfig @@ -0,0 +1,14 @@ +# SPDX-License-Identifier: GPL-2.0+ + +config AMD_ISP4 + tristate "AMD ISP4 and camera driver" + depends on DRM_AMD_ISP && VIDEO_DEV + select VIDEOBUF2_CORE + select VIDEOBUF2_MEMOPS + select VIDEOBUF2_V4L2 + select VIDEO_V4L2_SUBDEV_API + help + This is support for AMD ISP4 and camera subsystem driver. + Say Y here to enable the ISP4 and camera device for video capture. + To compile this driver as a module, choose M here. The module will + be called amd_capture. diff --git a/drivers/media/platform/amd/isp4/Makefile b/drivers/media/platform/amd/isp4/Makefile new file mode 100644 index 000000000000..607151c0a2be --- /dev/null +++ b/drivers/media/platform/amd/isp4/Makefile @@ -0,0 +1,10 @@ +# SPDX-License-Identifier: GPL-2.0+ +# +# Copyright (C) 2025 Advanced Micro Devices, Inc. + +obj-$(CONFIG_AMD_ISP4) += amd_capture.o +amd_capture-objs := isp4.o \ + isp4_debug.o \ + isp4_interface.o \ + isp4_subdev.o \ + isp4_video.o diff --git a/drivers/media/platform/amd/isp4/isp4.c b/drivers/media/platform/amd/isp4/isp4.c new file mode 100644 index 000000000000..93cb2d41320e --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4.c @@ -0,0 +1,235 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#include +#include +#include +#include +#include + +#include "isp4.h" +#include "isp4_debug.h" +#include "isp4_hw_reg.h" + +#define ISP4_DRV_NAME "amd_isp_capture" +#define ISP4_FW_RESP_RB_IRQ_STATUS_MASK \ + (ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT9_INT_MASK | \ + ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT12_INT_MASK) + +static const struct { + const char *name; + u32 status_mask; + u32 en_mask; + u32 ack_mask; + u32 rb_int_num; +} isp4_irq[ISP4SD_MAX_FW_RESP_STREAM_NUM] = { + /* The IRQ order is aligned with the isp4_subdev.fw_resp_thread order */ + { + .name = "isp_irq_global", + .status_mask = ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT12_INT_MASK, + .en_mask = ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT12_EN_MASK, + .ack_mask = ISP_SYS_INT0_ACK__SYS_INT_RINGBUFFER_WPT12_ACK_MASK, + .rb_int_num = 4, /* ISP_4_1__SRCID__ISP_RINGBUFFER_WPT12 */ + }, + { + .name = "isp_irq_stream1", + .status_mask = ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT9_INT_MASK, + .en_mask = ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT9_EN_MASK, + .ack_mask = ISP_SYS_INT0_ACK__SYS_INT_RINGBUFFER_WPT9_ACK_MASK, + .rb_int_num = 0, /* ISP_4_1__SRCID__ISP_RINGBUFFER_WPT9 */ + }, +}; + +void isp4_intr_enable(struct isp4_subdev *isp_subdev, u32 index, bool enable) +{ + u32 intr_en; + + /* Synchronize ISP_SYS_INT0_EN writes with the IRQ handler's writes */ + spin_lock_irq(&isp_subdev->irq_lock); + intr_en = isp4hw_rreg(isp_subdev->mmio, ISP_SYS_INT0_EN); + if (enable) + intr_en |= isp4_irq[index].en_mask; + else + intr_en &= ~isp4_irq[index].en_mask; + + isp4hw_wreg(isp_subdev->mmio, ISP_SYS_INT0_EN, intr_en); + spin_unlock_irq(&isp_subdev->irq_lock); +} + +static void isp4_wake_up_resp_thread(struct isp4_subdev *isp_subdev, u32 index) +{ + struct isp4sd_thread_handler *thread_ctx = &isp_subdev->fw_resp_thread[index]; + + thread_ctx->resp_ready = true; + wake_up_interruptible(&thread_ctx->waitq); +} + +static irqreturn_t isp4_irq_handler(int irq, void *arg) +{ + struct isp4_subdev *isp_subdev = arg; + u32 intr_ack = 0, intr_en = 0, intr_status; + int seen = 0; + + /* Get the ISP_SYS interrupt status */ + intr_status = isp4hw_rreg(isp_subdev->mmio, ISP_SYS_INT0_STATUS); + intr_status &= ISP4_FW_RESP_RB_IRQ_STATUS_MASK; + + /* Find which ISP_SYS interrupts fired */ + for (size_t i = 0; i < ARRAY_SIZE(isp4_irq); i++) { + if (intr_status & isp4_irq[i].status_mask) { + intr_ack |= isp4_irq[i].ack_mask; + intr_en |= isp4_irq[i].en_mask; + seen |= BIT(i); + } + } + + /* + * Disable the ISP_SYS interrupts that fired. Must be done before waking + * the response threads, since they re-enable interrupts when finished. + * The lock synchronizes RMW of INT0_EN with isp4_enable_interrupt(). + */ + spin_lock(&isp_subdev->irq_lock); + intr_en = isp4hw_rreg(isp_subdev->mmio, ISP_SYS_INT0_EN) & ~intr_en; + isp4hw_wreg(isp_subdev->mmio, ISP_SYS_INT0_EN, intr_en); + spin_unlock(&isp_subdev->irq_lock); + + /* + * Clear the ISP_SYS interrupts. This must be done after the interrupts + * are disabled, so that ISP FW won't flag any new interrupts on these + * streams, and thus we don't need to clear interrupts again before + * re-enabling them in the response thread. + */ + isp4hw_wreg(isp_subdev->mmio, ISP_SYS_INT0_ACK, intr_ack); + + /* Wake up the response threads */ + for (int i; (i = ffs(seen)); seen = (seen >> i) << i) + isp4_wake_up_resp_thread(isp_subdev, i - 1); + + return IRQ_HANDLED; +} + +static int isp4_capture_probe(struct platform_device *pdev) +{ + int irq[ISP4SD_MAX_FW_RESP_STREAM_NUM]; + struct device *dev = &pdev->dev; + struct isp4_subdev *isp_subdev; + struct isp4_device *isp_dev; + size_t i; + int ret; + + isp_dev = devm_kzalloc(dev, sizeof(*isp_dev), GFP_KERNEL); + if (!isp_dev) + return -ENOMEM; + + dev->init_name = ISP4_DRV_NAME; + + isp_subdev = &isp_dev->isp_subdev; + isp_subdev->mmio = devm_platform_ioremap_resource(pdev, 0); + if (IS_ERR(isp_subdev->mmio)) + return dev_err_probe(dev, PTR_ERR(isp_subdev->mmio), "isp ioremap fail\n"); + + for (i = 0; i < ARRAY_SIZE(isp4_irq); i++) { + irq[i] = platform_get_irq(pdev, isp4_irq[i].rb_int_num); + if (irq[i] < 0) + return dev_err_probe(dev, irq[i], "fail to get irq %d\n", + isp4_irq[i].rb_int_num); + + ret = devm_request_irq(dev, irq[i], isp4_irq_handler, + IRQF_NO_AUTOEN, isp4_irq[i].name, isp_subdev); + if (ret) + return dev_err_probe(dev, ret, "fail to req irq %d\n", irq[i]); + } + + /* Link the media device within the v4l2_device */ + isp_dev->v4l2_dev.mdev = &isp_dev->mdev; + + /* Initialize media device */ + strscpy(isp_dev->mdev.model, "amd_isp41_mdev", sizeof(isp_dev->mdev.model)); + snprintf(isp_dev->mdev.bus_info, sizeof(isp_dev->mdev.bus_info), + "platform:%s", ISP4_DRV_NAME); + isp_dev->mdev.dev = dev; + media_device_init(&isp_dev->mdev); + + /* register v4l2 device */ + snprintf(isp_dev->v4l2_dev.name, sizeof(isp_dev->v4l2_dev.name), + "AMD-V4L2-ROOT"); + ret = v4l2_device_register(dev, &isp_dev->v4l2_dev); + if (ret) { + dev_err_probe(dev, ret, "fail register v4l2 device\n"); + goto err_clean_media; + } + + pm_runtime_set_suspended(dev); + pm_runtime_enable(dev); + spin_lock_init(&isp_subdev->irq_lock); + ret = isp4sd_init(&isp_dev->isp_subdev, &isp_dev->v4l2_dev, irq); + if (ret) { + dev_err_probe(dev, ret, "fail init isp4 sub dev\n"); + goto err_pm_disable; + } + + ret = media_create_pad_link(&isp_dev->isp_subdev.sdev.entity, + 0, &isp_dev->isp_subdev.isp_vdev.vdev.entity, + 0, + MEDIA_LNK_FL_ENABLED | + MEDIA_LNK_FL_IMMUTABLE); + if (ret) { + dev_err_probe(dev, ret, "fail to create pad link\n"); + goto err_isp4_deinit; + } + + ret = media_device_register(&isp_dev->mdev); + if (ret) { + dev_err_probe(dev, ret, "fail to register media device\n"); + goto err_isp4_deinit; + } + + platform_set_drvdata(pdev, isp_dev); + isp_debugfs_create(isp_dev); + + return 0; + +err_isp4_deinit: + isp4sd_deinit(&isp_dev->isp_subdev); +err_pm_disable: + pm_runtime_disable(dev); + v4l2_device_unregister(&isp_dev->v4l2_dev); +err_clean_media: + media_device_cleanup(&isp_dev->mdev); + + return ret; +} + +static void isp4_capture_remove(struct platform_device *pdev) +{ + struct isp4_device *isp_dev = platform_get_drvdata(pdev); + struct device *dev = &pdev->dev; + + isp_debugfs_remove(isp_dev); + + media_device_unregister(&isp_dev->mdev); + isp4sd_deinit(&isp_dev->isp_subdev); + pm_runtime_disable(dev); + v4l2_device_unregister(&isp_dev->v4l2_dev); + media_device_cleanup(&isp_dev->mdev); +} + +static struct platform_driver isp4_capture_drv = { + .probe = isp4_capture_probe, + .remove = isp4_capture_remove, + .driver = { + .name = ISP4_DRV_NAME, + } +}; + +module_platform_driver(isp4_capture_drv); + +MODULE_ALIAS("platform:" ISP4_DRV_NAME); +MODULE_IMPORT_NS("DMA_BUF"); + +MODULE_DESCRIPTION("AMD ISP4 Driver"); +MODULE_AUTHOR("Bin Du "); +MODULE_AUTHOR("Pratap Nirujogi "); +MODULE_LICENSE("GPL"); diff --git a/drivers/media/platform/amd/isp4/isp4.h b/drivers/media/platform/amd/isp4/isp4.h new file mode 100644 index 000000000000..2db6683d6d8b --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4.h @@ -0,0 +1,20 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#ifndef _ISP4_H_ +#define _ISP4_H_ + +#include +#include "isp4_subdev.h" + +struct isp4_device { + struct v4l2_device v4l2_dev; + struct isp4_subdev isp_subdev; + struct media_device mdev; +}; + +void isp4_intr_enable(struct isp4_subdev *isp_subdev, u32 index, bool enable); + +#endif /* _ISP4_H_ */ diff --git a/drivers/media/platform/amd/isp4/isp4_debug.c b/drivers/media/platform/amd/isp4/isp4_debug.c new file mode 100644 index 000000000000..c6d957ea9132 --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_debug.c @@ -0,0 +1,271 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#include "isp4.h" +#include "isp4_debug.h" +#include "isp4_hw_reg.h" +#include "isp4_interface.h" + +#define ISP4DBG_FW_LOG_RINGBUF_SIZE (2 * 1024 * 1024) +#define ISP4DBG_MACRO_2_STR(X) #X +#define ISP4DBG_MAX_ONE_TIME_LOG_LEN 510 + +#ifdef CONFIG_DEBUG_FS + +void isp_debugfs_create(struct isp4_device *isp_dev) +{ + isp_dev->isp_subdev.debugfs_dir = debugfs_create_dir("amd_isp", NULL); + debugfs_create_bool("fw_log_enable", 0644, + isp_dev->isp_subdev.debugfs_dir, + &isp_dev->isp_subdev.enable_fw_log); + isp_dev->isp_subdev.fw_log_output = + devm_kzalloc(isp_dev->isp_subdev.dev, + ISP4DBG_FW_LOG_RINGBUF_SIZE + 32, + GFP_KERNEL); +} + +void isp_debugfs_remove(struct isp4_device *isp_dev) +{ + debugfs_remove_recursive(isp_dev->isp_subdev.debugfs_dir); + isp_dev->isp_subdev.debugfs_dir = NULL; +} + +static u32 isp_fw_fill_rb_log(struct isp4_subdev *isp, void *sys, u32 rb_size) +{ + struct isp4_interface *ispif = &isp->ispif; + char *buf = isp->fw_log_output; + struct device *dev = isp->dev; + u32 rd_ptr, wr_ptr; + u32 total_cnt = 0; + u32 offset = 0; + u32 cnt; + + if (!sys || !rb_size) + return 0; + + guard(mutex)(&ispif->isp4if_mutex); + + rd_ptr = isp4hw_rreg(isp->mmio, ISP_LOG_RB_RPTR0); + wr_ptr = isp4hw_rreg(isp->mmio, ISP_LOG_RB_WPTR0); + + do { + if (wr_ptr > rd_ptr) + cnt = wr_ptr - rd_ptr; + else if (wr_ptr < rd_ptr) + cnt = rb_size - rd_ptr; + else + goto quit; + + if (cnt > rb_size) { + dev_err(dev, "fail bad fw log size %u\n", cnt); + goto quit; + } + + memcpy(buf + offset, sys + rd_ptr, cnt); + + offset += cnt; + total_cnt += cnt; + rd_ptr = (rd_ptr + cnt) % rb_size; + } while (rd_ptr < wr_ptr); + + isp4hw_wreg(isp->mmio, ISP_LOG_RB_RPTR0, rd_ptr); + +quit: + return total_cnt; +} + +void isp_fw_log_print(struct isp4_subdev *isp) +{ + struct isp4_interface *ispif = &isp->ispif; + char *fw_log_buf = isp->fw_log_output; + u32 cnt; + + if (!isp->enable_fw_log || !fw_log_buf) + return; + + cnt = isp_fw_fill_rb_log(isp, ispif->fw_log_buf->sys_addr, + ispif->fw_log_buf->mem_size); + + if (cnt) { + char *line_end; + char temp_ch; + char *str; + char *end; + + str = (char *)fw_log_buf; + end = ((char *)fw_log_buf + cnt); + fw_log_buf[cnt] = 0; + + while (str < end) { + line_end = strchr(str, 0x0A); + if ((line_end && (str + ISP4DBG_MAX_ONE_TIME_LOG_LEN) >= line_end) || + (!line_end && (str + ISP4DBG_MAX_ONE_TIME_LOG_LEN) >= end)) { + if (line_end) + *line_end = 0; + + if (*str != '\0') + dev_dbg(isp->dev, + "%s", str); + + if (line_end) { + *line_end = 0x0A; + str = line_end + 1; + } else { + break; + } + } else { + u32 tmp_len = ISP4DBG_MAX_ONE_TIME_LOG_LEN; + + temp_ch = str[tmp_len]; + str[tmp_len] = 0; + dev_dbg(isp->dev, "%s", str); + str[tmp_len] = temp_ch; + str = &str[tmp_len]; + } + } + } +} +#endif + +char *isp4dbg_get_buf_src_str(u32 src) +{ + switch (src) { + case BUFFER_SOURCE_STREAM: + return ISP4DBG_MACRO_2_STR(BUFFER_SOURCE_STREAM); + default: + return "Unknown buf source"; + } +} + +char *isp4dbg_get_buf_done_str(u32 status) +{ + switch (status) { + case BUFFER_STATUS_INVALID: + return ISP4DBG_MACRO_2_STR(BUFFER_STATUS_INVALID); + case BUFFER_STATUS_SKIPPED: + return ISP4DBG_MACRO_2_STR(BUFFER_STATUS_SKIPPED); + case BUFFER_STATUS_EXIST: + return ISP4DBG_MACRO_2_STR(BUFFER_STATUS_EXIST); + case BUFFER_STATUS_DONE: + return ISP4DBG_MACRO_2_STR(BUFFER_STATUS_DONE); + case BUFFER_STATUS_LACK: + return ISP4DBG_MACRO_2_STR(BUFFER_STATUS_LACK); + case BUFFER_STATUS_DIRTY: + return ISP4DBG_MACRO_2_STR(BUFFER_STATUS_DIRTY); + case BUFFER_STATUS_MAX: + return ISP4DBG_MACRO_2_STR(BUFFER_STATUS_MAX); + default: + return "Unknown Buf Done Status"; + } +} + +char *isp4dbg_get_img_fmt_str(int fmt /* enum isp4fw_image_format * */) +{ + switch (fmt) { + case IMAGE_FORMAT_NV12: + return "NV12"; + case IMAGE_FORMAT_YUV422INTERLEAVED: + return "YUV422INTERLEAVED"; + default: + return "unknown fmt"; + } +} + +void isp4dbg_show_bufmeta_info(struct device *dev, char *pre, + void *in, void *orig_buf) +{ + struct isp4fw_buffer_meta_info *p; + struct isp4if_img_buf_info *orig; + + if (!in) + return; + + if (!pre) + pre = ""; + + p = in; + orig = orig_buf; + + dev_dbg(dev, "%s(%s) en:%d,stat:%s(%u),src:%s\n", pre, + isp4dbg_get_img_fmt_str(p->image_prop.image_format), + p->enabled, isp4dbg_get_buf_done_str(p->status), p->status, + isp4dbg_get_buf_src_str(p->source)); + + dev_dbg(dev, "%p,0x%llx(%u) %p,0x%llx(%u) %p,0x%llx(%u)\n", + orig->planes[0].sys_addr, orig->planes[0].mc_addr, + orig->planes[0].len, orig->planes[1].sys_addr, + orig->planes[1].mc_addr, orig->planes[1].len, + orig->planes[2].sys_addr, orig->planes[2].mc_addr, + orig->planes[2].len); +} + +char *isp4dbg_get_buf_type(u32 type) +{ + /* enum isp4fw_buffer_type */ + switch (type) { + case BUFFER_TYPE_PREVIEW: + return ISP4DBG_MACRO_2_STR(BUFFER_TYPE_PREVIEW); + case BUFFER_TYPE_META_INFO: + return ISP4DBG_MACRO_2_STR(BUFFER_TYPE_META_INFO); + case BUFFER_TYPE_MEM_POOL: + return ISP4DBG_MACRO_2_STR(BUFFER_TYPE_MEM_POOL); + default: + return "unknown type"; + } +} + +char *isp4dbg_get_cmd_str(u32 cmd) +{ + switch (cmd) { + case CMD_ID_START_STREAM: + return ISP4DBG_MACRO_2_STR(CMD_ID_START_STREAM); + case CMD_ID_STOP_STREAM: + return ISP4DBG_MACRO_2_STR(CMD_ID_STOP_STREAM); + case CMD_ID_SEND_BUFFER: + return ISP4DBG_MACRO_2_STR(CMD_ID_SEND_BUFFER); + case CMD_ID_SET_STREAM_CONFIG: + return ISP4DBG_MACRO_2_STR(CMD_ID_SET_STREAM_CONFIG); + case CMD_ID_SET_OUT_CHAN_PROP: + return ISP4DBG_MACRO_2_STR(CMD_ID_SET_OUT_CHAN_PROP); + case CMD_ID_ENABLE_OUT_CHAN: + return ISP4DBG_MACRO_2_STR(CMD_ID_ENABLE_OUT_CHAN); + default: + return "unknown cmd"; + } +} + +char *isp4dbg_get_resp_str(u32 cmd) +{ + switch (cmd) { + case RESP_ID_CMD_DONE: + return ISP4DBG_MACRO_2_STR(RESP_ID_CMD_DONE); + case RESP_ID_NOTI_FRAME_DONE: + return ISP4DBG_MACRO_2_STR(RESP_ID_NOTI_FRAME_DONE); + default: + return "unknown respid"; + } +} + +char *isp4dbg_get_if_stream_str(u32 stream /* enum fw_cmd_resp_stream_id */) +{ + switch (stream) { + case ISP4IF_STREAM_ID_GLOBAL: + return "STREAM_GLOBAL"; + case ISP4IF_STREAM_ID_1: + return "STREAM1"; + default: + return "unknown streamID"; + } +} + +char *isp4dbg_get_out_ch_str(int ch /* enum isp4fw_pipe_out_ch */) +{ + switch ((enum isp4fw_pipe_out_ch)ch) { + case ISP_PIPE_OUT_CH_PREVIEW: + return "prev"; + default: + return "unknown channel"; + } +} diff --git a/drivers/media/platform/amd/isp4/isp4_debug.h b/drivers/media/platform/amd/isp4/isp4_debug.h new file mode 100644 index 000000000000..1a13762af502 --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_debug.h @@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#ifndef _ISP4_DEBUG_H_ +#define _ISP4_DEBUG_H_ + +#include +#include + +#include "isp4_subdev.h" + +#ifdef CONFIG_DEBUG_FS +struct isp4_device; + +void isp_debugfs_create(struct isp4_device *isp_dev); +void isp_debugfs_remove(struct isp4_device *isp_dev); +void isp_fw_log_print(struct isp4_subdev *isp); + +#else + +/* to avoid checkpatch warning */ +#define isp_debugfs_create(cam) ((void)(cam)) +#define isp_debugfs_remove(cam) ((void)(cam)) +#define isp_fw_log_print(isp) ((void)(isp)) + +#endif /* CONFIG_DEBUG_FS */ + +void isp4dbg_show_bufmeta_info(struct device *dev, char *pre, void *p, + void *orig_buf /* struct sys_img_buf_handle */); +char *isp4dbg_get_img_fmt_str(int fmt /* enum _image_format_t */); +char *isp4dbg_get_out_ch_str(int ch /* enum _isp_pipe_out_ch_t */); +char *isp4dbg_get_cmd_str(u32 cmd); +char *isp4dbg_get_buf_type(u32 type);/* enum _buffer_type_t */ +char *isp4dbg_get_resp_str(u32 resp); +char *isp4dbg_get_buf_src_str(u32 src); +char *isp4dbg_get_buf_done_str(u32 status); +char *isp4dbg_get_if_stream_str(u32 stream); + +#endif /* _ISP4_DEBUG_H_ */ diff --git a/drivers/media/platform/amd/isp4/isp4_fw_cmd_resp.h b/drivers/media/platform/amd/isp4/isp4_fw_cmd_resp.h new file mode 100644 index 000000000000..d571b3873edb --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_fw_cmd_resp.h @@ -0,0 +1,314 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#ifndef _ISP4_CMD_RESP_H_ +#define _ISP4_CMD_RESP_H_ + +/* + * @brief Host and Firmware command & response channel. + * Two types of command/response channel. + * Type Global Command has one command/response channel. + * Type Stream Command has one command/response channel. + *----------- ------------ + *| | --------------------------- | | + *| | ---->| Global Command |----> | | + *| | --------------------------- | | + *| | | | + *| | | | + *| | --------------------------- | | + *| | ---->| Stream Command |----> | | + *| | --------------------------- | | + *| | | | + *| | | | + *| | | | + *| HOST | | Firmware | + *| | | | + *| | | | + *| | -------------------------- | | + *| | <----| Global Response |<---- | | + *| | -------------------------- | | + *| | | | + *| | | | + *| | -------------------------- | | + *| | <----| Stream Response |<---- | | + *| | -------------------------- | | + *| | | | + *| | | | + *----------- ------------ + */ + +/* + * @brief command ID format + * cmd_id is in the format of following type: + * type: indicate command type, global/stream commands. + * group: indicate the command group. + * id: A unique command identification in one type and group. + * |<-Bit31 ~ Bit24->|<-Bit23 ~ Bit16->|<-Bit15 ~ Bit0->| + * | type | group | id | + */ + +#define CMD_TYPE_SHIFT 24 +#define CMD_GROUP_SHIFT 16 +#define CMD_TYPE_STREAM_CTRL (0x2U << CMD_TYPE_SHIFT) + +#define CMD_GROUP_STREAM_CTRL (0x1U << CMD_GROUP_SHIFT) +#define CMD_GROUP_STREAM_BUFFER (0x4U << CMD_GROUP_SHIFT) + +/* Stream Command */ +#define CMD_ID_SET_STREAM_CONFIG (CMD_TYPE_STREAM_CTRL | CMD_GROUP_STREAM_CTRL | 0x1) +#define CMD_ID_SET_OUT_CHAN_PROP (CMD_TYPE_STREAM_CTRL | CMD_GROUP_STREAM_CTRL | 0x3) +#define CMD_ID_ENABLE_OUT_CHAN (CMD_TYPE_STREAM_CTRL | CMD_GROUP_STREAM_CTRL | 0x5) +#define CMD_ID_START_STREAM (CMD_TYPE_STREAM_CTRL | CMD_GROUP_STREAM_CTRL | 0x7) +#define CMD_ID_STOP_STREAM (CMD_TYPE_STREAM_CTRL | CMD_GROUP_STREAM_CTRL | 0x8) + +/* Stream Buffer Command */ +#define CMD_ID_SEND_BUFFER (CMD_TYPE_STREAM_CTRL | CMD_GROUP_STREAM_BUFFER | 0x1) + +/* + * @brief response ID format + * resp_id is in the format of following type: + * type: indicate command type, global/stream commands. + * group: indicate the command group. + * id: A unique command identification in one type and group. + * |<-Bit31 ~ Bit24->|<-Bit23 ~ Bit16->|<-Bit15 ~ Bit0->| + * | type | group | id | + */ + +#define RESP_GROUP_SHIFT 16 +#define RESP_GROUP_MASK (0xff << RESP_GROUP_SHIFT) + +#define GET_RESP_GROUP_VALUE(resp_id) (((resp_id) & RESP_GROUP_MASK) >> RESP_GROUP_SHIFT) +#define GET_RESP_ID_VALUE(resp_id) ((resp_id) & 0xffff) + +#define RESP_GROUP_GENERAL (0x1 << RESP_GROUP_SHIFT) +#define RESP_GROUP_NOTIFICATION (0x3 << RESP_GROUP_SHIFT) + +/* General Response */ +#define RESP_ID_CMD_DONE (RESP_GROUP_GENERAL | 0x1) + +/* Notification */ +#define RESP_ID_NOTI_FRAME_DONE (RESP_GROUP_NOTIFICATION | 0x1) + +#define CMD_STATUS_SUCCESS 0 +#define CMD_STATUS_FAIL 1 +#define CMD_STATUS_SKIPPED 2 + +#define ADDR_SPACE_TYPE_GPU_VA 4 + +#define FW_MEMORY_POOL_SIZE (100 * 1024 * 1024) + +/* + * standard ISP mipicsi=>isp + */ +#define MIPI0_ISP_PIPELINE_ID 0x5f91 + +enum isp4fw_sensor_id { + SENSOR_ID_ON_MIPI0 = 0, /* Sensor id for ISP input from MIPI port 0 */ +}; + +enum isp4fw_stream_id { + STREAM_ID_INVALID = -1, /* STREAM_ID_INVALID. */ + STREAM_ID_1 = 0, /* STREAM_ID_1. */ + STREAM_ID_2 = 1, /* STREAM_ID_2. */ + STREAM_ID_3 = 2, /* STREAM_ID_3. */ + STREAM_ID_MAXIMUM /* STREAM_ID_MAXIMUM. */ +}; + +enum isp4fw_image_format { + IMAGE_FORMAT_NV12 = 1, /* 4:2:0,semi-planar, 8-bit */ + IMAGE_FORMAT_YUV422INTERLEAVED = 7, /* interleave, 4:2:2, 8-bit */ +}; + +enum isp4fw_pipe_out_ch { + ISP_PIPE_OUT_CH_PREVIEW = 0, +}; + +enum isp4fw_yuv_range { + ISP_YUV_RANGE_FULL = 0, /* YUV value range in 0~255 */ + ISP_YUV_RANGE_NARROW = 1, /* YUV value range in 16~235 */ + ISP_YUV_RANGE_MAX +}; + +enum isp4fw_buffer_type { + BUFFER_TYPE_PREVIEW = 8, + BUFFER_TYPE_META_INFO = 10, + BUFFER_TYPE_MEM_POOL = 15, +}; + +enum isp4fw_buffer_status { + BUFFER_STATUS_INVALID, /* The buffer is INVALID */ + BUFFER_STATUS_SKIPPED, /* The buffer is not filled with image data */ + BUFFER_STATUS_EXIST, /* The buffer is exist and waiting for filled */ + BUFFER_STATUS_DONE, /* The buffer is filled with image data */ + BUFFER_STATUS_LACK, /* The buffer is unavailable */ + BUFFER_STATUS_DIRTY, /* The buffer is dirty, probably caused by + * LMI leakage + */ + BUFFER_STATUS_MAX /* The buffer STATUS_MAX */ +}; + +enum isp4fw_buffer_source { + /* The buffer is from the stream buffer queue */ + BUFFER_SOURCE_STREAM, +}; + +struct isp4fw_error_code { + u32 code1; + u32 code2; + u32 code3; + u32 code4; + u32 code5; +}; + +/* + * Command Structure for FW + */ + +struct isp4fw_cmd { + u32 cmd_seq_num; + u32 cmd_id; + u32 cmd_param[12]; + u16 cmd_stream_id; + u8 cmd_silent_resp; + u8 reserved; + u32 cmd_check_sum; +}; + +struct isp4fw_resp_cmd_done { + /* + * The host2fw command seqNum. + * To indicate which command this response refers to. + */ + u32 cmd_seq_num; + /* The host2fw command id for host double check. */ + u32 cmd_id; + /* + * Indicate the command process status. + * 0 means success. 1 means fail. 2 means skipped + */ + u16 cmd_status; + /* + * If the cmd_status is 1, that means the command is processed fail, + * host can check the isp4fw_error_code to get the details + * error information + */ + u16 isp4fw_error_code; + /* The response payload will be in different struct type */ + /* according to different cmd done response. */ + u8 payload[36]; +}; + +struct isp4fw_resp_param_package { + u32 package_addr_lo; /* The low 32 bit addr of the pkg address. */ + u32 package_addr_hi; /* The high 32 bit addr of the pkg address. */ + u32 package_size; /* The total pkg size in bytes. */ + u32 package_check_sum; /* The byte sum of the pkg. */ +}; + +struct isp4fw_resp { + u32 resp_seq_num; + u32 resp_id; + union { + struct isp4fw_resp_cmd_done cmd_done; + struct isp4fw_resp_param_package frame_done; + u32 resp_param[12]; + } param; + u8 reserved[4]; + u32 resp_check_sum; +}; + +struct isp4fw_mipi_pipe_path_cfg { + u32 b_enable; + enum isp4fw_sensor_id isp4fw_sensor_id; +}; + +struct isp4fw_isp_pipe_path_cfg { + u32 isp_pipe_id; /* pipe ids for pipeline construction */ +}; + +struct isp4fw_isp_stream_cfg { + /* Isp mipi path */ + struct isp4fw_mipi_pipe_path_cfg mipi_pipe_path_cfg; + /* Isp pipe path */ + struct isp4fw_isp_pipe_path_cfg isp_pipe_path_cfg; + /* enable TNR */ + u32 b_enable_tnr; + /* + * number of frame rta per-processing, + * set to 0 to use fw default value + */ + u32 rta_frames_per_proc; +}; + +struct isp4fw_image_prop { + enum isp4fw_image_format image_format; /* Image format */ + u32 width; /* Width */ + u32 height; /* Height */ + u32 luma_pitch; /* Luma pitch */ + u32 chroma_pitch; /* Chrom pitch */ + enum isp4fw_yuv_range yuv_range; /* YUV value range */ +}; + +struct isp4fw_buffer { + /* A check num for debug usage, host need to */ + /* set the buf_tags to different number */ + u32 buf_tags; + union { + u32 value; + struct { + u32 space : 16; + u32 vmid : 16; + } bit; + } vmid_space; + u32 buf_base_a_lo; /* Low address of buffer A */ + u32 buf_base_a_hi; /* High address of buffer A */ + u32 buf_size_a; /* Buffer size of buffer A */ + + u32 buf_base_b_lo; /* Low address of buffer B */ + u32 buf_base_b_hi; /* High address of buffer B */ + u32 buf_size_b; /* Buffer size of buffer B */ + + u32 buf_base_c_lo; /* Low address of buffer C */ + u32 buf_base_c_hi; /* High address of buffer C */ + u32 buf_size_c; /* Buffer size of buffer C */ +}; + +struct isp4fw_buffer_meta_info { + u32 enabled; /* enabled flag */ + enum isp4fw_buffer_status status; /* BufferStatus */ + struct isp4fw_error_code err; /* err code */ + enum isp4fw_buffer_source source; /* BufferSource */ + struct isp4fw_image_prop image_prop; /* image_prop */ + struct isp4fw_buffer buffer; /* buffer */ +}; + +struct isp4fw_meta_info { + u32 poc; /* frame id */ + u32 fc_id; /* frame ctl id */ + u32 time_stamp_lo; /* time_stamp_lo */ + u32 time_stamp_hi; /* time_stamp_hi */ + struct isp4fw_buffer_meta_info preview; /* preview BufferMetaInfo */ +}; + +struct isp4fw_cmd_send_buffer { + enum isp4fw_buffer_type buffer_type; /* buffer Type */ + struct isp4fw_buffer buffer; /* buffer info */ +}; + +struct isp4fw_cmd_set_out_ch_prop { + enum isp4fw_pipe_out_ch ch; /* ISP pipe out channel */ + struct isp4fw_image_prop image_prop; /* image property */ +}; + +struct isp4fw_cmd_enable_out_ch { + enum isp4fw_pipe_out_ch ch; /* ISP pipe out channel */ + u32 is_enable; /* If enable channel or not */ +}; + +struct isp4fw_cmd_set_stream_cfg { + struct isp4fw_isp_stream_cfg stream_cfg; /* stream path config */ +}; + +#endif /* _ISP4_CMD_RESP_H_ */ diff --git a/drivers/media/platform/amd/isp4/isp4_hw_reg.h b/drivers/media/platform/amd/isp4/isp4_hw_reg.h new file mode 100644 index 000000000000..09c76f75c5ee --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_hw_reg.h @@ -0,0 +1,124 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#ifndef _ISP4_HW_REG_H_ +#define _ISP4_HW_REG_H_ + +#include + +#define ISP_SOFT_RESET 0x62000 +#define ISP_SYS_INT0_EN 0x62010 +#define ISP_SYS_INT0_STATUS 0x62014 +#define ISP_SYS_INT0_ACK 0x62018 +#define ISP_CCPU_CNTL 0x62054 +#define ISP_STATUS 0x62058 +#define ISP_LOG_RB_BASE_LO0 0x62148 +#define ISP_LOG_RB_BASE_HI0 0x6214c +#define ISP_LOG_RB_SIZE0 0x62150 +#define ISP_LOG_RB_RPTR0 0x62154 +#define ISP_LOG_RB_WPTR0 0x62158 +#define ISP_RB_BASE_LO1 0x62170 +#define ISP_RB_BASE_HI1 0x62174 +#define ISP_RB_SIZE1 0x62178 +#define ISP_RB_RPTR1 0x6217c +#define ISP_RB_WPTR1 0x62180 +#define ISP_RB_BASE_LO2 0x62184 +#define ISP_RB_BASE_HI2 0x62188 +#define ISP_RB_SIZE2 0x6218c +#define ISP_RB_RPTR2 0x62190 +#define ISP_RB_WPTR2 0x62194 +#define ISP_RB_BASE_LO3 0x62198 +#define ISP_RB_BASE_HI3 0x6219c +#define ISP_RB_SIZE3 0x621a0 +#define ISP_RB_RPTR3 0x621a4 +#define ISP_RB_WPTR3 0x621a8 +#define ISP_RB_BASE_LO4 0x621ac +#define ISP_RB_BASE_HI4 0x621b0 +#define ISP_RB_SIZE4 0x621b4 +#define ISP_RB_RPTR4 0x621b8 +#define ISP_RB_WPTR4 0x621bc +#define ISP_RB_BASE_LO5 0x621c0 +#define ISP_RB_BASE_HI5 0x621c4 +#define ISP_RB_SIZE5 0x621c8 +#define ISP_RB_RPTR5 0x621cc +#define ISP_RB_WPTR5 0x621d0 +#define ISP_RB_BASE_LO6 0x621d4 +#define ISP_RB_BASE_HI6 0x621d8 +#define ISP_RB_SIZE6 0x621dc +#define ISP_RB_RPTR6 0x621e0 +#define ISP_RB_WPTR6 0x621e4 +#define ISP_RB_BASE_LO7 0x621e8 +#define ISP_RB_BASE_HI7 0x621ec +#define ISP_RB_SIZE7 0x621f0 +#define ISP_RB_RPTR7 0x621f4 +#define ISP_RB_WPTR7 0x621f8 +#define ISP_RB_BASE_LO8 0x621fc +#define ISP_RB_BASE_HI8 0x62200 +#define ISP_RB_SIZE8 0x62204 +#define ISP_RB_RPTR8 0x62208 +#define ISP_RB_WPTR8 0x6220c +#define ISP_RB_BASE_LO9 0x62210 +#define ISP_RB_BASE_HI9 0x62214 +#define ISP_RB_SIZE9 0x62218 +#define ISP_RB_RPTR9 0x6221c +#define ISP_RB_WPTR9 0x62220 +#define ISP_RB_BASE_LO10 0x62224 +#define ISP_RB_BASE_HI10 0x62228 +#define ISP_RB_SIZE10 0x6222c +#define ISP_RB_RPTR10 0x62230 +#define ISP_RB_WPTR10 0x62234 +#define ISP_RB_BASE_LO11 0x62238 +#define ISP_RB_BASE_HI11 0x6223c +#define ISP_RB_SIZE11 0x62240 +#define ISP_RB_RPTR11 0x62244 +#define ISP_RB_WPTR11 0x62248 +#define ISP_RB_BASE_LO12 0x6224c +#define ISP_RB_BASE_HI12 0x62250 +#define ISP_RB_SIZE12 0x62254 +#define ISP_RB_RPTR12 0x62258 +#define ISP_RB_WPTR12 0x6225c + +#define ISP_POWER_STATUS 0x60000 + +/* ISP_SOFT_RESET */ +#define ISP_SOFT_RESET__CCPU_SOFT_RESET_MASK 0x00000001UL + +/* ISP_CCPU_CNTL */ +#define ISP_CCPU_CNTL__CCPU_HOST_SOFT_RST_MASK 0x00040000UL + +/* ISP_STATUS */ +#define ISP_STATUS__CCPU_REPORT_MASK 0x000000feUL + +/* ISP_SYS_INT0_STATUS */ +#define ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT9_INT_MASK 0x00010000UL +#define ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT10_INT_MASK 0x00040000UL +#define ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT11_INT_MASK 0x00100000UL +#define ISP_SYS_INT0_STATUS__SYS_INT_RINGBUFFER_WPT12_INT_MASK 0x00400000UL + +/* ISP_SYS_INT0_EN */ +#define ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT9_EN_MASK 0x00010000UL +#define ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT10_EN_MASK 0x00040000UL +#define ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT11_EN_MASK 0x00100000UL +#define ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT12_EN_MASK 0x00400000UL + +/* ISP_SYS_INT0_ACK */ +#define ISP_SYS_INT0_ACK__SYS_INT_RINGBUFFER_WPT9_ACK_MASK 0x00010000UL +#define ISP_SYS_INT0_ACK__SYS_INT_RINGBUFFER_WPT10_ACK_MASK 0x00040000UL +#define ISP_SYS_INT0_ACK__SYS_INT_RINGBUFFER_WPT11_ACK_MASK 0x00100000UL +#define ISP_SYS_INT0_ACK__SYS_INT_RINGBUFFER_WPT12_ACK_MASK 0x00400000UL + +/* Helper functions for reading isp registers */ +static inline u32 isp4hw_rreg(void __iomem *base, u32 reg) +{ + return readl(base + reg); +} + +/* Helper functions for writing isp registers */ +static inline void isp4hw_wreg(void __iomem *base, u32 reg, u32 val) +{ + return writel(val, base + reg); +} + +#endif /* _ISP4_HW_REG_H_ */ diff --git a/drivers/media/platform/amd/isp4/isp4_interface.c b/drivers/media/platform/amd/isp4/isp4_interface.c new file mode 100644 index 000000000000..c3b19bab7a02 --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_interface.c @@ -0,0 +1,789 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#include + +#include "isp4_debug.h" +#include "isp4_fw_cmd_resp.h" +#include "isp4_hw_reg.h" +#include "isp4_interface.h" + +#define ISP4IF_FW_RESP_RB_IRQ_EN_MASK \ + (ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT9_EN_MASK | \ + ISP_SYS_INT0_EN__SYS_INT_RINGBUFFER_WPT12_EN_MASK) + +#define ISP4IF_FW_CMD_TIMEOUT (HZ / 2) + +struct isp4if_rb_config { + const char *name; + u32 index; + u32 reg_rptr; + u32 reg_wptr; + u32 reg_base_lo; + u32 reg_base_hi; + u32 reg_size; + u32 val_size; + u64 base_mc_addr; + void *base_sys_addr; +}; + +/* FW cmd ring buffer configuration */ +static struct isp4if_rb_config + isp4if_cmd_rb_config[ISP4IF_STREAM_ID_MAX] = { + { + .name = "CMD_RB_GBL0", + .index = 3, + .reg_rptr = ISP_RB_RPTR4, + .reg_wptr = ISP_RB_WPTR4, + .reg_base_lo = ISP_RB_BASE_LO4, + .reg_base_hi = ISP_RB_BASE_HI4, + .reg_size = ISP_RB_SIZE4, + }, + { + .name = "CMD_RB_STR1", + .index = 0, + .reg_rptr = ISP_RB_RPTR1, + .reg_wptr = ISP_RB_WPTR1, + .reg_base_lo = ISP_RB_BASE_LO1, + .reg_base_hi = ISP_RB_BASE_HI1, + .reg_size = ISP_RB_SIZE1, + }, + { + .name = "CMD_RB_STR2", + .index = 1, + .reg_rptr = ISP_RB_RPTR2, + .reg_wptr = ISP_RB_WPTR2, + .reg_base_lo = ISP_RB_BASE_LO2, + .reg_base_hi = ISP_RB_BASE_HI2, + .reg_size = ISP_RB_SIZE2, + }, + { + .name = "CMD_RB_STR3", + .index = 2, + .reg_rptr = ISP_RB_RPTR3, + .reg_wptr = ISP_RB_WPTR3, + .reg_base_lo = ISP_RB_BASE_LO3, + .reg_base_hi = ISP_RB_BASE_HI3, + .reg_size = ISP_RB_SIZE3, + }, +}; + +/* FW resp ring buffer configuration */ +static struct isp4if_rb_config + isp4if_resp_rb_config[ISP4IF_STREAM_ID_MAX] = { + { + .name = "RES_RB_GBL0", + .index = 3, + .reg_rptr = ISP_RB_RPTR12, + .reg_wptr = ISP_RB_WPTR12, + .reg_base_lo = ISP_RB_BASE_LO12, + .reg_base_hi = ISP_RB_BASE_HI12, + .reg_size = ISP_RB_SIZE12, + }, + { + .name = "RES_RB_STR1", + .index = 0, + .reg_rptr = ISP_RB_RPTR9, + .reg_wptr = ISP_RB_WPTR9, + .reg_base_lo = ISP_RB_BASE_LO9, + .reg_base_hi = ISP_RB_BASE_HI9, + .reg_size = ISP_RB_SIZE9, + }, + { + .name = "RES_RB_STR2", + .index = 1, + .reg_rptr = ISP_RB_RPTR10, + .reg_wptr = ISP_RB_WPTR10, + .reg_base_lo = ISP_RB_BASE_LO10, + .reg_base_hi = ISP_RB_BASE_HI10, + .reg_size = ISP_RB_SIZE10, + }, + { + .name = "RES_RB_STR3", + .index = 2, + .reg_rptr = ISP_RB_RPTR11, + .reg_wptr = ISP_RB_WPTR11, + .reg_base_lo = ISP_RB_BASE_LO11, + .reg_base_hi = ISP_RB_BASE_HI11, + .reg_size = ISP_RB_SIZE11, + }, +}; + +/* FW log ring buffer configuration */ +static struct isp4if_rb_config isp4if_log_rb_config = { + .name = "LOG_RB", + .index = 0, + .reg_rptr = ISP_LOG_RB_RPTR0, + .reg_wptr = ISP_LOG_RB_WPTR0, + .reg_base_lo = ISP_LOG_RB_BASE_LO0, + .reg_base_hi = ISP_LOG_RB_BASE_HI0, + .reg_size = ISP_LOG_RB_SIZE0, +}; + +static struct isp4if_gpu_mem_info *isp4if_gpu_mem_alloc(struct isp4_interface *ispif, u32 mem_size) +{ + struct isp4if_gpu_mem_info *mem_info; + struct device *dev = ispif->dev; + int ret; + + mem_info = kmalloc(sizeof(*mem_info), GFP_KERNEL); + if (!mem_info) + return NULL; + + mem_info->mem_size = mem_size; + ret = isp_kernel_buffer_alloc(dev, mem_info->mem_size, &mem_info->mem_handle, + &mem_info->gpu_mc_addr, &mem_info->sys_addr); + if (ret) { + kfree(mem_info); + return NULL; + } + + return mem_info; +} + +static void isp4if_gpu_mem_free(struct isp4_interface *ispif, + struct isp4if_gpu_mem_info **mem_info_ptr) +{ + struct isp4if_gpu_mem_info *mem_info = *mem_info_ptr; + struct device *dev = ispif->dev; + + if (!mem_info) { + dev_err(dev, "invalid mem_info\n"); + return; + } + + *mem_info_ptr = NULL; + isp_kernel_buffer_free(&mem_info->mem_handle, &mem_info->gpu_mc_addr, &mem_info->sys_addr); + kfree(mem_info); +} + +static void isp4if_dealloc_fw_gpumem(struct isp4_interface *ispif) +{ + int i; + + isp4if_gpu_mem_free(ispif, &ispif->fw_mem_pool); + isp4if_gpu_mem_free(ispif, &ispif->fw_cmd_resp_buf); + isp4if_gpu_mem_free(ispif, &ispif->fw_log_buf); + + for (i = 0; i < ISP4IF_MAX_STREAM_BUF_COUNT; i++) + isp4if_gpu_mem_free(ispif, &ispif->meta_info_buf[i]); +} + +static int isp4if_alloc_fw_gpumem(struct isp4_interface *ispif) +{ + struct device *dev = ispif->dev; + int i; + + ispif->fw_mem_pool = isp4if_gpu_mem_alloc(ispif, FW_MEMORY_POOL_SIZE); + if (!ispif->fw_mem_pool) + goto error_no_memory; + + ispif->fw_cmd_resp_buf = + isp4if_gpu_mem_alloc(ispif, ISP4IF_RB_PMBMAP_MEM_SIZE); + if (!ispif->fw_cmd_resp_buf) + goto error_no_memory; + + ispif->fw_log_buf = + isp4if_gpu_mem_alloc(ispif, ISP4IF_FW_LOG_RINGBUF_SIZE); + if (!ispif->fw_log_buf) + goto error_no_memory; + + for (i = 0; i < ISP4IF_MAX_STREAM_BUF_COUNT; i++) { + ispif->meta_info_buf[i] = + isp4if_gpu_mem_alloc(ispif, ISP4IF_META_INFO_BUF_SIZE); + if (!ispif->meta_info_buf[i]) + goto error_no_memory; + } + + return 0; + +error_no_memory: + dev_err(dev, "failed to allocate gpu memory\n"); + return -ENOMEM; +} + +static u32 isp4if_compute_check_sum(const void *buf, size_t buf_size) +{ + const u8 *surplus_ptr; + const u32 *buffer; + u32 checksum = 0; + size_t i; + + buffer = (const u32 *)buf; + for (i = 0; i < buf_size / sizeof(u32); i++) + checksum += buffer[i]; + + surplus_ptr = (const u8 *)&buffer[i]; + /* add surplus data crc checksum */ + for (i = 0; i < buf_size % sizeof(u32); i++) + checksum += surplus_ptr[i]; + + return checksum; +} + +void isp4if_clear_cmdq(struct isp4_interface *ispif) +{ + struct isp4if_cmd_element *buf_node, *tmp_node; + LIST_HEAD(free_list); + + scoped_guard(spinlock, &ispif->cmdq_lock) + list_splice_init(&ispif->cmdq, &free_list); + + list_for_each_entry_safe(buf_node, tmp_node, &free_list, list) + kfree(buf_node); +} + +static bool isp4if_is_cmdq_rb_full(struct isp4_interface *ispif, enum isp4if_stream_id stream) +{ + struct isp4if_rb_config *rb_config = &isp4if_cmd_rb_config[stream]; + u32 rreg = rb_config->reg_rptr, wreg = rb_config->reg_wptr; + u32 len = rb_config->val_size; + u32 rd_ptr, wr_ptr; + u32 bytes_free; + + rd_ptr = isp4hw_rreg(ispif->mmio, rreg); + wr_ptr = isp4hw_rreg(ispif->mmio, wreg); + + /* Read and write pointers are equal, indicating the ringbuf is empty */ + if (wr_ptr == rd_ptr) + return false; + + if (wr_ptr > rd_ptr) + bytes_free = len - (wr_ptr - rd_ptr); + else + bytes_free = rd_ptr - wr_ptr; + + /* + * Ignore one byte from the bytes free to prevent rd_ptr from equaling + * wr_ptr when the ringbuf is full, because rd_ptr == wr_ptr is supposed + * to indicate that the ringbuf is empty. + */ + return bytes_free <= sizeof(struct isp4fw_cmd); +} + +struct isp4if_cmd_element *isp4if_rm_cmd_from_cmdq(struct isp4_interface *ispif, u32 seq_num, + u32 cmd_id) +{ + struct isp4if_cmd_element *buf_node; + + guard(spinlock)(&ispif->cmdq_lock); + + list_for_each_entry(buf_node, &ispif->cmdq, list) { + if (buf_node->seq_num == seq_num && buf_node->cmd_id == cmd_id) { + list_del(&buf_node->list); + return buf_node; + } + } + + return NULL; +} + +/* Must check that isp4if_is_cmdq_rb_full() == false before calling */ +static int isp4if_insert_isp_fw_cmd(struct isp4_interface *ispif, enum isp4if_stream_id stream, + const struct isp4fw_cmd *cmd) +{ + struct isp4if_rb_config *rb_config = &isp4if_cmd_rb_config[stream]; + u32 rreg = rb_config->reg_rptr, wreg = rb_config->reg_wptr; + void *mem_sys = rb_config->base_sys_addr; + const u32 cmd_sz = sizeof(*cmd); + struct device *dev = ispif->dev; + u32 len = rb_config->val_size; + const void *src = cmd; + u32 rd_ptr, wr_ptr; + u32 bytes_to_end; + + rd_ptr = isp4hw_rreg(ispif->mmio, rreg); + wr_ptr = isp4hw_rreg(ispif->mmio, wreg); + if (rd_ptr >= len || wr_ptr >= len) { + dev_err(dev, "rb invalid: stream=%u(%s), rd=%u, wr=%u, len=%u, cmd_sz=%u\n", + stream, isp4dbg_get_if_stream_str(stream), rd_ptr, wr_ptr, len, cmd_sz); + return -EINVAL; + } + + bytes_to_end = len - wr_ptr; + if (bytes_to_end >= cmd_sz) { + /* FW cmd is just a straight copy to the write pointer */ + memcpy(mem_sys + wr_ptr, src, cmd_sz); + isp4hw_wreg(ispif->mmio, wreg, (wr_ptr + cmd_sz) % len); + } else { + /* FW cmd is split because the ringbuf needs to wrap around */ + memcpy(mem_sys + wr_ptr, src, bytes_to_end); + memcpy(mem_sys, src + bytes_to_end, cmd_sz - bytes_to_end); + isp4hw_wreg(ispif->mmio, wreg, cmd_sz - bytes_to_end); + } + + return 0; +} + +static inline enum isp4if_stream_id isp4if_get_fw_stream(u32 cmd_id) +{ + return ISP4IF_STREAM_ID_1; +} + +static int isp4if_send_fw_cmd(struct isp4_interface *ispif, u32 cmd_id, const void *package, + u32 package_size, bool sync) +{ + enum isp4if_stream_id stream = isp4if_get_fw_stream(cmd_id); + struct isp4if_cmd_element *ele = NULL; + struct device *dev = ispif->dev; + struct isp4fw_cmd cmd; + u32 seq_num; + int ret; + + if (package_size > sizeof(cmd.cmd_param)) { + dev_err(dev, "fail pkgsize(%u)>%zu cmd:0x%x,stream %d\n", + package_size, sizeof(cmd.cmd_param), cmd_id, stream); + return -EINVAL; + } + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&cmd, 0, sizeof(cmd)); + cmd.cmd_id = cmd_id; + switch (stream) { + case ISP4IF_STREAM_ID_GLOBAL: + cmd.cmd_stream_id = STREAM_ID_INVALID; + break; + case ISP4IF_STREAM_ID_1: + cmd.cmd_stream_id = STREAM_ID_1; + break; + default: + dev_err(dev, "fail bad stream id %d\n", stream); + return -EINVAL; + } + + /* Allocate the sync command object early and outside of the lock */ + if (sync) { + ele = kmalloc(sizeof(*ele), GFP_KERNEL); + if (!ele) + return -ENOMEM; + + /* Get two references: one for the resp thread, one for us */ + atomic_set(&ele->refcnt, 2); + init_completion(&ele->cmd_done); + } + + if (package && package_size) + memcpy(cmd.cmd_param, package, package_size); + + scoped_guard(mutex, &ispif->isp4if_mutex) { + ret = read_poll_timeout(isp4if_is_cmdq_rb_full, ret, !ret, ISP4IF_RB_FULL_SLEEP_US, + ISP4IF_RB_FULL_TIMEOUT_US, false, ispif, stream); + if (ret) { + struct isp4if_rb_config *rb_config = &isp4if_resp_rb_config[stream]; + u32 rd_ptr = isp4hw_rreg(ispif->mmio, rb_config->reg_rptr); + u32 wr_ptr = isp4hw_rreg(ispif->mmio, rb_config->reg_wptr); + + dev_err(dev, + "failed to get free cmdq slot, stream %s(%d),rd %u, wr %u\n", + isp4dbg_get_if_stream_str(stream), + stream, rd_ptr, wr_ptr); + ret = -ETIMEDOUT; + goto free_ele; + } + + seq_num = ispif->host2fw_seq_num++; + cmd.cmd_seq_num = seq_num; + cmd.cmd_check_sum = isp4if_compute_check_sum(&cmd, sizeof(cmd) - sizeof(u32)); + + /* + * only append the fw cmd to queue when its response needs to be waited for, + * currently there are only two such commands, disable channel and stop stream + * which are only sent after close camera + */ + if (ele) { + ele->seq_num = seq_num; + ele->cmd_id = cmd_id; + scoped_guard(spinlock, &ispif->cmdq_lock) + list_add_tail(&ele->list, &ispif->cmdq); + } + + ret = isp4if_insert_isp_fw_cmd(ispif, stream, &cmd); + if (ret) { + dev_err(dev, "fail for insert_isp_fw_cmd cmd_id %s(0x%08x)\n", + isp4dbg_get_cmd_str(cmd_id), cmd_id); + goto err_dequeue_ele; + } + } + + if (ele) { + ret = wait_for_completion_timeout(&ele->cmd_done, ISP4IF_FW_CMD_TIMEOUT); + if (!ret) { + ret = -ETIMEDOUT; + goto err_dequeue_ele; + } + + ret = 0; + goto put_ele_ref; + } + + return 0; + +err_dequeue_ele: + /* + * Try to remove the command from the queue. If that fails, then it + * means the response thread is currently using the object, and we need + * to use the refcount to avoid a use-after-free by either side. + */ + if (ele && isp4if_rm_cmd_from_cmdq(ispif, seq_num, cmd_id)) + goto free_ele; + +put_ele_ref: + /* Don't free the command if we didn't put the last reference */ + if (ele && atomic_dec_return(&ele->refcnt)) + ele = NULL; + +free_ele: + kfree(ele); + return ret; +} + +static int isp4if_send_buffer(struct isp4_interface *ispif, struct isp4if_img_buf_info *buf_info) +{ + struct isp4fw_cmd_send_buffer cmd; + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&cmd, 0, sizeof(cmd)); + cmd.buffer_type = BUFFER_TYPE_PREVIEW; + cmd.buffer.vmid_space.bit.space = ADDR_SPACE_TYPE_GPU_VA; + isp4if_split_addr64(buf_info->planes[0].mc_addr, + &cmd.buffer.buf_base_a_lo, + &cmd.buffer.buf_base_a_hi); + cmd.buffer.buf_size_a = buf_info->planes[0].len; + + isp4if_split_addr64(buf_info->planes[1].mc_addr, + &cmd.buffer.buf_base_b_lo, + &cmd.buffer.buf_base_b_hi); + cmd.buffer.buf_size_b = buf_info->planes[1].len; + + isp4if_split_addr64(buf_info->planes[2].mc_addr, + &cmd.buffer.buf_base_c_lo, + &cmd.buffer.buf_base_c_hi); + cmd.buffer.buf_size_c = buf_info->planes[2].len; + + return isp4if_send_fw_cmd(ispif, CMD_ID_SEND_BUFFER, &cmd, sizeof(cmd), false); +} + +static void isp4if_init_rb_config(struct isp4_interface *ispif, struct isp4if_rb_config *rb_config) +{ + isp4hw_wreg(ispif->mmio, rb_config->reg_rptr, 0x0); + isp4hw_wreg(ispif->mmio, rb_config->reg_wptr, 0x0); + isp4hw_wreg(ispif->mmio, rb_config->reg_base_lo, rb_config->base_mc_addr); + isp4hw_wreg(ispif->mmio, rb_config->reg_base_hi, rb_config->base_mc_addr >> 32); + isp4hw_wreg(ispif->mmio, rb_config->reg_size, rb_config->val_size); +} + +static int isp4if_fw_init(struct isp4_interface *ispif) +{ + u32 aligned_rb_chunk_size = ISP4IF_RB_PMBMAP_MEM_CHUNK & 0xffffffc0; + struct isp4if_rb_config *rb_config; + u32 offset; + int i; + + /* initialize CMD_RB streams */ + for (i = 0; i < ISP4IF_STREAM_ID_MAX; i++) { + rb_config = (isp4if_cmd_rb_config + i); + offset = aligned_rb_chunk_size * rb_config->index; + + rb_config->val_size = ISP4IF_FW_CMD_BUF_SIZE; + rb_config->base_sys_addr = + ispif->fw_cmd_resp_buf->sys_addr + offset; + rb_config->base_mc_addr = + ispif->fw_cmd_resp_buf->gpu_mc_addr + offset; + + isp4if_init_rb_config(ispif, rb_config); + } + + /* initialize RESP_RB streams */ + for (i = 0; i < ISP4IF_STREAM_ID_MAX; i++) { + rb_config = (isp4if_resp_rb_config + i); + offset = aligned_rb_chunk_size * + (rb_config->index + ISP4IF_RESP_CHAN_TO_RB_OFFSET - 1); + + rb_config->val_size = ISP4IF_FW_CMD_BUF_SIZE; + rb_config->base_sys_addr = + ispif->fw_cmd_resp_buf->sys_addr + offset; + rb_config->base_mc_addr = + ispif->fw_cmd_resp_buf->gpu_mc_addr + offset; + + isp4if_init_rb_config(ispif, rb_config); + } + + /* initialize LOG_RB stream */ + rb_config = &isp4if_log_rb_config; + rb_config->val_size = ISP4IF_FW_LOG_RINGBUF_SIZE; + rb_config->base_mc_addr = ispif->fw_log_buf->gpu_mc_addr; + rb_config->base_sys_addr = ispif->fw_log_buf->sys_addr; + + isp4if_init_rb_config(ispif, rb_config); + + return 0; +} + +static int isp4if_wait_fw_ready(struct isp4_interface *ispif, u32 isp_status_addr) +{ + struct device *dev = ispif->dev; + u32 timeout_ms = 100; + u32 interval_ms = 1; + u32 reg_val; + + /* wait for FW initialize done! */ + if (!read_poll_timeout(isp4hw_rreg, reg_val, reg_val & ISP_STATUS__CCPU_REPORT_MASK, + interval_ms * 1000, timeout_ms * 1000, false, + ispif->mmio, isp_status_addr)) + return 0; + + dev_err(dev, "ISP CCPU FW boot failed\n"); + + return -ETIME; +} + +static void isp4if_enable_ccpu(struct isp4_interface *ispif) +{ + u32 reg_val; + + reg_val = isp4hw_rreg(ispif->mmio, ISP_SOFT_RESET); + reg_val &= (~ISP_SOFT_RESET__CCPU_SOFT_RESET_MASK); + isp4hw_wreg(ispif->mmio, ISP_SOFT_RESET, reg_val); + + usleep_range(100, 150); + + reg_val = isp4hw_rreg(ispif->mmio, ISP_CCPU_CNTL); + reg_val &= (~ISP_CCPU_CNTL__CCPU_HOST_SOFT_RST_MASK); + isp4hw_wreg(ispif->mmio, ISP_CCPU_CNTL, reg_val); +} + +static void isp4if_disable_ccpu(struct isp4_interface *ispif) +{ + u32 reg_val; + + reg_val = isp4hw_rreg(ispif->mmio, ISP_CCPU_CNTL); + reg_val |= ISP_CCPU_CNTL__CCPU_HOST_SOFT_RST_MASK; + isp4hw_wreg(ispif->mmio, ISP_CCPU_CNTL, reg_val); + + usleep_range(100, 150); + + reg_val = isp4hw_rreg(ispif->mmio, ISP_SOFT_RESET); + reg_val |= ISP_SOFT_RESET__CCPU_SOFT_RESET_MASK; + isp4hw_wreg(ispif->mmio, ISP_SOFT_RESET, reg_val); +} + +static int isp4if_fw_boot(struct isp4_interface *ispif) +{ + struct device *dev = ispif->dev; + + if (ispif->status != ISP4IF_STATUS_PWR_ON) { + dev_err(dev, "invalid isp power status %d\n", ispif->status); + return -EINVAL; + } + + isp4if_disable_ccpu(ispif); + + isp4if_fw_init(ispif); + + /* clear ccpu status */ + isp4hw_wreg(ispif->mmio, ISP_STATUS, 0x0); + + isp4if_enable_ccpu(ispif); + + if (isp4if_wait_fw_ready(ispif, ISP_STATUS)) { + isp4if_disable_ccpu(ispif); + return -EINVAL; + } + + /* enable interrupts */ + isp4hw_wreg(ispif->mmio, ISP_SYS_INT0_EN, ISP4IF_FW_RESP_RB_IRQ_EN_MASK); + + ispif->status = ISP4IF_STATUS_FW_RUNNING; + + dev_dbg(dev, "ISP CCPU FW boot success\n"); + + return 0; +} + +int isp4if_f2h_resp(struct isp4_interface *ispif, enum isp4if_stream_id stream, + struct isp4fw_resp *resp) +{ + struct isp4if_rb_config *rb_config = &isp4if_resp_rb_config[stream]; + u32 rreg = rb_config->reg_rptr, wreg = rb_config->reg_wptr; + void *mem_sys = rb_config->base_sys_addr; + const u32 resp_sz = sizeof(*resp); + struct device *dev = ispif->dev; + u32 len = rb_config->val_size; + u32 rd_ptr, wr_ptr; + u32 bytes_to_end; + void *dst = resp; + u32 checksum; + + rd_ptr = isp4hw_rreg(ispif->mmio, rreg); + wr_ptr = isp4hw_rreg(ispif->mmio, wreg); + if (rd_ptr >= len || wr_ptr >= len) + goto err_rb_invalid; + + /* Read and write pointers are equal, indicating the ringbuf is empty */ + if (rd_ptr == wr_ptr) + return -ENODATA; + + bytes_to_end = len - rd_ptr; + if (bytes_to_end >= resp_sz) { + /* FW response is just a straight copy from the read pointer */ + if (wr_ptr > rd_ptr && wr_ptr - rd_ptr < resp_sz) + goto err_rb_invalid; + + memcpy(dst, mem_sys + rd_ptr, resp_sz); + isp4hw_wreg(ispif->mmio, rreg, (rd_ptr + resp_sz) % len); + } else { + /* FW response is split because the ringbuf wrapped around */ + if (wr_ptr > rd_ptr || wr_ptr < resp_sz - bytes_to_end) + goto err_rb_invalid; + + memcpy(dst, mem_sys + rd_ptr, bytes_to_end); + memcpy(dst + bytes_to_end, mem_sys, resp_sz - bytes_to_end); + isp4hw_wreg(ispif->mmio, rreg, resp_sz - bytes_to_end); + } + + checksum = isp4if_compute_check_sum(resp, resp_sz - sizeof(u32)); + if (checksum != resp->resp_check_sum) { + dev_err(dev, "resp checksum 0x%x,should 0x%x,rptr %u,wptr %u\n", + checksum, resp->resp_check_sum, rd_ptr, wr_ptr); + dev_err(dev, "%s(%u), seqNo %u, resp_id %s(0x%x)\n", + isp4dbg_get_if_stream_str(stream), stream, resp->resp_seq_num, + isp4dbg_get_resp_str(resp->resp_id), resp->resp_id); + return -EINVAL; + } + + return 0; + +err_rb_invalid: + dev_err(dev, "rb invalid: stream=%u(%s), rd=%u, wr=%u, len=%u, resp_sz=%u\n", + stream, isp4dbg_get_if_stream_str(stream), rd_ptr, wr_ptr, len, resp_sz); + return -EINVAL; +} + +int isp4if_send_command(struct isp4_interface *ispif, u32 cmd_id, const void *package, + u32 package_size) +{ + return isp4if_send_fw_cmd(ispif, cmd_id, package, package_size, false); +} + +int isp4if_send_command_sync(struct isp4_interface *ispif, u32 cmd_id, const void *package, + u32 package_size) +{ + return isp4if_send_fw_cmd(ispif, cmd_id, package, package_size, true); +} + +void isp4if_clear_bufq(struct isp4_interface *ispif) +{ + struct isp4if_img_buf_node *buf_node, *tmp_node; + LIST_HEAD(free_list); + + scoped_guard(spinlock, &ispif->bufq_lock) + list_splice_init(&ispif->bufq, &free_list); + + list_for_each_entry_safe(buf_node, tmp_node, &free_list, node) + kfree(buf_node); +} + +void isp4if_dealloc_buffer_node(struct isp4if_img_buf_node *buf_node) +{ + kfree(buf_node); +} + +struct isp4if_img_buf_node *isp4if_alloc_buffer_node(struct isp4if_img_buf_info *buf_info) +{ + struct isp4if_img_buf_node *node; + + node = kmalloc(sizeof(*node), GFP_KERNEL); + if (node) + node->buf_info = *buf_info; + + return node; +} + +struct isp4if_img_buf_node *isp4if_dequeue_buffer(struct isp4_interface *ispif) +{ + struct isp4if_img_buf_node *buf_node; + + guard(spinlock)(&ispif->bufq_lock); + + buf_node = list_first_entry_or_null(&ispif->bufq, typeof(*buf_node), node); + if (buf_node) + list_del(&buf_node->node); + + return buf_node; +} + +int isp4if_queue_buffer(struct isp4_interface *ispif, struct isp4if_img_buf_node *buf_node) +{ + int ret; + + ret = isp4if_send_buffer(ispif, &buf_node->buf_info); + if (ret) + return ret; + + scoped_guard(spinlock, &ispif->bufq_lock) + list_add_tail(&buf_node->node, &ispif->bufq); + + return 0; +} + +int isp4if_stop(struct isp4_interface *ispif) +{ + isp4if_disable_ccpu(ispif); + + isp4if_dealloc_fw_gpumem(ispif); + + return 0; +} + +int isp4if_start(struct isp4_interface *ispif) +{ + int ret; + + ret = isp4if_alloc_fw_gpumem(ispif); + if (ret) + return ret; + + ret = isp4if_fw_boot(ispif); + if (ret) + goto failed_fw_boot; + + return 0; + +failed_fw_boot: + isp4if_dealloc_fw_gpumem(ispif); + return ret; +} + +int isp4if_deinit(struct isp4_interface *ispif) +{ + isp4if_clear_cmdq(ispif); + + isp4if_clear_bufq(ispif); + + mutex_destroy(&ispif->isp4if_mutex); + + return 0; +} + +int isp4if_init(struct isp4_interface *ispif, struct device *dev, void __iomem *isp_mmio) +{ + ispif->dev = dev; + ispif->mmio = isp_mmio; + + spin_lock_init(&ispif->cmdq_lock); /* used for cmdq access */ + spin_lock_init(&ispif->bufq_lock); /* used for bufq access */ + mutex_init(&ispif->isp4if_mutex); /* used for commands sent to ispfw */ + + INIT_LIST_HEAD(&ispif->cmdq); + INIT_LIST_HEAD(&ispif->bufq); + + return 0; +} diff --git a/drivers/media/platform/amd/isp4/isp4_interface.h b/drivers/media/platform/amd/isp4/isp4_interface.h new file mode 100644 index 000000000000..01d5268f7d4c --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_interface.h @@ -0,0 +1,141 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#ifndef _ISP4_INTERFACE_H_ +#define _ISP4_INTERFACE_H_ + +#include +#include +#include +#include + +struct isp4fw_resp; + +#define ISP4IF_RB_MAX 25 +#define ISP4IF_RESP_CHAN_TO_RB_OFFSET 9 +#define ISP4IF_RB_PMBMAP_MEM_SIZE (SZ_16M - 1) +#define ISP4IF_RB_PMBMAP_MEM_CHUNK \ + (ISP4IF_RB_PMBMAP_MEM_SIZE / (ISP4IF_RB_MAX - 1)) +#define ISP4IF_HOST2FW_COMMAND_SIZE sizeof(struct isp4fw_cmd) +#define ISP4IF_MAX_NUM_HOST2FW_COMMAND 40 +#define ISP4IF_FW_CMD_BUF_SIZE \ + (ISP4IF_MAX_NUM_HOST2FW_COMMAND * ISP4IF_HOST2FW_COMMAND_SIZE) +#define ISP4IF_RB_FULL_SLEEP_US (33 * USEC_PER_MSEC) +#define ISP4IF_RB_FULL_TIMEOUT_US (10 * ISP4IF_RB_FULL_SLEEP_US) + +#define ISP4IF_META_INFO_BUF_SIZE ALIGN(sizeof(struct isp4fw_meta_info), 0x8000) +#define ISP4IF_MAX_STREAM_BUF_COUNT 8 + +#define ISP4IF_FW_LOG_RINGBUF_SIZE SZ_2M + +enum isp4if_stream_id { + ISP4IF_STREAM_ID_GLOBAL = 0, + ISP4IF_STREAM_ID_1 = 1, + ISP4IF_STREAM_ID_MAX = 4 +}; + +enum isp4if_status { + ISP4IF_STATUS_PWR_OFF, + ISP4IF_STATUS_PWR_ON, + ISP4IF_STATUS_FW_RUNNING, + ISP4IF_FSM_STATUS_MAX +}; + +struct isp4if_gpu_mem_info { + u64 mem_size; + u64 gpu_mc_addr; + void *sys_addr; + void *mem_handle; +}; + +struct isp4if_img_buf_info { + struct { + void *sys_addr; + u64 mc_addr; + u32 len; + } planes[3]; +}; + +struct isp4if_img_buf_node { + struct list_head node; + struct isp4if_img_buf_info buf_info; +}; + +struct isp4if_cmd_element { + struct list_head list; + u32 seq_num; + u32 cmd_id; + struct completion cmd_done; + atomic_t refcnt; +}; + +struct isp4_interface { + struct device *dev; + void __iomem *mmio; + + spinlock_t cmdq_lock; /* used for cmdq access */ + spinlock_t bufq_lock; /* used for bufq access */ + struct mutex isp4if_mutex; /* used to send fw cmd and read fw log */ + + struct list_head cmdq; /* commands sent to fw */ + struct list_head bufq; /* buffers sent to fw */ + + enum isp4if_status status; + u32 host2fw_seq_num; + + /* ISP fw buffers */ + struct isp4if_gpu_mem_info *fw_log_buf; + struct isp4if_gpu_mem_info *fw_cmd_resp_buf; + struct isp4if_gpu_mem_info *fw_mem_pool; + struct isp4if_gpu_mem_info *meta_info_buf[ISP4IF_MAX_STREAM_BUF_COUNT]; +}; + +static inline void isp4if_split_addr64(u64 addr, u32 *lo, u32 *hi) +{ + if (lo) + *lo = addr & 0xffffffff; + + if (hi) + *hi = addr >> 32; +} + +static inline u64 isp4if_join_addr64(u32 lo, u32 hi) +{ + return (((u64)hi) << 32) | (u64)lo; +} + +int isp4if_f2h_resp(struct isp4_interface *ispif, enum isp4if_stream_id stream, + struct isp4fw_resp *resp); + +int isp4if_send_command(struct isp4_interface *ispif, u32 cmd_id, const void *package, + u32 package_size); + +int isp4if_send_command_sync(struct isp4_interface *ispif, u32 cmd_id, const void *package, + u32 package_size); + +struct isp4if_cmd_element *isp4if_rm_cmd_from_cmdq(struct isp4_interface *ispif, u32 seq_num, + u32 cmd_id); + +void isp4if_clear_cmdq(struct isp4_interface *ispif); + +void isp4if_clear_bufq(struct isp4_interface *ispif); + +void isp4if_dealloc_buffer_node(struct isp4if_img_buf_node *buf_node); + +struct isp4if_img_buf_node *isp4if_alloc_buffer_node(struct isp4if_img_buf_info *buf_info); + +struct isp4if_img_buf_node *isp4if_dequeue_buffer(struct isp4_interface *ispif); + +int isp4if_queue_buffer(struct isp4_interface *ispif, struct isp4if_img_buf_node *buf_node); + +int isp4if_stop(struct isp4_interface *ispif); + +int isp4if_start(struct isp4_interface *ispif); + +int isp4if_deinit(struct isp4_interface *ispif); + +int isp4if_init(struct isp4_interface *ispif, struct device *dev, void __iomem *isp_mmio); + +#endif /* _ISP4_INTERFACE_H_ */ diff --git a/drivers/media/platform/amd/isp4/isp4_subdev.c b/drivers/media/platform/amd/isp4/isp4_subdev.c new file mode 100644 index 000000000000..21de6bc7fce0 --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_subdev.c @@ -0,0 +1,1057 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#include +#include + +#include "isp4_debug.h" +#include "isp4_fw_cmd_resp.h" +#include "isp4_interface.h" +#include "isp4.h" + +#define ISP4SD_MIN_BUF_CNT_BEF_START_STREAM 4 + +#define ISP4SD_PERFORMANCE_STATE_LOW 0 +#define ISP4SD_PERFORMANCE_STATE_HIGH 1 + +/* align 32KB */ +#define ISP4SD_META_BUF_SIZE ALIGN(sizeof(struct isp4fw_meta_info), 0x8000) + +#define to_isp4_subdev(v4l2_sdev) \ + container_of(v4l2_sdev, struct isp4_subdev, sdev) + +static const char *isp4sd_entity_name = "amd isp4"; + +static const char *isp4sd_thread_name[ISP4SD_MAX_FW_RESP_STREAM_NUM] = { + "amd_isp4_thread_global", + "amd_isp4_thread_stream1", +}; + +static void isp4sd_module_enable(struct isp4_subdev *isp_subdev, bool enable) +{ + if (isp_subdev->enable_gpio) { + gpiod_set_value(isp_subdev->enable_gpio, enable ? 1 : 0); + dev_dbg(isp_subdev->dev, "%s isp_subdev module\n", + enable ? "enable" : "disable"); + } +} + +static int isp4sd_setup_fw_mem_pool(struct isp4_subdev *isp_subdev) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct isp4fw_cmd_send_buffer buf_type; + struct device *dev = isp_subdev->dev; + int ret; + + if (!ispif->fw_mem_pool) { + dev_err(dev, "fail to alloc mem pool\n"); + return -ENOMEM; + } + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&buf_type, 0, sizeof(buf_type)); + buf_type.buffer_type = BUFFER_TYPE_MEM_POOL; + buf_type.buffer.vmid_space.bit.space = ADDR_SPACE_TYPE_GPU_VA; + isp4if_split_addr64(ispif->fw_mem_pool->gpu_mc_addr, + &buf_type.buffer.buf_base_a_lo, + &buf_type.buffer.buf_base_a_hi); + buf_type.buffer.buf_size_a = ispif->fw_mem_pool->mem_size; + + ret = isp4if_send_command(ispif, CMD_ID_SEND_BUFFER, + &buf_type, sizeof(buf_type)); + if (ret) { + dev_err(dev, "send fw mem pool 0x%llx(%u) fail %d\n", + ispif->fw_mem_pool->gpu_mc_addr, buf_type.buffer.buf_size_a, ret); + return ret; + } + + dev_dbg(dev, "send fw mem pool 0x%llx(%u) suc\n", + ispif->fw_mem_pool->gpu_mc_addr, buf_type.buffer.buf_size_a); + + return 0; +} + +static int isp4sd_set_stream_path(struct isp4_subdev *isp_subdev) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct isp4fw_cmd_set_stream_cfg cmd; + struct device *dev = isp_subdev->dev; + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&cmd, 0, sizeof(cmd)); + cmd.stream_cfg.mipi_pipe_path_cfg.isp4fw_sensor_id = SENSOR_ID_ON_MIPI0; + cmd.stream_cfg.mipi_pipe_path_cfg.b_enable = true; + cmd.stream_cfg.isp_pipe_path_cfg.isp_pipe_id = MIPI0_ISP_PIPELINE_ID; + + cmd.stream_cfg.b_enable_tnr = true; + dev_dbg(dev, "isp4fw_sensor_id %d, pipeId 0x%x EnableTnr %u\n", + cmd.stream_cfg.mipi_pipe_path_cfg.isp4fw_sensor_id, + cmd.stream_cfg.isp_pipe_path_cfg.isp_pipe_id, + cmd.stream_cfg.b_enable_tnr); + + return isp4if_send_command(ispif, CMD_ID_SET_STREAM_CONFIG, + &cmd, sizeof(cmd)); +} + +static int isp4sd_send_meta_buf(struct isp4_subdev *isp_subdev) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct isp4fw_cmd_send_buffer buf_type; + struct device *dev = isp_subdev->dev; + int i; + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&buf_type, 0, sizeof(buf_type)); + for (i = 0; i < ISP4IF_MAX_STREAM_BUF_COUNT; i++) { + struct isp4if_gpu_mem_info *meta_info_buf = + isp_subdev->ispif.meta_info_buf[i]; + int ret; + + if (!meta_info_buf) { + dev_err(dev, "fail for no meta info buf(%u)\n", i); + return -ENOMEM; + } + + buf_type.buffer_type = BUFFER_TYPE_META_INFO; + buf_type.buffer.vmid_space.bit.space = ADDR_SPACE_TYPE_GPU_VA; + isp4if_split_addr64(meta_info_buf->gpu_mc_addr, + &buf_type.buffer.buf_base_a_lo, + &buf_type.buffer.buf_base_a_hi); + buf_type.buffer.buf_size_a = meta_info_buf->mem_size; + ret = isp4if_send_command(ispif, CMD_ID_SEND_BUFFER, + &buf_type, sizeof(buf_type)); + if (ret) { + dev_err(dev, "send meta info(%u) fail\n", i); + return ret; + } + } + + dev_dbg(dev, "send meta info suc\n"); + return 0; +} + +static bool isp4sd_get_str_out_prop(struct isp4_subdev *isp_subdev, + struct isp4fw_image_prop *out_prop, + struct v4l2_subdev_state *state, u32 pad) +{ + struct device *dev = isp_subdev->dev; + struct v4l2_mbus_framefmt *format; + + format = v4l2_subdev_state_get_format(state, pad, 0); + if (!format) { + dev_err(dev, "fail get subdev state format\n"); + return false; + } + + switch (format->code) { + case MEDIA_BUS_FMT_YUYV8_1_5X8: + out_prop->image_format = IMAGE_FORMAT_NV12; + out_prop->width = format->width; + out_prop->height = format->height; + out_prop->luma_pitch = format->width; + out_prop->chroma_pitch = out_prop->width; + break; + case MEDIA_BUS_FMT_YUYV8_1X16: + out_prop->image_format = IMAGE_FORMAT_YUV422INTERLEAVED; + out_prop->width = format->width; + out_prop->height = format->height; + out_prop->luma_pitch = format->width * 2; + out_prop->chroma_pitch = 0; + break; + default: + dev_err(dev, "fail for bad image format:0x%x\n", + format->code); + return false; + } + + if (!out_prop->width || !out_prop->height) + return false; + + return true; +} + +static int isp4sd_kickoff_stream(struct isp4_subdev *isp_subdev, u32 w, u32 h) +{ + struct isp4sd_sensor_info *sensor_info = &isp_subdev->sensor_info; + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = isp_subdev->dev; + + if (sensor_info->status == ISP4SD_START_STATUS_STARTED) + return 0; + + if (sensor_info->status == ISP4SD_START_STATUS_START_FAIL) { + dev_err(dev, "fail for previous start fail\n"); + return -EINVAL; + } + + dev_dbg(dev, "w:%u,h:%u\n", w, h); + + if (isp4sd_send_meta_buf(isp_subdev)) { + dev_err(dev, "fail to send meta buf\n"); + sensor_info->status = ISP4SD_START_STATUS_START_FAIL; + return -EINVAL; + } + + sensor_info->status = ISP4SD_START_STATUS_OFF; + + if (!sensor_info->start_stream_cmd_sent && + sensor_info->buf_sent_cnt >= ISP4SD_MIN_BUF_CNT_BEF_START_STREAM) { + int ret = isp4if_send_command(ispif, CMD_ID_START_STREAM, + NULL, 0); + if (ret) { + dev_err(dev, "fail to start stream\n"); + return ret; + } + + sensor_info->start_stream_cmd_sent = true; + } else { + dev_dbg(dev, + "no send START_STREAM, start_sent %u, buf_sent %u\n", + sensor_info->start_stream_cmd_sent, + sensor_info->buf_sent_cnt); + } + + return 0; +} + +static int isp4sd_setup_output(struct isp4_subdev *isp_subdev, + struct v4l2_subdev_state *state, u32 pad) +{ + struct isp4sd_output_info *output_info = &isp_subdev->sensor_info.output_info; + struct isp4sd_sensor_info *sensor_info = &isp_subdev->sensor_info; + struct isp4_interface *ispif = &isp_subdev->ispif; + struct isp4fw_cmd_set_out_ch_prop cmd_ch_prop; + struct isp4fw_cmd_enable_out_ch cmd_ch_en; + struct device *dev = isp_subdev->dev; + int ret; + + if (output_info->start_status == ISP4SD_START_STATUS_STARTED) + return 0; + + if (output_info->start_status == ISP4SD_START_STATUS_START_FAIL) { + dev_err(dev, "fail for previous start fail\n"); + return -EINVAL; + } + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&cmd_ch_prop, 0, sizeof(cmd_ch_prop)); + cmd_ch_prop.ch = ISP_PIPE_OUT_CH_PREVIEW; + + if (!isp4sd_get_str_out_prop(isp_subdev, &cmd_ch_prop.image_prop, state, pad)) { + dev_err(dev, "fail to get out prop\n"); + return -EINVAL; + } + + dev_dbg(dev, "channel:%s,fmt %s,w:h=%u:%u,lp:%u,cp%u\n", + isp4dbg_get_out_ch_str(cmd_ch_prop.ch), + isp4dbg_get_img_fmt_str(cmd_ch_prop.image_prop.image_format), + cmd_ch_prop.image_prop.width, cmd_ch_prop.image_prop.height, + cmd_ch_prop.image_prop.luma_pitch, + cmd_ch_prop.image_prop.chroma_pitch); + + ret = isp4if_send_command(ispif, CMD_ID_SET_OUT_CHAN_PROP, + &cmd_ch_prop, sizeof(cmd_ch_prop)); + if (ret) { + output_info->start_status = ISP4SD_START_STATUS_START_FAIL; + dev_err(dev, "fail to set out prop\n"); + return ret; + } + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&cmd_ch_en, 0, sizeof(cmd_ch_en)); + cmd_ch_en.ch = ISP_PIPE_OUT_CH_PREVIEW; + cmd_ch_en.is_enable = true; + ret = isp4if_send_command(ispif, CMD_ID_ENABLE_OUT_CHAN, + &cmd_ch_en, sizeof(cmd_ch_en)); + if (ret) { + output_info->start_status = ISP4SD_START_STATUS_START_FAIL; + dev_err(dev, "fail to enable channel\n"); + return ret; + } + + dev_dbg(dev, "enable channel %s\n", isp4dbg_get_out_ch_str(cmd_ch_en.ch)); + + if (!sensor_info->start_stream_cmd_sent) { + ret = isp4sd_kickoff_stream(isp_subdev, + cmd_ch_prop.image_prop.width, + cmd_ch_prop.image_prop.height); + if (ret) { + dev_err(dev, "kickoff stream fail %d\n", ret); + return ret; + } + /* + * sensor_info->start_stream_cmd_sent will be set to true + * 1. in isp4sd_kickoff_stream, if app first send buffer then + * start stream + * 2. in isp_set_stream_buf, if app first start stream, then + * send buffer + * because ISP FW has the requirement, host needs to send buffer + * before send start stream cmd + */ + if (sensor_info->start_stream_cmd_sent) { + sensor_info->status = ISP4SD_START_STATUS_STARTED; + output_info->start_status = ISP4SD_START_STATUS_STARTED; + dev_dbg(dev, "kickoff stream suc,start cmd sent\n"); + } + } else { + dev_dbg(dev, "stream running, no need kickoff\n"); + output_info->start_status = ISP4SD_START_STATUS_STARTED; + } + + dev_dbg(dev, "setup output suc\n"); + return 0; +} + +static int isp4sd_init_stream(struct isp4_subdev *isp_subdev) +{ + struct device *dev = isp_subdev->dev; + int ret; + + ret = isp4sd_setup_fw_mem_pool(isp_subdev); + if (ret) { + dev_err(dev, "fail to setup fw mem pool\n"); + return ret; + } + + ret = isp4sd_set_stream_path(isp_subdev); + if (ret) { + dev_err(dev, "fail to setup stream path\n"); + return ret; + } + + return 0; +} + +static void isp4sd_uninit_stream(struct isp4_subdev *isp_subdev, + struct v4l2_subdev_state *state, u32 pad) +{ + struct isp4sd_sensor_info *sensor_info = &isp_subdev->sensor_info; + struct isp4sd_output_info *output_info = &sensor_info->output_info; + struct isp4_interface *ispif = &isp_subdev->ispif; + struct v4l2_mbus_framefmt *format; + + format = v4l2_subdev_state_get_format(state, pad, 0); + if (!format) { + dev_err(isp_subdev->dev, "fail to get v4l2 format\n"); + } else { + memset(format, 0, sizeof(*format)); + format->code = MEDIA_BUS_FMT_YUYV8_1_5X8; + } + + isp4if_clear_bufq(ispif); + isp4if_clear_cmdq(ispif); + + sensor_info->start_stream_cmd_sent = false; + sensor_info->buf_sent_cnt = 0; + + sensor_info->status = ISP4SD_START_STATUS_OFF; + output_info->start_status = ISP4SD_START_STATUS_OFF; +} + +static void isp4sd_fw_resp_cmd_done(struct isp4_subdev *isp_subdev, + enum isp4if_stream_id stream_id, + struct isp4fw_resp_cmd_done *para) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct isp4if_cmd_element *ele = + isp4if_rm_cmd_from_cmdq(ispif, para->cmd_seq_num, para->cmd_id); + struct device *dev = isp_subdev->dev; + + dev_dbg(dev, "stream %d,cmd %s(0x%08x)(%d),seq %u, ele %p\n", + stream_id, + isp4dbg_get_cmd_str(para->cmd_id), + para->cmd_id, para->cmd_status, para->cmd_seq_num, + ele); + + if (ele) { + complete(&ele->cmd_done); + if (atomic_dec_and_test(&ele->refcnt)) + kfree(ele); + } +} + +static struct isp4fw_meta_info *isp4sd_get_meta_by_mc(struct isp4_subdev *isp_subdev, + u64 mc) +{ + for (int i = 0; i < ISP4IF_MAX_STREAM_BUF_COUNT; i++) { + struct isp4if_gpu_mem_info *meta_info_buf = + isp_subdev->ispif.meta_info_buf[i]; + + if (meta_info_buf->gpu_mc_addr == mc) + return meta_info_buf->sys_addr; + } + + return NULL; +} + +static void isp4sd_send_meta_info(struct isp4_subdev *isp_subdev, + u64 meta_info_mc) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct isp4fw_cmd_send_buffer buf_type; + struct device *dev = isp_subdev->dev; + + if (isp_subdev->sensor_info.status != ISP4SD_START_STATUS_STARTED) { + dev_warn(dev, "not working status %i, meta_info 0x%llx\n", + isp_subdev->sensor_info.status, meta_info_mc); + return; + } + + /* + * The struct will be shared with ISP FW, use memset() to guarantee padding bits are + * zeroed, since this is not guaranteed on all compilers. + */ + memset(&buf_type, 0, sizeof(buf_type)); + buf_type.buffer_type = BUFFER_TYPE_META_INFO; + buf_type.buffer.vmid_space.bit.space = ADDR_SPACE_TYPE_GPU_VA; + isp4if_split_addr64(meta_info_mc, + &buf_type.buffer.buf_base_a_lo, + &buf_type.buffer.buf_base_a_hi); + buf_type.buffer.buf_size_a = ISP4SD_META_BUF_SIZE; + + if (isp4if_send_command(ispif, CMD_ID_SEND_BUFFER, + &buf_type, sizeof(buf_type))) + dev_err(dev, "fail send meta_info 0x%llx\n", + meta_info_mc); + else + dev_dbg(dev, "resend meta_info 0x%llx\n", meta_info_mc); +} + +static void isp4sd_fw_resp_frame_done(struct isp4_subdev *isp_subdev, + enum isp4if_stream_id stream_id, + struct isp4fw_resp_param_package *para) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = isp_subdev->dev; + struct isp4if_img_buf_node *prev; + struct isp4fw_meta_info *meta; + u64 mc; + + mc = isp4if_join_addr64(para->package_addr_lo, para->package_addr_hi); + meta = isp4sd_get_meta_by_mc(isp_subdev, mc); + if (!meta) { + dev_err(dev, "fail to get meta from mc %llx\n", mc); + return; + } + + dev_dbg(dev, "ts:%llu,streamId:%d,poc:%u,preview_en:%u,%s(%i)\n", + ktime_get_ns(), stream_id, meta->poc, + meta->preview.enabled, + isp4dbg_get_buf_done_str(meta->preview.status), + meta->preview.status); + + if (meta->preview.enabled && + (meta->preview.status == BUFFER_STATUS_SKIPPED || + meta->preview.status == BUFFER_STATUS_DONE || + meta->preview.status == BUFFER_STATUS_DIRTY)) { + prev = isp4if_dequeue_buffer(ispif); + if (prev) { + isp4dbg_show_bufmeta_info(dev, "prev", &meta->preview, + &prev->buf_info); + isp4vid_handle_frame_done(&isp_subdev->isp_vdev, + &prev->buf_info); + isp4if_dealloc_buffer_node(prev); + } else { + dev_err(dev, "fail null prev buf\n"); + } + } else if (meta->preview.enabled) { + dev_err(dev, "fail bad preview status %u(%s)\n", + meta->preview.status, + isp4dbg_get_buf_done_str(meta->preview.status)); + } + + if (isp_subdev->sensor_info.status == ISP4SD_START_STATUS_STARTED) + isp4sd_send_meta_info(isp_subdev, mc); + + dev_dbg(dev, "stream_id:%d, status:%d\n", stream_id, + isp_subdev->sensor_info.status); +} + +static void isp4sd_fw_resp_func(struct isp4_subdev *isp_subdev, + enum isp4if_stream_id stream_id) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = isp_subdev->dev; + struct isp4fw_resp resp; + + if (stream_id == ISP4IF_STREAM_ID_1) + isp_fw_log_print(isp_subdev); + + while (true) { + if (isp4if_f2h_resp(ispif, stream_id, &resp)) { + /* Re-enable the interrupt */ + isp4_intr_enable(isp_subdev, stream_id, true); + /* + * Recheck to see if there is a new response. + * To ensure that an in-flight interrupt is not lost, + * enabling the interrupt must occur _before_ checking + * for a new response, hence a memory barrier is needed. + * Disable the interrupt again if there was a new response. + */ + mb(); + if (likely(isp4if_f2h_resp(ispif, stream_id, &resp))) + break; + + isp4_intr_enable(isp_subdev, stream_id, false); + } + + switch (resp.resp_id) { + case RESP_ID_CMD_DONE: + isp4sd_fw_resp_cmd_done(isp_subdev, stream_id, + &resp.param.cmd_done); + break; + case RESP_ID_NOTI_FRAME_DONE: + isp4sd_fw_resp_frame_done(isp_subdev, stream_id, + &resp.param.frame_done); + break; + default: + dev_err(dev, "-><- fail respid %s(0x%x)\n", + isp4dbg_get_resp_str(resp.resp_id), + resp.resp_id); + break; + } + } +} + +static s32 isp4sd_fw_resp_thread(void *context) +{ + struct isp4_subdev_thread_param *para = context; + struct isp4_subdev *isp_subdev = para->isp_subdev; + struct isp4sd_thread_handler *thread_ctx = + &isp_subdev->fw_resp_thread[para->idx]; + struct device *dev = isp_subdev->dev; + + dev_dbg(dev, "[%u] fw resp thread started\n", para->idx); + while (true) { + wait_event_interruptible(thread_ctx->waitq, thread_ctx->resp_ready); + thread_ctx->resp_ready = false; + + if (kthread_should_stop()) { + dev_dbg(dev, "[%u] fw resp thread quit\n", para->idx); + break; + } + + isp4sd_fw_resp_func(isp_subdev, para->idx); + } + + return 0; +} + +static int isp4sd_stop_resp_proc_threads(struct isp4_subdev *isp_subdev) +{ + int i; + + for (i = 0; i < ISP4SD_MAX_FW_RESP_STREAM_NUM; i++) { + struct isp4sd_thread_handler *thread_ctx = + &isp_subdev->fw_resp_thread[i]; + + if (thread_ctx->thread) { + kthread_stop(thread_ctx->thread); + thread_ctx->thread = NULL; + } + } + + return 0; +} + +static int isp4sd_start_resp_proc_threads(struct isp4_subdev *isp_subdev) +{ + struct device *dev = isp_subdev->dev; + int i; + + for (i = 0; i < ISP4SD_MAX_FW_RESP_STREAM_NUM; i++) { + struct isp4sd_thread_handler *thread_ctx = &isp_subdev->fw_resp_thread[i]; + + isp_subdev->isp_resp_para[i].idx = i; + isp_subdev->isp_resp_para[i].isp_subdev = isp_subdev; + init_waitqueue_head(&thread_ctx->waitq); + thread_ctx->resp_ready = false; + + thread_ctx->thread = kthread_run(isp4sd_fw_resp_thread, + &isp_subdev->isp_resp_para[i], + isp4sd_thread_name[i]); + if (IS_ERR(thread_ctx->thread)) { + dev_err(dev, "create thread [%d] fail\n", i); + thread_ctx->thread = NULL; + isp4sd_stop_resp_proc_threads(isp_subdev); + return -EINVAL; + } + } + + return 0; +} + +static int isp4sd_pwroff_and_deinit(struct isp4_subdev *isp_subdev) +{ + struct isp4sd_sensor_info *sensor_info = &isp_subdev->sensor_info; + unsigned int perf_state = ISP4SD_PERFORMANCE_STATE_LOW; + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = isp_subdev->dev; + int ret; + + if (sensor_info->status == ISP4SD_START_STATUS_STARTED) { + dev_err(dev, "fail for stream still running\n"); + return -EINVAL; + } + + sensor_info->status = ISP4SD_START_STATUS_OFF; + + if (isp_subdev->irq_enabled) { + for (int i = 0; i < ISP4SD_MAX_FW_RESP_STREAM_NUM; i++) + disable_irq(isp_subdev->irq[i]); + isp_subdev->irq_enabled = false; + } + + isp4sd_stop_resp_proc_threads(isp_subdev); + dev_dbg(dev, "isp_subdev stop resp proc streads suc"); + + isp4if_stop(ispif); + + ret = dev_pm_genpd_set_performance_state(dev, perf_state); + if (ret) + dev_err(dev, "fail to set isp_subdev performance state %u,ret %d\n", + perf_state, ret); + + /* hold ccpu reset */ + isp4hw_wreg(isp_subdev->mmio, ISP_SOFT_RESET, 0); + isp4hw_wreg(isp_subdev->mmio, ISP_POWER_STATUS, 0); + ret = pm_runtime_put_sync(dev); + if (ret) + dev_err(dev, "power off isp_subdev fail %d\n", ret); + else + dev_dbg(dev, "power off isp_subdev suc\n"); + + ispif->status = ISP4IF_STATUS_PWR_OFF; + isp4if_clear_cmdq(ispif); + isp4sd_module_enable(isp_subdev, false); + + /* + * When opening the camera, isp4sd_module_enable(isp_subdev, true) is called. + * Hardware requires at least a 20ms delay between disabling and enabling the module, + * so a sleep is added to ensure ISP stability during quick reopen scenarios. + */ + msleep(20); + + return 0; +} + +static int isp4sd_pwron_and_init(struct isp4_subdev *isp_subdev) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = isp_subdev->dev; + int ret; + + if (ispif->status == ISP4IF_STATUS_FW_RUNNING) { + dev_dbg(dev, "camera already opened, do nothing\n"); + return 0; + } + + isp4sd_module_enable(isp_subdev, true); + + if (ispif->status < ISP4IF_STATUS_PWR_ON) { + unsigned int perf_state = ISP4SD_PERFORMANCE_STATE_HIGH; + + ret = pm_runtime_resume_and_get(dev); + if (ret) { + dev_err(dev, "fail to power on isp_subdev ret %d\n", + ret); + goto err_deinit; + } + + /* ISPPG ISP Power Status */ + isp4hw_wreg(isp_subdev->mmio, ISP_POWER_STATUS, 0x7FF); + ret = dev_pm_genpd_set_performance_state(dev, perf_state); + if (ret) { + dev_err(dev, + "fail to set performance state %u, ret %d\n", + perf_state, ret); + goto err_deinit; + } + + ispif->status = ISP4IF_STATUS_PWR_ON; + } + + isp_subdev->sensor_info.start_stream_cmd_sent = false; + isp_subdev->sensor_info.buf_sent_cnt = 0; + + ret = isp4if_start(ispif); + if (ret) { + dev_err(dev, "fail to start isp_subdev interface\n"); + goto err_deinit; + } + + if (isp4sd_start_resp_proc_threads(isp_subdev)) { + dev_err(dev, "isp_start_resp_proc_threads fail"); + goto err_deinit; + } + + dev_dbg(dev, "create resp threads ok"); + + for (int i = 0; i < ISP4SD_MAX_FW_RESP_STREAM_NUM; i++) + enable_irq(isp_subdev->irq[i]); + isp_subdev->irq_enabled = true; + + return 0; +err_deinit: + isp4sd_pwroff_and_deinit(isp_subdev); + return -EINVAL; +} + +static int isp4sd_stop_stream(struct isp4_subdev *isp_subdev, + struct v4l2_subdev_state *state, u32 pad) +{ + struct isp4sd_sensor_info *sensor_info = &isp_subdev->sensor_info; + struct isp4sd_output_info *output_info = &sensor_info->output_info; + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = isp_subdev->dev; + + guard(mutex)(&isp_subdev->ops_mutex); + dev_dbg(dev, "status %i\n", output_info->start_status); + + if (output_info->start_status == ISP4SD_START_STATUS_STARTED) { + struct isp4fw_cmd_enable_out_ch cmd_ch_disable; + int ret; + + /* + * The struct will be shared with ISP FW, use memset() to guarantee + * padding bits are zeroed, since this is not guaranteed on all compilers. + */ + memset(&cmd_ch_disable, 0, sizeof(cmd_ch_disable)); + cmd_ch_disable.ch = ISP_PIPE_OUT_CH_PREVIEW; + /* `cmd_ch_disable.is_enable` is already false */ + ret = isp4if_send_command_sync(ispif, CMD_ID_ENABLE_OUT_CHAN, + &cmd_ch_disable, + sizeof(cmd_ch_disable)); + if (ret) + dev_err(dev, "fail to disable stream\n"); + else + dev_dbg(dev, "wait disable stream suc\n"); + + ret = isp4if_send_command_sync(ispif, CMD_ID_STOP_STREAM, + NULL, 0); + if (ret) + dev_err(dev, "fail to stop steam\n"); + else + dev_dbg(dev, "wait stop stream suc\n"); + } + + isp4sd_uninit_stream(isp_subdev, state, pad); + + /* + * Return success to ensure the stop process proceeds, + * and disregard any errors since they are not fatal. + */ + return 0; +} + +static int isp4sd_start_stream(struct isp4_subdev *isp_subdev, + struct v4l2_subdev_state *state, u32 pad) +{ + struct isp4sd_output_info *output_info = + &isp_subdev->sensor_info.output_info; + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = isp_subdev->dev; + int ret; + + guard(mutex)(&isp_subdev->ops_mutex); + + if (ispif->status != ISP4IF_STATUS_FW_RUNNING) { + dev_err(dev, "fail, bad fsm %d", ispif->status); + return -EINVAL; + } + + switch (output_info->start_status) { + case ISP4SD_START_STATUS_OFF: + break; + case ISP4SD_START_STATUS_STARTED: + dev_dbg(dev, "stream already started, do nothing\n"); + return 0; + case ISP4SD_START_STATUS_START_FAIL: + dev_err(dev, "stream previously failed to start\n"); + return -EINVAL; + } + + ret = isp4sd_init_stream(isp_subdev); + if (ret) { + dev_err(dev, "fail to init isp_subdev stream\n"); + goto err_stop_stream; + } + + ret = isp4sd_setup_output(isp_subdev, state, pad); + if (ret) { + dev_err(dev, "fail to setup output\n"); + goto err_stop_stream; + } + + return 0; + +err_stop_stream: + isp4sd_stop_stream(isp_subdev, state, pad); + return ret; +} + +static int isp4sd_ioc_send_img_buf(struct v4l2_subdev *sd, + struct isp4if_img_buf_info *buf_info) +{ + struct isp4_subdev *isp_subdev = to_isp4_subdev(sd); + struct isp4_interface *ispif = &isp_subdev->ispif; + struct isp4if_img_buf_node *buf_node; + struct device *dev = isp_subdev->dev; + int ret; + + guard(mutex)(&isp_subdev->ops_mutex); + + if (ispif->status != ISP4IF_STATUS_FW_RUNNING) { + dev_err(dev, "fail send img buf for bad fsm %d\n", + ispif->status); + return -EINVAL; + } + + buf_node = isp4if_alloc_buffer_node(buf_info); + if (!buf_node) { + dev_err(dev, "fail alloc sys img buf info node\n"); + return -ENOMEM; + } + + ret = isp4if_queue_buffer(ispif, buf_node); + if (ret) { + dev_err(dev, "fail to queue image buf, %d\n", ret); + goto error_release_buf_node; + } + + if (!isp_subdev->sensor_info.start_stream_cmd_sent) { + isp_subdev->sensor_info.buf_sent_cnt++; + + if (isp_subdev->sensor_info.buf_sent_cnt >= + ISP4SD_MIN_BUF_CNT_BEF_START_STREAM) { + ret = isp4if_send_command(ispif, CMD_ID_START_STREAM, + NULL, 0); + if (ret) { + dev_err(dev, "fail to START_STREAM"); + goto error_release_buf_node; + } + isp_subdev->sensor_info.start_stream_cmd_sent = true; + isp_subdev->sensor_info.output_info.start_status = + ISP4SD_START_STATUS_STARTED; + isp_subdev->sensor_info.status = + ISP4SD_START_STATUS_STARTED; + } else { + dev_dbg(dev, "no send start, required %u, buf sent %u\n", + ISP4SD_MIN_BUF_CNT_BEF_START_STREAM, + isp_subdev->sensor_info.buf_sent_cnt); + } + } + + return 0; + +error_release_buf_node: + isp4if_dealloc_buffer_node(buf_node); + return ret; +} + +static int isp4sd_set_power(struct v4l2_subdev *sd, int on) +{ + struct isp4_subdev *isp_subdev = to_isp4_subdev(sd); + + guard(mutex)(&isp_subdev->ops_mutex); + if (on) + return isp4sd_pwron_and_init(isp_subdev); + else + return isp4sd_pwroff_and_deinit(isp_subdev); +} + +static const struct v4l2_subdev_core_ops isp4sd_core_ops = { + .s_power = isp4sd_set_power, +}; + +static const struct v4l2_subdev_video_ops isp4sd_video_ops = { + .s_stream = v4l2_subdev_s_stream_helper, +}; + +static int isp4sd_set_pad_format(struct v4l2_subdev *sd, + struct v4l2_subdev_state *sd_state, + struct v4l2_subdev_format *fmt) +{ + struct isp4sd_output_info *steam_info = + &(to_isp4_subdev(sd)->sensor_info.output_info); + struct v4l2_mbus_framefmt *format; + + format = v4l2_subdev_state_get_format(sd_state, fmt->pad); + + if (!format) { + dev_err(sd->dev, "fail to get state format\n"); + return -EINVAL; + } + + *format = fmt->format; + switch (format->code) { + case MEDIA_BUS_FMT_YUYV8_1_5X8: + steam_info->image_size = format->width * format->height * 3 / 2; + break; + case MEDIA_BUS_FMT_YUYV8_1X16: + steam_info->image_size = format->width * format->height * 2; + break; + default: + steam_info->image_size = 0; + break; + } + if (!steam_info->image_size) { + dev_err(sd->dev, "fail set pad format,code 0x%x,width %u, height %u\n", + format->code, format->width, format->height); + return -EINVAL; + } + + dev_dbg(sd->dev, "set pad format suc, code:%x w:%u h:%u size:%u\n", + format->code, format->width, format->height, steam_info->image_size); + + return 0; +} + +static int isp4sd_enable_streams(struct v4l2_subdev *sd, + struct v4l2_subdev_state *state, u32 pad, + u64 streams_mask) +{ + struct isp4_subdev *isp_subdev = to_isp4_subdev(sd); + + return isp4sd_start_stream(isp_subdev, state, pad); +} + +static int isp4sd_disable_streams(struct v4l2_subdev *sd, + struct v4l2_subdev_state *state, u32 pad, + u64 streams_mask) +{ + struct isp4_subdev *isp_subdev = to_isp4_subdev(sd); + + return isp4sd_stop_stream(isp_subdev, state, pad); +} + +static const struct v4l2_subdev_pad_ops isp4sd_pad_ops = { + .get_fmt = v4l2_subdev_get_fmt, + .set_fmt = isp4sd_set_pad_format, + .enable_streams = isp4sd_enable_streams, + .disable_streams = isp4sd_disable_streams, +}; + +static const struct v4l2_subdev_ops isp4sd_subdev_ops = { + .core = &isp4sd_core_ops, + .video = &isp4sd_video_ops, + .pad = &isp4sd_pad_ops, +}; + +static int isp4sd_sdev_link_validate(struct media_link *link) +{ + return 0; +} + +static const struct media_entity_operations isp4sd_sdev_ent_ops = { + .link_validate = isp4sd_sdev_link_validate, +}; + +static const struct isp4vid_ops isp4sd_isp4vid_ops = { + .send_buffer = isp4sd_ioc_send_img_buf, +}; + +int isp4sd_init(struct isp4_subdev *isp_subdev, struct v4l2_device *v4l2_dev, + int irq[ISP4SD_MAX_FW_RESP_STREAM_NUM]) +{ + struct isp4sd_sensor_info *sensor_info = &isp_subdev->sensor_info; + struct isp4_interface *ispif = &isp_subdev->ispif; + struct device *dev = v4l2_dev->dev; + int ret; + + isp_subdev->dev = dev; + v4l2_subdev_init(&isp_subdev->sdev, &isp4sd_subdev_ops); + isp_subdev->sdev.owner = THIS_MODULE; + isp_subdev->sdev.dev = dev; + snprintf(isp_subdev->sdev.name, sizeof(isp_subdev->sdev.name), "%s", + dev_name(dev)); + + isp_subdev->sdev.entity.name = isp4sd_entity_name; + isp_subdev->sdev.entity.function = MEDIA_ENT_F_PROC_VIDEO_ISP; + isp_subdev->sdev.entity.ops = &isp4sd_sdev_ent_ops; + isp_subdev->sdev_pad.flags = MEDIA_PAD_FL_SOURCE; + ret = media_entity_pads_init(&isp_subdev->sdev.entity, 1, + &isp_subdev->sdev_pad); + if (ret) { + dev_err(dev, "fail to init isp4 subdev entity pad %d\n", ret); + return ret; + } + + ret = v4l2_subdev_init_finalize(&isp_subdev->sdev); + if (ret < 0) { + dev_err(dev, "fail to init finalize isp4 subdev %d\n", + ret); + return ret; + } + + ret = v4l2_device_register_subdev(v4l2_dev, &isp_subdev->sdev); + if (ret) { + dev_err(dev, "fail to register isp4 subdev to V4L2 device %d\n", ret); + goto err_media_clean_up; + } + + isp4if_init(ispif, dev, isp_subdev->mmio); + + mutex_init(&isp_subdev->ops_mutex); + sensor_info->status = ISP4SD_START_STATUS_OFF; + + /* create ISP enable gpio control */ + isp_subdev->enable_gpio = devm_gpiod_get(isp_subdev->dev, + "enable_isp", + GPIOD_OUT_LOW); + if (IS_ERR(isp_subdev->enable_gpio)) { + ret = PTR_ERR(isp_subdev->enable_gpio); + dev_err(dev, "fail to get gpiod %d\n", ret); + goto err_subdev_unreg; + } + + for (int i = 0; i < ISP4SD_MAX_FW_RESP_STREAM_NUM; i++) + isp_subdev->irq[i] = irq[i]; + + isp_subdev->host2fw_seq_num = 1; + ispif->status = ISP4IF_STATUS_PWR_OFF; + + ret = isp4vid_dev_init(&isp_subdev->isp_vdev, &isp_subdev->sdev, + &isp4sd_isp4vid_ops); + if (ret) + goto err_subdev_unreg; + + return 0; + +err_subdev_unreg: + v4l2_device_unregister_subdev(&isp_subdev->sdev); +err_media_clean_up: + v4l2_subdev_cleanup(&isp_subdev->sdev); + media_entity_cleanup(&isp_subdev->sdev.entity); + return ret; +} + +void isp4sd_deinit(struct isp4_subdev *isp_subdev) +{ + struct isp4_interface *ispif = &isp_subdev->ispif; + + isp4vid_dev_deinit(&isp_subdev->isp_vdev); + v4l2_device_unregister_subdev(&isp_subdev->sdev); + media_entity_cleanup(&isp_subdev->sdev.entity); + isp4if_deinit(ispif); + isp4sd_module_enable(isp_subdev, false); + + ispif->status = ISP4IF_STATUS_PWR_OFF; +} diff --git a/drivers/media/platform/amd/isp4/isp4_subdev.h b/drivers/media/platform/amd/isp4/isp4_subdev.h new file mode 100644 index 000000000000..942245a6f356 --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_subdev.h @@ -0,0 +1,131 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#ifndef _ISP4_SUBDEV_H_ +#define _ISP4_SUBDEV_H_ + +#include +#include +#include +#include +#include +#include +#include + +#include "isp4_fw_cmd_resp.h" +#include "isp4_hw_reg.h" +#include "isp4_interface.h" +#include "isp4_video.h" + +/* + * one is for none sesnor specefic response which is not used now + * another is for sensor specific response + */ +#define ISP4SD_MAX_FW_RESP_STREAM_NUM 2 + +/* + * cmd used to register frame done callback, parameter is + * struct isp4sd_register_framedone_cb_param * + * when a image buffer is filled by ISP, ISP will call the registered callback. + * callback func prototype is isp4sd_framedone_cb, cb_ctx can be anything + * provided by caller which will be provided back as the first parameter of the + * callback function. + * both cb_func and cb_ctx are provide by caller, set cb_func to NULL to + * unregister the callback + */ + +/* used to indicate the ISP status */ +enum isp4sd_status { + ISP4SD_STATUS_PWR_OFF, + ISP4SD_STATUS_PWR_ON, + ISP4SD_STATUS_FW_RUNNING, + ISP4SD_STATUS_MAX +}; + +/* used to indicate the status of sensor, output stream */ +enum isp4sd_start_status { + ISP4SD_START_STATUS_OFF, + ISP4SD_START_STATUS_STARTED, + ISP4SD_START_STATUS_START_FAIL, +}; + +struct isp4sd_img_buf_node { + struct list_head node; + struct isp4if_img_buf_info buf_info; +}; + +/* this is isp output after processing bayer raw input from sensor */ +struct isp4sd_output_info { + enum isp4sd_start_status start_status; + u32 image_size; +}; + +/* + * This struct represents the sensor info which is input or source of ISP, + * status is the sensor status + * output_info is the isp output info after ISP processing the sensor input, + * start_stream_cmd_sent mean if CMD_ID_START_STREAM has sent to fw. + * buf_sent_cnt is buffer count app has sent to receive the images + */ +struct isp4sd_sensor_info { + struct isp4sd_output_info output_info; + enum isp4sd_start_status status; + bool start_stream_cmd_sent; + u32 buf_sent_cnt; +}; + +/* + * Thread created by driver to receive fw response + * thread will be wakeup by fw to driver response interrupt + */ +struct isp4sd_thread_handler { + struct task_struct *thread; + wait_queue_head_t waitq; + bool resp_ready; +}; + +struct isp4_subdev_thread_param { + u32 idx; + struct isp4_subdev *isp_subdev; +}; + +struct isp4_subdev { + struct v4l2_subdev sdev; + struct isp4_interface ispif; + struct isp4vid_dev isp_vdev; + + struct media_pad sdev_pad; + + enum isp4sd_status isp_status; + struct mutex ops_mutex; /* ops_mutex */ + + /* Used to store fw cmds sent to FW whose response driver needs to wait for */ + struct isp4sd_thread_handler fw_resp_thread[ISP4SD_MAX_FW_RESP_STREAM_NUM]; + + u32 host2fw_seq_num; + + struct isp4sd_sensor_info sensor_info; + + /* gpio descriptor */ + struct gpio_desc *enable_gpio; + struct device *dev; + void __iomem *mmio; + struct isp4_subdev_thread_param isp_resp_para[ISP4SD_MAX_FW_RESP_STREAM_NUM]; + int irq[ISP4SD_MAX_FW_RESP_STREAM_NUM]; + bool irq_enabled; + /* spin lock to access ISP_SYS_INT0_EN exclusively */ + spinlock_t irq_lock; +#ifdef CONFIG_DEBUG_FS + bool enable_fw_log; + struct dentry *debugfs_dir; + char *fw_log_output; +#endif +}; + +int isp4sd_init(struct isp4_subdev *isp_subdev, struct v4l2_device *v4l2_dev, + int irq[ISP4SD_MAX_FW_RESP_STREAM_NUM]); +void isp4sd_deinit(struct isp4_subdev *isp_subdev); + +#endif /* _ISP4_SUBDEV_H_ */ diff --git a/drivers/media/platform/amd/isp4/isp4_video.c b/drivers/media/platform/amd/isp4/isp4_video.c new file mode 100644 index 000000000000..5006e5021155 --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_video.c @@ -0,0 +1,1165 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#include +#include +#include + +#include "isp4_interface.h" +#include "isp4_subdev.h" +#include "isp4_video.h" + +#define ISP4VID_ISP_DRV_NAME "amd_isp_capture" +#define ISP4VID_MAX_PREVIEW_FPS 30 +#define ISP4VID_DEFAULT_FMT isp4vid_formats[0] + +#define ISP4VID_PAD_VIDEO_OUTPUT 0 + +/* timeperframe default */ +#define ISP4VID_ISP_TPF_DEFAULT isp4vid_tpfs[0] + +/* amdisp buffer for vb2 operations */ +struct isp4vid_vb2_buf { + struct device *dev; + void *vaddr; + unsigned long size; + refcount_t refcount; + struct dma_buf *dbuf; + void *bo; + u64 gpu_addr; + struct vb2_vmarea_handler handler; +}; + +static void isp4vid_vb2_put(void *buf_priv); + +static const char *const isp4vid_video_dev_name = "Preview"; + +/* Sizes must be in increasing order */ +static const struct v4l2_frmsize_discrete isp4vid_frmsize[] = { + {640, 360}, + {640, 480}, + {1280, 720}, + {1280, 960}, + {1920, 1080}, + {1920, 1440}, + {2560, 1440}, + {2880, 1620}, + {2880, 1624}, + {2888, 1808}, +}; + +static const u32 isp4vid_formats[] = { + V4L2_PIX_FMT_NV12, + V4L2_PIX_FMT_YUYV +}; + +/* timeperframe list */ +static const struct v4l2_fract isp4vid_tpfs[] = { + { 1, ISP4VID_MAX_PREVIEW_FPS } +}; + +void isp4vid_handle_frame_done(struct isp4vid_dev *isp_vdev, + const struct isp4if_img_buf_info *img_buf) +{ + struct isp4vid_capture_buffer *isp4vid_buf; + void *vbuf; + + scoped_guard(mutex, &isp_vdev->buf_list_lock) { + /* Get the first entry of the list */ + isp4vid_buf = list_first_entry_or_null(&isp_vdev->buf_list, typeof(*isp4vid_buf), + list); + if (!isp4vid_buf) + return; + + vbuf = vb2_plane_vaddr(&isp4vid_buf->vb2.vb2_buf, 0); + + if (vbuf != img_buf->planes[0].sys_addr) { + dev_err(isp_vdev->dev, "Invalid vbuf"); + return; + } + + /* Remove this entry from the list */ + list_del(&isp4vid_buf->list); + } + + /* Fill the buffer */ + isp4vid_buf->vb2.vb2_buf.timestamp = ktime_get_ns(); + isp4vid_buf->vb2.sequence = isp_vdev->sequence++; + isp4vid_buf->vb2.field = V4L2_FIELD_ANY; + + /* at most 2 planes */ + vb2_set_plane_payload(&isp4vid_buf->vb2.vb2_buf, + 0, isp_vdev->format.sizeimage); + + vb2_buffer_done(&isp4vid_buf->vb2.vb2_buf, VB2_BUF_STATE_DONE); + + dev_dbg(isp_vdev->dev, "call vb2_buffer_done(size=%u)\n", + isp_vdev->format.sizeimage); +} + +static unsigned int isp4vid_vb2_num_users(void *buf_priv) +{ + struct isp4vid_vb2_buf *buf = buf_priv; + + return refcount_read(&buf->refcount); +} + +static int isp4vid_vb2_mmap(void *buf_priv, struct vm_area_struct *vma) +{ + struct isp4vid_vb2_buf *buf = buf_priv; + int ret; + + if (!buf) { + pr_err("fail no memory to map\n"); + return -EINVAL; + } + + ret = remap_vmalloc_range(vma, buf->vaddr, 0); + if (ret) { + dev_err(buf->dev, "fail remap vmalloc mem, %d\n", ret); + return ret; + } + + /* + * Make sure that vm_areas for 2 buffers won't be merged together + */ + vm_flags_set(vma, VM_DONTEXPAND); + + /* + * Use common vm_area operations to track buffer refcount. + */ + vma->vm_private_data = &buf->handler; + vma->vm_ops = &vb2_common_vm_ops; + + vma->vm_ops->open(vma); + + dev_dbg(buf->dev, "mmap isp user bo 0x%llx size %ld refcount %d\n", + buf->gpu_addr, buf->size, refcount_read(&buf->refcount)); + + return 0; +} + +static void *isp4vid_vb2_vaddr(struct vb2_buffer *vb, void *buf_priv) +{ + struct isp4vid_vb2_buf *buf = buf_priv; + + if (!buf->vaddr) { + dev_err(buf->dev, + "fail for buf vaddr is null\n"); + return NULL; + } + + return buf->vaddr; +} + +static void isp4vid_vb2_detach_dmabuf(void *mem_priv) +{ + struct isp4vid_vb2_buf *buf = mem_priv; + + dev_dbg(buf->dev, "detach dmabuf of isp user bo 0x%llx size %ld", + buf->gpu_addr, buf->size); + + kfree(buf); +} + +static void *isp4vid_vb2_attach_dmabuf(struct vb2_buffer *vb, struct device *dev, + struct dma_buf *dbuf, + unsigned long size) +{ + struct isp4vid_vb2_buf *dbg_buf = dbuf->priv; + struct isp4vid_vb2_buf *buf; + + if (dbuf->size < size) { + dev_err(dev, "Invalid dmabuf size %zu %lu", dbuf->size, size); + return ERR_PTR(-EFAULT); + } + + buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (!buf) + return ERR_PTR(-ENOMEM); + + buf->dev = dev; + buf->dbuf = dbuf; + buf->size = size; + + dev_dbg(dev, "attach dmabuf of isp user bo 0x%llx size %ld", + dbg_buf->gpu_addr, dbg_buf->size); + + return buf; +} + +static void isp4vid_vb2_unmap_dmabuf(void *mem_priv) +{ + struct isp4vid_vb2_buf *buf = mem_priv; + struct iosys_map map = IOSYS_MAP_INIT_VADDR(buf->vaddr); + + dev_dbg(buf->dev, "unmap dmabuf of isp user bo 0x%llx size %ld", + buf->gpu_addr, buf->size); + + dma_buf_vunmap_unlocked(buf->dbuf, &map); + buf->vaddr = NULL; +} + +static int isp4vid_vb2_map_dmabuf(void *mem_priv) +{ + struct isp4vid_vb2_buf *buf = mem_priv, *mmap_buf; + struct iosys_map map; + int ret; + + ret = dma_buf_vmap_unlocked(buf->dbuf, &map); + if (ret) { + dev_err(buf->dev, "vmap_unlocked fail"); + return -EFAULT; + } + + buf->vaddr = map.vaddr; + mmap_buf = buf->dbuf->priv; + buf->gpu_addr = mmap_buf->gpu_addr; + + dev_dbg(buf->dev, "map dmabuf of isp user bo 0x%llx size %ld", + buf->gpu_addr, buf->size); + + return 0; +} + +#ifdef CONFIG_HAS_DMA +struct isp4vid_vb2_amdgpu_attachment { + struct sg_table sgt; + enum dma_data_direction dma_dir; +}; + +static int isp4vid_dmabuf_ops_attach(struct dma_buf *dbuf, + struct dma_buf_attachment *dbuf_attach) +{ + struct isp4vid_vb2_buf *buf = dbuf->priv; + int num_pages = PAGE_ALIGN(buf->size) / PAGE_SIZE; + struct isp4vid_vb2_amdgpu_attachment *attach; + void *vaddr = buf->vaddr; + struct scatterlist *sg; + struct sg_table *sgt; + int ret; + int i; + + attach = kzalloc(sizeof(*attach), GFP_KERNEL); + if (!attach) + return -ENOMEM; + + sgt = &attach->sgt; + ret = sg_alloc_table(sgt, num_pages, GFP_KERNEL); + if (ret) { + kfree(attach); + return ret; + } + + for_each_sgtable_sg(sgt, sg, i) { + struct page *page = vmalloc_to_page(vaddr); + + if (!page) { + sg_free_table(sgt); + kfree(attach); + return -ENOMEM; + } + + sg_set_page(sg, page, PAGE_SIZE, 0); + vaddr = ((char *)vaddr) + PAGE_SIZE; + } + + attach->dma_dir = DMA_NONE; + dbuf_attach->priv = attach; + + return 0; +} + +static void isp4vid_dmabuf_ops_detach(struct dma_buf *dbuf, + struct dma_buf_attachment *dbuf_attach) +{ + struct isp4vid_vb2_amdgpu_attachment *attach = dbuf_attach->priv; + struct sg_table *sgt; + + if (!attach) { + pr_err("fail invalid attach handler\n"); + return; + } + + sgt = &attach->sgt; + + /* release the scatterlist cache */ + if (attach->dma_dir != DMA_NONE) + dma_unmap_sgtable(dbuf_attach->dev, sgt, attach->dma_dir, 0); + + sg_free_table(sgt); + kfree(attach); + dbuf_attach->priv = NULL; +} + +static struct sg_table *isp4vid_dmabuf_ops_map(struct dma_buf_attachment *dbuf_attach, + enum dma_data_direction dma_dir) +{ + struct isp4vid_vb2_amdgpu_attachment *attach = dbuf_attach->priv; + struct sg_table *sgt; + + sgt = &attach->sgt; + /* return previously mapped sg table */ + if (attach->dma_dir == dma_dir) + return sgt; + + /* release any previous cache */ + if (attach->dma_dir != DMA_NONE) { + dma_unmap_sgtable(dbuf_attach->dev, sgt, attach->dma_dir, 0); + attach->dma_dir = DMA_NONE; + } + + /* mapping to the client with new direction */ + if (dma_map_sgtable(dbuf_attach->dev, sgt, dma_dir, 0)) { + dev_err(dbuf_attach->dev, "fail to map scatterlist"); + return ERR_PTR(-EIO); + } + + attach->dma_dir = dma_dir; + + return sgt; +} + +static void isp4vid_dmabuf_ops_unmap(struct dma_buf_attachment *dbuf_attach, + struct sg_table *sgt, + enum dma_data_direction dma_dir) +{ + /* nothing to be done here */ +} + +static int isp4vid_dmabuf_ops_vmap(struct dma_buf *dbuf, + struct iosys_map *map) +{ + struct isp4vid_vb2_buf *buf = dbuf->priv; + + iosys_map_set_vaddr(map, buf->vaddr); + + return 0; +} + +static void isp4vid_dmabuf_ops_release(struct dma_buf *dbuf) +{ + struct isp4vid_vb2_buf *buf = dbuf->priv; + + /* drop reference obtained in isp4vid_vb2_get_dmabuf */ + if (dbuf != buf->dbuf) + isp4vid_vb2_put(buf); + else + kfree(buf); +} + +static int isp4vid_dmabuf_ops_mmap(struct dma_buf *dbuf, struct vm_area_struct *vma) +{ + return isp4vid_vb2_mmap(dbuf->priv, vma); +} + +static const struct dma_buf_ops isp4vid_dmabuf_ops = { + .attach = isp4vid_dmabuf_ops_attach, + .detach = isp4vid_dmabuf_ops_detach, + .map_dma_buf = isp4vid_dmabuf_ops_map, + .unmap_dma_buf = isp4vid_dmabuf_ops_unmap, + .vmap = isp4vid_dmabuf_ops_vmap, + .mmap = isp4vid_dmabuf_ops_mmap, + .release = isp4vid_dmabuf_ops_release, +}; + +static struct dma_buf *isp4vid_get_dmabuf(struct isp4vid_vb2_buf *buf, unsigned long flags) +{ + DEFINE_DMA_BUF_EXPORT_INFO(exp_info); + struct dma_buf *dbuf; + + if (WARN_ON(!buf->vaddr)) + return NULL; + + exp_info.ops = &isp4vid_dmabuf_ops; + exp_info.size = buf->size; + exp_info.flags = flags; + exp_info.priv = buf; + + dbuf = dma_buf_export(&exp_info); + if (IS_ERR(dbuf)) + return NULL; + + return dbuf; +} + +static struct dma_buf *isp4vid_vb2_get_dmabuf(struct vb2_buffer *vb, void *buf_priv, + unsigned long flags) +{ + struct isp4vid_vb2_buf *buf = buf_priv; + struct dma_buf *dbuf; + + dbuf = isp4vid_get_dmabuf(buf, flags); + if (!dbuf) { + dev_err(buf->dev, "fail to get isp dma buffer\n"); + return NULL; + } + + refcount_inc(&buf->refcount); + + dev_dbg(buf->dev, "buf exported, refcount %d\n", + refcount_read(&buf->refcount)); + + return dbuf; +} +#endif /* CONFIG_HAS_DMA */ + +static void isp4vid_vb2_put(void *buf_priv) +{ + struct isp4vid_vb2_buf *buf = buf_priv; + + dev_dbg(buf->dev, + "release isp user bo 0x%llx size %ld refcount %d", + buf->gpu_addr, buf->size, + refcount_read(&buf->refcount)); + + if (refcount_dec_and_test(&buf->refcount)) { + isp_user_buffer_free(buf->bo); + vfree(buf->vaddr); + /* + * Putting the implicit dmabuf frees `buf`. Freeing `buf` must + * be done from the dmabuf release callback since dma_buf_put() + * isn't always synchronous; it's just an fput(), which may be + * deferred. Since the dmabuf release callback needs to access + * `buf`, this means `buf` cannot be freed until then. + */ + dma_buf_put(buf->dbuf); + } +} + +static void *isp4vid_vb2_alloc(struct vb2_buffer *vb, struct device *dev, unsigned long size) +{ + struct isp4vid_vb2_buf *buf; + u64 gpu_addr; + void *bo; + int ret; + + buf = kzalloc(sizeof(*buf), GFP_KERNEL | vb->vb2_queue->gfp_flags); + if (!buf) + return ERR_PTR(-ENOMEM); + + buf->dev = dev; + buf->size = size; + buf->vaddr = vmalloc_user(buf->size); + if (!buf->vaddr) { + dev_err(dev, "fail to vmalloc buffer\n"); + goto free_buf; + } + + buf->handler.refcount = &buf->refcount; + buf->handler.put = isp4vid_vb2_put; + buf->handler.arg = buf; + + /* get implicit dmabuf */ + buf->dbuf = isp4vid_get_dmabuf(buf, 0); + if (!buf->dbuf) { + dev_err(dev, "fail to get implicit dmabuf\n"); + goto free_user_vmem; + } + + /* create isp user BO and obtain gpu_addr */ + ret = isp_user_buffer_alloc(dev, buf->dbuf, &bo, &gpu_addr); + if (ret) { + dev_err(dev, "fail to create isp user BO\n"); + goto put_dmabuf; + } + + buf->bo = bo; + buf->gpu_addr = gpu_addr; + + refcount_set(&buf->refcount, 1); + + dev_dbg(dev, "allocated isp user bo 0x%llx size %ld refcount %d\n", + buf->gpu_addr, buf->size, refcount_read(&buf->refcount)); + + return buf; + +put_dmabuf: + dma_buf_put(buf->dbuf); +free_user_vmem: + vfree(buf->vaddr); +free_buf: + ret = buf->vaddr ? -EINVAL : -ENOMEM; + kfree(buf); + return ERR_PTR(ret); +} + +static const struct vb2_mem_ops isp4vid_vb2_memops = { + .alloc = isp4vid_vb2_alloc, + .put = isp4vid_vb2_put, +#ifdef CONFIG_HAS_DMA + .get_dmabuf = isp4vid_vb2_get_dmabuf, +#endif + .map_dmabuf = isp4vid_vb2_map_dmabuf, + .unmap_dmabuf = isp4vid_vb2_unmap_dmabuf, + .attach_dmabuf = isp4vid_vb2_attach_dmabuf, + .detach_dmabuf = isp4vid_vb2_detach_dmabuf, + .vaddr = isp4vid_vb2_vaddr, + .mmap = isp4vid_vb2_mmap, + .num_users = isp4vid_vb2_num_users, +}; + +static const struct v4l2_pix_format isp4vid_fmt_default = { + .width = 1920, + .height = 1080, + .pixelformat = ISP4VID_DEFAULT_FMT, + .field = V4L2_FIELD_NONE, + .colorspace = V4L2_COLORSPACE_SRGB, +}; + +static void isp4vid_capture_return_all_buffers(struct isp4vid_dev *isp_vdev, + enum vb2_buffer_state state) +{ + struct isp4vid_capture_buffer *vbuf, *node; + + scoped_guard(mutex, &isp_vdev->buf_list_lock) { + list_for_each_entry_safe(vbuf, node, &isp_vdev->buf_list, list) + vb2_buffer_done(&vbuf->vb2.vb2_buf, state); + INIT_LIST_HEAD(&isp_vdev->buf_list); + } + + dev_dbg(isp_vdev->dev, "call vb2_buffer_done(%d)\n", state); +} + +static int isp4vid_vdev_link_validate(struct media_link *link) +{ + return 0; +} + +static const struct media_entity_operations isp4vid_vdev_ent_ops = { + .link_validate = isp4vid_vdev_link_validate, +}; + +static const struct v4l2_file_operations isp4vid_vdev_fops = { + .owner = THIS_MODULE, + .open = v4l2_fh_open, + .release = vb2_fop_release, + .read = vb2_fop_read, + .poll = vb2_fop_poll, + .unlocked_ioctl = video_ioctl2, + .mmap = vb2_fop_mmap, +}; + +static int isp4vid_ioctl_querycap(struct file *file, void *fh, struct v4l2_capability *cap) +{ + struct isp4vid_dev *isp_vdev = video_drvdata(file); + + strscpy(cap->driver, ISP4VID_ISP_DRV_NAME, sizeof(cap->driver)); + snprintf(cap->card, sizeof(cap->card), "%s", ISP4VID_ISP_DRV_NAME); + cap->capabilities |= V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE; + + dev_dbg(isp_vdev->dev, "%s|capabilities=0x%X", isp_vdev->vdev.name, cap->capabilities); + + return 0; +} + +static int isp4vid_g_fmt_vid_cap(struct file *file, void *priv, struct v4l2_format *f) +{ + struct isp4vid_dev *isp_vdev = video_drvdata(file); + + f->fmt.pix = isp_vdev->format; + + return 0; +} + +static int isp4vid_fill_buffer_size(struct v4l2_pix_format *fmt) +{ + int ret = 0; + + switch (fmt->pixelformat) { + case V4L2_PIX_FMT_NV12: + fmt->bytesperline = fmt->width; + fmt->sizeimage = fmt->bytesperline * fmt->height * 3 / 2; + break; + case V4L2_PIX_FMT_YUYV: + fmt->bytesperline = fmt->width * 2; + fmt->sizeimage = fmt->bytesperline * fmt->height; + break; + default: + ret = -EINVAL; + break; + } + + return ret; +} + +static int isp4vid_try_fmt_vid_cap(struct file *file, void *priv, struct v4l2_format *f) +{ + struct isp4vid_dev *isp_vdev = video_drvdata(file); + struct v4l2_pix_format *format = &f->fmt.pix; + const struct v4l2_frmsize_discrete *fsz; + int i; + + if (f->type != V4L2_BUF_TYPE_VIDEO_CAPTURE) + return -EINVAL; + + /* + * Check if the hardware supports the requested format, use the default + * format otherwise. + */ + for (i = 0; i < ARRAY_SIZE(isp4vid_formats); i++) + if (isp4vid_formats[i] == format->pixelformat) + break; + + if (i == ARRAY_SIZE(isp4vid_formats)) + format->pixelformat = ISP4VID_DEFAULT_FMT; + + switch (format->pixelformat) { + case V4L2_PIX_FMT_NV12: + case V4L2_PIX_FMT_YUYV: + fsz = v4l2_find_nearest_size(isp4vid_frmsize, ARRAY_SIZE(isp4vid_frmsize), + width, height, format->width, format->height); + format->width = fsz->width; + format->height = fsz->height; + break; + default: + dev_err(isp_vdev->dev, "%s|unsupported fmt=%u", isp_vdev->vdev.name, + format->pixelformat); + return -EINVAL; + } + + /* There is no need to check the return value, as failure will never happen here */ + isp4vid_fill_buffer_size(format); + + if (format->field == V4L2_FIELD_ANY) + format->field = isp4vid_fmt_default.field; + + if (format->colorspace == V4L2_COLORSPACE_DEFAULT) + format->colorspace = isp4vid_fmt_default.colorspace; + + return 0; +} + +static int isp4vid_set_fmt_2_isp(struct v4l2_subdev *sdev, struct v4l2_pix_format *pix_fmt) +{ + struct v4l2_subdev_format fmt = {}; + + switch (pix_fmt->pixelformat) { + case V4L2_PIX_FMT_NV12: + fmt.format.code = MEDIA_BUS_FMT_YUYV8_1_5X8; + break; + case V4L2_PIX_FMT_YUYV: + fmt.format.code = MEDIA_BUS_FMT_YUYV8_1X16; + break; + default: + return -EINVAL; + } + fmt.which = V4L2_SUBDEV_FORMAT_ACTIVE; + fmt.pad = ISP4VID_PAD_VIDEO_OUTPUT; + fmt.format.width = pix_fmt->width; + fmt.format.height = pix_fmt->height; + return v4l2_subdev_call(sdev, pad, set_fmt, NULL, &fmt); +} + +static int isp4vid_s_fmt_vid_cap(struct file *file, void *priv, struct v4l2_format *f) +{ + struct isp4vid_dev *isp_vdev = video_drvdata(file); + int ret; + + /* Do not change the format while stream is on */ + if (vb2_is_busy(&isp_vdev->vbq)) + return -EBUSY; + + if (f->type != V4L2_BUF_TYPE_VIDEO_CAPTURE) + return -EINVAL; + + ret = isp4vid_try_fmt_vid_cap(file, priv, f); + if (ret) + return ret; + + dev_dbg(isp_vdev->dev, "%s|width height:%ux%u->%ux%u", + isp_vdev->vdev.name, + isp_vdev->format.width, isp_vdev->format.height, + f->fmt.pix.width, f->fmt.pix.height); + dev_dbg(isp_vdev->dev, "%s|pixelformat:0x%x-0x%x", + isp_vdev->vdev.name, isp_vdev->format.pixelformat, + f->fmt.pix.pixelformat); + dev_dbg(isp_vdev->dev, "%s|bytesperline:%u->%u", + isp_vdev->vdev.name, isp_vdev->format.bytesperline, + f->fmt.pix.bytesperline); + dev_dbg(isp_vdev->dev, "%s|sizeimage:%u->%u", + isp_vdev->vdev.name, isp_vdev->format.sizeimage, + f->fmt.pix.sizeimage); + + isp_vdev->format = f->fmt.pix; + ret = isp4vid_set_fmt_2_isp(isp_vdev->isp_sdev, &isp_vdev->format); + + return ret; +} + +static int isp4vid_enum_fmt_vid_cap(struct file *file, void *priv, struct v4l2_fmtdesc *f) +{ + struct isp4vid_dev *isp_vdev = video_drvdata(file); + + switch (f->index) { + case 0: + f->pixelformat = V4L2_PIX_FMT_NV12; + break; + case 1: + f->pixelformat = V4L2_PIX_FMT_YUYV; + break; + default: + return -EINVAL; + } + + dev_dbg(isp_vdev->dev, "%s|index=%d, pixelformat=0x%X", + isp_vdev->vdev.name, f->index, f->pixelformat); + + return 0; +} + +static int isp4vid_enum_framesizes(struct file *file, void *fh, struct v4l2_frmsizeenum *fsize) +{ + struct isp4vid_dev *isp_vdev = video_drvdata(file); + unsigned int i; + + for (i = 0; i < ARRAY_SIZE(isp4vid_formats); i++) { + if (isp4vid_formats[i] == fsize->pixel_format) + break; + } + if (i == ARRAY_SIZE(isp4vid_formats)) + return -EINVAL; + + if (fsize->index < ARRAY_SIZE(isp4vid_frmsize)) { + fsize->type = V4L2_FRMSIZE_TYPE_DISCRETE; + fsize->discrete = isp4vid_frmsize[fsize->index]; + dev_dbg(isp_vdev->dev, "%s|size[%d]=%dx%d", + isp_vdev->vdev.name, fsize->index, + fsize->discrete.width, fsize->discrete.height); + } else { + return -EINVAL; + } + + return 0; +} + +static int isp4vid_ioctl_enum_frameintervals(struct file *file, void *priv, + struct v4l2_frmivalenum *fival) +{ + struct isp4vid_dev *isp_vdev = video_drvdata(file); + int i; + + if (fival->index >= ARRAY_SIZE(isp4vid_tpfs)) + return -EINVAL; + + for (i = 0; i < ARRAY_SIZE(isp4vid_formats); i++) + if (isp4vid_formats[i] == fival->pixel_format) + break; + + if (i == ARRAY_SIZE(isp4vid_formats)) + return -EINVAL; + + for (i = 0; i < ARRAY_SIZE(isp4vid_frmsize); i++) + if (isp4vid_frmsize[i].width == fival->width && + isp4vid_frmsize[i].height == fival->height) + break; + + if (i == ARRAY_SIZE(isp4vid_frmsize)) + return -EINVAL; + + fival->type = V4L2_FRMIVAL_TYPE_DISCRETE; + fival->discrete = isp4vid_tpfs[fival->index]; + v4l2_simplify_fraction(&fival->discrete.numerator, + &fival->discrete.denominator, 8, 333); + + dev_dbg(isp_vdev->dev, "%s|interval[%d]=%d/%d", + isp_vdev->vdev.name, fival->index, + fival->discrete.numerator, + fival->discrete.denominator); + + return 0; +} + +static int isp4vid_ioctl_g_param(struct file *file, void *priv, struct v4l2_streamparm *param) +{ + struct v4l2_captureparm *capture = ¶m->parm.capture; + struct isp4vid_dev *isp_vdev = video_drvdata(file); + + if (param->type != V4L2_BUF_TYPE_VIDEO_CAPTURE) + return -EINVAL; + + capture->capability = V4L2_CAP_TIMEPERFRAME; + capture->timeperframe = isp_vdev->timeperframe; + capture->readbuffers = 0; + + dev_dbg(isp_vdev->dev, "%s|timeperframe=%d/%d", isp_vdev->vdev.name, + capture->timeperframe.numerator, + capture->timeperframe.denominator); + + return 0; +} + +static const struct v4l2_ioctl_ops isp4vid_vdev_ioctl_ops = { + /* VIDIOC_QUERYCAP handler */ + .vidioc_querycap = isp4vid_ioctl_querycap, + + /* VIDIOC_ENUM_FMT handlers */ + .vidioc_enum_fmt_vid_cap = isp4vid_enum_fmt_vid_cap, + + /* VIDIOC_G_FMT handlers */ + .vidioc_g_fmt_vid_cap = isp4vid_g_fmt_vid_cap, + + /* VIDIOC_S_FMT handlers */ + .vidioc_s_fmt_vid_cap = isp4vid_s_fmt_vid_cap, + + /* VIDIOC_TRY_FMT handlers */ + .vidioc_try_fmt_vid_cap = isp4vid_try_fmt_vid_cap, + + /* Buffer handlers */ + .vidioc_reqbufs = vb2_ioctl_reqbufs, + .vidioc_querybuf = vb2_ioctl_querybuf, + .vidioc_qbuf = vb2_ioctl_qbuf, + .vidioc_expbuf = vb2_ioctl_expbuf, + .vidioc_dqbuf = vb2_ioctl_dqbuf, + .vidioc_create_bufs = vb2_ioctl_create_bufs, + .vidioc_prepare_buf = vb2_ioctl_prepare_buf, + + /* Stream on/off */ + .vidioc_streamon = vb2_ioctl_streamon, + .vidioc_streamoff = vb2_ioctl_streamoff, + + /* Stream type-dependent parameter ioctls */ + .vidioc_g_parm = isp4vid_ioctl_g_param, + .vidioc_s_parm = isp4vid_ioctl_g_param, + + .vidioc_enum_framesizes = isp4vid_enum_framesizes, + .vidioc_enum_frameintervals = isp4vid_ioctl_enum_frameintervals, + +}; + +static unsigned int isp4vid_get_image_size(struct v4l2_pix_format *fmt) +{ + switch (fmt->pixelformat) { + case V4L2_PIX_FMT_NV12: + return fmt->width * fmt->height * 3 / 2; + case V4L2_PIX_FMT_YUYV: + return fmt->width * fmt->height * 2; + default: + return 0; + } +} + +static int isp4vid_qops_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, + unsigned int *nplanes, unsigned int sizes[], + struct device *alloc_devs[]) +{ + struct isp4vid_dev *isp_vdev = vb2_get_drv_priv(vq); + unsigned int q_num_bufs = vb2_get_num_buffers(vq); + + if (*nplanes > 1) { + dev_err(isp_vdev->dev, + "fail to setup queue, no mplane supported %u\n", + *nplanes); + return -EINVAL; + } + + if (*nplanes == 1) { + unsigned int size; + + size = isp4vid_get_image_size(&isp_vdev->format); + if (sizes[0] < size) { + dev_err(isp_vdev->dev, + "fail for small plane size %u, %u expected\n", + sizes[0], size); + return -EINVAL; + } + } + + if (q_num_bufs + *nbuffers < ISP4IF_MAX_STREAM_BUF_COUNT) + *nbuffers = ISP4IF_MAX_STREAM_BUF_COUNT - q_num_bufs; + + switch (isp_vdev->format.pixelformat) { + case V4L2_PIX_FMT_NV12: + case V4L2_PIX_FMT_YUYV: { + *nplanes = 1; + sizes[0] = max(sizes[0], isp_vdev->format.sizeimage); + isp_vdev->format.sizeimage = sizes[0]; + } + break; + default: + dev_err(isp_vdev->dev, "%s|unsupported fmt=%u\n", + isp_vdev->vdev.name, isp_vdev->format.pixelformat); + return -EINVAL; + } + + dev_dbg(isp_vdev->dev, "%s|*nbuffers=%u *nplanes=%u sizes[0]=%u\n", + isp_vdev->vdev.name, + *nbuffers, *nplanes, sizes[0]); + + return 0; +} + +static void isp4vid_qops_buffer_queue(struct vb2_buffer *vb) +{ + struct isp4vid_capture_buffer *buf = + container_of(vb, struct isp4vid_capture_buffer, vb2.vb2_buf); + struct isp4vid_dev *isp_vdev = vb2_get_drv_priv(vb->vb2_queue); + + struct isp4vid_vb2_buf *priv_buf = vb->planes[0].mem_priv; + struct isp4if_img_buf_info *img_buf = &buf->img_buf; + + dev_dbg(isp_vdev->dev, "%s|index=%u", isp_vdev->vdev.name, vb->index); + + dev_dbg(isp_vdev->dev, "queue isp user bo 0x%llx size=%lu", + priv_buf->gpu_addr, + priv_buf->size); + + switch (isp_vdev->format.pixelformat) { + case V4L2_PIX_FMT_NV12: { + u32 y_size = isp_vdev->format.sizeimage / 3 * 2; + u32 uv_size = isp_vdev->format.sizeimage / 3; + + img_buf->planes[0].len = y_size; + img_buf->planes[0].sys_addr = priv_buf->vaddr; + img_buf->planes[0].mc_addr = priv_buf->gpu_addr; + + dev_dbg(isp_vdev->dev, "img_buf[0]: mc=0x%llx size=%u", + img_buf->planes[0].mc_addr, + img_buf->planes[0].len); + + img_buf->planes[1].len = uv_size; + img_buf->planes[1].sys_addr = priv_buf->vaddr + y_size; + img_buf->planes[1].mc_addr = priv_buf->gpu_addr + y_size; + + dev_dbg(isp_vdev->dev, "img_buf[1]: mc=0x%llx size=%u", + img_buf->planes[1].mc_addr, + img_buf->planes[1].len); + + img_buf->planes[2].len = 0; + } + break; + case V4L2_PIX_FMT_YUYV: { + img_buf->planes[0].len = isp_vdev->format.sizeimage; + img_buf->planes[0].sys_addr = priv_buf->vaddr; + img_buf->planes[0].mc_addr = priv_buf->gpu_addr; + + dev_dbg(isp_vdev->dev, "img_buf[0]: mc=0x%llx size=%u", + img_buf->planes[0].mc_addr, + img_buf->planes[0].len); + + img_buf->planes[1].len = 0; + img_buf->planes[2].len = 0; + } + break; + default: + dev_err(isp_vdev->dev, "%s|unsupported fmt=%u", + isp_vdev->vdev.name, isp_vdev->format.pixelformat); + return; + } + + if (isp_vdev->stream_started) + isp_vdev->ops->send_buffer(isp_vdev->isp_sdev, img_buf); + + scoped_guard(mutex, &isp_vdev->buf_list_lock) + list_add_tail(&buf->list, &isp_vdev->buf_list); +} + +static int isp4vid_qops_start_streaming(struct vb2_queue *vq, unsigned int count) +{ + struct isp4vid_dev *isp_vdev = vb2_get_drv_priv(vq); + struct isp4vid_capture_buffer *isp4vid_buf; + struct media_entity *entity; + struct v4l2_subdev *subdev; + struct media_pad *pad; + int ret = 0; + + isp_vdev->sequence = 0; + ret = v4l2_pipeline_pm_get(&isp_vdev->vdev.entity); + if (ret) { + dev_err(isp_vdev->dev, "power up isp fail %d\n", ret); + goto release_buffers; + } + + entity = &isp_vdev->vdev.entity; + while (1) { + pad = &entity->pads[0]; + if (!(pad->flags & MEDIA_PAD_FL_SINK)) + break; + + pad = media_pad_remote_pad_first(pad); + if (!pad || !is_media_entity_v4l2_subdev(pad->entity)) + break; + + entity = pad->entity; + subdev = media_entity_to_v4l2_subdev(entity); + + ret = v4l2_subdev_call(subdev, video, s_stream, 1); + if (ret < 0 && ret != -ENOIOCTLCMD) { + dev_dbg(isp_vdev->dev, "fail start streaming: %s %d\n", + subdev->name, ret); + goto release_buffers; + } + } + + list_for_each_entry(isp4vid_buf, &isp_vdev->buf_list, list) + isp_vdev->ops->send_buffer(isp_vdev->isp_sdev, &isp4vid_buf->img_buf); + + /* Start the media pipeline */ + ret = video_device_pipeline_start(&isp_vdev->vdev, &isp_vdev->pipe); + if (ret) { + dev_err(isp_vdev->dev, "video_device_pipeline_start fail:%d", + ret); + goto release_buffers; + } + + isp_vdev->stream_started = true; + + return 0; + +release_buffers: + isp4vid_capture_return_all_buffers(isp_vdev, VB2_BUF_STATE_QUEUED); + return ret; +} + +static void isp4vid_qops_stop_streaming(struct vb2_queue *vq) +{ + struct isp4vid_dev *isp_vdev = vb2_get_drv_priv(vq); + struct v4l2_subdev *subdev; + struct media_entity *entity; + struct media_pad *pad; + int ret; + + entity = &isp_vdev->vdev.entity; + while (1) { + pad = &entity->pads[0]; + if (!(pad->flags & MEDIA_PAD_FL_SINK)) + break; + + pad = media_pad_remote_pad_first(pad); + if (!pad || !is_media_entity_v4l2_subdev(pad->entity)) + break; + + entity = pad->entity; + subdev = media_entity_to_v4l2_subdev(entity); + + ret = v4l2_subdev_call(subdev, video, s_stream, 0); + + if (ret < 0 && ret != -ENOIOCTLCMD) + dev_dbg(isp_vdev->dev, "fail stop streaming: %s %d\n", + subdev->name, ret); + } + + isp_vdev->stream_started = false; + v4l2_pipeline_pm_put(&isp_vdev->vdev.entity); + + /* Stop the media pipeline */ + video_device_pipeline_stop(&isp_vdev->vdev); + + /* Release all active buffers */ + isp4vid_capture_return_all_buffers(isp_vdev, VB2_BUF_STATE_ERROR); +} + +static const struct vb2_ops isp4vid_qops = { + .queue_setup = isp4vid_qops_queue_setup, + .buf_queue = isp4vid_qops_buffer_queue, + .start_streaming = isp4vid_qops_start_streaming, + .stop_streaming = isp4vid_qops_stop_streaming, + .wait_prepare = vb2_ops_wait_prepare, + .wait_finish = vb2_ops_wait_finish, +}; + +int isp4vid_dev_init(struct isp4vid_dev *isp_vdev, struct v4l2_subdev *isp_sdev, + const struct isp4vid_ops *ops) +{ + const char *vdev_name = isp4vid_video_dev_name; + struct v4l2_device *v4l2_dev; + struct video_device *vdev; + struct vb2_queue *q; + int ret; + + if (!isp_vdev || !isp_sdev || !isp_sdev->v4l2_dev) + return -EINVAL; + + v4l2_dev = isp_sdev->v4l2_dev; + vdev = &isp_vdev->vdev; + + isp_vdev->isp_sdev = isp_sdev; + isp_vdev->dev = v4l2_dev->dev; + isp_vdev->ops = ops; + + /* Initialize the vb2_queue struct */ + mutex_init(&isp_vdev->vbq_lock); + q = &isp_vdev->vbq; + q->type = V4L2_BUF_TYPE_VIDEO_CAPTURE; + q->io_modes = VB2_MMAP | VB2_DMABUF; + q->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC; + q->buf_struct_size = sizeof(struct isp4vid_capture_buffer); + q->min_queued_buffers = 2; + q->ops = &isp4vid_qops; + q->drv_priv = isp_vdev; + q->mem_ops = &isp4vid_vb2_memops; + q->lock = &isp_vdev->vbq_lock; + q->dev = v4l2_dev->dev; + ret = vb2_queue_init(q); + if (ret) { + dev_err(v4l2_dev->dev, "vb2_queue_init error:%d", ret); + return ret; + } + /* Initialize buffer list and its lock */ + mutex_init(&isp_vdev->buf_list_lock); + INIT_LIST_HEAD(&isp_vdev->buf_list); + + /* Set default frame format */ + isp_vdev->format = isp4vid_fmt_default; + isp_vdev->timeperframe = ISP4VID_ISP_TPF_DEFAULT; + v4l2_simplify_fraction(&isp_vdev->timeperframe.numerator, + &isp_vdev->timeperframe.denominator, 8, 333); + + ret = isp4vid_fill_buffer_size(&isp_vdev->format); + if (ret) { + dev_err(v4l2_dev->dev, "fail to fill buffer size: %d\n", ret); + goto err_release_vb2_queue; + } + + ret = isp4vid_set_fmt_2_isp(isp_sdev, &isp_vdev->format); + if (ret) { + dev_err(v4l2_dev->dev, "fail init format :%d\n", ret); + goto err_release_vb2_queue; + } + + /* Initialize the video_device struct */ + isp_vdev->vdev.entity.name = vdev_name; + isp_vdev->vdev.entity.function = MEDIA_ENT_F_IO_V4L; + isp_vdev->vdev_pad.flags = MEDIA_PAD_FL_SINK; + ret = media_entity_pads_init(&isp_vdev->vdev.entity, 1, + &isp_vdev->vdev_pad); + + if (ret) { + dev_err(v4l2_dev->dev, "init media entity pad fail:%d\n", ret); + goto err_release_vb2_queue; + } + + vdev->device_caps = V4L2_CAP_VIDEO_CAPTURE | + V4L2_CAP_STREAMING | V4L2_CAP_IO_MC; + vdev->entity.ops = &isp4vid_vdev_ent_ops; + vdev->release = video_device_release_empty; + vdev->fops = &isp4vid_vdev_fops; + vdev->ioctl_ops = &isp4vid_vdev_ioctl_ops; + vdev->lock = NULL; + vdev->queue = q; + vdev->v4l2_dev = v4l2_dev; + vdev->vfl_dir = VFL_DIR_RX; + strscpy(vdev->name, vdev_name, sizeof(vdev->name)); + video_set_drvdata(vdev, isp_vdev); + + ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1); + if (ret) { + dev_err(v4l2_dev->dev, "register video device fail:%d\n", ret); + goto err_entity_cleanup; + } + + return 0; + +err_entity_cleanup: + media_entity_cleanup(&isp_vdev->vdev.entity); +err_release_vb2_queue: + vb2_queue_release(q); + return ret; +} + +void isp4vid_dev_deinit(struct isp4vid_dev *isp_vdev) +{ + vb2_video_unregister_device(&isp_vdev->vdev); +} diff --git a/drivers/media/platform/amd/isp4/isp4_video.h b/drivers/media/platform/amd/isp4/isp4_video.h new file mode 100644 index 000000000000..b87316d2a2e5 --- /dev/null +++ b/drivers/media/platform/amd/isp4/isp4_video.h @@ -0,0 +1,65 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (C) 2025 Advanced Micro Devices, Inc. + */ + +#ifndef _ISP4_VIDEO_H_ +#define _ISP4_VIDEO_H_ + +#include +#include + +#include "isp4_interface.h" + +struct isp4vid_capture_buffer { + /* + * struct vb2_v4l2_buffer must be the first element + * the videobuf2 framework will allocate this struct based on + * buf_struct_size and use the first sizeof(struct vb2_buffer) bytes of + * memory as a vb2_buffer + */ + struct vb2_v4l2_buffer vb2; + struct isp4if_img_buf_info img_buf; + struct list_head list; +}; + +struct isp4vid_ops { + int (*send_buffer)(struct v4l2_subdev *sd, + struct isp4if_img_buf_info *img_buf); +}; + +struct isp4vid_dev { + struct video_device vdev; + struct media_pad vdev_pad; + struct v4l2_pix_format format; + + /* mutex that protects vbq */ + struct mutex vbq_lock; + struct vb2_queue vbq; + + /* mutex that protects buf_list */ + struct mutex buf_list_lock; + struct list_head buf_list; + + u32 sequence; + bool stream_started; + + struct media_pipeline pipe; + struct device *dev; + struct v4l2_subdev *isp_sdev; + struct v4l2_fract timeperframe; + + /* Callback operations */ + const struct isp4vid_ops *ops; +}; + +int isp4vid_dev_init(struct isp4vid_dev *isp_vdev, + struct v4l2_subdev *isp_sdev, + const struct isp4vid_ops *ops); + +void isp4vid_dev_deinit(struct isp4vid_dev *isp_vdev); + +void isp4vid_handle_frame_done(struct isp4vid_dev *isp_vdev, + const struct isp4if_img_buf_info *img_buf); + +#endif /* _ISP4_VIDEO_H_ */ -- 2.53.0 From 6f38e9eee87cc602425a108b7ff44b023be06f4c Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Mon, 16 Feb 2026 13:50:47 +0100 Subject: [PATCH 2/7] bbr3 Signed-off-by: Peter Jung --- include/linux/tcp.h | 6 +- include/net/inet_connection_sock.h | 4 +- include/net/tcp.h | 72 +- include/net/tcp_ecn.h | 6 +- include/uapi/linux/inet_diag.h | 23 + include/uapi/linux/rtnetlink.h | 4 +- include/uapi/linux/tcp.h | 1 + net/ipv4/Kconfig | 21 +- net/ipv4/bpf_tcp_ca.c | 4 +- net/ipv4/tcp.c | 3 + net/ipv4/tcp_bbr.c | 2233 +++++++++++++++++++++------- net/ipv4/tcp_cong.c | 1 + net/ipv4/tcp_input.c | 40 +- net/ipv4/tcp_minisocks.c | 2 + net/ipv4/tcp_output.c | 42 +- net/ipv4/tcp_rate.c | 30 +- net/ipv4/tcp_timer.c | 1 + 17 files changed, 1939 insertions(+), 554 deletions(-) diff --git a/include/linux/tcp.h b/include/linux/tcp.h index 20b8c6e21fef..e334b7a7aac2 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -236,7 +236,8 @@ struct tcp_sock { tcp_usec_ts : 1, /* TSval values in usec */ is_sack_reneg:1, /* in recovery from loss with SACK reneg? */ is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */ - recvmsg_inq : 1;/* Indicate # of bytes in queue upon recvmsg */ + recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */ + fast_ack_mode:1;/* ack ASAP if >1 rcv_mss received? */ __cacheline_group_end(tcp_sock_read_txrx); /* RX read-mostly hotpath cache lines */ @@ -292,7 +293,8 @@ struct tcp_sock { * 0x5?10 << 16 + snd_wnd in net byte order */ u8 nonagle : 4,/* Disable Nagle algorithm? */ - rate_app_limited:1; /* rate_{delivered,interval_us} limited? */ + rate_app_limited:1, /* rate_{delivered,interval_us} limited? */ + tlp_orig_data_app_limited:1; /* app-limited before TLP rtx? */ u8 received_ce_pending:4, /* Not yet transmit cnt of received_ce */ unused2:4; u8 accecn_minlen:2,/* Minimum length of AccECN option sent */ diff --git a/include/net/inet_connection_sock.h b/include/net/inet_connection_sock.h index ecb362025c4e..9de884b7fe01 100644 --- a/include/net/inet_connection_sock.h +++ b/include/net/inet_connection_sock.h @@ -137,8 +137,8 @@ struct inet_connection_sock { u32 icsk_probes_tstamp; u32 icsk_user_timeout; - u64 icsk_ca_priv[104 / sizeof(u64)]; -#define ICSK_CA_PRIV_SIZE sizeof_field(struct inet_connection_sock, icsk_ca_priv) +#define ICSK_CA_PRIV_SIZE (144) + u64 icsk_ca_priv[ICSK_CA_PRIV_SIZE / sizeof(u64)]; }; #define ICSK_TIME_RETRANS 1 /* Retransmit timer */ diff --git a/include/net/tcp.h b/include/net/tcp.h index 0deb5e9dd911..4e01320786b3 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -406,6 +406,8 @@ static inline void tcp_dec_quickack_mode(struct sock *sk) #define TCP_ECN_DEMAND_CWR BIT(2) #define TCP_ECN_SEEN BIT(3) #define TCP_ECN_MODE_ACCECN BIT(4) +#define TCP_ECN_LOW BIT(5) +#define TCP_ECN_ECT_PERMANENT BIT(6) #define TCP_ECN_DISABLED 0 #define TCP_ECN_MODE_PENDING (TCP_ECN_MODE_RFC3168 | TCP_ECN_MODE_ACCECN) @@ -851,6 +853,15 @@ static inline unsigned long tcp_reqsk_timeout(struct request_sock *req) u32 tcp_delack_max(const struct sock *sk); +static inline void tcp_set_ecn_low_from_dst(struct sock *sk, + const struct dst_entry *dst) +{ + struct tcp_sock *tp = tcp_sk(sk); + + if (dst_feature(dst, RTAX_FEATURE_ECN_LOW)) + tp->ecn_flags |= TCP_ECN_LOW; +} + /* Compute the actual rto_min value */ static inline u32 tcp_rto_min(const struct sock *sk) { @@ -956,6 +967,11 @@ static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) return max_t(s64, t1 - t0, 0); } +static inline u32 tcp_stamp32_us_delta(u32 t1, u32 t0) +{ + return max_t(s32, t1 - t0, 0); +} + /* provide the departure time in us unit */ static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) { @@ -1066,9 +1082,14 @@ struct tcp_skb_cb { /* pkts S/ACKed so far upon tx of skb, incl retrans: */ __u32 delivered; /* start of send pipeline phase */ - u64 first_tx_mstamp; + u32 first_tx_mstamp; /* when we reached the "delivered" count */ - u64 delivered_mstamp; + u32 delivered_mstamp; +#define TCPCB_IN_FLIGHT_BITS 20 +#define TCPCB_IN_FLIGHT_MAX ((1U << TCPCB_IN_FLIGHT_BITS) - 1) + u32 in_flight:20, /* packets in flight at transmit */ + unused2:12; + u32 lost; /* packets lost so far upon tx of skb */ } tx; /* only used for outgoing skbs */ union { struct inet_skb_parm h4; @@ -1181,6 +1202,7 @@ enum tcp_ca_event { CA_EVENT_LOSS, /* loss timeout */ CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ + CA_EVENT_TLP_RECOVERY, /* a lost segment was repaired by TLP probe */ }; /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ @@ -1203,7 +1225,11 @@ enum tcp_ca_ack_event_flags { #define TCP_CONG_NON_RESTRICTED BIT(0) /* Requires ECN/ECT set on all packets */ #define TCP_CONG_NEEDS_ECN BIT(1) -#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) +/* Wants notification of CE events (CA_EVENT_ECN_IS_CE, CA_EVENT_ECN_NO_CE). */ +#define TCP_CONG_WANTS_CE_EVENTS BIT(2) +#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | \ + TCP_CONG_NEEDS_ECN | \ + TCP_CONG_WANTS_CE_EVENTS) union tcp_cc_info; @@ -1223,10 +1249,13 @@ struct ack_sample { */ struct rate_sample { u64 prior_mstamp; /* starting timestamp for interval */ + u32 prior_lost; /* tp->lost at "prior_mstamp" */ u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ + u32 tx_in_flight; /* packets in flight at starting timestamp */ + s32 lost; /* number of packets lost over interval */ s32 delivered; /* number of packets delivered over interval */ - s32 delivered_ce; /* number of packets delivered w/ CE marks*/ + s32 delivered_ce; /* packets delivered w/ CE mark over interval */ long interval_us; /* time for tp->delivered to incr "delivered" */ u32 snd_interval_us; /* snd interval for delivered packets */ u32 rcv_interval_us; /* rcv interval for delivered packets */ @@ -1237,7 +1266,9 @@ struct rate_sample { u32 last_end_seq; /* end_seq of most recently ACKed packet */ bool is_app_limited; /* is sample from packet with bubble in pipe? */ bool is_retrans; /* is sample from retransmission? */ + bool is_acking_tlp_retrans_seq; /* ACKed a TLP retransmit sequence? */ bool is_ack_delayed; /* is this (likely) a delayed ACK? */ + bool is_ece; /* did this ACK have ECN marked? */ }; struct tcp_congestion_ops { @@ -1261,8 +1292,11 @@ struct tcp_congestion_ops { /* hook for packet ack accounting (optional) */ void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); - /* override sysctl_tcp_min_tso_segs */ - u32 (*min_tso_segs)(struct sock *sk); + /* pick target number of segments per TSO/GSO skb (optional): */ + u32 (*tso_segs)(struct sock *sk, unsigned int mss_now); + + /* react to a specific lost skb (optional) */ + void (*skb_marked_lost)(struct sock *sk, const struct sk_buff *skb); /* call when packets are delivered to update cwnd and pacing rate, * after all the ca_state processing. (optional) @@ -1328,6 +1362,14 @@ static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) } #endif +static inline bool tcp_ca_wants_ce_events(const struct sock *sk) +{ + const struct inet_connection_sock *icsk = inet_csk(sk); + + return icsk->icsk_ca_ops->flags & (TCP_CONG_NEEDS_ECN | + TCP_CONG_WANTS_CE_EVENTS); +} + static inline bool tcp_ca_needs_ecn(const struct sock *sk) { const struct inet_connection_sock *icsk = inet_csk(sk); @@ -1347,6 +1389,7 @@ static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) void tcp_set_ca_state(struct sock *sk, const u8 ca_state); /* From tcp_rate.c */ +void tcp_set_tx_in_flight(struct sock *sk, struct sk_buff *skb); void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, struct rate_sample *rs); @@ -1359,6 +1402,21 @@ static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) return t1 > t2 || (t1 == t2 && after(seq1, seq2)); } +/* If a retransmit failed due to local qdisc congestion or other local issues, + * then we may have called tcp_set_skb_tso_segs() to increase the number of + * segments in the skb without increasing the tx.in_flight. In all other cases, + * the tx.in_flight should be at least as big as the pcount of the sk_buff. We + * do not have the state to know whether a retransmit failed due to local qdisc + * congestion or other local issues, so to avoid spurious warnings we consider + * that any skb marked lost may have suffered that fate. + */ +static inline bool tcp_skb_tx_in_flight_is_suspicious(u32 skb_pcount, + u32 skb_sacked_flags, + u32 tx_in_flight) +{ + return (skb_pcount > tx_in_flight) && !(skb_sacked_flags & TCPCB_LOST); +} + /* These functions determine how the current flow behaves in respect of SACK * handling. SACK is negotiated with the peer, and therefore it can vary * between different flows. @@ -2533,7 +2591,7 @@ struct tcp_plb_state { u8 consec_cong_rounds:5, /* consecutive congested rounds */ unused:3; u32 pause_until; /* jiffies32 when PLB can resume rerouting */ -}; +} __attribute__ ((__packed__)); static inline void tcp_plb_init(const struct sock *sk, struct tcp_plb_state *plb) diff --git a/include/net/tcp_ecn.h b/include/net/tcp_ecn.h index f13e5cd2b1ac..bc5de05260eb 100644 --- a/include/net/tcp_ecn.h +++ b/include/net/tcp_ecn.h @@ -583,10 +583,9 @@ static inline void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb) use_ecn = tcp_ecn == TCP_ECN_IN_ECN_OUT_ECN || tcp_ecn == TCP_ECN_IN_ACCECN_OUT_ECN || tcp_ca_needs_ecn(sk) || bpf_needs_ecn || use_accecn; + const struct dst_entry *dst = __sk_dst_get(sk); if (!use_ecn) { - const struct dst_entry *dst = __sk_dst_get(sk); - if (dst && dst_feature(dst, RTAX_FEATURE_ECN)) use_ecn = true; } @@ -604,6 +603,9 @@ static inline void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb) tp->syn_ect_snt = inet_sk(sk)->tos & INET_ECN_MASK; } else { tcp_ecn_mode_set(tp, TCP_ECN_MODE_RFC3168); + + if (dst) + tcp_set_ecn_low_from_dst(sk, dst); } } } diff --git a/include/uapi/linux/inet_diag.h b/include/uapi/linux/inet_diag.h index 86bb2e8b17c9..9d9a3eb2ce9b 100644 --- a/include/uapi/linux/inet_diag.h +++ b/include/uapi/linux/inet_diag.h @@ -229,6 +229,29 @@ struct tcp_bbr_info { __u32 bbr_min_rtt; /* min-filtered RTT in uSec */ __u32 bbr_pacing_gain; /* pacing gain shifted left 8 bits */ __u32 bbr_cwnd_gain; /* cwnd gain shifted left 8 bits */ + __u32 bbr_bw_hi_lsb; /* lower 32 bits of bw_hi */ + __u32 bbr_bw_hi_msb; /* upper 32 bits of bw_hi */ + __u32 bbr_bw_lo_lsb; /* lower 32 bits of bw_lo */ + __u32 bbr_bw_lo_msb; /* upper 32 bits of bw_lo */ + __u8 bbr_mode; /* current bbr_mode in state machine */ + __u8 bbr_phase; /* current state machine phase */ + __u8 unused1; /* alignment padding; not used yet */ + __u8 bbr_version; /* BBR algorithm version */ + __u32 bbr_inflight_lo; /* lower short-term data volume bound */ + __u32 bbr_inflight_hi; /* higher long-term data volume bound */ + __u32 bbr_extra_acked; /* max excess packets ACKed in epoch */ +}; + +/* TCP BBR congestion control bbr_phase as reported in netlink/ss stats. */ +enum tcp_bbr_phase { + BBR_PHASE_INVALID = 0, + BBR_PHASE_STARTUP = 1, + BBR_PHASE_DRAIN = 2, + BBR_PHASE_PROBE_RTT = 3, + BBR_PHASE_PROBE_BW_UP = 4, + BBR_PHASE_PROBE_BW_DOWN = 5, + BBR_PHASE_PROBE_BW_CRUISE = 6, + BBR_PHASE_PROBE_BW_REFILL = 7, }; union tcp_cc_info { diff --git a/include/uapi/linux/rtnetlink.h b/include/uapi/linux/rtnetlink.h index dab9493c791b..cce4975fdcfe 100644 --- a/include/uapi/linux/rtnetlink.h +++ b/include/uapi/linux/rtnetlink.h @@ -517,12 +517,14 @@ enum { #define RTAX_FEATURE_TIMESTAMP (1 << 2) /* unused */ #define RTAX_FEATURE_ALLFRAG (1 << 3) /* unused */ #define RTAX_FEATURE_TCP_USEC_TS (1 << 4) +#define RTAX_FEATURE_ECN_LOW (1 << 5) #define RTAX_FEATURE_MASK (RTAX_FEATURE_ECN | \ RTAX_FEATURE_SACK | \ RTAX_FEATURE_TIMESTAMP | \ RTAX_FEATURE_ALLFRAG | \ - RTAX_FEATURE_TCP_USEC_TS) + RTAX_FEATURE_TCP_USEC_TS | \ + RTAX_FEATURE_ECN_LOW) struct rta_session { __u8 proto; diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h index dce3113787a7..6efba4f74f6f 100644 --- a/include/uapi/linux/tcp.h +++ b/include/uapi/linux/tcp.h @@ -185,6 +185,7 @@ enum tcp_fastopen_client_fail { #define TCPI_OPT_SYN_DATA 32 /* SYN-ACK acked data in SYN sent or rcvd */ #define TCPI_OPT_USEC_TS 64 /* usec timestamps */ #define TCPI_OPT_TFO_CHILD 128 /* child from a Fast Open option on SYN */ +#define TCPI_OPT_ECN_LOW 256 /* Low-latency ECN enabled at conn init */ /* * Sender's congestion state indicating normal or abnormal situations diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig index b71c22475c51..85d95a59708e 100644 --- a/net/ipv4/Kconfig +++ b/net/ipv4/Kconfig @@ -669,15 +669,18 @@ config TCP_CONG_BBR default n help - BBR (Bottleneck Bandwidth and RTT) TCP congestion control aims to - maximize network utilization and minimize queues. It builds an explicit - model of the bottleneck delivery rate and path round-trip propagation - delay. It tolerates packet loss and delay unrelated to congestion. It - can operate over LAN, WAN, cellular, wifi, or cable modem links. It can - coexist with flows that use loss-based congestion control, and can - operate with shallow buffers, deep buffers, bufferbloat, policers, or - AQM schemes that do not provide a delay signal. It requires the fq - ("Fair Queue") pacing packet scheduler. + BBR (Bottleneck Bandwidth and RTT) TCP congestion control is a + model-based congestion control algorithm that aims to maximize + network utilization, keep queues and retransmit rates low, and to be + able to coexist with Reno/CUBIC in common scenarios. It builds an + explicit model of the network path. It tolerates a targeted degree + of random packet loss and delay. It can operate over LAN, WAN, + cellular, wifi, or cable modem links, and can use shallow-threshold + ECN signals. It can coexist to some degree with flows that use + loss-based congestion control, and can operate with shallow buffers, + deep buffers, bufferbloat, policers, or AQM schemes that do not + provide a delay signal. It requires pacing, using either TCP internal + pacing or the fq ("Fair Queue") pacing packet scheduler. choice prompt "Default TCP congestion control" diff --git a/net/ipv4/bpf_tcp_ca.c b/net/ipv4/bpf_tcp_ca.c index e01492234b0b..27893b774e08 100644 --- a/net/ipv4/bpf_tcp_ca.c +++ b/net/ipv4/bpf_tcp_ca.c @@ -280,7 +280,7 @@ static void bpf_tcp_ca_pkts_acked(struct sock *sk, const struct ack_sample *samp { } -static u32 bpf_tcp_ca_min_tso_segs(struct sock *sk) +static u32 bpf_tcp_ca_tso_segs(struct sock *sk, unsigned int mss_now) { return 0; } @@ -315,7 +315,7 @@ static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = { .cwnd_event = bpf_tcp_ca_cwnd_event, .in_ack_event = bpf_tcp_ca_in_ack_event, .pkts_acked = bpf_tcp_ca_pkts_acked, - .min_tso_segs = bpf_tcp_ca_min_tso_segs, + .tso_segs = bpf_tcp_ca_tso_segs, .cong_control = bpf_tcp_ca_cong_control, .undo_cwnd = bpf_tcp_ca_undo_cwnd, .sndbuf_expand = bpf_tcp_ca_sndbuf_expand, diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index d5319ebe2452..4e4c7ea4eb2e 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -3468,6 +3468,7 @@ int tcp_disconnect(struct sock *sk, int flags) tp->rx_opt.dsack = 0; tp->rx_opt.num_sacks = 0; tp->rcv_ooopack = 0; + tp->fast_ack_mode = 0; /* Clean up fastopen related fields */ @@ -4248,6 +4249,8 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info) info->tcpi_options |= TCPI_OPT_ECN; if (tp->ecn_flags & TCP_ECN_SEEN) info->tcpi_options |= TCPI_OPT_ECN_SEEN; + if (tp->ecn_flags & TCP_ECN_LOW) + info->tcpi_options |= TCPI_OPT_ECN_LOW; if (tp->syn_data_acked) info->tcpi_options |= TCPI_OPT_SYN_DATA; if (tp->tcp_usec_ts) diff --git a/net/ipv4/tcp_bbr.c b/net/ipv4/tcp_bbr.c index 760941e55153..9279be755c16 100644 --- a/net/ipv4/tcp_bbr.c +++ b/net/ipv4/tcp_bbr.c @@ -1,18 +1,19 @@ -/* Bottleneck Bandwidth and RTT (BBR) congestion control +/* BBR (Bottleneck Bandwidth and RTT) congestion control * - * BBR congestion control computes the sending rate based on the delivery - * rate (throughput) estimated from ACKs. In a nutshell: + * BBR is a model-based congestion control algorithm that aims for low queues, + * low loss, and (bounded) Reno/CUBIC coexistence. To maintain a model of the + * network path, it uses measurements of bandwidth and RTT, as well as (if they + * occur) packet loss and/or shallow-threshold ECN signals. Note that although + * it can use ECN or loss signals explicitly, it does not require either; it + * can bound its in-flight data based on its estimate of the BDP. * - * On each ACK, update our model of the network path: - * bottleneck_bandwidth = windowed_max(delivered / elapsed, 10 round trips) - * min_rtt = windowed_min(rtt, 10 seconds) - * pacing_rate = pacing_gain * bottleneck_bandwidth - * cwnd = max(cwnd_gain * bottleneck_bandwidth * min_rtt, 4) - * - * The core algorithm does not react directly to packet losses or delays, - * although BBR may adjust the size of next send per ACK when loss is - * observed, or adjust the sending rate if it estimates there is a - * traffic policer, in order to keep the drop rate reasonable. + * The model has both higher and lower bounds for the operating range: + * lo: bw_lo, inflight_lo: conservative short-term lower bound + * hi: bw_hi, inflight_hi: robust long-term upper bound + * The bandwidth-probing time scale is (a) extended dynamically based on + * estimated BDP to improve coexistence with Reno/CUBIC; (b) bounded by + * an interactive wall-clock time-scale to be more scalable and responsive + * than Reno and CUBIC. * * Here is a state transition diagram for BBR: * @@ -65,6 +66,13 @@ #include #include +#include +#include "tcp_dctcp.h" + +#define BBR_VERSION 3 + +#define bbr_param(sk,name) (bbr_ ## name) + /* Scale factor for rate in pkt/uSec unit to avoid truncation in bandwidth * estimation. The rate unit ~= (1500 bytes / 1 usec / 2^24) ~= 715 bps. * This handles bandwidths from 0.06pps (715bps) to 256Mpps (3Tbps) in a u32. @@ -85,36 +93,41 @@ enum bbr_mode { BBR_PROBE_RTT, /* cut inflight to min to probe min_rtt */ }; +/* How does the incoming ACK stream relate to our bandwidth probing? */ +enum bbr_ack_phase { + BBR_ACKS_INIT, /* not probing; not getting probe feedback */ + BBR_ACKS_REFILLING, /* sending at est. bw to fill pipe */ + BBR_ACKS_PROBE_STARTING, /* inflight rising to probe bw */ + BBR_ACKS_PROBE_FEEDBACK, /* getting feedback from bw probing */ + BBR_ACKS_PROBE_STOPPING, /* stopped probing; still getting feedback */ +}; + /* BBR congestion control block */ struct bbr { u32 min_rtt_us; /* min RTT in min_rtt_win_sec window */ u32 min_rtt_stamp; /* timestamp of min_rtt_us */ u32 probe_rtt_done_stamp; /* end time for BBR_PROBE_RTT mode */ - struct minmax bw; /* Max recent delivery rate in pkts/uS << 24 */ - u32 rtt_cnt; /* count of packet-timed rounds elapsed */ + u32 probe_rtt_min_us; /* min RTT in probe_rtt_win_ms win */ + u32 probe_rtt_min_stamp; /* timestamp of probe_rtt_min_us*/ u32 next_rtt_delivered; /* scb->tx.delivered at end of round */ u64 cycle_mstamp; /* time of this cycle phase start */ - u32 mode:3, /* current bbr_mode in state machine */ + u32 mode:2, /* current bbr_mode in state machine */ prev_ca_state:3, /* CA state on previous ACK */ - packet_conservation:1, /* use packet conservation? */ round_start:1, /* start of packet-timed tx->ack round? */ + ce_state:1, /* If most recent data has CE bit set */ + bw_probe_up_rounds:5, /* cwnd-limited rounds in PROBE_UP */ + try_fast_path:1, /* can we take fast path? */ idle_restart:1, /* restarting after idle? */ probe_rtt_round_done:1, /* a BBR_PROBE_RTT round at 4 pkts? */ - unused:13, - lt_is_sampling:1, /* taking long-term ("LT") samples now? */ - lt_rtt_cnt:7, /* round trips in long-term interval */ - lt_use_bw:1; /* use lt_bw as our bw estimate? */ - u32 lt_bw; /* LT est delivery rate in pkts/uS << 24 */ - u32 lt_last_delivered; /* LT intvl start: tp->delivered */ - u32 lt_last_stamp; /* LT intvl start: tp->delivered_mstamp */ - u32 lt_last_lost; /* LT intvl start: tp->lost */ + init_cwnd:7, /* initial cwnd */ + unused_1:10; u32 pacing_gain:10, /* current gain for setting pacing rate */ cwnd_gain:10, /* current gain for setting cwnd */ full_bw_reached:1, /* reached full bw in Startup? */ full_bw_cnt:2, /* number of rounds without large bw gains */ - cycle_idx:3, /* current index in pacing_gain cycle array */ + cycle_idx:2, /* current index in pacing_gain cycle array */ has_seen_rtt:1, /* have we seen an RTT sample yet? */ - unused_b:5; + unused_2:6; u32 prior_cwnd; /* prior cwnd upon entering loss recovery */ u32 full_bw; /* recent bw, to estimate if pipe is full */ @@ -124,19 +137,67 @@ struct bbr { u32 ack_epoch_acked:20, /* packets (S)ACKed in sampling epoch */ extra_acked_win_rtts:5, /* age of extra_acked, in round trips */ extra_acked_win_idx:1, /* current index in extra_acked array */ - unused_c:6; + /* BBR v3 state: */ + full_bw_now:1, /* recently reached full bw plateau? */ + startup_ecn_rounds:2, /* consecutive hi ECN STARTUP rounds */ + loss_in_cycle:1, /* packet loss in this cycle? */ + ecn_in_cycle:1, /* ECN in this cycle? */ + unused_3:1; + u32 loss_round_delivered; /* scb->tx.delivered ending loss round */ + u32 undo_bw_lo; /* bw_lo before latest losses */ + u32 undo_inflight_lo; /* inflight_lo before latest losses */ + u32 undo_inflight_hi; /* inflight_hi before latest losses */ + u32 bw_latest; /* max delivered bw in last round trip */ + u32 bw_lo; /* lower bound on sending bandwidth */ + u32 bw_hi[2]; /* max recent measured bw sample */ + u32 inflight_latest; /* max delivered data in last round trip */ + u32 inflight_lo; /* lower bound of inflight data range */ + u32 inflight_hi; /* upper bound of inflight data range */ + u32 bw_probe_up_cnt; /* packets delivered per inflight_hi incr */ + u32 bw_probe_up_acks; /* packets (S)ACKed since inflight_hi incr */ + u32 probe_wait_us; /* PROBE_DOWN until next clock-driven probe */ + u32 prior_rcv_nxt; /* tp->rcv_nxt when CE state last changed */ + u32 ecn_eligible:1, /* sender can use ECN (RTT, handshake)? */ + ecn_alpha:9, /* EWMA delivered_ce/delivered; 0..256 */ + bw_probe_samples:1, /* rate samples reflect bw probing? */ + prev_probe_too_high:1, /* did last PROBE_UP go too high? */ + stopped_risky_probe:1, /* last PROBE_UP stopped due to risk? */ + rounds_since_probe:8, /* packet-timed rounds since probed bw */ + loss_round_start:1, /* loss_round_delivered round trip? */ + loss_in_round:1, /* loss marked in this round trip? */ + ecn_in_round:1, /* ECN marked in this round trip? */ + ack_phase:3, /* bbr_ack_phase: meaning of ACKs */ + loss_events_in_round:4,/* losses in STARTUP round */ + initialized:1; /* has bbr_init() been called? */ + u32 alpha_last_delivered; /* tp->delivered at alpha update */ + u32 alpha_last_delivered_ce; /* tp->delivered_ce at alpha update */ + + u8 unused_4; /* to preserve alignment */ + struct tcp_plb_state plb; }; -#define CYCLE_LEN 8 /* number of phases in a pacing gain cycle */ +struct bbr_context { + u32 sample_bw; +}; -/* Window length of bw filter (in rounds): */ -static const int bbr_bw_rtts = CYCLE_LEN + 2; /* Window length of min_rtt filter (in sec): */ static const u32 bbr_min_rtt_win_sec = 10; /* Minimum time (in ms) spent at bbr_cwnd_min_target in BBR_PROBE_RTT mode: */ static const u32 bbr_probe_rtt_mode_ms = 200; -/* Skip TSO below the following bandwidth (bits/sec): */ -static const int bbr_min_tso_rate = 1200000; +/* Window length of probe_rtt_min_us filter (in ms), and consequently the + * typical interval between PROBE_RTT mode entries. The default is 5000ms. + * Note that bbr_probe_rtt_win_ms must be <= bbr_min_rtt_win_sec * MSEC_PER_SEC + */ +static const u32 bbr_probe_rtt_win_ms = 5000; +/* Proportion of cwnd to estimated BDP in PROBE_RTT, in units of BBR_UNIT: */ +static const u32 bbr_probe_rtt_cwnd_gain = BBR_UNIT * 1 / 2; + +/* Use min_rtt to help adapt TSO burst size, with smaller min_rtt resulting + * in bigger TSO bursts. We cut the RTT-based allowance in half + * for every 2^9 usec (aka 512 us) of RTT, so that the RTT-based allowance + * is below 1500 bytes after 6 * ~500 usec = 3ms. + */ +static const u32 bbr_tso_rtt_shift = 9; /* Pace at ~1% below estimated bw, on average, to reduce queue at bottleneck. * In order to help drive the network toward lower queues and low latency while @@ -146,13 +207,15 @@ static const int bbr_min_tso_rate = 1200000; */ static const int bbr_pacing_margin_percent = 1; -/* We use a high_gain value of 2/ln(2) because it's the smallest pacing gain +/* We use a startup_pacing_gain of 4*ln(2) because it's the smallest value * that will allow a smoothly increasing pacing rate that will double each RTT * and send the same number of packets per RTT that an un-paced, slow-starting * Reno or CUBIC flow would: */ -static const int bbr_high_gain = BBR_UNIT * 2885 / 1000 + 1; -/* The pacing gain of 1/high_gain in BBR_DRAIN is calculated to typically drain +static const int bbr_startup_pacing_gain = BBR_UNIT * 277 / 100 + 1; +/* The gain for deriving startup cwnd: */ +static const int bbr_startup_cwnd_gain = BBR_UNIT * 2; +/* The pacing gain in BBR_DRAIN is calculated to typically drain * the queue created in BBR_STARTUP in a single round: */ static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885; @@ -160,13 +223,17 @@ static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885; static const int bbr_cwnd_gain = BBR_UNIT * 2; /* The pacing_gain values for the PROBE_BW gain cycle, to discover/share bw: */ static const int bbr_pacing_gain[] = { - BBR_UNIT * 5 / 4, /* probe for more available bw */ - BBR_UNIT * 3 / 4, /* drain queue and/or yield bw to other flows */ - BBR_UNIT, BBR_UNIT, BBR_UNIT, /* cruise at 1.0*bw to utilize pipe, */ - BBR_UNIT, BBR_UNIT, BBR_UNIT /* without creating excess queue... */ + BBR_UNIT * 5 / 4, /* UP: probe for more available bw */ + BBR_UNIT * 91 / 100, /* DOWN: drain queue and/or yield bw */ + BBR_UNIT, /* CRUISE: try to use pipe w/ some headroom */ + BBR_UNIT, /* REFILL: refill pipe to estimated 100% */ +}; +enum bbr_pacing_gain_phase { + BBR_BW_PROBE_UP = 0, /* push up inflight to probe for bw/vol */ + BBR_BW_PROBE_DOWN = 1, /* drain excess inflight from the queue */ + BBR_BW_PROBE_CRUISE = 2, /* use pipe, w/ headroom in queue/pipe */ + BBR_BW_PROBE_REFILL = 3, /* refill the pipe again to 100% */ }; -/* Randomize the starting gain cycling phase over N phases: */ -static const u32 bbr_cycle_rand = 7; /* Try to keep at least this many packets in flight, if things go smoothly. For * smooth functioning, a sliding window protocol ACKing every other packet @@ -174,24 +241,12 @@ static const u32 bbr_cycle_rand = 7; */ static const u32 bbr_cwnd_min_target = 4; -/* To estimate if BBR_STARTUP mode (i.e. high_gain) has filled pipe... */ +/* To estimate if BBR_STARTUP or BBR_BW_PROBE_UP has filled pipe... */ /* If bw has increased significantly (1.25x), there may be more bw available: */ static const u32 bbr_full_bw_thresh = BBR_UNIT * 5 / 4; /* But after 3 rounds w/o significant bw growth, estimate pipe is full: */ static const u32 bbr_full_bw_cnt = 3; -/* "long-term" ("LT") bandwidth estimator parameters... */ -/* The minimum number of rounds in an LT bw sampling interval: */ -static const u32 bbr_lt_intvl_min_rtts = 4; -/* If lost/delivered ratio > 20%, interval is "lossy" and we may be policed: */ -static const u32 bbr_lt_loss_thresh = 50; -/* If 2 intervals have a bw ratio <= 1/8, their bw is "consistent": */ -static const u32 bbr_lt_bw_ratio = BBR_UNIT / 8; -/* If 2 intervals have a bw diff <= 4 Kbit/sec their bw is "consistent": */ -static const u32 bbr_lt_bw_diff = 4000 / 8; -/* If we estimate we're policed, use lt_bw for this many round trips: */ -static const u32 bbr_lt_bw_max_rtts = 48; - /* Gain factor for adding extra_acked to target cwnd: */ static const int bbr_extra_acked_gain = BBR_UNIT; /* Window length of extra_acked window. */ @@ -201,8 +256,123 @@ static const u32 bbr_ack_epoch_acked_reset_thresh = 1U << 20; /* Time period for clamping cwnd increment due to ack aggregation */ static const u32 bbr_extra_acked_max_us = 100 * 1000; +/* Flags to control BBR ECN-related behavior... */ + +/* Ensure ACKs only ACK packets with consistent ECN CE status? */ +static const bool bbr_precise_ece_ack = true; + +/* Max RTT (in usec) at which to use sender-side ECN logic. + * Disabled when 0 (ECN allowed at any RTT). + */ +static const u32 bbr_ecn_max_rtt_us = 5000; + +/* On losses, scale down inflight and pacing rate by beta scaled by BBR_SCALE. + * No loss response when 0. + */ +static const u32 bbr_beta = BBR_UNIT * 30 / 100; + +/* Gain factor for ECN mark ratio samples, scaled by BBR_SCALE (1/16 = 6.25%) */ +static const u32 bbr_ecn_alpha_gain = BBR_UNIT * 1 / 16; + +/* The initial value for ecn_alpha; 1.0 allows a flow to respond quickly + * to congestion if the bottleneck is congested when the flow starts up. + */ +static const u32 bbr_ecn_alpha_init = BBR_UNIT; + +/* On ECN, cut inflight_lo to (1 - ecn_factor * ecn_alpha) scaled by BBR_SCALE. + * No ECN based bounding when 0. + */ +static const u32 bbr_ecn_factor = BBR_UNIT * 1 / 3; /* 1/3 = 33% */ + +/* Estimate bw probing has gone too far if CE ratio exceeds this threshold. + * Scaled by BBR_SCALE. Disabled when 0. + */ +static const u32 bbr_ecn_thresh = BBR_UNIT * 1 / 2; /* 1/2 = 50% */ + +/* If non-zero, if in a cycle with no losses but some ECN marks, after ECN + * clears then make the first round's increment to inflight_hi the following + * fraction of inflight_hi. + */ +static const u32 bbr_ecn_reprobe_gain = BBR_UNIT * 1 / 2; + +/* Estimate bw probing has gone too far if loss rate exceeds this level. */ +static const u32 bbr_loss_thresh = BBR_UNIT * 2 / 100; /* 2% loss */ + +/* Slow down for a packet loss recovered by TLP? */ +static const bool bbr_loss_probe_recovery = true; + +/* Exit STARTUP if number of loss marking events in a Recovery round is >= N, + * and loss rate is higher than bbr_loss_thresh. + * Disabled if 0. + */ +static const u32 bbr_full_loss_cnt = 6; + +/* Exit STARTUP if number of round trips with ECN mark rate above ecn_thresh + * meets this count. + */ +static const u32 bbr_full_ecn_cnt = 2; + +/* Fraction of unutilized headroom to try to leave in path upon high loss. */ +static const u32 bbr_inflight_headroom = BBR_UNIT * 15 / 100; + +/* How much do we increase cwnd_gain when probing for bandwidth in + * BBR_BW_PROBE_UP? This specifies the increment in units of + * BBR_UNIT/4. The default is 1, meaning 0.25. + * The min value is 0 (meaning 0.0); max is 3 (meaning 0.75). + */ +static const u32 bbr_bw_probe_cwnd_gain = 1; + +/* Max number of packet-timed rounds to wait before probing for bandwidth. If + * we want to tolerate 1% random loss per round, and not have this cut our + * inflight too much, we must probe for bw periodically on roughly this scale. + * If low, limits Reno/CUBIC coexistence; if high, limits loss tolerance. + * We aim to be fair with Reno/CUBIC up to a BDP of at least: + * BDP = 25Mbps * .030sec /(1514bytes) = 61.9 packets + */ +static const u32 bbr_bw_probe_max_rounds = 63; + +/* Max amount of randomness to inject in round counting for Reno-coexistence. + */ +static const u32 bbr_bw_probe_rand_rounds = 2; + +/* Use BBR-native probe time scale starting at this many usec. + * We aim to be fair with Reno/CUBIC up to an inter-loss time epoch of at least: + * BDP*RTT = 25Mbps * .030sec /(1514bytes) * 0.030sec = 1.9 secs + */ +static const u32 bbr_bw_probe_base_us = 2 * USEC_PER_SEC; /* 2 secs */ + +/* Use BBR-native probes spread over this many usec: */ +static const u32 bbr_bw_probe_rand_us = 1 * USEC_PER_SEC; /* 1 secs */ + +/* Use fast path if app-limited, no loss/ECN, and target cwnd was reached? */ +static const bool bbr_fast_path = true; + +/* Use fast ack mode? */ +static const bool bbr_fast_ack_mode = true; + +static u32 bbr_max_bw(const struct sock *sk); +static u32 bbr_bw(const struct sock *sk); +static void bbr_exit_probe_rtt(struct sock *sk); +static void bbr_reset_congestion_signals(struct sock *sk); +static void bbr_run_loss_probe_recovery(struct sock *sk); + static void bbr_check_probe_rtt_done(struct sock *sk); +/* This connection can use ECN if both endpoints have signaled ECN support in + * the handshake and the per-route settings indicated this is a + * shallow-threshold ECN environment, meaning both: + * (a) ECN CE marks indicate low-latency/shallow-threshold congestion, and + * (b) TCP endpoints provide precise ACKs that only ACK data segments + * with consistent ECN CE status + */ +static bool bbr_can_use_ecn(const struct sock *sk) +{ + const struct tcp_sock *tp = tcp_sk(sk); + + return tcp_ecn_mode_any(tp) && + (tcp_sk(sk)->ecn_flags & TCP_ECN_LOW); +} + /* Do we estimate that STARTUP filled the pipe? */ static bool bbr_full_bw_reached(const struct sock *sk) { @@ -214,17 +384,17 @@ static bool bbr_full_bw_reached(const struct sock *sk) /* Return the windowed max recent bandwidth sample, in pkts/uS << BW_SCALE. */ static u32 bbr_max_bw(const struct sock *sk) { - struct bbr *bbr = inet_csk_ca(sk); + const struct bbr *bbr = inet_csk_ca(sk); - return minmax_get(&bbr->bw); + return max(bbr->bw_hi[0], bbr->bw_hi[1]); } /* Return the estimated bandwidth of the path, in pkts/uS << BW_SCALE. */ static u32 bbr_bw(const struct sock *sk) { - struct bbr *bbr = inet_csk_ca(sk); + const struct bbr *bbr = inet_csk_ca(sk); - return bbr->lt_use_bw ? bbr->lt_bw : bbr_max_bw(sk); + return min(bbr_max_bw(sk), bbr->bw_lo); } /* Return maximum extra acked in past k-2k round trips, @@ -241,15 +411,23 @@ static u16 bbr_extra_acked(const struct sock *sk) * The order here is chosen carefully to avoid overflow of u64. This should * work for input rates of up to 2.9Tbit/sec and gain of 2.89x. */ -static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain) +static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain, + int margin) { unsigned int mss = tcp_sk(sk)->mss_cache; rate *= mss; rate *= gain; rate >>= BBR_SCALE; - rate *= USEC_PER_SEC / 100 * (100 - bbr_pacing_margin_percent); - return rate >> BW_SCALE; + rate *= USEC_PER_SEC / 100 * (100 - margin); + rate >>= BW_SCALE; + rate = max(rate, 1ULL); + return rate; +} + +static u64 bbr_bw_bytes_per_sec(struct sock *sk, u64 rate) +{ + return bbr_rate_bytes_per_sec(sk, rate, BBR_UNIT, 0); } /* Convert a BBR bw and gain factor to a pacing rate in bytes per second. */ @@ -257,12 +435,13 @@ static unsigned long bbr_bw_to_pacing_rate(struct sock *sk, u32 bw, int gain) { u64 rate = bw; - rate = bbr_rate_bytes_per_sec(sk, rate, gain); + rate = bbr_rate_bytes_per_sec(sk, rate, gain, + bbr_pacing_margin_percent); rate = min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate)); return rate; } -/* Initialize pacing rate to: high_gain * init_cwnd / RTT. */ +/* Initialize pacing rate to: startup_pacing_gain * init_cwnd / RTT. */ static void bbr_init_pacing_rate_from_rtt(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); @@ -279,7 +458,8 @@ static void bbr_init_pacing_rate_from_rtt(struct sock *sk) bw = (u64)tcp_snd_cwnd(tp) * BW_UNIT; do_div(bw, rtt_us); WRITE_ONCE(sk->sk_pacing_rate, - bbr_bw_to_pacing_rate(sk, bw, bbr_high_gain)); + bbr_bw_to_pacing_rate(sk, bw, + bbr_param(sk, startup_pacing_gain))); } /* Pace using current bw estimate and a gain factor. */ @@ -295,26 +475,48 @@ static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain) WRITE_ONCE(sk->sk_pacing_rate, rate); } -/* override sysctl_tcp_min_tso_segs */ -__bpf_kfunc static u32 bbr_min_tso_segs(struct sock *sk) +/* Return the number of segments BBR would like in a TSO/GSO skb, given a + * particular max gso size as a constraint. TODO: make this simpler and more + * consistent by switching bbr to just call tcp_tso_autosize(). + */ +static u32 bbr_tso_segs_generic(struct sock *sk, unsigned int mss_now, + u32 gso_max_size) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 segs, r; + u64 bytes; + + /* Budget a TSO/GSO burst size allowance based on bw (pacing_rate). */ + bytes = READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift); + + /* Budget a TSO/GSO burst size allowance based on min_rtt. For every + * K = 2^tso_rtt_shift microseconds of min_rtt, halve the burst. + * The min_rtt-based burst allowance is: 64 KBytes / 2^(min_rtt/K) + */ + if (bbr_param(sk, tso_rtt_shift)) { + r = bbr->min_rtt_us >> bbr_param(sk, tso_rtt_shift); + if (r < BITS_PER_TYPE(u32)) /* prevent undefined behavior */ + bytes += GSO_LEGACY_MAX_SIZE >> r; + } + + bytes = min_t(u32, bytes, gso_max_size - 1 - MAX_TCP_HEADER); + segs = max_t(u32, bytes / mss_now, + sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); + return segs; +} + +/* Custom tcp_tso_autosize() for BBR, used at transmit time to cap skb size. */ +__bpf_kfunc static u32 bbr_tso_segs(struct sock *sk, unsigned int mss_now) { - return READ_ONCE(sk->sk_pacing_rate) < (bbr_min_tso_rate >> 3) ? 1 : 2; + return bbr_tso_segs_generic(sk, mss_now, sk->sk_gso_max_size); } +/* Like bbr_tso_segs(), using mss_cache, ignoring driver's sk_gso_max_size. */ static u32 bbr_tso_segs_goal(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); - u32 segs, bytes; - - /* Sort of tcp_tso_autosize() but ignoring - * driver provided sk_gso_max_size. - */ - bytes = min_t(unsigned long, - READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift), - GSO_LEGACY_MAX_SIZE - 1 - MAX_TCP_HEADER); - segs = max_t(u32, bytes / tp->mss_cache, bbr_min_tso_segs(sk)); - return min(segs, 0x7FU); + return bbr_tso_segs_generic(sk, tp->mss_cache, GSO_LEGACY_MAX_SIZE); } /* Save "last known good" cwnd so we can restore it after losses or PROBE_RTT */ @@ -334,7 +536,9 @@ __bpf_kfunc static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event) struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); - if (event == CA_EVENT_TX_START && tp->app_limited) { + if (event == CA_EVENT_TX_START) { + if (!tp->app_limited) + return; bbr->idle_restart = 1; bbr->ack_epoch_mstamp = tp->tcp_mstamp; bbr->ack_epoch_acked = 0; @@ -345,6 +549,16 @@ __bpf_kfunc static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event) bbr_set_pacing_rate(sk, bbr_bw(sk), BBR_UNIT); else if (bbr->mode == BBR_PROBE_RTT) bbr_check_probe_rtt_done(sk); + } else if ((event == CA_EVENT_ECN_IS_CE || + event == CA_EVENT_ECN_NO_CE) && + bbr_can_use_ecn(sk) && + bbr_param(sk, precise_ece_ack)) { + u32 state = bbr->ce_state; + dctcp_ece_ack_update(sk, event, &bbr->prior_rcv_nxt, &state); + bbr->ce_state = state; + } else if (event == CA_EVENT_TLP_RECOVERY && + bbr_param(sk, loss_probe_recovery)) { + bbr_run_loss_probe_recovery(sk); } } @@ -367,10 +581,10 @@ static u32 bbr_bdp(struct sock *sk, u32 bw, int gain) * default. This should only happen when the connection is not using TCP * timestamps and has retransmitted all of the SYN/SYNACK/data packets * ACKed so far. In this case, an RTO can cut cwnd to 1, in which - * case we need to slow-start up toward something safe: TCP_INIT_CWND. + * case we need to slow-start up toward something safe: initial cwnd. */ if (unlikely(bbr->min_rtt_us == ~0U)) /* no valid RTT samples yet? */ - return TCP_INIT_CWND; /* be safe: cap at default initial cwnd*/ + return bbr->init_cwnd; /* be safe: cap at initial cwnd */ w = (u64)bw * bbr->min_rtt_us; @@ -387,23 +601,23 @@ static u32 bbr_bdp(struct sock *sk, u32 bw, int gain) * - one skb in sending host Qdisc, * - one skb in sending host TSO/GSO engine * - one skb being received by receiver host LRO/GRO/delayed-ACK engine - * Don't worry, at low rates (bbr_min_tso_rate) this won't bloat cwnd because - * in such cases tso_segs_goal is 1. The minimum cwnd is 4 packets, + * Don't worry, at low rates this won't bloat cwnd because + * in such cases tso_segs_goal is small. The minimum cwnd is 4 packets, * which allows 2 outstanding 2-packet sequences, to try to keep pipe * full even with ACK-every-other-packet delayed ACKs. */ static u32 bbr_quantization_budget(struct sock *sk, u32 cwnd) { struct bbr *bbr = inet_csk_ca(sk); + u32 tso_segs_goal; - /* Allow enough full-sized skbs in flight to utilize end systems. */ - cwnd += 3 * bbr_tso_segs_goal(sk); - - /* Reduce delayed ACKs by rounding up cwnd to the next even number. */ - cwnd = (cwnd + 1) & ~1U; + tso_segs_goal = 3 * bbr_tso_segs_goal(sk); + /* Allow enough full-sized skbs in flight to utilize end systems. */ + cwnd = max_t(u32, cwnd, tso_segs_goal); + cwnd = max_t(u32, cwnd, bbr_param(sk, cwnd_min_target)); /* Ensure gain cycling gets inflight above BDP even for small BDPs. */ - if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == 0) + if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == BBR_BW_PROBE_UP) cwnd += 2; return cwnd; @@ -458,10 +672,10 @@ static u32 bbr_ack_aggregation_cwnd(struct sock *sk) { u32 max_aggr_cwnd, aggr_cwnd = 0; - if (bbr_extra_acked_gain && bbr_full_bw_reached(sk)) { + if (bbr_param(sk, extra_acked_gain)) { max_aggr_cwnd = ((u64)bbr_bw(sk) * bbr_extra_acked_max_us) / BW_UNIT; - aggr_cwnd = (bbr_extra_acked_gain * bbr_extra_acked(sk)) + aggr_cwnd = (bbr_param(sk, extra_acked_gain) * bbr_extra_acked(sk)) >> BBR_SCALE; aggr_cwnd = min(aggr_cwnd, max_aggr_cwnd); } @@ -469,66 +683,27 @@ static u32 bbr_ack_aggregation_cwnd(struct sock *sk) return aggr_cwnd; } -/* An optimization in BBR to reduce losses: On the first round of recovery, we - * follow the packet conservation principle: send P packets per P packets acked. - * After that, we slow-start and send at most 2*P packets per P packets acked. - * After recovery finishes, or upon undo, we restore the cwnd we had when - * recovery started (capped by the target cwnd based on estimated BDP). - * - * TODO(ycheng/ncardwell): implement a rate-based approach. - */ -static bool bbr_set_cwnd_to_recover_or_restore( - struct sock *sk, const struct rate_sample *rs, u32 acked, u32 *new_cwnd) +/* Returns the cwnd for PROBE_RTT mode. */ +static u32 bbr_probe_rtt_cwnd(struct sock *sk) { - struct tcp_sock *tp = tcp_sk(sk); - struct bbr *bbr = inet_csk_ca(sk); - u8 prev_state = bbr->prev_ca_state, state = inet_csk(sk)->icsk_ca_state; - u32 cwnd = tcp_snd_cwnd(tp); - - /* An ACK for P pkts should release at most 2*P packets. We do this - * in two steps. First, here we deduct the number of lost packets. - * Then, in bbr_set_cwnd() we slow start up toward the target cwnd. - */ - if (rs->losses > 0) - cwnd = max_t(s32, cwnd - rs->losses, 1); - - if (state == TCP_CA_Recovery && prev_state != TCP_CA_Recovery) { - /* Starting 1st round of Recovery, so do packet conservation. */ - bbr->packet_conservation = 1; - bbr->next_rtt_delivered = tp->delivered; /* start round now */ - /* Cut unused cwnd from app behavior, TSQ, or TSO deferral: */ - cwnd = tcp_packets_in_flight(tp) + acked; - } else if (prev_state >= TCP_CA_Recovery && state < TCP_CA_Recovery) { - /* Exiting loss recovery; restore cwnd saved before recovery. */ - cwnd = max(cwnd, bbr->prior_cwnd); - bbr->packet_conservation = 0; - } - bbr->prev_ca_state = state; - - if (bbr->packet_conservation) { - *new_cwnd = max(cwnd, tcp_packets_in_flight(tp) + acked); - return true; /* yes, using packet conservation */ - } - *new_cwnd = cwnd; - return false; + return max_t(u32, bbr_param(sk, cwnd_min_target), + bbr_bdp(sk, bbr_bw(sk), bbr_param(sk, probe_rtt_cwnd_gain))); } /* Slow-start up toward target cwnd (if bw estimate is growing, or packet loss * has drawn us down below target), or snap down to target if we're above it. */ static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs, - u32 acked, u32 bw, int gain) + u32 acked, u32 bw, int gain, u32 cwnd, + struct bbr_context *ctx) { struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); - u32 cwnd = tcp_snd_cwnd(tp), target_cwnd = 0; + u32 target_cwnd = 0; if (!acked) goto done; /* no packet fully ACKed; just apply caps */ - if (bbr_set_cwnd_to_recover_or_restore(sk, rs, acked, &cwnd)) - goto done; - target_cwnd = bbr_bdp(sk, bw, gain); /* Increment the cwnd to account for excess ACKed data that seems @@ -537,74 +712,26 @@ static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs, target_cwnd += bbr_ack_aggregation_cwnd(sk); target_cwnd = bbr_quantization_budget(sk, target_cwnd); - /* If we're below target cwnd, slow start cwnd toward target cwnd. */ - if (bbr_full_bw_reached(sk)) /* only cut cwnd if we filled the pipe */ - cwnd = min(cwnd + acked, target_cwnd); - else if (cwnd < target_cwnd || tp->delivered < TCP_INIT_CWND) - cwnd = cwnd + acked; - cwnd = max(cwnd, bbr_cwnd_min_target); + /* Update cwnd and enable fast path if cwnd reaches target_cwnd. */ + bbr->try_fast_path = 0; + if (bbr_full_bw_reached(sk)) { /* only cut cwnd if we filled the pipe */ + cwnd += acked; + if (cwnd >= target_cwnd) { + cwnd = target_cwnd; + bbr->try_fast_path = 1; + } + } else if (cwnd < target_cwnd || cwnd < 2 * bbr->init_cwnd) { + cwnd += acked; + } else { + bbr->try_fast_path = 1; + } + cwnd = max_t(u32, cwnd, bbr_param(sk, cwnd_min_target)); done: - tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp)); /* apply global cap */ + tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp)); /* global cap */ if (bbr->mode == BBR_PROBE_RTT) /* drain queue, refresh min_rtt */ - tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp), bbr_cwnd_min_target)); -} - -/* End cycle phase if it's time and/or we hit the phase's in-flight target. */ -static bool bbr_is_next_cycle_phase(struct sock *sk, - const struct rate_sample *rs) -{ - struct tcp_sock *tp = tcp_sk(sk); - struct bbr *bbr = inet_csk_ca(sk); - bool is_full_length = - tcp_stamp_us_delta(tp->delivered_mstamp, bbr->cycle_mstamp) > - bbr->min_rtt_us; - u32 inflight, bw; - - /* The pacing_gain of 1.0 paces at the estimated bw to try to fully - * use the pipe without increasing the queue. - */ - if (bbr->pacing_gain == BBR_UNIT) - return is_full_length; /* just use wall clock time */ - - inflight = bbr_packets_in_net_at_edt(sk, rs->prior_in_flight); - bw = bbr_max_bw(sk); - - /* A pacing_gain > 1.0 probes for bw by trying to raise inflight to at - * least pacing_gain*BDP; this may take more than min_rtt if min_rtt is - * small (e.g. on a LAN). We do not persist if packets are lost, since - * a path with small buffers may not hold that much. - */ - if (bbr->pacing_gain > BBR_UNIT) - return is_full_length && - (rs->losses || /* perhaps pacing_gain*BDP won't fit */ - inflight >= bbr_inflight(sk, bw, bbr->pacing_gain)); - - /* A pacing_gain < 1.0 tries to drain extra queue we added if bw - * probing didn't find more bw. If inflight falls to match BDP then we - * estimate queue is drained; persisting would underutilize the pipe. - */ - return is_full_length || - inflight <= bbr_inflight(sk, bw, BBR_UNIT); -} - -static void bbr_advance_cycle_phase(struct sock *sk) -{ - struct tcp_sock *tp = tcp_sk(sk); - struct bbr *bbr = inet_csk_ca(sk); - - bbr->cycle_idx = (bbr->cycle_idx + 1) & (CYCLE_LEN - 1); - bbr->cycle_mstamp = tp->delivered_mstamp; -} - -/* Gain cycling: cycle pacing gain to converge to fair share of available bw. */ -static void bbr_update_cycle_phase(struct sock *sk, - const struct rate_sample *rs) -{ - struct bbr *bbr = inet_csk_ca(sk); - - if (bbr->mode == BBR_PROBE_BW && bbr_is_next_cycle_phase(sk, rs)) - bbr_advance_cycle_phase(sk); + tcp_snd_cwnd_set(tp, min_t(u32, tcp_snd_cwnd(tp), + bbr_probe_rtt_cwnd(sk))); } static void bbr_reset_startup_mode(struct sock *sk) @@ -614,191 +741,49 @@ static void bbr_reset_startup_mode(struct sock *sk) bbr->mode = BBR_STARTUP; } -static void bbr_reset_probe_bw_mode(struct sock *sk) -{ - struct bbr *bbr = inet_csk_ca(sk); - - bbr->mode = BBR_PROBE_BW; - bbr->cycle_idx = CYCLE_LEN - 1 - get_random_u32_below(bbr_cycle_rand); - bbr_advance_cycle_phase(sk); /* flip to next phase of gain cycle */ -} - -static void bbr_reset_mode(struct sock *sk) -{ - if (!bbr_full_bw_reached(sk)) - bbr_reset_startup_mode(sk); - else - bbr_reset_probe_bw_mode(sk); -} - -/* Start a new long-term sampling interval. */ -static void bbr_reset_lt_bw_sampling_interval(struct sock *sk) -{ - struct tcp_sock *tp = tcp_sk(sk); - struct bbr *bbr = inet_csk_ca(sk); - - bbr->lt_last_stamp = div_u64(tp->delivered_mstamp, USEC_PER_MSEC); - bbr->lt_last_delivered = tp->delivered; - bbr->lt_last_lost = tp->lost; - bbr->lt_rtt_cnt = 0; -} - -/* Completely reset long-term bandwidth sampling. */ -static void bbr_reset_lt_bw_sampling(struct sock *sk) -{ - struct bbr *bbr = inet_csk_ca(sk); - - bbr->lt_bw = 0; - bbr->lt_use_bw = 0; - bbr->lt_is_sampling = false; - bbr_reset_lt_bw_sampling_interval(sk); -} - -/* Long-term bw sampling interval is done. Estimate whether we're policed. */ -static void bbr_lt_bw_interval_done(struct sock *sk, u32 bw) -{ - struct bbr *bbr = inet_csk_ca(sk); - u32 diff; - - if (bbr->lt_bw) { /* do we have bw from a previous interval? */ - /* Is new bw close to the lt_bw from the previous interval? */ - diff = abs(bw - bbr->lt_bw); - if ((diff * BBR_UNIT <= bbr_lt_bw_ratio * bbr->lt_bw) || - (bbr_rate_bytes_per_sec(sk, diff, BBR_UNIT) <= - bbr_lt_bw_diff)) { - /* All criteria are met; estimate we're policed. */ - bbr->lt_bw = (bw + bbr->lt_bw) >> 1; /* avg 2 intvls */ - bbr->lt_use_bw = 1; - bbr->pacing_gain = BBR_UNIT; /* try to avoid drops */ - bbr->lt_rtt_cnt = 0; - return; - } - } - bbr->lt_bw = bw; - bbr_reset_lt_bw_sampling_interval(sk); -} - -/* Token-bucket traffic policers are common (see "An Internet-Wide Analysis of - * Traffic Policing", SIGCOMM 2016). BBR detects token-bucket policers and - * explicitly models their policed rate, to reduce unnecessary losses. We - * estimate that we're policed if we see 2 consecutive sampling intervals with - * consistent throughput and high packet loss. If we think we're being policed, - * set lt_bw to the "long-term" average delivery rate from those 2 intervals. +/* See if we have reached next round trip. Upon start of the new round, + * returns packets delivered since previous round start plus this ACK. */ -static void bbr_lt_bw_sampling(struct sock *sk, const struct rate_sample *rs) -{ - struct tcp_sock *tp = tcp_sk(sk); - struct bbr *bbr = inet_csk_ca(sk); - u32 lost, delivered; - u64 bw; - u32 t; - - if (bbr->lt_use_bw) { /* already using long-term rate, lt_bw? */ - if (bbr->mode == BBR_PROBE_BW && bbr->round_start && - ++bbr->lt_rtt_cnt >= bbr_lt_bw_max_rtts) { - bbr_reset_lt_bw_sampling(sk); /* stop using lt_bw */ - bbr_reset_probe_bw_mode(sk); /* restart gain cycling */ - } - return; - } - - /* Wait for the first loss before sampling, to let the policer exhaust - * its tokens and estimate the steady-state rate allowed by the policer. - * Starting samples earlier includes bursts that over-estimate the bw. - */ - if (!bbr->lt_is_sampling) { - if (!rs->losses) - return; - bbr_reset_lt_bw_sampling_interval(sk); - bbr->lt_is_sampling = true; - } - - /* To avoid underestimates, reset sampling if we run out of data. */ - if (rs->is_app_limited) { - bbr_reset_lt_bw_sampling(sk); - return; - } - - if (bbr->round_start) - bbr->lt_rtt_cnt++; /* count round trips in this interval */ - if (bbr->lt_rtt_cnt < bbr_lt_intvl_min_rtts) - return; /* sampling interval needs to be longer */ - if (bbr->lt_rtt_cnt > 4 * bbr_lt_intvl_min_rtts) { - bbr_reset_lt_bw_sampling(sk); /* interval is too long */ - return; - } - - /* End sampling interval when a packet is lost, so we estimate the - * policer tokens were exhausted. Stopping the sampling before the - * tokens are exhausted under-estimates the policed rate. - */ - if (!rs->losses) - return; - - /* Calculate packets lost and delivered in sampling interval. */ - lost = tp->lost - bbr->lt_last_lost; - delivered = tp->delivered - bbr->lt_last_delivered; - /* Is loss rate (lost/delivered) >= lt_loss_thresh? If not, wait. */ - if (!delivered || (lost << BBR_SCALE) < bbr_lt_loss_thresh * delivered) - return; - - /* Find average delivery rate in this sampling interval. */ - t = div_u64(tp->delivered_mstamp, USEC_PER_MSEC) - bbr->lt_last_stamp; - if ((s32)t < 1) - return; /* interval is less than one ms, so wait */ - /* Check if can multiply without overflow */ - if (t >= ~0U / USEC_PER_MSEC) { - bbr_reset_lt_bw_sampling(sk); /* interval too long; reset */ - return; - } - t *= USEC_PER_MSEC; - bw = (u64)delivered * BW_UNIT; - do_div(bw, t); - bbr_lt_bw_interval_done(sk, bw); -} - -/* Estimate the bandwidth based on how fast packets are delivered */ -static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs) +static u32 bbr_update_round_start(struct sock *sk, + const struct rate_sample *rs, struct bbr_context *ctx) { struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); - u64 bw; + u32 round_delivered = 0; bbr->round_start = 0; - if (rs->delivered < 0 || rs->interval_us <= 0) - return; /* Not a valid observation */ /* See if we've reached the next RTT */ - if (!before(rs->prior_delivered, bbr->next_rtt_delivered)) { + if (rs->interval_us > 0 && + !before(rs->prior_delivered, bbr->next_rtt_delivered)) { + round_delivered = tp->delivered - bbr->next_rtt_delivered; bbr->next_rtt_delivered = tp->delivered; - bbr->rtt_cnt++; bbr->round_start = 1; - bbr->packet_conservation = 0; } + return round_delivered; +} - bbr_lt_bw_sampling(sk, rs); +/* Calculate the bandwidth based on how fast packets are delivered */ +static void bbr_calculate_bw_sample(struct sock *sk, + const struct rate_sample *rs, struct bbr_context *ctx) +{ + u64 bw = 0; /* Divide delivered by the interval to find a (lower bound) bottleneck * bandwidth sample. Delivered is in packets and interval_us in uS and * ratio will be <<1 for most connections. So delivered is first scaled. + * Round up to allow growth at low rates, even with integer division. */ - bw = div64_long((u64)rs->delivered * BW_UNIT, rs->interval_us); - - /* If this sample is application-limited, it is likely to have a very - * low delivered count that represents application behavior rather than - * the available network rate. Such a sample could drag down estimated - * bw, causing needless slow-down. Thus, to continue to send at the - * last measured network rate, we filter out app-limited samples unless - * they describe the path bw at least as well as our bw model. - * - * So the goal during app-limited phase is to proceed with the best - * network rate no matter how long. We automatically leave this - * phase when app writes faster than the network can deliver :) - */ - if (!rs->is_app_limited || bw >= bbr_max_bw(sk)) { - /* Incorporate new sample into our max bw filter. */ - minmax_running_max(&bbr->bw, bbr_bw_rtts, bbr->rtt_cnt, bw); + if (rs->interval_us > 0) { + if (WARN_ONCE(rs->delivered < 0, + "negative delivered: %d interval_us: %ld\n", + rs->delivered, rs->interval_us)) + return; + + bw = DIV_ROUND_UP_ULL((u64)rs->delivered * BW_UNIT, rs->interval_us); } + + ctx->sample_bw = bw; } /* Estimates the windowed max degree of ack aggregation. @@ -812,7 +797,7 @@ static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs) * * Max extra_acked is clamped by cwnd and bw * bbr_extra_acked_max_us (100 ms). * Max filter is an approximate sliding window of 5-10 (packet timed) round - * trips. + * trips for non-startup phase, and 1-2 round trips for startup. */ static void bbr_update_ack_aggregation(struct sock *sk, const struct rate_sample *rs) @@ -820,15 +805,19 @@ static void bbr_update_ack_aggregation(struct sock *sk, u32 epoch_us, expected_acked, extra_acked; struct bbr *bbr = inet_csk_ca(sk); struct tcp_sock *tp = tcp_sk(sk); + u32 extra_acked_win_rtts_thresh = bbr_param(sk, extra_acked_win_rtts); - if (!bbr_extra_acked_gain || rs->acked_sacked <= 0 || + if (!bbr_param(sk, extra_acked_gain) || rs->acked_sacked <= 0 || rs->delivered < 0 || rs->interval_us <= 0) return; if (bbr->round_start) { bbr->extra_acked_win_rtts = min(0x1F, bbr->extra_acked_win_rtts + 1); - if (bbr->extra_acked_win_rtts >= bbr_extra_acked_win_rtts) { + if (!bbr_full_bw_reached(sk)) + extra_acked_win_rtts_thresh = 1; + if (bbr->extra_acked_win_rtts >= + extra_acked_win_rtts_thresh) { bbr->extra_acked_win_rtts = 0; bbr->extra_acked_win_idx = bbr->extra_acked_win_idx ? 0 : 1; @@ -862,49 +851,6 @@ static void bbr_update_ack_aggregation(struct sock *sk, bbr->extra_acked[bbr->extra_acked_win_idx] = extra_acked; } -/* Estimate when the pipe is full, using the change in delivery rate: BBR - * estimates that STARTUP filled the pipe if the estimated bw hasn't changed by - * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited - * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the - * higher rwin, 3: we get higher delivery rate samples. Or transient - * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar - * design goal, but uses delay and inter-ACK spacing instead of bandwidth. - */ -static void bbr_check_full_bw_reached(struct sock *sk, - const struct rate_sample *rs) -{ - struct bbr *bbr = inet_csk_ca(sk); - u32 bw_thresh; - - if (bbr_full_bw_reached(sk) || !bbr->round_start || rs->is_app_limited) - return; - - bw_thresh = (u64)bbr->full_bw * bbr_full_bw_thresh >> BBR_SCALE; - if (bbr_max_bw(sk) >= bw_thresh) { - bbr->full_bw = bbr_max_bw(sk); - bbr->full_bw_cnt = 0; - return; - } - ++bbr->full_bw_cnt; - bbr->full_bw_reached = bbr->full_bw_cnt >= bbr_full_bw_cnt; -} - -/* If pipe is probably full, drain the queue and then enter steady-state. */ -static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs) -{ - struct bbr *bbr = inet_csk_ca(sk); - - if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) { - bbr->mode = BBR_DRAIN; /* drain queue we created */ - tcp_sk(sk)->snd_ssthresh = - bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT); - } /* fall through to check if in-flight is already small: */ - if (bbr->mode == BBR_DRAIN && - bbr_packets_in_net_at_edt(sk, tcp_packets_in_flight(tcp_sk(sk))) <= - bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT)) - bbr_reset_probe_bw_mode(sk); /* we estimate queue is drained */ -} - static void bbr_check_probe_rtt_done(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); @@ -914,9 +860,9 @@ static void bbr_check_probe_rtt_done(struct sock *sk) after(tcp_jiffies32, bbr->probe_rtt_done_stamp))) return; - bbr->min_rtt_stamp = tcp_jiffies32; /* wait a while until PROBE_RTT */ + bbr->probe_rtt_min_stamp = tcp_jiffies32; /* schedule next PROBE_RTT */ tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp), bbr->prior_cwnd)); - bbr_reset_mode(sk); + bbr_exit_probe_rtt(sk); } /* The goal of PROBE_RTT mode is to have BBR flows cooperatively and @@ -942,23 +888,35 @@ static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs) { struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); - bool filter_expired; + bool probe_rtt_expired, min_rtt_expired; + u32 expire; - /* Track min RTT seen in the min_rtt_win_sec filter window: */ - filter_expired = after(tcp_jiffies32, - bbr->min_rtt_stamp + bbr_min_rtt_win_sec * HZ); + /* Track min RTT in probe_rtt_win_ms to time next PROBE_RTT state. */ + expire = bbr->probe_rtt_min_stamp + + msecs_to_jiffies(bbr_param(sk, probe_rtt_win_ms)); + probe_rtt_expired = after(tcp_jiffies32, expire); if (rs->rtt_us >= 0 && - (rs->rtt_us < bbr->min_rtt_us || - (filter_expired && !rs->is_ack_delayed))) { - bbr->min_rtt_us = rs->rtt_us; - bbr->min_rtt_stamp = tcp_jiffies32; + (rs->rtt_us < bbr->probe_rtt_min_us || + (probe_rtt_expired && !rs->is_ack_delayed))) { + bbr->probe_rtt_min_us = rs->rtt_us; + bbr->probe_rtt_min_stamp = tcp_jiffies32; + } + /* Track min RTT seen in the min_rtt_win_sec filter window: */ + expire = bbr->min_rtt_stamp + bbr_param(sk, min_rtt_win_sec) * HZ; + min_rtt_expired = after(tcp_jiffies32, expire); + if (bbr->probe_rtt_min_us <= bbr->min_rtt_us || + min_rtt_expired) { + bbr->min_rtt_us = bbr->probe_rtt_min_us; + bbr->min_rtt_stamp = bbr->probe_rtt_min_stamp; } - if (bbr_probe_rtt_mode_ms > 0 && filter_expired && + if (bbr_param(sk, probe_rtt_mode_ms) > 0 && probe_rtt_expired && !bbr->idle_restart && bbr->mode != BBR_PROBE_RTT) { bbr->mode = BBR_PROBE_RTT; /* dip, drain queue */ bbr_save_cwnd(sk); /* note cwnd so we can restore it */ bbr->probe_rtt_done_stamp = 0; + bbr->ack_phase = BBR_ACKS_PROBE_STOPPING; + bbr->next_rtt_delivered = tp->delivered; } if (bbr->mode == BBR_PROBE_RTT) { @@ -967,9 +925,9 @@ static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs) (tp->delivered + tcp_packets_in_flight(tp)) ? : 1; /* Maintain min packets in flight for max(200 ms, 1 round). */ if (!bbr->probe_rtt_done_stamp && - tcp_packets_in_flight(tp) <= bbr_cwnd_min_target) { + tcp_packets_in_flight(tp) <= bbr_probe_rtt_cwnd(sk)) { bbr->probe_rtt_done_stamp = tcp_jiffies32 + - msecs_to_jiffies(bbr_probe_rtt_mode_ms); + msecs_to_jiffies(bbr_param(sk, probe_rtt_mode_ms)); bbr->probe_rtt_round_done = 0; bbr->next_rtt_delivered = tp->delivered; } else if (bbr->probe_rtt_done_stamp) { @@ -990,18 +948,20 @@ static void bbr_update_gains(struct sock *sk) switch (bbr->mode) { case BBR_STARTUP: - bbr->pacing_gain = bbr_high_gain; - bbr->cwnd_gain = bbr_high_gain; + bbr->pacing_gain = bbr_param(sk, startup_pacing_gain); + bbr->cwnd_gain = bbr_param(sk, startup_cwnd_gain); break; case BBR_DRAIN: - bbr->pacing_gain = bbr_drain_gain; /* slow, to drain */ - bbr->cwnd_gain = bbr_high_gain; /* keep cwnd */ + bbr->pacing_gain = bbr_param(sk, drain_gain); /* slow, to drain */ + bbr->cwnd_gain = bbr_param(sk, startup_cwnd_gain); /* keep cwnd */ break; case BBR_PROBE_BW: - bbr->pacing_gain = (bbr->lt_use_bw ? - BBR_UNIT : - bbr_pacing_gain[bbr->cycle_idx]); - bbr->cwnd_gain = bbr_cwnd_gain; + bbr->pacing_gain = bbr_pacing_gain[bbr->cycle_idx]; + bbr->cwnd_gain = bbr_param(sk, cwnd_gain); + if (bbr_param(sk, bw_probe_cwnd_gain) && + bbr->cycle_idx == BBR_BW_PROBE_UP) + bbr->cwnd_gain += + BBR_UNIT * bbr_param(sk, bw_probe_cwnd_gain) / 4; break; case BBR_PROBE_RTT: bbr->pacing_gain = BBR_UNIT; @@ -1013,144 +973,1387 @@ static void bbr_update_gains(struct sock *sk) } } -static void bbr_update_model(struct sock *sk, const struct rate_sample *rs) +__bpf_kfunc static u32 bbr_sndbuf_expand(struct sock *sk) { - bbr_update_bw(sk, rs); - bbr_update_ack_aggregation(sk, rs); - bbr_update_cycle_phase(sk, rs); - bbr_check_full_bw_reached(sk, rs); - bbr_check_drain(sk, rs); - bbr_update_min_rtt(sk, rs); - bbr_update_gains(sk); + /* Provision 3 * cwnd since BBR may slow-start even during recovery. */ + return 3; } -__bpf_kfunc static void bbr_main(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs) +/* Incorporate a new bw sample into the current window of our max filter. */ +static void bbr_take_max_bw_sample(struct sock *sk, u32 bw) { struct bbr *bbr = inet_csk_ca(sk); - u32 bw; - - bbr_update_model(sk, rs); - bw = bbr_bw(sk); - bbr_set_pacing_rate(sk, bw, bbr->pacing_gain); - bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain); + bbr->bw_hi[1] = max(bw, bbr->bw_hi[1]); } -__bpf_kfunc static void bbr_init(struct sock *sk) +/* Keep max of last 1-2 cycles. Each PROBE_BW cycle, flip filter window. */ +static void bbr_advance_max_bw_filter(struct sock *sk) { - struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); - bbr->prior_cwnd = 0; - tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; - bbr->rtt_cnt = 0; - bbr->next_rtt_delivered = tp->delivered; - bbr->prev_ca_state = TCP_CA_Open; - bbr->packet_conservation = 0; - - bbr->probe_rtt_done_stamp = 0; - bbr->probe_rtt_round_done = 0; - bbr->min_rtt_us = tcp_min_rtt(tp); - bbr->min_rtt_stamp = tcp_jiffies32; - - minmax_reset(&bbr->bw, bbr->rtt_cnt, 0); /* init max bw to 0 */ + if (!bbr->bw_hi[1]) + return; /* no samples in this window; remember old window */ + bbr->bw_hi[0] = bbr->bw_hi[1]; + bbr->bw_hi[1] = 0; +} - bbr->has_seen_rtt = 0; - bbr_init_pacing_rate_from_rtt(sk); +/* Reset the estimator for reaching full bandwidth based on bw plateau. */ +static void bbr_reset_full_bw(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); - bbr->round_start = 0; - bbr->idle_restart = 0; - bbr->full_bw_reached = 0; bbr->full_bw = 0; bbr->full_bw_cnt = 0; - bbr->cycle_mstamp = 0; - bbr->cycle_idx = 0; - bbr_reset_lt_bw_sampling(sk); - bbr_reset_startup_mode(sk); + bbr->full_bw_now = 0; +} - bbr->ack_epoch_mstamp = tp->tcp_mstamp; - bbr->ack_epoch_acked = 0; - bbr->extra_acked_win_rtts = 0; - bbr->extra_acked_win_idx = 0; - bbr->extra_acked[0] = 0; - bbr->extra_acked[1] = 0; +/* How much do we want in flight? Our BDP, unless congestion cut cwnd. */ +static u32 bbr_target_inflight(struct sock *sk) +{ + u32 bdp = bbr_inflight(sk, bbr_bw(sk), BBR_UNIT); - cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED); + return min(bdp, tcp_sk(sk)->snd_cwnd); } -__bpf_kfunc static u32 bbr_sndbuf_expand(struct sock *sk) +static bool bbr_is_probing_bandwidth(struct sock *sk) { - /* Provision 3 * cwnd since BBR may slow-start even during recovery. */ - return 3; + struct bbr *bbr = inet_csk_ca(sk); + + return (bbr->mode == BBR_STARTUP) || + (bbr->mode == BBR_PROBE_BW && + (bbr->cycle_idx == BBR_BW_PROBE_REFILL || + bbr->cycle_idx == BBR_BW_PROBE_UP)); +} + +/* Has the given amount of time elapsed since we marked the phase start? */ +static bool bbr_has_elapsed_in_phase(const struct sock *sk, u32 interval_us) +{ + const struct tcp_sock *tp = tcp_sk(sk); + const struct bbr *bbr = inet_csk_ca(sk); + + return tcp_stamp_us_delta(tp->tcp_mstamp, + bbr->cycle_mstamp + interval_us) > 0; +} + +static void bbr_handle_queue_too_high_in_startup(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 bdp; /* estimated BDP in packets, with quantization budget */ + + bbr->full_bw_reached = 1; + + bdp = bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT); + bbr->inflight_hi = max(bdp, bbr->inflight_latest); +} + +/* Exit STARTUP upon N consecutive rounds with ECN mark rate > ecn_thresh. */ +static void bbr_check_ecn_too_high_in_startup(struct sock *sk, u32 ce_ratio) +{ + struct bbr *bbr = inet_csk_ca(sk); + + if (bbr_full_bw_reached(sk) || !bbr->ecn_eligible || + !bbr_param(sk, full_ecn_cnt) || !bbr_param(sk, ecn_thresh)) + return; + + if (ce_ratio >= bbr_param(sk, ecn_thresh)) + bbr->startup_ecn_rounds++; + else + bbr->startup_ecn_rounds = 0; + + if (bbr->startup_ecn_rounds >= bbr_param(sk, full_ecn_cnt)) { + bbr_handle_queue_too_high_in_startup(sk); + return; + } +} + +/* Updates ecn_alpha and returns ce_ratio. -1 if not available. */ +static int bbr_update_ecn_alpha(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct net *net = sock_net(sk); + struct bbr *bbr = inet_csk_ca(sk); + s32 delivered, delivered_ce; + u64 alpha, ce_ratio; + u32 gain; + bool want_ecn_alpha; + + /* See if we should use ECN sender logic for this connection. */ + if (!bbr->ecn_eligible && bbr_can_use_ecn(sk) && + !!bbr_param(sk, ecn_factor) && + (bbr->min_rtt_us <= bbr_ecn_max_rtt_us || + !bbr_ecn_max_rtt_us)) + bbr->ecn_eligible = 1; + + /* Skip updating alpha only if not ECN-eligible and PLB is disabled. */ + want_ecn_alpha = (bbr->ecn_eligible || + (bbr_can_use_ecn(sk) && + READ_ONCE(net->ipv4.sysctl_tcp_plb_enabled))); + if (!want_ecn_alpha) + return -1; + + delivered = tp->delivered - bbr->alpha_last_delivered; + delivered_ce = tp->delivered_ce - bbr->alpha_last_delivered_ce; + + if (delivered == 0 || /* avoid divide by zero */ + WARN_ON_ONCE(delivered < 0 || delivered_ce < 0)) /* backwards? */ + return -1; + + BUILD_BUG_ON(BBR_SCALE != TCP_PLB_SCALE); + ce_ratio = (u64)delivered_ce << BBR_SCALE; + do_div(ce_ratio, delivered); + + gain = bbr_param(sk, ecn_alpha_gain); + alpha = ((BBR_UNIT - gain) * bbr->ecn_alpha) >> BBR_SCALE; + alpha += (gain * ce_ratio) >> BBR_SCALE; + bbr->ecn_alpha = min_t(u32, alpha, BBR_UNIT); + + bbr->alpha_last_delivered = tp->delivered; + bbr->alpha_last_delivered_ce = tp->delivered_ce; + + bbr_check_ecn_too_high_in_startup(sk, ce_ratio); + return (int)ce_ratio; } -/* In theory BBR does not need to undo the cwnd since it does not - * always reduce cwnd on losses (see bbr_main()). Keep it for now. +/* Protective Load Balancing (PLB). PLB rehashes outgoing data (to a new IPv6 + * flow label) if it encounters sustained congestion in the form of ECN marks. */ -__bpf_kfunc static u32 bbr_undo_cwnd(struct sock *sk) +static void bbr_plb(struct sock *sk, const struct rate_sample *rs, int ce_ratio) +{ + struct bbr *bbr = inet_csk_ca(sk); + + if (bbr->round_start && ce_ratio >= 0) + tcp_plb_update_state(sk, &bbr->plb, ce_ratio); + + tcp_plb_check_rehash(sk, &bbr->plb); +} + +/* Each round trip of BBR_BW_PROBE_UP, double volume of probing data. */ +static void bbr_raise_inflight_hi_slope(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + u32 growth_this_round, cnt; + + /* Calculate "slope": packets S/Acked per inflight_hi increment. */ + growth_this_round = 1 << bbr->bw_probe_up_rounds; + bbr->bw_probe_up_rounds = min(bbr->bw_probe_up_rounds + 1, 30); + cnt = tcp_snd_cwnd(tp) / growth_this_round; + cnt = max(cnt, 1U); + bbr->bw_probe_up_cnt = cnt; +} + +/* In BBR_BW_PROBE_UP, not seeing high loss/ECN/queue, so raise inflight_hi. */ +static void bbr_probe_inflight_hi_upward(struct sock *sk, + const struct rate_sample *rs) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + u32 delta; + + if (!tp->is_cwnd_limited || tcp_snd_cwnd(tp) < bbr->inflight_hi) + return; /* not fully using inflight_hi, so don't grow it */ + + /* For each bw_probe_up_cnt packets ACKed, increase inflight_hi by 1. */ + bbr->bw_probe_up_acks += rs->acked_sacked; + if (bbr->bw_probe_up_acks >= bbr->bw_probe_up_cnt) { + delta = bbr->bw_probe_up_acks / bbr->bw_probe_up_cnt; + bbr->bw_probe_up_acks -= delta * bbr->bw_probe_up_cnt; + bbr->inflight_hi += delta; + bbr->try_fast_path = 0; /* Need to update cwnd */ + } + + if (bbr->round_start) + bbr_raise_inflight_hi_slope(sk); +} + +/* Does loss/ECN rate for this sample say inflight is "too high"? + * This is used by both the bbr_check_loss_too_high_in_startup() function, + * and in PROBE_UP. + */ +static bool bbr_is_inflight_too_high(const struct sock *sk, + const struct rate_sample *rs) +{ + const struct bbr *bbr = inet_csk_ca(sk); + u32 loss_thresh, ecn_thresh; + + if (rs->lost > 0 && rs->tx_in_flight) { + loss_thresh = (u64)rs->tx_in_flight * bbr_param(sk, loss_thresh) >> + BBR_SCALE; + if (rs->lost > loss_thresh) { + return true; + } + } + + if (rs->delivered_ce > 0 && rs->delivered > 0 && + bbr->ecn_eligible && !!bbr_param(sk, ecn_thresh)) { + ecn_thresh = (u64)rs->delivered * bbr_param(sk, ecn_thresh) >> + BBR_SCALE; + if (rs->delivered_ce > ecn_thresh) { + return true; + } + } + + return false; +} + +/* Calculate the tx_in_flight level that corresponded to excessive loss. + * We find "lost_prefix" segs of the skb where loss rate went too high, + * by solving for "lost_prefix" in the following equation: + * lost / inflight >= loss_thresh + * (lost_prev + lost_prefix) / (inflight_prev + lost_prefix) >= loss_thresh + * Then we take that equation, convert it to fixed point, and + * round up to the nearest packet. + */ +static u32 bbr_inflight_hi_from_lost_skb(const struct sock *sk, + const struct rate_sample *rs, + const struct sk_buff *skb) +{ + const struct tcp_sock *tp = tcp_sk(sk); + u32 loss_thresh = bbr_param(sk, loss_thresh); + u32 pcount, divisor, inflight_hi; + s32 inflight_prev, lost_prev; + u64 loss_budget, lost_prefix; + + pcount = tcp_skb_pcount(skb); + + /* How much data was in flight before this skb? */ + inflight_prev = rs->tx_in_flight - pcount; + if (inflight_prev < 0) { + WARN_ONCE(tcp_skb_tx_in_flight_is_suspicious( + pcount, + TCP_SKB_CB(skb)->sacked, + rs->tx_in_flight), + "tx_in_flight: %u pcount: %u reneg: %u", + rs->tx_in_flight, pcount, tcp_sk(sk)->is_sack_reneg); + return ~0U; + } + + /* How much inflight data was marked lost before this skb? */ + lost_prev = rs->lost - pcount; + if (WARN_ONCE(lost_prev < 0, + "cwnd: %u ca: %d out: %u lost: %u pif: %u " + "tx_in_flight: %u tx.lost: %u tp->lost: %u rs->lost: %d " + "lost_prev: %d pcount: %d seq: %u end_seq: %u reneg: %u", + tcp_snd_cwnd(tp), inet_csk(sk)->icsk_ca_state, + tp->packets_out, tp->lost_out, tcp_packets_in_flight(tp), + rs->tx_in_flight, TCP_SKB_CB(skb)->tx.lost, tp->lost, + rs->lost, lost_prev, pcount, + TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, + tp->is_sack_reneg)) + return ~0U; + + /* At what prefix of this lost skb did losss rate exceed loss_thresh? */ + loss_budget = (u64)inflight_prev * loss_thresh + BBR_UNIT - 1; + loss_budget >>= BBR_SCALE; + if (lost_prev >= loss_budget) { + lost_prefix = 0; /* previous losses crossed loss_thresh */ + } else { + lost_prefix = loss_budget - lost_prev; + lost_prefix <<= BBR_SCALE; + divisor = BBR_UNIT - loss_thresh; + if (WARN_ON_ONCE(!divisor)) /* loss_thresh is 8 bits */ + return ~0U; + do_div(lost_prefix, divisor); + } + + inflight_hi = inflight_prev + lost_prefix; + return inflight_hi; +} + +/* If loss/ECN rates during probing indicated we may have overfilled a + * buffer, return an operating point that tries to leave unutilized headroom in + * the path for other flows, for fairness convergence and lower RTTs and loss. + */ +static u32 bbr_inflight_with_headroom(const struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 headroom, headroom_fraction; + + if (bbr->inflight_hi == ~0U) + return ~0U; + + headroom_fraction = bbr_param(sk, inflight_headroom); + headroom = ((u64)bbr->inflight_hi * headroom_fraction) >> BBR_SCALE; + headroom = max(headroom, 1U); + return max_t(s32, bbr->inflight_hi - headroom, + bbr_param(sk, cwnd_min_target)); +} + +/* Bound cwnd to a sensible level, based on our current probing state + * machine phase and model of a good inflight level (inflight_lo, inflight_hi). + */ +static void bbr_bound_cwnd_for_inflight_model(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + u32 cap; + + /* tcp_rcv_synsent_state_process() currently calls tcp_ack() + * and thus cong_control() without first initializing us(!). + */ + if (!bbr->initialized) + return; + + cap = ~0U; + if (bbr->mode == BBR_PROBE_BW && + bbr->cycle_idx != BBR_BW_PROBE_CRUISE) { + /* Probe to see if more packets fit in the path. */ + cap = bbr->inflight_hi; + } else { + if (bbr->mode == BBR_PROBE_RTT || + (bbr->mode == BBR_PROBE_BW && + bbr->cycle_idx == BBR_BW_PROBE_CRUISE)) + cap = bbr_inflight_with_headroom(sk); + } + /* Adapt to any loss/ECN since our last bw probe. */ + cap = min(cap, bbr->inflight_lo); + + cap = max_t(u32, cap, bbr_param(sk, cwnd_min_target)); + tcp_snd_cwnd_set(tp, min(cap, tcp_snd_cwnd(tp))); +} + +/* How should we multiplicatively cut bw or inflight limits based on ECN? */ +static u32 bbr_ecn_cut(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + + return BBR_UNIT - + ((bbr->ecn_alpha * bbr_param(sk, ecn_factor)) >> BBR_SCALE); +} + +/* Init lower bounds if have not inited yet. */ +static void bbr_init_lower_bounds(struct sock *sk, bool init_bw) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + if (init_bw && bbr->bw_lo == ~0U) + bbr->bw_lo = bbr_max_bw(sk); + if (bbr->inflight_lo == ~0U) + bbr->inflight_lo = tcp_snd_cwnd(tp); +} + +/* Reduce bw and inflight to (1 - beta). */ +static void bbr_loss_lower_bounds(struct sock *sk, u32 *bw, u32 *inflight) +{ + struct bbr* bbr = inet_csk_ca(sk); + u32 loss_cut = BBR_UNIT - bbr_param(sk, beta); + + *bw = max_t(u32, bbr->bw_latest, + (u64)bbr->bw_lo * loss_cut >> BBR_SCALE); + *inflight = max_t(u32, bbr->inflight_latest, + (u64)bbr->inflight_lo * loss_cut >> BBR_SCALE); +} + +/* Reduce inflight to (1 - alpha*ecn_factor). */ +static void bbr_ecn_lower_bounds(struct sock *sk, u32 *inflight) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 ecn_cut = bbr_ecn_cut(sk); + + *inflight = (u64)bbr->inflight_lo * ecn_cut >> BBR_SCALE; +} + +/* Estimate a short-term lower bound on the capacity available now, based + * on measurements of the current delivery process and recent history. When we + * are seeing loss/ECN at times when we are not probing bw, then conservatively + * move toward flow balance by multiplicatively cutting our short-term + * estimated safe rate and volume of data (bw_lo and inflight_lo). We use a + * multiplicative decrease in order to converge to a lower capacity in time + * logarithmic in the magnitude of the decrease. + * + * However, we do not cut our short-term estimates lower than the current rate + * and volume of delivered data from this round trip, since from the current + * delivery process we can estimate the measured capacity available now. + * + * Anything faster than that approach would knowingly risk high loss, which can + * cause low bw for Reno/CUBIC and high loss recovery latency for + * request/response flows using any congestion control. + */ +static void bbr_adapt_lower_bounds(struct sock *sk, + const struct rate_sample *rs) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 ecn_inflight_lo = ~0U; + + /* We only use lower-bound estimates when not probing bw. + * When probing we need to push inflight higher to probe bw. + */ + if (bbr_is_probing_bandwidth(sk)) + return; + + /* ECN response. */ + if (bbr->ecn_in_round && !!bbr_param(sk, ecn_factor)) { + bbr_init_lower_bounds(sk, false); + bbr_ecn_lower_bounds(sk, &ecn_inflight_lo); + } + + /* Loss response. */ + if (bbr->loss_in_round) { + bbr_init_lower_bounds(sk, true); + bbr_loss_lower_bounds(sk, &bbr->bw_lo, &bbr->inflight_lo); + } + + /* Adjust to the lower of the levels implied by loss/ECN. */ + bbr->inflight_lo = min(bbr->inflight_lo, ecn_inflight_lo); + bbr->bw_lo = max(1U, bbr->bw_lo); +} + +/* Reset any short-term lower-bound adaptation to congestion, so that we can + * push our inflight up. + */ +static void bbr_reset_lower_bounds(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + + bbr->bw_lo = ~0U; + bbr->inflight_lo = ~0U; +} + +/* After bw probing (STARTUP/PROBE_UP), reset signals before entering a state + * machine phase where we adapt our lower bound based on congestion signals. + */ +static void bbr_reset_congestion_signals(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + + bbr->loss_in_round = 0; + bbr->ecn_in_round = 0; + bbr->loss_in_cycle = 0; + bbr->ecn_in_cycle = 0; + bbr->bw_latest = 0; + bbr->inflight_latest = 0; +} + +static void bbr_exit_loss_recovery(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp), bbr->prior_cwnd)); + bbr->try_fast_path = 0; /* bound cwnd using latest model */ +} + +/* Update rate and volume of delivered data from latest round trip. */ +static void bbr_update_latest_delivery_signals( + struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + bbr->loss_round_start = 0; + if (rs->interval_us <= 0 || !rs->acked_sacked) + return; /* Not a valid observation */ + + bbr->bw_latest = max_t(u32, bbr->bw_latest, ctx->sample_bw); + bbr->inflight_latest = max_t(u32, bbr->inflight_latest, rs->delivered); + + if (!before(rs->prior_delivered, bbr->loss_round_delivered)) { + bbr->loss_round_delivered = tp->delivered; + bbr->loss_round_start = 1; /* mark start of new round trip */ + } +} + +/* Once per round, reset filter for latest rate and volume of delivered data. */ +static void bbr_advance_latest_delivery_signals( + struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx) +{ + struct bbr *bbr = inet_csk_ca(sk); + + /* If ACK matches a TLP retransmit, persist the filter. If we detect + * that a TLP retransmit plugged a tail loss, we'll want to remember + * how much data the path delivered before the tail loss. + */ + if (bbr->loss_round_start && !rs->is_acking_tlp_retrans_seq) { + bbr->bw_latest = ctx->sample_bw; + bbr->inflight_latest = rs->delivered; + } +} + +/* Update (most of) our congestion signals: track the recent rate and volume of + * delivered data, presence of loss, and EWMA degree of ECN marking. + */ +static void bbr_update_congestion_signals( + struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx) +{ + struct bbr *bbr = inet_csk_ca(sk); + u64 bw; + + if (rs->interval_us <= 0 || !rs->acked_sacked) + return; /* Not a valid observation */ + bw = ctx->sample_bw; + + if (!rs->is_app_limited || bw >= bbr_max_bw(sk)) + bbr_take_max_bw_sample(sk, bw); + + bbr->loss_in_round |= (rs->losses > 0); + + if (!bbr->loss_round_start) + return; /* skip the per-round-trip updates */ + /* Now do per-round-trip updates. */ + bbr_adapt_lower_bounds(sk, rs); + + bbr->loss_in_round = 0; + bbr->ecn_in_round = 0; +} + +/* Bandwidth probing can cause loss. To help coexistence with loss-based + * congestion control we spread out our probing in a Reno-conscious way. Due to + * the shape of the Reno sawtooth, the time required between loss epochs for an + * idealized Reno flow is a number of round trips that is the BDP of that + * flow. We count packet-timed round trips directly, since measured RTT can + * vary widely, and Reno is driven by packet-timed round trips. + */ +static bool bbr_is_reno_coexistence_probe_time(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 rounds; + + /* Random loss can shave some small percentage off of our inflight + * in each round. To survive this, flows need robust periodic probes. + */ + rounds = min_t(u32, bbr_param(sk, bw_probe_max_rounds), bbr_target_inflight(sk)); + return bbr->rounds_since_probe >= rounds; +} + +/* How long do we want to wait before probing for bandwidth (and risking + * loss)? We randomize the wait, for better mixing and fairness convergence. + * + * We bound the Reno-coexistence inter-bw-probe time to be 62-63 round trips. + * This is calculated to allow fairness with a 25Mbps, 30ms Reno flow, + * (eg 4K video to a broadband user): + * BDP = 25Mbps * .030sec /(1514bytes) = 61.9 packets + * + * We bound the BBR-native inter-bw-probe wall clock time to be: + * (a) higher than 2 sec: to try to avoid causing loss for a long enough time + * to allow Reno at 30ms to get 4K video bw, the inter-bw-probe time must + * be at least: 25Mbps * .030sec / (1514bytes) * 0.030sec = 1.9secs + * (b) lower than 3 sec: to ensure flows can start probing in a reasonable + * amount of time to discover unutilized bw on human-scale interactive + * time-scales (e.g. perhaps traffic from a web page download that we + * were competing with is now complete). + */ +static void bbr_pick_probe_wait(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + + /* Decide the random round-trip bound for wait until probe: */ + bbr->rounds_since_probe = + get_random_u32_below(bbr_param(sk, bw_probe_rand_rounds)); + /* Decide the random wall clock bound for wait until probe: */ + bbr->probe_wait_us = bbr_param(sk, bw_probe_base_us) + + get_random_u32_below(bbr_param(sk, bw_probe_rand_us)); +} + +static void bbr_set_cycle_idx(struct sock *sk, int cycle_idx) +{ + struct bbr *bbr = inet_csk_ca(sk); + + bbr->cycle_idx = cycle_idx; + /* New phase, so need to update cwnd and pacing rate. */ + bbr->try_fast_path = 0; +} + +/* Send at estimated bw to fill the pipe, but not queue. We need this phase + * before PROBE_UP, because as soon as we send faster than the available bw + * we will start building a queue, and if the buffer is shallow we can cause + * loss. If we do not fill the pipe before we cause this loss, our bw_hi and + * inflight_hi estimates will underestimate. + */ +static void bbr_start_bw_probe_refill(struct sock *sk, u32 bw_probe_up_rounds) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + bbr_reset_lower_bounds(sk); + bbr->bw_probe_up_rounds = bw_probe_up_rounds; + bbr->bw_probe_up_acks = 0; + bbr->stopped_risky_probe = 0; + bbr->ack_phase = BBR_ACKS_REFILLING; + bbr->next_rtt_delivered = tp->delivered; + bbr_set_cycle_idx(sk, BBR_BW_PROBE_REFILL); +} + +/* Now probe max deliverable data rate and volume. */ +static void bbr_start_bw_probe_up(struct sock *sk, struct bbr_context *ctx) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + bbr->ack_phase = BBR_ACKS_PROBE_STARTING; + bbr->next_rtt_delivered = tp->delivered; + bbr->cycle_mstamp = tp->tcp_mstamp; + bbr_reset_full_bw(sk); + bbr->full_bw = ctx->sample_bw; + bbr_set_cycle_idx(sk, BBR_BW_PROBE_UP); + bbr_raise_inflight_hi_slope(sk); +} + +/* Start a new PROBE_BW probing cycle of some wall clock length. Pick a wall + * clock time at which to probe beyond an inflight that we think to be + * safe. This will knowingly risk packet loss, so we want to do this rarely, to + * keep packet loss rates low. Also start a round-trip counter, to probe faster + * if we estimate a Reno flow at our BDP would probe faster. + */ +static void bbr_start_bw_probe_down(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + bbr_reset_congestion_signals(sk); + bbr->bw_probe_up_cnt = ~0U; /* not growing inflight_hi any more */ + bbr_pick_probe_wait(sk); + bbr->cycle_mstamp = tp->tcp_mstamp; /* start wall clock */ + bbr->ack_phase = BBR_ACKS_PROBE_STOPPING; + bbr->next_rtt_delivered = tp->delivered; + bbr_set_cycle_idx(sk, BBR_BW_PROBE_DOWN); +} + +/* Cruise: maintain what we estimate to be a neutral, conservative + * operating point, without attempting to probe up for bandwidth or down for + * RTT, and only reducing inflight in response to loss/ECN signals. + */ +static void bbr_start_bw_probe_cruise(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + + if (bbr->inflight_lo != ~0U) + bbr->inflight_lo = min(bbr->inflight_lo, bbr->inflight_hi); + + bbr_set_cycle_idx(sk, BBR_BW_PROBE_CRUISE); +} + +/* Loss and/or ECN rate is too high while probing. + * Adapt (once per bw probe) by cutting inflight_hi and then restarting cycle. + */ +static void bbr_handle_inflight_too_high(struct sock *sk, + const struct rate_sample *rs) +{ + struct bbr *bbr = inet_csk_ca(sk); + const u32 beta = bbr_param(sk, beta); + + bbr->prev_probe_too_high = 1; + bbr->bw_probe_samples = 0; /* only react once per probe */ + /* If we are app-limited then we are not robustly + * probing the max volume of inflight data we think + * might be safe (analogous to how app-limited bw + * samples are not known to be robustly probing bw). + */ + if (!rs->is_app_limited) { + bbr->inflight_hi = max_t(u32, rs->tx_in_flight, + (u64)bbr_target_inflight(sk) * + (BBR_UNIT - beta) >> BBR_SCALE); + } + if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == BBR_BW_PROBE_UP) + bbr_start_bw_probe_down(sk); +} + +/* If we're seeing bw and loss samples reflecting our bw probing, adapt + * using the signals we see. If loss or ECN mark rate gets too high, then adapt + * inflight_hi downward. If we're able to push inflight higher without such + * signals, push higher: adapt inflight_hi upward. + */ +static bool bbr_adapt_upper_bounds(struct sock *sk, + const struct rate_sample *rs, + struct bbr_context *ctx) +{ + struct bbr *bbr = inet_csk_ca(sk); + + /* Track when we'll see bw/loss samples resulting from our bw probes. */ + if (bbr->ack_phase == BBR_ACKS_PROBE_STARTING && bbr->round_start) + bbr->ack_phase = BBR_ACKS_PROBE_FEEDBACK; + if (bbr->ack_phase == BBR_ACKS_PROBE_STOPPING && bbr->round_start) { + /* End of samples from bw probing phase. */ + bbr->bw_probe_samples = 0; + bbr->ack_phase = BBR_ACKS_INIT; + /* At this point in the cycle, our current bw sample is also + * our best recent chance at finding the highest available bw + * for this flow. So now is the best time to forget the bw + * samples from the previous cycle, by advancing the window. + */ + if (bbr->mode == BBR_PROBE_BW && !rs->is_app_limited) + bbr_advance_max_bw_filter(sk); + /* If we had an inflight_hi, then probed and pushed inflight all + * the way up to hit that inflight_hi without seeing any + * high loss/ECN in all the resulting ACKs from that probing, + * then probe up again, this time letting inflight persist at + * inflight_hi for a round trip, then accelerating beyond. + */ + if (bbr->mode == BBR_PROBE_BW && + bbr->stopped_risky_probe && !bbr->prev_probe_too_high) { + bbr_start_bw_probe_refill(sk, 0); + return true; /* yes, decided state transition */ + } + } + if (bbr_is_inflight_too_high(sk, rs)) { + if (bbr->bw_probe_samples) /* sample is from bw probing? */ + bbr_handle_inflight_too_high(sk, rs); + } else { + /* Loss/ECN rate is declared safe. Adjust upper bound upward. */ + + if (bbr->inflight_hi == ~0U) + return false; /* no excess queue signals yet */ + + /* To be resilient to random loss, we must raise bw/inflight_hi + * if we observe in any phase that a higher level is safe. + */ + if (rs->tx_in_flight > bbr->inflight_hi) { + bbr->inflight_hi = rs->tx_in_flight; + } + + if (bbr->mode == BBR_PROBE_BW && + bbr->cycle_idx == BBR_BW_PROBE_UP) + bbr_probe_inflight_hi_upward(sk, rs); + } + + return false; +} + +/* Check if it's time to probe for bandwidth now, and if so, kick it off. */ +static bool bbr_check_time_to_probe_bw(struct sock *sk, + const struct rate_sample *rs) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 n; + + /* If we seem to be at an operating point where we are not seeing loss + * but we are seeing ECN marks, then when the ECN marks cease we reprobe + * quickly (in case cross-traffic has ceased and freed up bw). + */ + if (bbr_param(sk, ecn_reprobe_gain) && bbr->ecn_eligible && + bbr->ecn_in_cycle && !bbr->loss_in_cycle && + inet_csk(sk)->icsk_ca_state == TCP_CA_Open) { + /* Calculate n so that when bbr_raise_inflight_hi_slope() + * computes growth_this_round as 2^n it will be roughly the + * desired volume of data (inflight_hi*ecn_reprobe_gain). + */ + n = ilog2((((u64)bbr->inflight_hi * + bbr_param(sk, ecn_reprobe_gain)) >> BBR_SCALE)); + bbr_start_bw_probe_refill(sk, n); + return true; + } + + if (bbr_has_elapsed_in_phase(sk, bbr->probe_wait_us) || + bbr_is_reno_coexistence_probe_time(sk)) { + bbr_start_bw_probe_refill(sk, 0); + return true; + } + return false; +} + +/* Is it time to transition from PROBE_DOWN to PROBE_CRUISE? */ +static bool bbr_check_time_to_cruise(struct sock *sk, u32 inflight, u32 bw) +{ + /* Always need to pull inflight down to leave headroom in queue. */ + if (inflight > bbr_inflight_with_headroom(sk)) + return false; + + return inflight <= bbr_inflight(sk, bw, BBR_UNIT); +} + +/* PROBE_BW state machine: cruise, refill, probe for bw, or drain? */ +static void bbr_update_cycle_phase(struct sock *sk, + const struct rate_sample *rs, + struct bbr_context *ctx) { + struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); + bool is_bw_probe_done = false; + u32 inflight, bw; + + if (!bbr_full_bw_reached(sk)) + return; + + /* In DRAIN, PROBE_BW, or PROBE_RTT, adjust upper bounds. */ + if (bbr_adapt_upper_bounds(sk, rs, ctx)) + return; /* already decided state transition */ + + if (bbr->mode != BBR_PROBE_BW) + return; + + inflight = bbr_packets_in_net_at_edt(sk, rs->prior_in_flight); + bw = bbr_max_bw(sk); + + switch (bbr->cycle_idx) { + /* First we spend most of our time cruising with a pacing_gain of 1.0, + * which paces at the estimated bw, to try to fully use the pipe + * without building queue. If we encounter loss/ECN marks, we adapt + * by slowing down. + */ + case BBR_BW_PROBE_CRUISE: + if (bbr_check_time_to_probe_bw(sk, rs)) + return; /* already decided state transition */ + break; + + /* After cruising, when it's time to probe, we first "refill": we send + * at the estimated bw to fill the pipe, before probing higher and + * knowingly risking overflowing the bottleneck buffer (causing loss). + */ + case BBR_BW_PROBE_REFILL: + if (bbr->round_start) { + /* After one full round trip of sending in REFILL, we + * start to see bw samples reflecting our REFILL, which + * may be putting too much data in flight. + */ + bbr->bw_probe_samples = 1; + bbr_start_bw_probe_up(sk, ctx); + } + break; - bbr->full_bw = 0; /* spurious slow-down; reset full pipe detection */ + /* After we refill the pipe, we probe by using a pacing_gain > 1.0, to + * probe for bw. If we have not seen loss/ECN, we try to raise inflight + * to at least pacing_gain*BDP; note that this may take more than + * min_rtt if min_rtt is small (e.g. on a LAN). + * + * We terminate PROBE_UP bandwidth probing upon any of the following: + * + * (1) We've pushed inflight up to hit the inflight_hi target set in the + * most recent previous bw probe phase. Thus we want to start + * draining the queue immediately because it's very likely the most + * recently sent packets will fill the queue and cause drops. + * (2) If inflight_hi has not limited bandwidth growth recently, and + * yet delivered bandwidth has not increased much recently + * (bbr->full_bw_now). + * (3) Loss filter says loss rate is "too high". + * (4) ECN filter says ECN mark rate is "too high". + * + * (1) (2) checked here, (3) (4) checked in bbr_is_inflight_too_high() + */ + case BBR_BW_PROBE_UP: + if (bbr->prev_probe_too_high && + inflight >= bbr->inflight_hi) { + bbr->stopped_risky_probe = 1; + is_bw_probe_done = true; + } else { + if (tp->is_cwnd_limited && + tcp_snd_cwnd(tp) >= bbr->inflight_hi) { + /* inflight_hi is limiting bw growth */ + bbr_reset_full_bw(sk); + bbr->full_bw = ctx->sample_bw; + } else if (bbr->full_bw_now) { + /* Plateau in estimated bw. Pipe looks full. */ + is_bw_probe_done = true; + } + } + if (is_bw_probe_done) { + bbr->prev_probe_too_high = 0; /* no loss/ECN (yet) */ + bbr_start_bw_probe_down(sk); /* restart w/ down */ + } + break; + + /* After probing in PROBE_UP, we have usually accumulated some data in + * the bottleneck buffer (if bw probing didn't find more bw). We next + * enter PROBE_DOWN to try to drain any excess data from the queue. To + * do this, we use a pacing_gain < 1.0. We hold this pacing gain until + * our inflight is less then that target cruising point, which is the + * minimum of (a) the amount needed to leave headroom, and (b) the + * estimated BDP. Once inflight falls to match the target, we estimate + * the queue is drained; persisting would underutilize the pipe. + */ + case BBR_BW_PROBE_DOWN: + if (bbr_check_time_to_probe_bw(sk, rs)) + return; /* already decided state transition */ + if (bbr_check_time_to_cruise(sk, inflight, bw)) + bbr_start_bw_probe_cruise(sk); + break; + + default: + WARN_ONCE(1, "BBR invalid cycle index %u\n", bbr->cycle_idx); + } +} + +/* Exiting PROBE_RTT, so return to bandwidth probing in STARTUP or PROBE_BW. */ +static void bbr_exit_probe_rtt(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + + bbr_reset_lower_bounds(sk); + if (bbr_full_bw_reached(sk)) { + bbr->mode = BBR_PROBE_BW; + /* Raising inflight after PROBE_RTT may cause loss, so reset + * the PROBE_BW clock and schedule the next bandwidth probe for + * a friendly and randomized future point in time. + */ + bbr_start_bw_probe_down(sk); + /* Since we are exiting PROBE_RTT, we know inflight is + * below our estimated BDP, so it is reasonable to cruise. + */ + bbr_start_bw_probe_cruise(sk); + } else { + bbr->mode = BBR_STARTUP; + } +} + +/* Exit STARTUP based on loss rate > 1% and loss gaps in round >= N. Wait until + * the end of the round in recovery to get a good estimate of how many packets + * have been lost, and how many we need to drain with a low pacing rate. + */ +static void bbr_check_loss_too_high_in_startup(struct sock *sk, + const struct rate_sample *rs) +{ + struct bbr *bbr = inet_csk_ca(sk); + + if (bbr_full_bw_reached(sk)) + return; + + /* For STARTUP exit, check the loss rate at the end of each round trip + * of Recovery episodes in STARTUP. We check the loss rate at the end + * of the round trip to filter out noisy/low loss and have a better + * sense of inflight (extent of loss), so we can drain more accurately. + */ + if (rs->losses && bbr->loss_events_in_round < 0xf) + bbr->loss_events_in_round++; /* update saturating counter */ + if (bbr_param(sk, full_loss_cnt) && bbr->loss_round_start && + inet_csk(sk)->icsk_ca_state == TCP_CA_Recovery && + bbr->loss_events_in_round >= bbr_param(sk, full_loss_cnt) && + bbr_is_inflight_too_high(sk, rs)) { + bbr_handle_queue_too_high_in_startup(sk); + return; + } + if (bbr->loss_round_start) + bbr->loss_events_in_round = 0; +} + +/* Estimate when the pipe is full, using the change in delivery rate: BBR + * estimates bw probing filled the pipe if the estimated bw hasn't changed by + * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited + * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the + * higher rwin, 3: we get higher delivery rate samples. Or transient + * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar + * design goal, but uses delay and inter-ACK spacing instead of bandwidth. + */ +static void bbr_check_full_bw_reached(struct sock *sk, + const struct rate_sample *rs, + struct bbr_context *ctx) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 bw_thresh, full_cnt, thresh; + + if (bbr->full_bw_now || rs->is_app_limited) + return; + + thresh = bbr_param(sk, full_bw_thresh); + full_cnt = bbr_param(sk, full_bw_cnt); + bw_thresh = (u64)bbr->full_bw * thresh >> BBR_SCALE; + if (ctx->sample_bw >= bw_thresh) { + bbr_reset_full_bw(sk); + bbr->full_bw = ctx->sample_bw; + return; + } + if (!bbr->round_start) + return; + ++bbr->full_bw_cnt; + bbr->full_bw_now = bbr->full_bw_cnt >= full_cnt; + bbr->full_bw_reached |= bbr->full_bw_now; +} + +/* If pipe is probably full, drain the queue and then enter steady-state. */ +static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs, + struct bbr_context *ctx) +{ + struct bbr *bbr = inet_csk_ca(sk); + + if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) { + bbr->mode = BBR_DRAIN; /* drain queue we created */ + /* Set ssthresh to export purely for monitoring, to signal + * completion of initial STARTUP by setting to a non- + * TCP_INFINITE_SSTHRESH value (ssthresh is not used by BBR). + */ + tcp_sk(sk)->snd_ssthresh = + bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT); + bbr_reset_congestion_signals(sk); + } /* fall through to check if in-flight is already small: */ + if (bbr->mode == BBR_DRAIN && + bbr_packets_in_net_at_edt(sk, tcp_packets_in_flight(tcp_sk(sk))) <= + bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT)) { + bbr->mode = BBR_PROBE_BW; + bbr_start_bw_probe_down(sk); + } +} + +static void bbr_update_model(struct sock *sk, const struct rate_sample *rs, + struct bbr_context *ctx) +{ + bbr_update_congestion_signals(sk, rs, ctx); + bbr_update_ack_aggregation(sk, rs); + bbr_check_loss_too_high_in_startup(sk, rs); + bbr_check_full_bw_reached(sk, rs, ctx); + bbr_check_drain(sk, rs, ctx); + bbr_update_cycle_phase(sk, rs, ctx); + bbr_update_min_rtt(sk, rs); +} + +/* Fast path for app-limited case. + * + * On each ack, we execute bbr state machine, which primarily consists of: + * 1) update model based on new rate sample, and + * 2) update control based on updated model or state change. + * + * There are certain workload/scenarios, e.g. app-limited case, where + * either we can skip updating model or we can skip update of both model + * as well as control. This provides signifcant softirq cpu savings for + * processing incoming acks. + * + * In case of app-limited, if there is no congestion (loss/ecn) and + * if observed bw sample is less than current estimated bw, then we can + * skip some of the computation in bbr state processing: + * + * - if there is no rtt/mode/phase change: In this case, since all the + * parameters of the network model are constant, we can skip model + * as well control update. + * + * - else we can skip rest of the model update. But we still need to + * update the control to account for the new rtt/mode/phase. + * + * Returns whether we can take fast path or not. + */ +static bool bbr_run_fast_path(struct sock *sk, bool *update_model, + const struct rate_sample *rs, struct bbr_context *ctx) +{ + struct bbr *bbr = inet_csk_ca(sk); + u32 prev_min_rtt_us, prev_mode; + + if (bbr_param(sk, fast_path) && bbr->try_fast_path && + rs->is_app_limited && ctx->sample_bw < bbr_max_bw(sk) && + !bbr->loss_in_round && !bbr->ecn_in_round ) { + prev_mode = bbr->mode; + prev_min_rtt_us = bbr->min_rtt_us; + bbr_check_drain(sk, rs, ctx); + bbr_update_cycle_phase(sk, rs, ctx); + bbr_update_min_rtt(sk, rs); + + if (bbr->mode == prev_mode && + bbr->min_rtt_us == prev_min_rtt_us && + bbr->try_fast_path) { + return true; + } + + /* Skip model update, but control still needs to be updated */ + *update_model = false; + } + return false; +} + +__bpf_kfunc static void bbr_main(struct sock *sk, u32 ack, int flag, + const struct rate_sample *rs) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + struct bbr_context ctx = { 0 }; + bool update_model = true; + u32 bw, round_delivered; + int ce_ratio = -1; + + round_delivered = bbr_update_round_start(sk, rs, &ctx); + if (bbr->round_start) { + bbr->rounds_since_probe = + min_t(s32, bbr->rounds_since_probe + 1, 0xFF); + ce_ratio = bbr_update_ecn_alpha(sk); + } + bbr_plb(sk, rs, ce_ratio); + + bbr->ecn_in_round |= (bbr->ecn_eligible && rs->is_ece); + bbr_calculate_bw_sample(sk, rs, &ctx); + bbr_update_latest_delivery_signals(sk, rs, &ctx); + + if (bbr_run_fast_path(sk, &update_model, rs, &ctx)) + goto out; + + if (update_model) + bbr_update_model(sk, rs, &ctx); + + bbr_update_gains(sk); + bw = bbr_bw(sk); + bbr_set_pacing_rate(sk, bw, bbr->pacing_gain); + bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain, + tcp_snd_cwnd(tp), &ctx); + bbr_bound_cwnd_for_inflight_model(sk); + +out: + bbr_advance_latest_delivery_signals(sk, rs, &ctx); + bbr->prev_ca_state = inet_csk(sk)->icsk_ca_state; + bbr->loss_in_cycle |= rs->lost > 0; + bbr->ecn_in_cycle |= rs->delivered_ce > 0; +} + +__bpf_kfunc static void bbr_init(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + bbr->initialized = 1; + + bbr->init_cwnd = min(0x7FU, tcp_snd_cwnd(tp)); + bbr->prior_cwnd = tp->prior_cwnd; + tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; + bbr->next_rtt_delivered = tp->delivered; + bbr->prev_ca_state = TCP_CA_Open; + + bbr->probe_rtt_done_stamp = 0; + bbr->probe_rtt_round_done = 0; + bbr->probe_rtt_min_us = tcp_min_rtt(tp); + bbr->probe_rtt_min_stamp = tcp_jiffies32; + bbr->min_rtt_us = tcp_min_rtt(tp); + bbr->min_rtt_stamp = tcp_jiffies32; + + bbr->has_seen_rtt = 0; + bbr_init_pacing_rate_from_rtt(sk); + + bbr->round_start = 0; + bbr->idle_restart = 0; + bbr->full_bw_reached = 0; + bbr->full_bw = 0; bbr->full_bw_cnt = 0; - bbr_reset_lt_bw_sampling(sk); - return tcp_snd_cwnd(tcp_sk(sk)); + bbr->cycle_mstamp = 0; + bbr->cycle_idx = 0; + + bbr_reset_startup_mode(sk); + + bbr->ack_epoch_mstamp = tp->tcp_mstamp; + bbr->ack_epoch_acked = 0; + bbr->extra_acked_win_rtts = 0; + bbr->extra_acked_win_idx = 0; + bbr->extra_acked[0] = 0; + bbr->extra_acked[1] = 0; + + bbr->ce_state = 0; + bbr->prior_rcv_nxt = tp->rcv_nxt; + bbr->try_fast_path = 0; + + cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED); + + /* Start sampling ECN mark rate after first full flight is ACKed: */ + bbr->loss_round_delivered = tp->delivered + 1; + bbr->loss_round_start = 0; + bbr->undo_bw_lo = 0; + bbr->undo_inflight_lo = 0; + bbr->undo_inflight_hi = 0; + bbr->loss_events_in_round = 0; + bbr->startup_ecn_rounds = 0; + bbr_reset_congestion_signals(sk); + bbr->bw_lo = ~0U; + bbr->bw_hi[0] = 0; + bbr->bw_hi[1] = 0; + bbr->inflight_lo = ~0U; + bbr->inflight_hi = ~0U; + bbr_reset_full_bw(sk); + bbr->bw_probe_up_cnt = ~0U; + bbr->bw_probe_up_acks = 0; + bbr->bw_probe_up_rounds = 0; + bbr->probe_wait_us = 0; + bbr->stopped_risky_probe = 0; + bbr->ack_phase = BBR_ACKS_INIT; + bbr->rounds_since_probe = 0; + bbr->bw_probe_samples = 0; + bbr->prev_probe_too_high = 0; + bbr->ecn_eligible = 0; + bbr->ecn_alpha = bbr_param(sk, ecn_alpha_init); + bbr->alpha_last_delivered = 0; + bbr->alpha_last_delivered_ce = 0; + bbr->plb.pause_until = 0; + + tp->fast_ack_mode = bbr_fast_ack_mode ? 1 : 0; + + if (bbr_can_use_ecn(sk)) + tp->ecn_flags |= TCP_ECN_ECT_PERMANENT; +} + +/* BBR marks the current round trip as a loss round. */ +static void bbr_note_loss(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + /* Capture "current" data over the full round trip of loss, to + * have a better chance of observing the full capacity of the path. + */ + if (!bbr->loss_in_round) /* first loss in this round trip? */ + bbr->loss_round_delivered = tp->delivered; /* set round trip */ + bbr->loss_in_round = 1; + bbr->loss_in_cycle = 1; } -/* Entering loss recovery, so save cwnd for when we exit or undo recovery. */ +/* Core TCP stack informs us that the given skb was just marked lost. */ +__bpf_kfunc static void bbr_skb_marked_lost(struct sock *sk, + const struct sk_buff *skb) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + struct tcp_skb_cb *scb = TCP_SKB_CB(skb); + struct rate_sample rs = {}; + + bbr_note_loss(sk); + + if (!bbr->bw_probe_samples) + return; /* not an skb sent while probing for bandwidth */ + if (unlikely(!scb->tx.delivered_mstamp)) + return; /* skb was SACKed, reneged, marked lost; ignore it */ + /* We are probing for bandwidth. Construct a rate sample that + * estimates what happened in the flight leading up to this lost skb, + * then see if the loss rate went too high, and if so at which packet. + */ + rs.tx_in_flight = scb->tx.in_flight; + rs.lost = tp->lost - scb->tx.lost; + rs.is_app_limited = scb->tx.is_app_limited; + if (bbr_is_inflight_too_high(sk, &rs)) { + rs.tx_in_flight = bbr_inflight_hi_from_lost_skb(sk, &rs, skb); + bbr_handle_inflight_too_high(sk, &rs); + } +} + +static void bbr_run_loss_probe_recovery(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + struct rate_sample rs = {0}; + + bbr_note_loss(sk); + + if (!bbr->bw_probe_samples) + return; /* not sent while probing for bandwidth */ + /* We are probing for bandwidth. Construct a rate sample that + * estimates what happened in the flight leading up to this + * loss, then see if the loss rate went too high. + */ + rs.lost = 1; /* TLP probe repaired loss of a single segment */ + rs.tx_in_flight = bbr->inflight_latest + rs.lost; + rs.is_app_limited = tp->tlp_orig_data_app_limited; + if (bbr_is_inflight_too_high(sk, &rs)) + bbr_handle_inflight_too_high(sk, &rs); +} + +/* Revert short-term model if current loss recovery event was spurious. */ +__bpf_kfunc static u32 bbr_undo_cwnd(struct sock *sk) +{ + struct bbr *bbr = inet_csk_ca(sk); + + bbr_reset_full_bw(sk); /* spurious slow-down; reset full bw detector */ + bbr->loss_in_round = 0; + + /* Revert to cwnd and other state saved before loss episode. */ + bbr->bw_lo = max(bbr->bw_lo, bbr->undo_bw_lo); + bbr->inflight_lo = max(bbr->inflight_lo, bbr->undo_inflight_lo); + bbr->inflight_hi = max(bbr->inflight_hi, bbr->undo_inflight_hi); + bbr->try_fast_path = 0; /* take slow path to set proper cwnd, pacing */ + return bbr->prior_cwnd; +} + +/* Entering loss recovery, so save state for when we undo recovery. */ __bpf_kfunc static u32 bbr_ssthresh(struct sock *sk) { + struct bbr *bbr = inet_csk_ca(sk); + bbr_save_cwnd(sk); + /* For undo, save state that adapts based on loss signal. */ + bbr->undo_bw_lo = bbr->bw_lo; + bbr->undo_inflight_lo = bbr->inflight_lo; + bbr->undo_inflight_hi = bbr->inflight_hi; return tcp_sk(sk)->snd_ssthresh; } +static enum tcp_bbr_phase bbr_get_phase(struct bbr *bbr) +{ + switch (bbr->mode) { + case BBR_STARTUP: + return BBR_PHASE_STARTUP; + case BBR_DRAIN: + return BBR_PHASE_DRAIN; + case BBR_PROBE_BW: + break; + case BBR_PROBE_RTT: + return BBR_PHASE_PROBE_RTT; + default: + return BBR_PHASE_INVALID; + } + switch (bbr->cycle_idx) { + case BBR_BW_PROBE_UP: + return BBR_PHASE_PROBE_BW_UP; + case BBR_BW_PROBE_DOWN: + return BBR_PHASE_PROBE_BW_DOWN; + case BBR_BW_PROBE_CRUISE: + return BBR_PHASE_PROBE_BW_CRUISE; + case BBR_BW_PROBE_REFILL: + return BBR_PHASE_PROBE_BW_REFILL; + default: + return BBR_PHASE_INVALID; + } +} + static size_t bbr_get_info(struct sock *sk, u32 ext, int *attr, - union tcp_cc_info *info) + union tcp_cc_info *info) { if (ext & (1 << (INET_DIAG_BBRINFO - 1)) || ext & (1 << (INET_DIAG_VEGASINFO - 1))) { - struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); - u64 bw = bbr_bw(sk); - - bw = bw * tp->mss_cache * USEC_PER_SEC >> BW_SCALE; - memset(&info->bbr, 0, sizeof(info->bbr)); - info->bbr.bbr_bw_lo = (u32)bw; - info->bbr.bbr_bw_hi = (u32)(bw >> 32); - info->bbr.bbr_min_rtt = bbr->min_rtt_us; - info->bbr.bbr_pacing_gain = bbr->pacing_gain; - info->bbr.bbr_cwnd_gain = bbr->cwnd_gain; + u64 bw = bbr_bw_bytes_per_sec(sk, bbr_bw(sk)); + u64 bw_hi = bbr_bw_bytes_per_sec(sk, bbr_max_bw(sk)); + u64 bw_lo = bbr->bw_lo == ~0U ? + ~0ULL : bbr_bw_bytes_per_sec(sk, bbr->bw_lo); + struct tcp_bbr_info *bbr_info = &info->bbr; + + memset(bbr_info, 0, sizeof(*bbr_info)); + bbr_info->bbr_bw_lo = (u32)bw; + bbr_info->bbr_bw_hi = (u32)(bw >> 32); + bbr_info->bbr_min_rtt = bbr->min_rtt_us; + bbr_info->bbr_pacing_gain = bbr->pacing_gain; + bbr_info->bbr_cwnd_gain = bbr->cwnd_gain; + bbr_info->bbr_bw_hi_lsb = (u32)bw_hi; + bbr_info->bbr_bw_hi_msb = (u32)(bw_hi >> 32); + bbr_info->bbr_bw_lo_lsb = (u32)bw_lo; + bbr_info->bbr_bw_lo_msb = (u32)(bw_lo >> 32); + bbr_info->bbr_mode = bbr->mode; + bbr_info->bbr_phase = (__u8)bbr_get_phase(bbr); + bbr_info->bbr_version = (__u8)BBR_VERSION; + bbr_info->bbr_inflight_lo = bbr->inflight_lo; + bbr_info->bbr_inflight_hi = bbr->inflight_hi; + bbr_info->bbr_extra_acked = bbr_extra_acked(sk); *attr = INET_DIAG_BBRINFO; - return sizeof(info->bbr); + return sizeof(*bbr_info); } return 0; } __bpf_kfunc static void bbr_set_state(struct sock *sk, u8 new_state) { + struct tcp_sock *tp = tcp_sk(sk); struct bbr *bbr = inet_csk_ca(sk); if (new_state == TCP_CA_Loss) { - struct rate_sample rs = { .losses = 1 }; bbr->prev_ca_state = TCP_CA_Loss; - bbr->full_bw = 0; - bbr->round_start = 1; /* treat RTO like end of a round */ - bbr_lt_bw_sampling(sk, &rs); + tcp_plb_update_state_upon_rto(sk, &bbr->plb); + /* The tcp_write_timeout() call to sk_rethink_txhash() likely + * repathed this flow, so re-learn the min network RTT on the + * new path: + */ + bbr_reset_full_bw(sk); + if (!bbr_is_probing_bandwidth(sk) && bbr->inflight_lo == ~0U) { + /* bbr_adapt_lower_bounds() needs cwnd before + * we suffered an RTO, to update inflight_lo: + */ + bbr->inflight_lo = + max(tcp_snd_cwnd(tp), bbr->prior_cwnd); + } + } else if (bbr->prev_ca_state == TCP_CA_Loss && + new_state != TCP_CA_Loss) { + bbr_exit_loss_recovery(sk); } } + static struct tcp_congestion_ops tcp_bbr_cong_ops __read_mostly = { - .flags = TCP_CONG_NON_RESTRICTED, + .flags = TCP_CONG_NON_RESTRICTED | TCP_CONG_WANTS_CE_EVENTS, .name = "bbr", .owner = THIS_MODULE, .init = bbr_init, .cong_control = bbr_main, .sndbuf_expand = bbr_sndbuf_expand, + .skb_marked_lost = bbr_skb_marked_lost, .undo_cwnd = bbr_undo_cwnd, .cwnd_event = bbr_cwnd_event, .ssthresh = bbr_ssthresh, - .min_tso_segs = bbr_min_tso_segs, + .tso_segs = bbr_tso_segs, .get_info = bbr_get_info, .set_state = bbr_set_state, }; @@ -1159,10 +2362,11 @@ BTF_KFUNCS_START(tcp_bbr_check_kfunc_ids) BTF_ID_FLAGS(func, bbr_init) BTF_ID_FLAGS(func, bbr_main) BTF_ID_FLAGS(func, bbr_sndbuf_expand) +BTF_ID_FLAGS(func, bbr_skb_marked_lost) BTF_ID_FLAGS(func, bbr_undo_cwnd) BTF_ID_FLAGS(func, bbr_cwnd_event) BTF_ID_FLAGS(func, bbr_ssthresh) -BTF_ID_FLAGS(func, bbr_min_tso_segs) +BTF_ID_FLAGS(func, bbr_tso_segs) BTF_ID_FLAGS(func, bbr_set_state) BTF_KFUNCS_END(tcp_bbr_check_kfunc_ids) @@ -1195,5 +2399,12 @@ MODULE_AUTHOR("Van Jacobson "); MODULE_AUTHOR("Neal Cardwell "); MODULE_AUTHOR("Yuchung Cheng "); MODULE_AUTHOR("Soheil Hassas Yeganeh "); +MODULE_AUTHOR("Priyaranjan Jha "); +MODULE_AUTHOR("Yousuk Seung "); +MODULE_AUTHOR("Kevin Yang "); +MODULE_AUTHOR("Arjun Roy "); +MODULE_AUTHOR("David Morley "); + MODULE_LICENSE("Dual BSD/GPL"); MODULE_DESCRIPTION("TCP BBR (Bottleneck Bandwidth and RTT)"); +MODULE_VERSION(__stringify(BBR_VERSION)); diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c index df758adbb445..e98e5dbc050e 100644 --- a/net/ipv4/tcp_cong.c +++ b/net/ipv4/tcp_cong.c @@ -237,6 +237,7 @@ void tcp_init_congestion_control(struct sock *sk) struct inet_connection_sock *icsk = inet_csk(sk); tcp_sk(sk)->prior_ssthresh = 0; + tcp_sk(sk)->fast_ack_mode = 0; if (icsk->icsk_ca_ops->init) icsk->icsk_ca_ops->init(sk); if (tcp_ca_needs_ecn(sk)) diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 198f8a0d37be..a965bde56488 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -358,7 +358,7 @@ static void tcp_data_ecn_check(struct sock *sk, const struct sk_buff *skb) tcp_enter_quickack_mode(sk, 2); break; case INET_ECN_CE: - if (tcp_ca_needs_ecn(sk)) + if (tcp_ca_wants_ce_events(sk)) tcp_ca_event(sk, CA_EVENT_ECN_IS_CE); if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR) && @@ -376,7 +376,7 @@ static void tcp_data_ecn_check(struct sock *sk, const struct sk_buff *skb) tp->ecn_flags |= TCP_ECN_SEEN; break; default: - if (tcp_ca_needs_ecn(sk)) + if (tcp_ca_wants_ce_events(sk)) tcp_ca_event(sk, CA_EVENT_ECN_NO_CE); if (!tcp_ecn_mode_rfc3168(tp)) break; @@ -1305,7 +1305,12 @@ static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb) */ static void tcp_notify_skb_loss_event(struct tcp_sock *tp, const struct sk_buff *skb) { + struct sock *sk = (struct sock *)tp; + const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; + tp->lost += tcp_skb_pcount(skb); + if (ca_ops->skb_marked_lost) + ca_ops->skb_marked_lost(sk, skb); } void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb) @@ -1670,6 +1675,17 @@ static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *prev, WARN_ON_ONCE(tcp_skb_pcount(skb) < pcount); tcp_skb_pcount_add(skb, -pcount); + /* Adjust tx.in_flight as pcount is shifted from skb to prev. */ + if (WARN_ONCE(TCP_SKB_CB(skb)->tx.in_flight < pcount, + "prev in_flight: %u skb in_flight: %u pcount: %u", + TCP_SKB_CB(prev)->tx.in_flight, + TCP_SKB_CB(skb)->tx.in_flight, + pcount)) + TCP_SKB_CB(skb)->tx.in_flight = 0; + else + TCP_SKB_CB(skb)->tx.in_flight -= pcount; + TCP_SKB_CB(prev)->tx.in_flight += pcount; + /* When we're adding to gso_segs == 1, gso_size will be zero, * in theory this shouldn't be necessary but as long as DSACK * code can come after this skb later on it's better to keep @@ -3905,7 +3921,8 @@ static int tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq) /* This routine deals with acks during a TLP episode and ends an episode by * resetting tlp_high_seq. Ref: TLP algorithm in RFC8985 */ -static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) +static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag, + struct rate_sample *rs) { struct tcp_sock *tp = tcp_sk(sk); @@ -3922,6 +3939,7 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) /* ACK advances: there was a loss, so reduce cwnd. Reset * tlp_high_seq in tcp_init_cwnd_reduction() */ + tcp_ca_event(sk, CA_EVENT_TLP_RECOVERY); tcp_init_cwnd_reduction(sk); tcp_set_ca_state(sk, TCP_CA_CWR); tcp_end_cwnd_reduction(sk); @@ -3932,6 +3950,11 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) FLAG_NOT_DUP | FLAG_DATA_SACKED))) { /* Pure dupack: original and TLP probe arrived; no loss */ tp->tlp_high_seq = 0; + } else { + /* This ACK matches a TLP retransmit. We cannot yet tell if + * this ACK is for the original or the TLP retransmit. + */ + rs->is_acking_tlp_retrans_seq = 1; } } @@ -4058,6 +4081,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una; rs.prior_in_flight = tcp_packets_in_flight(tp); + tcp_rate_check_app_limited(sk); /* ts_recent update must be made after we are sure that the packet * is in window. @@ -4130,7 +4154,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) tcp_in_ack_event(sk, flag); if (tp->tlp_high_seq) - tcp_process_tlp_ack(sk, ack, flag); + tcp_process_tlp_ack(sk, ack, flag, &rs); if (tcp_ack_is_dubious(sk, flag)) { if (!(flag & (FLAG_SND_UNA_ADVANCED | @@ -4155,6 +4179,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) lost = tp->lost - lost; /* freshly marked lost */ rs.is_ack_delayed = !!(flag & FLAG_ACK_MAYBE_DELAYED); + rs.is_ece = !!(flag & FLAG_ECE); tcp_rate_gen(sk, delivered, lost, is_sack_reneg, sack_state.rate); tcp_cong_control(sk, ack, delivered, flag, sack_state.rate); tcp_xmit_recovery(sk, rexmit); @@ -4180,7 +4205,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) tcp_ack_probe(sk); if (tp->tlp_high_seq) - tcp_process_tlp_ack(sk, ack, flag); + tcp_process_tlp_ack(sk, ack, flag, &rs); return 1; old_ack: @@ -5909,13 +5934,14 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible) /* More than one full frame received... */ if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss && + (tp->fast_ack_mode == 1 || /* ... and right edge of window advances far enough. * (tcp_recvmsg() will send ACK otherwise). * If application uses SO_RCVLOWAT, we want send ack now if * we have not received enough bytes to satisfy the condition. */ - (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat || - __tcp_select_window(sk) >= tp->rcv_wnd)) || + (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat || + __tcp_select_window(sk) >= tp->rcv_wnd))) || /* We ACK each frame or... */ tcp_in_quickack_mode(sk) || /* Protocol state mandates a one-time immediate ACK */ diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c index bd5462154f97..0be0e7a79f1e 100644 --- a/net/ipv4/tcp_minisocks.c +++ b/net/ipv4/tcp_minisocks.c @@ -497,6 +497,8 @@ void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst) u32 ca_key = dst_metric(dst, RTAX_CC_ALGO); bool ca_got_dst = false; + tcp_set_ecn_low_from_dst(sk, dst); + if (ca_key != TCP_CA_UNSPEC) { const struct tcp_congestion_ops *ca; diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 479afb714bdf..a9eb14d0cf47 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -348,7 +348,8 @@ static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb, th->cwr = 1; skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; } - } else if (!tcp_ca_needs_ecn(sk)) { + } else if (!(tp->ecn_flags & TCP_ECN_ECT_PERMANENT) && + !tcp_ca_needs_ecn(sk)) { /* ACK or retransmitted segment: clear ECT|CE */ INET_ECN_dontxmit(sk); } @@ -1762,7 +1763,7 @@ int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, { struct tcp_sock *tp = tcp_sk(sk); struct sk_buff *buff; - int old_factor; + int old_factor, inflight_prev; long limit; u16 flags; int nlen; @@ -1837,6 +1838,30 @@ int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, if (diff) tcp_adjust_pcount(sk, skb, diff); + + inflight_prev = TCP_SKB_CB(skb)->tx.in_flight - old_factor; + if (inflight_prev < 0) { + WARN_ONCE(tcp_skb_tx_in_flight_is_suspicious( + old_factor, + TCP_SKB_CB(skb)->sacked, + TCP_SKB_CB(skb)->tx.in_flight), + "inconsistent: tx.in_flight: %u " + "old_factor: %d mss: %u sacked: %u " + "1st pcount: %d 2nd pcount: %d " + "1st len: %u 2nd len: %u ", + TCP_SKB_CB(skb)->tx.in_flight, old_factor, + mss_now, TCP_SKB_CB(skb)->sacked, + tcp_skb_pcount(skb), tcp_skb_pcount(buff), + skb->len, buff->len); + inflight_prev = 0; + } + /* Set 1st tx.in_flight as if 1st were sent by itself: */ + TCP_SKB_CB(skb)->tx.in_flight = inflight_prev + + tcp_skb_pcount(skb); + /* Set 2nd tx.in_flight with new 1st and 2nd pcounts: */ + TCP_SKB_CB(buff)->tx.in_flight = inflight_prev + + tcp_skb_pcount(skb) + + tcp_skb_pcount(buff); } /* Link BUFF into the send queue. */ @@ -2193,13 +2218,12 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now) { const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; - u32 min_tso, tso_segs; - - min_tso = ca_ops->min_tso_segs ? - ca_ops->min_tso_segs(sk) : - READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); + u32 tso_segs; - tso_segs = tcp_tso_autosize(sk, mss_now, min_tso); + tso_segs = ca_ops->tso_segs ? + ca_ops->tso_segs(sk, mss_now) : + tcp_tso_autosize(sk, mss_now, + sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); return min_t(u32, tso_segs, sk->sk_gso_max_segs); } @@ -2940,6 +2964,7 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle, skb_set_delivery_time(skb, tp->tcp_wstamp_ns, SKB_CLOCK_MONOTONIC); list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); tcp_init_tso_segs(skb, mss_now); + tcp_set_tx_in_flight(sk, skb); goto repair; /* Skip network transmission */ } @@ -3152,6 +3177,7 @@ void tcp_send_loss_probe(struct sock *sk) if (WARN_ON(!skb || !tcp_skb_pcount(skb))) goto rearm_timer; + tp->tlp_orig_data_app_limited = TCP_SKB_CB(skb)->tx.is_app_limited; if (__tcp_retransmit_skb(sk, skb, 1)) goto rearm_timer; diff --git a/net/ipv4/tcp_rate.c b/net/ipv4/tcp_rate.c index a8f6d9d06f2e..8737f2134648 100644 --- a/net/ipv4/tcp_rate.c +++ b/net/ipv4/tcp_rate.c @@ -34,6 +34,24 @@ * ready to send in the write queue. */ +void tcp_set_tx_in_flight(struct sock *sk, struct sk_buff *skb) +{ + struct tcp_sock *tp = tcp_sk(sk); + u32 in_flight; + + /* Check, sanitize, and record packets in flight after skb was sent. */ + in_flight = tcp_packets_in_flight(tp) + tcp_skb_pcount(skb); + if (WARN_ONCE(in_flight > TCPCB_IN_FLIGHT_MAX, + "insane in_flight %u cc %s mss %u " + "cwnd %u pif %u %u %u %u\n", + in_flight, inet_csk(sk)->icsk_ca_ops->name, + tp->mss_cache, tp->snd_cwnd, + tp->packets_out, tp->retrans_out, + tp->sacked_out, tp->lost_out)) + in_flight = TCPCB_IN_FLIGHT_MAX; + TCP_SKB_CB(skb)->tx.in_flight = in_flight; +} + /* Snapshot the current delivery information in the skb, to generate * a rate sample later when the skb is (s)acked in tcp_rate_skb_delivered(). */ @@ -66,7 +84,9 @@ void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb) TCP_SKB_CB(skb)->tx.delivered_mstamp = tp->delivered_mstamp; TCP_SKB_CB(skb)->tx.delivered = tp->delivered; TCP_SKB_CB(skb)->tx.delivered_ce = tp->delivered_ce; + TCP_SKB_CB(skb)->tx.lost = tp->lost; TCP_SKB_CB(skb)->tx.is_app_limited = tp->app_limited ? 1 : 0; + tcp_set_tx_in_flight(sk, skb); } /* When an skb is sacked or acked, we fill in the rate sample with the (prior) @@ -91,18 +111,21 @@ void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, if (!rs->prior_delivered || tcp_skb_sent_after(tx_tstamp, tp->first_tx_mstamp, scb->end_seq, rs->last_end_seq)) { + rs->prior_lost = scb->tx.lost; rs->prior_delivered_ce = scb->tx.delivered_ce; rs->prior_delivered = scb->tx.delivered; rs->prior_mstamp = scb->tx.delivered_mstamp; rs->is_app_limited = scb->tx.is_app_limited; rs->is_retrans = scb->sacked & TCPCB_RETRANS; + rs->tx_in_flight = scb->tx.in_flight; rs->last_end_seq = scb->end_seq; /* Record send time of most recently ACKed packet: */ tp->first_tx_mstamp = tx_tstamp; /* Find the duration of the "send phase" of this window: */ - rs->interval_us = tcp_stamp_us_delta(tp->first_tx_mstamp, - scb->tx.first_tx_mstamp); + rs->interval_us = tcp_stamp32_us_delta( + tp->first_tx_mstamp, + scb->tx.first_tx_mstamp); } /* Mark off the skb delivered once it's sacked to avoid being @@ -144,6 +167,7 @@ void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, return; } rs->delivered = tp->delivered - rs->prior_delivered; + rs->lost = tp->lost - rs->prior_lost; rs->delivered_ce = tp->delivered_ce - rs->prior_delivered_ce; /* delivered_ce occupies less than 32 bits in the skb control block */ @@ -155,7 +179,7 @@ void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, * longer phase. */ snd_us = rs->interval_us; /* send phase */ - ack_us = tcp_stamp_us_delta(tp->tcp_mstamp, + ack_us = tcp_stamp32_us_delta(tp->tcp_mstamp, rs->prior_mstamp); /* ack phase */ rs->interval_us = max(snd_us, ack_us); diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c index 160080c9021d..06ee74f2c01d 100644 --- a/net/ipv4/tcp_timer.c +++ b/net/ipv4/tcp_timer.c @@ -702,6 +702,7 @@ void tcp_write_timer_handler(struct sock *sk) tcp_timeout_expires(sk)); return; } + tcp_rate_check_app_limited(sk); tcp_mstamp_refresh(tcp_sk(sk)); event = icsk->icsk_pending; -- 2.53.0 From 2c17440fd39b3d57beb31b3527781edea3f57bff Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Mon, 16 Feb 2026 13:50:59 +0100 Subject: [PATCH 3/7] cachy Signed-off-by: Peter Jung --- .gitignore | 2 + .../admin-guide/kernel-parameters.txt | 12 + Documentation/admin-guide/sysctl/vm.rst | 72 + Makefile | 33 +- arch/Kconfig | 19 + arch/x86/Kconfig.cpu | 46 + arch/x86/Makefile | 16 +- arch/x86/include/asm/pci.h | 6 + arch/x86/kernel/cpu/bus_lock.c | 4 + arch/x86/pci/common.c | 7 +- block/Kconfig.iosched | 14 + block/Makefile | 8 + block/adios.c | 2004 ++++++++++ block/elevator.c | 24 +- drivers/Makefile | 13 +- drivers/ata/ahci.c | 23 +- drivers/cpufreq/Kconfig.x86 | 2 - drivers/cpufreq/intel_pstate.c | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 10 + drivers/gpu/drm/amd/display/Kconfig | 6 + .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- .../amd/display/amdgpu_dm/amdgpu_dm_color.c | 2 +- .../amd/display/amdgpu_dm/amdgpu_dm_crtc.c | 6 +- .../amd/display/amdgpu_dm/amdgpu_dm_plane.c | 6 +- drivers/gpu/drm/amd/pm/amdgpu_pm.c | 3 + drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c | 15 +- drivers/input/evdev.c | 19 +- drivers/media/v4l2-core/Kconfig | 5 + drivers/media/v4l2-core/Makefile | 2 + drivers/media/v4l2-core/v4l2loopback.c | 3331 +++++++++++++++++ drivers/media/v4l2-core/v4l2loopback.h | 108 + .../media/v4l2-core/v4l2loopback_formats.h | 445 +++ drivers/pci/controller/Makefile | 6 + drivers/pci/controller/intel-nvme-remap.c | 462 +++ drivers/pci/quirks.c | 101 + drivers/scsi/Kconfig | 2 + drivers/scsi/Makefile | 1 + drivers/scsi/vhba/Kconfig | 9 + drivers/scsi/vhba/Makefile | 4 + drivers/scsi/vhba/vhba.c | 1132 ++++++ include/linux/mm.h | 8 + include/linux/pagemap.h | 2 +- init/Kconfig | 26 + kernel/Kconfig.hz | 24 + kernel/Kconfig.preempt | 2 +- kernel/fork.c | 28 + kernel/locking/rwsem.c | 4 +- kernel/sched/fair.c | 79 +- kernel/sched/sched.h | 2 +- mm/Kconfig | 65 +- mm/compaction.c | 4 + mm/huge_memory.c | 4 + mm/mm_init.c | 1 + mm/page-writeback.c | 8 + mm/page_alloc.c | 4 + mm/swap.c | 5 + mm/util.c | 34 + mm/vmscan.c | 157 +- scripts/Makefile.lib | 7 + scripts/Makefile.thinlto | 40 + scripts/Makefile.vmlinux_a | 83 + scripts/mod/modpost.c | 15 +- scripts/package/PKGBUILD | 5 + 64 files changed, 8493 insertions(+), 99 deletions(-) create mode 100644 block/adios.c create mode 100644 drivers/media/v4l2-core/v4l2loopback.c create mode 100644 drivers/media/v4l2-core/v4l2loopback.h create mode 100644 drivers/media/v4l2-core/v4l2loopback_formats.h create mode 100644 drivers/pci/controller/intel-nvme-remap.c create mode 100644 drivers/scsi/vhba/Kconfig create mode 100644 drivers/scsi/vhba/Makefile create mode 100644 drivers/scsi/vhba/vhba.c create mode 100644 scripts/Makefile.thinlto create mode 100644 scripts/Makefile.vmlinux_a diff --git a/.gitignore b/.gitignore index 3a7241c941f5..c12bdf5a97f9 100644 --- a/.gitignore +++ b/.gitignore @@ -56,6 +56,7 @@ *.zst Module.symvers dtbs-list +builtin.order modules.order # @@ -67,6 +68,7 @@ modules.order /vmlinux.32 /vmlinux.map /vmlinux.symvers +/vmlinux.thinlto-index /vmlinux.unstripped /vmlinux-gdb.py /vmlinuz diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt index aa0031108bc1..28f14d664aa3 100644 --- a/Documentation/admin-guide/kernel-parameters.txt +++ b/Documentation/admin-guide/kernel-parameters.txt @@ -2520,6 +2520,9 @@ Kernel parameters disable Do not enable intel_pstate as the default scaling driver for the supported processors + enable + Enable intel_pstate in-case "disable" was passed + previously in the kernel boot parameters active Use intel_pstate driver to bypass the scaling governors layer of cpufreq and provides it own @@ -5010,6 +5013,15 @@ Kernel parameters nomsi [MSI] If the PCI_MSI kernel config parameter is enabled, this kernel boot option can be used to disable the use of MSI interrupts system-wide. + pcie_acs_override = + [PCIE] Override missing PCIe ACS support for: + downstream + All downstream ports - full ACS capabilities + multfunction + All multifunction devices - multifunction ACS subset + id:nnnn:nnnn + Specfic device - full ACS capabilities + Specified as vid:did (vendor/device ID) in hex noioapicquirk [APIC] Disable all boot interrupt quirks. Safety option to keep boot IRQs enabled. This should never be necessary. diff --git a/Documentation/admin-guide/sysctl/vm.rst b/Documentation/admin-guide/sysctl/vm.rst index 06d0ebddeefa..c240a9b6bdbc 100644 --- a/Documentation/admin-guide/sysctl/vm.rst +++ b/Documentation/admin-guide/sysctl/vm.rst @@ -25,6 +25,9 @@ files can be found in mm/swap.c. Currently, these files are in /proc/sys/vm: - admin_reserve_kbytes +- anon_min_ratio +- clean_low_ratio +- clean_min_ratio - compact_memory - compaction_proactiveness - compact_unevictable_allowed @@ -110,6 +113,67 @@ On x86_64 this is about 128MB. Changing this takes effect whenever an application requests memory. +anon_min_ratio +============== + +This knob provides *hard* protection of anonymous pages. The anonymous pages +on the current node won't be reclaimed under any conditions when their amount +is below vm.anon_min_ratio. + +This knob may be used to prevent excessive swap thrashing when anonymous +memory is low (for example, when memory is going to be overfilled by +compressed data of zram module). + +Setting this value too high (close to 100) can result in inability to +swap and can lead to early OOM under memory pressure. + +The unit of measurement is the percentage of the total memory of the node. + +The default value is 1. + + +clean_low_ratio +================ + +This knob provides *best-effort* protection of clean file pages. The file pages +on the current node won't be reclaimed under memory pressure when the amount of +clean file pages is below vm.clean_low_ratio *unless* we threaten to OOM. + +Protection of clean file pages using this knob may be used when swapping is +still possible to + - prevent disk I/O thrashing under memory pressure; + - improve performance in disk cache-bound tasks under memory pressure. + +Setting it to a high value may result in a early eviction of anonymous pages +into the swap space by attempting to hold the protected amount of clean file +pages in memory. + +The unit of measurement is the percentage of the total memory of the node. + +The default value is 15. + + +clean_min_ratio +================ + +This knob provides *hard* protection of clean file pages. The file pages on the +current node won't be reclaimed under memory pressure when the amount of clean +file pages is below vm.clean_min_ratio. + +Hard protection of clean file pages using this knob may be used to + - prevent disk I/O thrashing under memory pressure even with no free swap space; + - improve performance in disk cache-bound tasks under memory pressure; + - avoid high latency and prevent livelock in near-OOM conditions. + +Setting it to a high value may result in a early out-of-memory condition due to +the inability to reclaim the protected amount of clean file pages when other +types of pages cannot be reclaimed. + +The unit of measurement is the percentage of the total memory of the node. + +The default value is 4. + + compact_memory ============== @@ -986,6 +1050,14 @@ be 133 (x + 2x = 200, 2x = 133.33). At 0, the kernel will not initiate swap until the amount of free and file-backed pages is less than the high watermark in a zone. +This knob has no effect if the amount of clean file pages on the current +node is below vm.clean_low_ratio or vm.clean_min_ratio. In this case, +only anonymous pages can be reclaimed. + +If the number of anonymous pages on the current node is below +vm.anon_min_ratio, then only file pages can be reclaimed with +any vm.swappiness value. + unprivileged_userfaultfd ======================== diff --git a/Makefile b/Makefile index 21df63507119..1d5c75295b56 100644 --- a/Makefile +++ b/Makefile @@ -888,11 +888,19 @@ KBUILD_CFLAGS += -fno-delete-null-pointer-checks ifdef CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE KBUILD_CFLAGS += -O2 KBUILD_RUSTFLAGS += -Copt-level=2 +else ifdef CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE_O3 +KBUILD_CFLAGS += -O3 +KBUILD_RUSTFLAGS += -Copt-level=3 else ifdef CONFIG_CC_OPTIMIZE_FOR_SIZE KBUILD_CFLAGS += -Os KBUILD_RUSTFLAGS += -Copt-level=s endif +# Perform swing modulo scheduling immediately before the first scheduling pass. +# This pass looks at innermost loops and reorders their instructions by +# overlapping different iterations. +KBUILD_CFLAGS += $(call cc-option,-fmodulo-sched -fmodulo-sched-allow-regmoves -fivopts) + # Always set `debug-assertions` and `overflow-checks` because their default # depends on `opt-level` and `debug-assertions`, respectively. KBUILD_RUSTFLAGS += -Cdebug-assertions=$(if $(CONFIG_RUST_DEBUG_ASSERTIONS),y,n) @@ -1022,10 +1030,10 @@ export CC_FLAGS_SCS endif ifdef CONFIG_LTO_CLANG -ifdef CONFIG_LTO_CLANG_THIN -CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit -else +ifdef CONFIG_LTO_CLANG_FULL CC_FLAGS_LTO := -flto +else +CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit endif CC_FLAGS_LTO += -fvisibility=hidden @@ -1232,7 +1240,7 @@ export ARCH_DRIVERS := $(drivers-y) $(drivers-m) KBUILD_VMLINUX_OBJS := built-in.a $(patsubst %/, %/lib.a, $(filter %/, $(libs-y))) KBUILD_VMLINUX_LIBS := $(filter-out %/, $(libs-y)) -export KBUILD_VMLINUX_LIBS +export KBUILD_VMLINUX_OBJS KBUILD_VMLINUX_LIBS export KBUILD_LDS := arch/$(SRCARCH)/kernel/vmlinux.lds ifdef CONFIG_TRIM_UNUSED_KSYMS @@ -1241,16 +1249,12 @@ ifdef CONFIG_TRIM_UNUSED_KSYMS KBUILD_MODULES := y endif -# '$(AR) mPi' needs 'T' to workaround the bug of llvm-ar <= 14 -quiet_cmd_ar_vmlinux.a = AR $@ - cmd_ar_vmlinux.a = \ - rm -f $@; \ - $(AR) cDPrST $@ $(KBUILD_VMLINUX_OBJS); \ - $(AR) mPiT $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt) +PHONY += vmlinux_a +vmlinux_a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE + $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_a -targets += vmlinux.a -vmlinux.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE - $(call if_changed,ar_vmlinux.a) +vmlinux.a: vmlinux_a + @: PHONY += vmlinux_o vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS) @@ -1610,6 +1614,7 @@ endif # CONFIG_MODULES CLEAN_FILES += vmlinux.symvers modules-only.symvers \ modules.builtin modules.builtin.modinfo modules.nsdeps \ modules.builtin.ranges vmlinux.o.map vmlinux.unstripped \ + vmlinux.thinlto-index builtin.order \ compile_commands.json rust/test \ rust-project.json .vmlinux.objs .vmlinux.export.c \ .builtin-dtbs-list .builtin-dtb.S @@ -2063,7 +2068,7 @@ clean: $(clean-dirs) $(call cmd,rmfiles) @find . $(RCS_FIND_IGNORE) \ \( -name '*.[aios]' -o -name '*.rsi' -o -name '*.ko' -o -name '.*.cmd' \ - -o -name '*.ko.*' \ + -o -name '*.ko.*' -o -name '*.o.thinlto.bc' \ -o -name '*.dtb' -o -name '*.dtbo' \ -o -name '*.dtb.S' -o -name '*.dtbo.S' \ -o -name '*.dt.yaml' -o -name 'dtbs-list' \ diff --git a/arch/Kconfig b/arch/Kconfig index 31220f512b16..5c54b2beb2d6 100644 --- a/arch/Kconfig +++ b/arch/Kconfig @@ -861,6 +861,25 @@ config LTO_CLANG_THIN https://clang.llvm.org/docs/ThinLTO.html If unsure, say Y. + +config LTO_CLANG_THIN_DIST + bool "Clang ThinLTO in distributed mode (EXPERIMENTAL)" + depends on HAS_LTO_CLANG && ARCH_SUPPORTS_LTO_CLANG_THIN + select LTO_CLANG + help + This option enables Clang's ThinLTO in distributed build mode. + In this mode, the linker performs the thin-link, generating + ThinLTO index files. Subsequently, the build system explicitly + invokes ThinLTO backend compilation using these index files + and pre-linked IR objects. The resulting native object files + are with the .thinlto-native.o suffix. + + This build mode offers improved visibility into the ThinLTO + process through explicit subcommand exposure. It also makes + final native object files directly available, benefiting + tools like objtool and kpatch. Additionally, it provides + crucial granular control over back-end options, enabling + module-specific compiler options, and simplifies debugging. endchoice config ARCH_SUPPORTS_AUTOFDO_CLANG diff --git a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu index f928cf6e3252..d4ce964d9713 100644 --- a/arch/x86/Kconfig.cpu +++ b/arch/x86/Kconfig.cpu @@ -255,6 +255,11 @@ config CC_HAS_MARCH_NATIVE # usage warnings that only appear wth '-march=native'. depends on CC_IS_GCC || CLANG_VERSION >= 190100 + +choice + prompt "x86_64 Compiler Build Optimization" + default GENERIC_CPU + config X86_NATIVE_CPU bool "Build and optimize for local/native CPU" depends on X86_64 @@ -269,6 +274,47 @@ config X86_NATIVE_CPU If unsure, say N. +config GENERIC_CPU + bool "Generic-x86-64" + depends on X86_64 + help + Generic x86-64 CPU. + Runs equally well on all x86-64 CPUs. + +config MZEN4 + bool "AMD Ryzen 4" + depends on (CC_IS_GCC && GCC_VERSION >= 130000) || (CC_IS_CLANG && CLANG_VERSION >= 160000) + help + Select this for AMD Family 19h Zen 4 processors. + + Enables -march=znver4 + +endchoice + +config X86_64_VERSION + int "x86-64 compiler ISA level" + range 1 4 + depends on (CC_IS_GCC && GCC_VERSION > 110000) || (CC_IS_CLANG && CLANG_VERSION >= 120000) + depends on X86_64 && GENERIC_CPU + help + Specify a specific x86-64 compiler ISA level. + + There are three x86-64 ISA levels that work on top of + the x86-64 baseline, namely: x86-64-v2 and x86-64-v3. + + x86-64-v2 brings support for vector instructions up to Streaming SIMD + Extensions 4.2 (SSE4.2) and Supplemental Streaming SIMD Extensions 3 + (SSSE3), the POPCNT instruction, and CMPXCHG16B. + + x86-64-v3 adds vector instructions up to AVX2, MOVBE, and additional + bit-manipulation instructions. + + x86-64-v4 is not included since the kernel does not use AVX512 instructions + + You can find the best version for your CPU by running one of the following: + /lib/ld-linux-x86-64.so.2 --help | grep supported + /lib64/ld-linux-x86-64.so.2 --help | grep supported + config X86_GENERIC bool "Generic x86 support" depends on X86_32 diff --git a/arch/x86/Makefile b/arch/x86/Makefile index 1d403a3612ea..d3a1ae544f2a 100644 --- a/arch/x86/Makefile +++ b/arch/x86/Makefile @@ -167,10 +167,22 @@ else ifdef CONFIG_X86_NATIVE_CPU KBUILD_CFLAGS += -march=native KBUILD_RUSTFLAGS += -Ctarget-cpu=native -else +endif + +ifdef CONFIG_MZEN4 + KBUILD_CFLAGS += -march=znver4 + KBUILD_RUSTFLAGS += -Ctarget-cpu=znver4 +endif + +ifdef CONFIG_GENERIC_CPU +ifeq ($(CONFIG_X86_64_VERSION),1) KBUILD_CFLAGS += -march=x86-64 -mtune=generic KBUILD_RUSTFLAGS += -Ctarget-cpu=x86-64 -Ztune-cpu=generic -endif +else + KBUILD_CFLAGS +=-march=x86-64-v$(CONFIG_X86_64_VERSION) + KBUILD_RUSTFLAGS += -Ctarget-cpu=x86-64-v$(CONFIG_X86_64_VERSION) +endif # CONFIG_X86_64_VERSION +endif # CONFIG_GENERIC_CPU KBUILD_CFLAGS += -mno-red-zone KBUILD_CFLAGS += -mcmodel=kernel diff --git a/arch/x86/include/asm/pci.h b/arch/x86/include/asm/pci.h index b3ab80a03365..5e883b397ff3 100644 --- a/arch/x86/include/asm/pci.h +++ b/arch/x86/include/asm/pci.h @@ -26,6 +26,7 @@ struct pci_sysdata { #if IS_ENABLED(CONFIG_VMD) struct pci_dev *vmd_dev; /* VMD Device if in Intel VMD domain */ #endif + struct pci_dev *nvme_remap_dev; /* AHCI Device if NVME remapped bus */ }; extern int pci_routeirq; @@ -69,6 +70,11 @@ static inline bool is_vmd(struct pci_bus *bus) #define is_vmd(bus) false #endif /* CONFIG_VMD */ +static inline bool is_nvme_remap(struct pci_bus *bus) +{ + return to_pci_sysdata(bus)->nvme_remap_dev != NULL; +} + /* Can be used to override the logic in pci_scan_bus for skipping already-configured bus numbers - to be used for buggy BIOSes or architectures with incomplete PCI setup by the loader */ diff --git a/arch/x86/kernel/cpu/bus_lock.c b/arch/x86/kernel/cpu/bus_lock.c index dbc99a47be45..e92b3f7c8cdd 100644 --- a/arch/x86/kernel/cpu/bus_lock.c +++ b/arch/x86/kernel/cpu/bus_lock.c @@ -47,7 +47,11 @@ static const struct { static struct ratelimit_state bld_ratelimit; +#ifdef CONFIG_CACHY +static unsigned int sysctl_sld_mitigate = 0; +#else static unsigned int sysctl_sld_mitigate = 1; +#endif /* CONFIG_CACHY */ static DEFINE_SEMAPHORE(buslock_sem, 1); #ifdef CONFIG_PROC_SYSCTL diff --git a/arch/x86/pci/common.c b/arch/x86/pci/common.c index ddb798603201..7c20387d8202 100644 --- a/arch/x86/pci/common.c +++ b/arch/x86/pci/common.c @@ -723,12 +723,15 @@ int pci_ext_cfg_avail(void) return 0; } -#if IS_ENABLED(CONFIG_VMD) struct pci_dev *pci_real_dma_dev(struct pci_dev *dev) { +#if IS_ENABLED(CONFIG_VMD) if (is_vmd(dev->bus)) return to_pci_sysdata(dev->bus)->vmd_dev; +#endif + + if (is_nvme_remap(dev->bus)) + return to_pci_sysdata(dev->bus)->nvme_remap_dev; return dev; } -#endif diff --git a/block/Kconfig.iosched b/block/Kconfig.iosched index 27f11320b8d1..e98585dd83e0 100644 --- a/block/Kconfig.iosched +++ b/block/Kconfig.iosched @@ -16,6 +16,20 @@ config MQ_IOSCHED_KYBER synchronous writes, it will self-tune queue depths to achieve that goal. +config MQ_IOSCHED_ADIOS + tristate "Adaptive Deadline I/O scheduler" + default m + help + The Adaptive Deadline I/O Scheduler (ADIOS) is a multi-queue I/O + scheduler with learning-based adaptive latency control. + +config MQ_IOSCHED_DEFAULT_ADIOS + bool "Enable ADIOS I/O scheduler as default MQ I/O scheduler" + depends on MQ_IOSCHED_ADIOS=y + default n + help + Enable the ADIOS I/O scheduler as the default scheduler for MQ I/O. + config IOSCHED_BFQ tristate "BFQ I/O scheduler" select BLK_ICQ diff --git a/block/Makefile b/block/Makefile index c65f4da93702..105b12fd86b8 100644 --- a/block/Makefile +++ b/block/Makefile @@ -22,6 +22,7 @@ obj-$(CONFIG_BLK_CGROUP_IOLATENCY) += blk-iolatency.o obj-$(CONFIG_BLK_CGROUP_IOCOST) += blk-iocost.o obj-$(CONFIG_MQ_IOSCHED_DEADLINE) += mq-deadline.o obj-$(CONFIG_MQ_IOSCHED_KYBER) += kyber-iosched.o +obj-$(CONFIG_MQ_IOSCHED_ADIOS) += adios.o bfq-y := bfq-iosched.o bfq-wf2q.o bfq-cgroup.o obj-$(CONFIG_IOSCHED_BFQ) += bfq.o @@ -36,3 +37,10 @@ obj-$(CONFIG_BLK_INLINE_ENCRYPTION) += blk-crypto.o blk-crypto-profile.o \ blk-crypto-sysfs.o obj-$(CONFIG_BLK_INLINE_ENCRYPTION_FALLBACK) += blk-crypto-fallback.o obj-$(CONFIG_BLOCK_HOLDER_DEPRECATED) += holder.o + +all: + make -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules + +clean: + make -C /lib/modules/$(shell uname -r)/build M=$(PWD) clean + diff --git a/block/adios.c b/block/adios.c new file mode 100644 index 000000000000..6ae4bce22463 --- /dev/null +++ b/block/adios.c @@ -0,0 +1,2004 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Adaptive Deadline I/O Scheduler (ADIOS) + * Copyright (C) 2025 Masahito Suzuki + */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "elevator.h" +#include "blk.h" +#include "blk-mq.h" +#include "blk-mq-sched.h" + +#define ADIOS_VERSION "3.1.9" + +/* Request Types: + * + * Tier 0 (Highest Priority): Emergency & System Integrity Requests + * ----------------------------------------------------------------- + * - Target: Requests with the BLK_MQ_INSERT_AT_HEAD flag. + * - Purpose: For critical, non-negotiable operations such as device error + * recovery or flush sequences that must bypass all other scheduling logic. + * - Implementation: Placed in a dedicated, high-priority FIFO queue + * (`prio_queue[0]`) for immediate dispatch. + * + * Tier 1 (High Priority): I/O Barrier Guarantees + * --------------------------------------------------------------- + * - Target: Requests with the REQ_OP_FLUSH flag. + * - Purpose: To enforce a strict I/O barrier. When a flush request is + * received, the scheduler stops processing new requests from its main + * queues until all preceding requests have been completed. This guarantees + * the order of operations required by filesystems for data integrity. + * - Implementation: A state flag (ADIOS_STATE_BARRIER) halts + * insertion into the main deadline tree. The barrier request and all + * subsequent requests are held in a temporary `barrier_queue`. Once the + * main queues are drained, the barrier request and the subsequent requests + * are released from the pending queue back into the scheduler. + * + * Tier 2 (Medium Priority): Application Responsiveness + * ---------------------------------------------------- + * - Target: Normal synchronous requests (e.g., from standard file reads). + * - Purpose: To ensure correct application behavior for operations that + * depend on sequential I/O completion (e.g., file system mounts) and to + * provide low latency for interactive applications. + * - Implementation: The deadline for these requests is set to their start + * time (`rq->start_time_ns`). This effectively enforces FIFO-like behavior + * within the deadline-sorted red-black tree, preventing out-of-order + * execution of dependent synchronous operations. + * + * Tier 3 (Normal Priority): Background Throughput + * ----------------------------------------------- + * - Target: Asynchronous requests. + * - Purpose: To maximize disk throughput for background tasks where latency + * is not critical. + * - Implementation: These are the only requests where ADIOS's adaptive + * latency prediction model is used. A dynamic deadline is calculated based + * on the predicted I/O latency, allowing for aggressive reordering to + * optimize I/O efficiency. + * + * Dispatch Logic: + * The scheduler always dispatches requests in strict priority order: + * 1. prio_queue[0] (Tier 0) + * 2. The deadline-sorted batch queue (which naturally prioritizes Tier 2 + * over Tier 3 due to their calculated deadlines). + * 3. Barrier-pending requests are handled only after the main queues are empty. + */ + +// Global variable to control the latency +static u64 default_global_latency_window = 16000000ULL; +static u64 default_global_latency_window_rotational = 22000000ULL; +// Ratio below which batch queues should be refilled +static u8 default_bq_refill_below_ratio = 20; +// Maximum latency sample to input +static u64 default_lat_model_latency_limit = 500000000ULL; +// Batch ordering strategy +static u64 default_batch_order = 0; + +/* Compliance Flags: + * 0x1: Async requests will not be reordered based on the predicted latency + */ +enum adios_compliance_flags { + ADIOS_CF_FIXORDER = 1U << 0, +}; + +// Flags to control compliance with block layer constraints +static u64 default_compliance_flags = 0x0; + +// Dynamic thresholds for shrinkage +static u32 default_lm_shrink_at_kreqs = 5000; +static u32 default_lm_shrink_at_gbytes = 50; +static u32 default_lm_shrink_resist = 2; + +enum adios_optype { + ADIOS_READ = 0, + ADIOS_WRITE = 1, + ADIOS_DISCARD = 2, + ADIOS_OTHER = 3, + ADIOS_OPTYPES = 4, +}; + +// Latency targets for each operation type +static u64 default_latency_target[ADIOS_OPTYPES] = { + [ADIOS_READ] = 2ULL * NSEC_PER_MSEC, + [ADIOS_WRITE] = 2000ULL * NSEC_PER_MSEC, + [ADIOS_DISCARD] = 8000ULL * NSEC_PER_MSEC, + [ADIOS_OTHER] = 0ULL * NSEC_PER_MSEC, +}; + +// Maximum batch size limits for each operation type +static u32 default_batch_limit[ADIOS_OPTYPES] = { + [ADIOS_READ] = 36, + [ADIOS_WRITE] = 72, + [ADIOS_DISCARD] = 1, + [ADIOS_OTHER] = 1, +}; + +enum adios_batch_order { + ADIOS_BO_OPTYPE = 0, + ADIOS_BO_ELEVATOR = 1, +}; + +// Thresholds for latency model control +#define LM_BLOCK_SIZE_THRESHOLD 4096 +#define LM_SAMPLES_THRESHOLD 1024 +#define LM_INTERVAL_THRESHOLD 1500 +#define LM_OUTLIER_PERCENTILE 99 +#define LM_LAT_BUCKET_COUNT 64 + +#define ADIOS_PQ_LEVELS 2 +#define ADIOS_DL_TYPES 2 +#define ADIOS_BQ_PAGES 2 + +static u32 default_dl_prio[ADIOS_DL_TYPES] = {8, 0}; + +// Bit flags for the atomic state variable, indicating which queues have requests. +enum adios_state_flags { + ADIOS_STATE_PQ_0 = 1U << 0, + ADIOS_STATE_PQ_1 = 1U << 1, + ADIOS_STATE_DL_0 = 1U << 2, + ADIOS_STATE_DL_1 = 1U << 3, + ADIOS_STATE_BQ_PAGE_0 = 1U << 4, + ADIOS_STATE_BQ_PAGE_1 = 1U << 5, + ADIOS_STATE_BARRIER = 1U << 6, +}; +#define ADIOS_STATE_PQ 0 +#define ADIOS_STATE_DL 2 +#define ADIOS_STATE_BQ 4 +#define ADIOS_STATE_BP 6 + +// Temporal granularity of the deadline tree node (dl_group) +#define ADIOS_QUANTUM_SHIFT 20 + +#define ADIOS_MAX_INSERTS_PER_LOCK 72 +#define ADIOS_MAX_DELETES_PER_LOCK 24 + +// Structure to hold latency bucket data for small requests +struct latency_bucket_small { + u64 weighted_sum_latency; + u64 sum_of_weights; +}; + +// Structure to hold latency bucket data for large requests +struct latency_bucket_large { + u64 weighted_sum_latency; + u64 weighted_sum_block_size; + u64 sum_of_weights; +}; + +// Structure to hold per-cpu buckets, improving data locality and code clarity. +struct lm_buckets { + struct latency_bucket_small small_bucket[LM_LAT_BUCKET_COUNT]; + struct latency_bucket_large large_bucket[LM_LAT_BUCKET_COUNT]; +}; + +// Structure to hold RCU-protected latency model parameters +struct latency_model_params { + u64 base; + u64 slope; + u64 small_sum_delay; + u64 small_count; + u64 large_sum_delay; + u64 large_sum_bsize; + u64 last_update_jiffies; + struct rcu_head rcu; +}; + +// Structure to hold the latency model context data +struct latency_model { + spinlock_t update_lock; + struct latency_model_params __rcu *params; + + // Per-CPU buckets to avoid lock contention on the completion path + struct lm_buckets __percpu *pcpu_buckets; + + u32 lm_shrink_at_kreqs; + u32 lm_shrink_at_gbytes; + u8 lm_shrink_resist; +}; + +union adios_in_flight_rqs { + atomic64_t atomic; + u64 scalar; + struct { + u64 count: 16; + u64 total_pred_lat: 48; + }; +}; + +// Adios scheduler data +struct adios_data { + spinlock_t pq_lock; + struct list_head prio_queue[2]; + + struct rb_root_cached dl_tree[2]; + spinlock_t lock; + s64 dl_bias; + s32 dl_prio[2]; + + atomic_t state; + u8 bq_state[ADIOS_BQ_PAGES]; + + bool models_stable; + + u64 global_latency_window; + u64 compliance_flags; + u64 latency_target[ADIOS_OPTYPES]; + u32 batch_limit[ADIOS_OPTYPES]; + u32 batch_actual_max_size[ADIOS_OPTYPES]; + u32 batch_actual_max_total; + u32 async_depth; + u32 lat_model_latency_limit; + u8 bq_refill_below_ratio; + u8 is_rotational; + u8 batch_order; + u8 elv_direction; + sector_t head_pos; + sector_t last_completed_pos; + + bool bq_page; + struct list_head batch_queue[ADIOS_BQ_PAGES][ADIOS_OPTYPES]; + u32 batch_count[ADIOS_BQ_PAGES][ADIOS_OPTYPES]; + u8 bq_batch_order[ADIOS_BQ_PAGES]; + spinlock_t bq_lock; + spinlock_t barrier_lock; + struct list_head barrier_queue; + + struct lm_buckets *aggr_buckets; + + struct latency_model latency_model[ADIOS_OPTYPES]; + struct timer_list update_timer; + + union adios_in_flight_rqs in_flight_rqs; + atomic64_t total_pred_lat; + u64 last_completed_time; + + struct kmem_cache *rq_data_pool; + struct kmem_cache *dl_group_pool; + + struct request_queue *queue; +}; + +// List of requests with the same deadline in the deadline-sorted tree +struct dl_group { + struct rb_node node; + struct list_head rqs; + u64 deadline; +} __attribute__((aligned(64))); + +// Structure to hold scheduler-specific data for each request +struct adios_rq_data { + struct list_head *dl_group; + struct list_head dl_node; + + struct request *rq; + u64 deadline; + u64 pred_lat; + u32 block_size; + bool managed; +} __attribute__((aligned(64))); + +static const int adios_prio_to_wmult[40] = { + /* -20 */ 88761, 71755, 56483, 46273, 36291, + /* -15 */ 29154, 23254, 18705, 14949, 11916, + /* -10 */ 9548, 7620, 6100, 4904, 3906, + /* -5 */ 3121, 2501, 1991, 1586, 1277, + /* 0 */ 1024, 820, 655, 526, 423, + /* 5 */ 335, 272, 215, 172, 137, + /* 10 */ 110, 87, 70, 56, 45, + /* 15 */ 36, 29, 23, 18, 15, +}; + +static inline bool compliant(struct adios_data *ad, u32 flag) { + return ad->compliance_flags & flag; +} + +// Count the number of entries in aggregated small buckets +static u64 lm_count_small_entries(struct latency_bucket_small *buckets) { + u64 total_weight = 0; + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) + total_weight += buckets[i].sum_of_weights; + return total_weight; +} + +// Update the small buckets in the latency model from aggregated data +static bool lm_update_small_buckets(struct latency_model *model, + struct latency_model_params *params, + struct latency_bucket_small *buckets, + u64 total_weight, bool count_all) { + u64 sum_latency = 0; + u64 sum_weight = 0; + u64 cumulative_weight = 0, threshold_weight = 0; + u8 outlier_threshold_bucket = 0; + u8 outlier_percentile = LM_OUTLIER_PERCENTILE; + u8 reduction; + + if (count_all) + outlier_percentile = 100; + + // Calculate the threshold weight for outlier detection + threshold_weight = (total_weight * outlier_percentile) / 100; + + // Identify the bucket that corresponds to the outlier threshold + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) { + cumulative_weight += buckets[i].sum_of_weights; + if (cumulative_weight >= threshold_weight) { + outlier_threshold_bucket = i; + break; + } + } + + // Calculate the average latency, excluding outliers + for (u8 i = 0; i <= outlier_threshold_bucket; i++) { + struct latency_bucket_small *bucket = &buckets[i]; + if (i < outlier_threshold_bucket) { + sum_latency += bucket->weighted_sum_latency; + sum_weight += bucket->sum_of_weights; + } else { + // The threshold bucket's contribution is proportional + u64 remaining_weight = + threshold_weight - (cumulative_weight - bucket->sum_of_weights); + if (bucket->sum_of_weights > 0) { + sum_latency += div_u64(bucket->weighted_sum_latency * + remaining_weight, bucket->sum_of_weights); + sum_weight += remaining_weight; + } + } + } + + // Shrink the model if it reaches at the readjustment threshold + if (params->small_count >= 1000ULL * model->lm_shrink_at_kreqs) { + reduction = model->lm_shrink_resist; + if (params->small_count >> reduction) { + params->small_sum_delay -= params->small_sum_delay >> reduction; + params->small_count -= params->small_count >> reduction; + } + } + + if (!sum_weight) + return false; + + // Accumulate the average latency into the statistics + params->small_sum_delay += sum_latency; + params->small_count += sum_weight; + + return true; +} + +// Count the number of entries in aggregated large buckets +static u64 lm_count_large_entries(struct latency_bucket_large *buckets) { + u64 total_weight = 0; + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) + total_weight += buckets[i].sum_of_weights; + return total_weight; +} + +// Update the large buckets in the latency model from aggregated data +static bool lm_update_large_buckets(struct latency_model *model, + struct latency_model_params *params, + struct latency_bucket_large *buckets, + u64 total_weight, bool count_all) { + s64 sum_latency = 0; + u64 sum_block_size = 0, intercept; + u64 cumulative_weight = 0, threshold_weight = 0; + u64 sum_weight = 0; + u8 outlier_threshold_bucket = 0; + u8 outlier_percentile = LM_OUTLIER_PERCENTILE; + u8 reduction; + + if (count_all) + outlier_percentile = 100; + + // Calculate the threshold weight for outlier detection + threshold_weight = (total_weight * outlier_percentile) / 100; + + // Identify the bucket that corresponds to the outlier threshold + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) { + cumulative_weight += buckets[i].sum_of_weights; + if (cumulative_weight >= threshold_weight) { + outlier_threshold_bucket = i; + break; + } + } + + // Calculate the average latency and block size, excluding outliers + for (u8 i = 0; i <= outlier_threshold_bucket; i++) { + struct latency_bucket_large *bucket = &buckets[i]; + if (i < outlier_threshold_bucket) { + sum_latency += bucket->weighted_sum_latency; + sum_block_size += bucket->weighted_sum_block_size; + sum_weight += bucket->sum_of_weights; + } else { + // The threshold bucket's contribution is proportional + u64 remaining_weight = + threshold_weight - (cumulative_weight - bucket->sum_of_weights); + if (bucket->sum_of_weights > 0) { + sum_latency += div_u64(bucket->weighted_sum_latency * + remaining_weight, bucket->sum_of_weights); + sum_block_size += div_u64(bucket->weighted_sum_block_size * + remaining_weight, bucket->sum_of_weights); + sum_weight += remaining_weight; + } + } + } + + if (!sum_weight) + return false; + + // Shrink the model if it reaches at the readjustment threshold + if (params->large_sum_bsize >= 0x40000000ULL * model->lm_shrink_at_gbytes) { + reduction = model->lm_shrink_resist; + if (params->large_sum_bsize >> reduction) { + params->large_sum_delay -= params->large_sum_delay >> reduction; + params->large_sum_bsize -= params->large_sum_bsize >> reduction; + } + } + + // Accumulate the average delay into the statistics + intercept = params->base; + if (sum_latency > intercept) + sum_latency -= intercept; + + params->large_sum_delay += sum_latency; + params->large_sum_bsize += sum_block_size; + + return true; +} + +static void reset_buckets(struct lm_buckets *buckets) +{ memset(buckets, 0, sizeof(*buckets)); } + +static void lm_reset_pcpu_buckets(struct latency_model *model) { + int cpu; + for_each_possible_cpu(cpu) + reset_buckets(per_cpu_ptr(model->pcpu_buckets, cpu)); +} + +// Update the latency model parameters and statistics +static void latency_model_update( + struct adios_data *ad, struct latency_model *model) { + u64 now; + u64 small_weight, large_weight; + bool time_elapsed; + bool small_processed = false, large_processed = false; + struct lm_buckets *aggr = ad->aggr_buckets; + struct latency_bucket_small *asb; + struct latency_bucket_large *alb; + struct lm_buckets *pcpu_b; + unsigned long flags; + int cpu; + struct latency_model_params *old_params, *new_params; + + spin_lock_irqsave(&model->update_lock, flags); + + old_params = rcu_dereference_protected(model->params, + lockdep_is_held(&model->update_lock)); + new_params = kmemdup(old_params, sizeof(*new_params), GFP_ATOMIC); + if (!new_params) { + spin_unlock_irqrestore(&model->update_lock, flags); + return; + } + + // Aggregate data from all CPUs and reset per-cpu buckets. + for_each_possible_cpu(cpu) { + pcpu_b = per_cpu_ptr(model->pcpu_buckets, cpu); + + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) { + if (pcpu_b->small_bucket[i].sum_of_weights) { + asb = &aggr->small_bucket[i]; + asb->sum_of_weights += + pcpu_b->small_bucket[i].sum_of_weights; + asb->weighted_sum_latency += + pcpu_b->small_bucket[i].weighted_sum_latency; + } + if (pcpu_b->large_bucket[i].sum_of_weights) { + alb = &aggr->large_bucket[i]; + alb->sum_of_weights += + pcpu_b->large_bucket[i].sum_of_weights; + alb->weighted_sum_latency += + pcpu_b->large_bucket[i].weighted_sum_latency; + alb->weighted_sum_block_size += + pcpu_b->large_bucket[i].weighted_sum_block_size; + } + } + // Reset per-cpu buckets after aggregating + reset_buckets(pcpu_b); + } + + // Count the number of entries in aggregated buckets + small_weight = lm_count_small_entries(aggr->small_bucket); + large_weight = lm_count_large_entries(aggr->large_bucket); + + // Whether enough time has elapsed since the last update + now = jiffies; + time_elapsed = unlikely(!new_params->base) || + new_params->last_update_jiffies + + msecs_to_jiffies(LM_INTERVAL_THRESHOLD) <= now; + + // Update small buckets + if (small_weight && (time_elapsed || + LM_SAMPLES_THRESHOLD <= small_weight || !new_params->base)) { + small_processed = lm_update_small_buckets(model, new_params, + aggr->small_bucket, small_weight, !new_params->base); + memset(&aggr->small_bucket[0], 0, sizeof(aggr->small_bucket)); + } + // Update large buckets + if (large_weight && (time_elapsed || + LM_SAMPLES_THRESHOLD <= large_weight || !new_params->slope)) { + large_processed = lm_update_large_buckets(model, new_params, + aggr->large_bucket, large_weight, !new_params->slope); + memset(&aggr->large_bucket[0], 0, sizeof(aggr->large_bucket)); + } + + // Update the base parameter if small bucket was processed + if (small_processed && likely(new_params->small_count)) + new_params->base = div_u64(new_params->small_sum_delay, + new_params->small_count); + + // Update the slope parameter if large bucket was processed + if (large_processed && likely(new_params->large_sum_bsize)) + new_params->slope = div_u64(new_params->large_sum_delay, + DIV_ROUND_UP_ULL(new_params->large_sum_bsize, 1024)); + + // Update last updated jiffies if update happened or time has elapsed + if (small_processed || large_processed || time_elapsed) + new_params->last_update_jiffies = now; + + rcu_assign_pointer(model->params, new_params); + spin_unlock_irqrestore(&model->update_lock, flags); + + kfree_rcu(old_params, rcu); +} + +// Determine the bucket index for a given measured and predicted latency +static u8 lm_input_bucket_index(u64 measured, u64 predicted) { + u8 bucket_index; + + if (measured < predicted * 2) + bucket_index = div_u64((measured * 20), predicted); + else if (measured < predicted * 5) + bucket_index = div_u64((measured * 10), predicted) + 20; + else + bucket_index = div_u64((measured * 3), predicted) + 40; + + return bucket_index; +} + +// Input latency data into the latency model +static void latency_model_input(struct adios_data *ad, + struct latency_model *model, + u32 block_size, u64 latency, u64 pred_lat, u32 weight) { + unsigned long flags; + u8 bucket_index; + struct lm_buckets *buckets; + u64 current_base; + struct latency_model_params *params; + + local_irq_save(flags); + buckets = per_cpu_ptr(model->pcpu_buckets, __smp_processor_id()); + + rcu_read_lock(); + params = rcu_dereference(model->params); + current_base = params->base; + rcu_read_unlock(); + + if (block_size <= LM_BLOCK_SIZE_THRESHOLD) { + // Handle small requests + bucket_index = lm_input_bucket_index(latency, current_base ?: 1); + + if (bucket_index >= LM_LAT_BUCKET_COUNT) + bucket_index = LM_LAT_BUCKET_COUNT - 1; + + buckets->small_bucket[bucket_index].sum_of_weights += weight; + buckets->small_bucket[bucket_index].weighted_sum_latency += + latency * weight; + + local_irq_restore(flags); + + if (unlikely(!current_base)) { + latency_model_update(ad, model); + return; + } + } else { + // Handle large requests + if (!current_base || !pred_lat) { + local_irq_restore(flags); + return; + } + + bucket_index = lm_input_bucket_index(latency, pred_lat); + + if (bucket_index >= LM_LAT_BUCKET_COUNT) + bucket_index = LM_LAT_BUCKET_COUNT - 1; + + buckets->large_bucket[bucket_index].sum_of_weights += weight; + buckets->large_bucket[bucket_index].weighted_sum_latency += + latency * weight; + buckets->large_bucket[bucket_index].weighted_sum_block_size += + block_size * weight; + + local_irq_restore(flags); + } +} + +// Predict the latency for a given block size using the latency model +static u64 latency_model_predict(struct latency_model *model, u32 block_size) { + u64 result; + struct latency_model_params *params; + + rcu_read_lock(); + params = rcu_dereference(model->params); + + result = params->base; + if (block_size > LM_BLOCK_SIZE_THRESHOLD) + result += params->slope * + DIV_ROUND_UP_ULL(block_size - LM_BLOCK_SIZE_THRESHOLD, 1024); + + rcu_read_unlock(); + + return result; +} + +// Determine the type of operation based on request flags +static u8 adios_optype(struct request *rq) { + switch (rq->cmd_flags & REQ_OP_MASK) { + case REQ_OP_READ: + return ADIOS_READ; + case REQ_OP_WRITE: + return ADIOS_WRITE; + case REQ_OP_DISCARD: + return ADIOS_DISCARD; + default: + return ADIOS_OTHER; + } +} + +static inline u8 adios_optype_not_read(struct request *rq) { + return (rq->cmd_flags & REQ_OP_MASK) != REQ_OP_READ; +} + +// Helper function to retrieve adios_rq_data from a request +static inline struct adios_rq_data *get_rq_data(struct request *rq) { + return rq->elv.priv[0]; +} + +static inline +void set_adios_state(struct adios_data *ad, u32 shift, u32 idx, bool flag) { + if (flag) + atomic_or(1U << (idx + shift), &ad->state); + else + atomic_andnot(1U << (idx + shift), &ad->state); +} + +static inline u32 get_adios_state(struct adios_data *ad) +{ return atomic_read(&ad->state); } + +static inline u32 eval_this_adios_state(u32 state, u32 shift) +{ return (state >> shift) & 0x3; } + +static inline u32 eval_adios_state(struct adios_data *ad, u32 shift) +{ return eval_this_adios_state(get_adios_state(ad), shift); } + +// Add a request to the deadline-sorted red-black tree +static void add_to_dl_tree( + struct adios_data *ad, bool dl_idx, struct request *rq) { + struct rb_root_cached *root = &ad->dl_tree[dl_idx]; + struct rb_node **link = &(root->rb_root.rb_node), *parent = NULL; + bool leftmost = true; + struct adios_rq_data *rd = get_rq_data(rq); + struct dl_group *dlg; + u64 deadline; + bool was_empty = RB_EMPTY_ROOT(&root->rb_root); + + /* Tier-2: Synchronous Requests + * - Needs to be FIFO within a same optype + * - Relaxed order between different optypes + * - basically needs to be processed in early time */ + rd->deadline = rq->start_time_ns; + + /* Tier-3: Aynchronous Requests + * - Can be reordered and delayed freely */ + if (!(rq->cmd_flags & REQ_SYNC)) { + rd->deadline += ad->latency_target[adios_optype(rq)]; + if (!compliant(ad, ADIOS_CF_FIXORDER)) + rd->deadline += rd->pred_lat; + } + + // Now quantize the deadline (-> dlg->deadline == RB-Tree key) + deadline = rd->deadline & ~((1ULL << ADIOS_QUANTUM_SHIFT) - 1); + + while (*link) { + dlg = rb_entry(*link, struct dl_group, node); + s64 diff = deadline - dlg->deadline; + + parent = *link; + if (diff < 0) { + link = &((*link)->rb_left); + } else if (diff > 0) { + link = &((*link)->rb_right); + leftmost = false; + } else { // diff == 0 + goto found; + } + } + + dlg = rb_entry_safe(parent, struct dl_group, node); + if (!dlg || dlg->deadline != deadline) { + dlg = kmem_cache_zalloc(ad->dl_group_pool, GFP_ATOMIC); + if (!dlg) + return; + dlg->deadline = deadline; + INIT_LIST_HEAD(&dlg->rqs); + rb_link_node(&dlg->node, parent, link); + rb_insert_color_cached(&dlg->node, root, leftmost); + } +found: + list_add_tail(&rd->dl_node, &dlg->rqs); + rd->dl_group = &dlg->rqs; + + if (was_empty) + set_adios_state(ad, ADIOS_STATE_DL, dl_idx, true); +} + +// Remove a request from the deadline-sorted red-black tree +static void del_from_dl_tree( + struct adios_data *ad, bool dl_idx, struct request *rq) { + struct rb_root_cached *root = &ad->dl_tree[dl_idx]; + struct adios_rq_data *rd = get_rq_data(rq); + struct dl_group *dlg = container_of(rd->dl_group, struct dl_group, rqs); + + list_del_init(&rd->dl_node); + if (list_empty(&dlg->rqs)) { + rb_erase_cached(&dlg->node, root); + kmem_cache_free(ad->dl_group_pool, dlg); + } + rd->dl_group = NULL; + + if (RB_EMPTY_ROOT(&ad->dl_tree[dl_idx].rb_root)) + set_adios_state(ad, ADIOS_STATE_DL, dl_idx, false); +} + +// Remove a request from the scheduler +static void remove_request(struct adios_data *ad, struct request *rq) { + bool dl_idx = adios_optype_not_read(rq); + struct request_queue *q = rq->q; + struct adios_rq_data *rd = get_rq_data(rq); + + list_del_init(&rq->queuelist); + + // We might not be on the rbtree, if we are doing an insert merge + if (rd->dl_group) + del_from_dl_tree(ad, dl_idx, rq); + + elv_rqhash_del(q, rq); + if (q->last_merge == rq) + q->last_merge = NULL; +} + +// Convert a queue depth to the corresponding word depth for shallow allocation +static int to_word_depth(struct blk_mq_hw_ctx *hctx, unsigned int qdepth) { + struct sbitmap_queue *bt = &hctx->sched_tags->bitmap_tags; + const unsigned int nrr = hctx->queue->nr_requests; + + return ((qdepth << bt->sb.shift) + nrr - 1) / nrr; +} + +// We limit the depth of request allocation for asynchronous and write requests +static void adios_limit_depth(blk_opf_t opf, struct blk_mq_alloc_data *data) { + struct adios_data *ad = data->q->elevator->elevator_data; + + // Do not throttle synchronous reads + if (op_is_sync(opf) && !op_is_write(opf)) + return; + + data->shallow_depth = to_word_depth(data->hctx, ad->async_depth); +} + +// The number of requests in the queue was notified from the block layer +static void adios_depth_updated(struct request_queue *q) { + struct adios_data *ad = q->elevator->elevator_data; + + ad->async_depth = q->nr_requests; + blk_mq_set_min_shallow_depth(q, 1); +} + +// Handle request merging after a merge operation +static void adios_request_merged(struct request_queue *q, struct request *req, + enum elv_merge type) { + bool dl_idx = adios_optype_not_read(req); + struct adios_data *ad = q->elevator->elevator_data; + + // Reposition request in the deadline-sorted tree + del_from_dl_tree(ad, dl_idx, req); + add_to_dl_tree(ad, dl_idx, req); +} + +// Handle merging of requests after one has been merged into another +static void adios_merged_requests(struct request_queue *q, struct request *req, + struct request *next) { + struct adios_data *ad = q->elevator->elevator_data; + + lockdep_assert_held(&ad->lock); + + // kill knowledge of next, this one is a goner + remove_request(ad, next); +} + +// Try to merge a bio into an existing rq before associating it with an rq +static bool adios_bio_merge(struct request_queue *q, struct bio *bio, + unsigned int nr_segs) { + unsigned long flags; + struct adios_data *ad = q->elevator->elevator_data; + struct request *free = NULL; + bool ret; + + if (eval_adios_state(ad, ADIOS_STATE_BP)) + return false; + + if (!spin_trylock_irqsave(&ad->lock, flags)) + return false; + + ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free); + spin_unlock_irqrestore(&ad->lock, flags); + + if (free) + blk_mq_free_request(free); + + return ret; +} + +static bool merge_or_insert_to_dl_tree(struct adios_data *ad, + struct request *rq, struct request_queue *q, struct list_head *free) { + if (blk_mq_sched_try_insert_merge(q, rq, free)) + return true; + + bool dl_idx = adios_optype_not_read(rq); + add_to_dl_tree(ad, dl_idx, rq); + + if (rq_mergeable(rq)) { + elv_rqhash_add(q, rq); + if (!q->last_merge) + q->last_merge = rq; + } + + return false; +} + +static void insert_to_prio_queue(struct adios_data *ad, + struct request *rq, bool pq_idx) { + struct adios_rq_data *rd = get_rq_data(rq); + + /* We're sure that rd->managed == true */ + union adios_in_flight_rqs ifr = { + .count = 1, + .total_pred_lat = rd->pred_lat, + }; + atomic64_add(ifr.scalar, &ad->in_flight_rqs.atomic); + + scoped_guard(spinlock_irqsave, &ad->pq_lock) { + bool was_empty = list_empty(&ad->prio_queue[pq_idx]); + list_add_tail(&rq->queuelist, &ad->prio_queue[pq_idx]); + if (was_empty) + set_adios_state(ad, ADIOS_STATE_PQ, pq_idx, true); + } +} + +// Insert a request into the scheduler (after Read & Write models stabilized) +static void insert_request_post_stability(struct blk_mq_hw_ctx *hctx, + struct request *rq, blk_insert_t insert_flags, struct list_head *free) { + struct request_queue *q = hctx->queue; + struct adios_data *ad = q->elevator->elevator_data; + struct adios_rq_data *rd = get_rq_data(rq); + u8 optype = adios_optype(rq); + bool rq_is_flush; + + rd->managed = true; + rd->block_size = blk_rq_bytes(rq); + rd->pred_lat = + latency_model_predict(&ad->latency_model[optype], rd->block_size); + + /* Tier-0: BLK_MQ_INSERT_AT_HEAD Requests */ + if (insert_flags & BLK_MQ_INSERT_AT_HEAD) { + insert_to_prio_queue(ad, rq, 0); + return; + } + + /* + * Strict Barrier Handling for REQ_OP_FLUSH: + * If a flush request arrives, or if the scheduler is already in a + * barrier-pending state, all subsequent requests are diverted to a + * separate barrier_queue. This ensures that no new requests are processed + * until all work preceding the barrier is complete. + */ + rq_is_flush = rq->cmd_flags & REQ_OP_FLUSH; + if (eval_adios_state(ad, ADIOS_STATE_BP) || rq_is_flush) { + scoped_guard(spinlock_irqsave, &ad->barrier_lock) { + if (rq_is_flush) + set_adios_state(ad, ADIOS_STATE_BP, 0, true); + list_add_tail(&rq->queuelist, &ad->barrier_queue); + } + return; + } + + if (merge_or_insert_to_dl_tree(ad, rq, q, free)) + return; +} + +// Insert a request into the scheduler (before Read & Write models stabilizes) +static void insert_request_pre_stability(struct blk_mq_hw_ctx *hctx, + struct request *rq, blk_insert_t insert_flags, struct list_head *free) { + struct adios_data *ad = hctx->queue->elevator->elevator_data; + struct adios_rq_data *rd = get_rq_data(rq); + u8 optype = adios_optype(rq); + u8 pq_idx = !(insert_flags & BLK_MQ_INSERT_AT_HEAD); + bool stable = false; + + rd->managed = true; + rd->block_size = blk_rq_bytes(rq); + rd->pred_lat = + latency_model_predict(&ad->latency_model[optype], rd->block_size); + + insert_to_prio_queue(ad, rq, pq_idx); + + rcu_read_lock(); + if (rcu_dereference(ad->latency_model[ADIOS_READ].params)->base > 0 && + rcu_dereference(ad->latency_model[ADIOS_WRITE].params)->base > 0) + stable = true; + rcu_read_unlock(); + + if (stable) + ad->models_stable = true; +} + +// Insert multiple requests into the scheduler +static void adios_insert_requests(struct blk_mq_hw_ctx *hctx, + struct list_head *list, + blk_insert_t insert_flags) { + struct request_queue *q = hctx->queue; + struct adios_data *ad = q->elevator->elevator_data; + struct request *rq; + bool stop = false; + LIST_HEAD(free); + + do { + scoped_guard(spinlock_irqsave, &ad->lock) + for (int i = 0; i < ADIOS_MAX_INSERTS_PER_LOCK; i++) { + if (list_empty(list)) { + stop = true; + break; + } + rq = list_first_entry(list, struct request, queuelist); + list_del_init(&rq->queuelist); + if (likely(ad->models_stable)) + insert_request_post_stability(hctx, rq, insert_flags, &free); + else + insert_request_pre_stability(hctx, rq, insert_flags, &free); + }} while (!stop); + + blk_mq_free_requests(&free); +} + +// Prepare a request before it is inserted into the scheduler +static void adios_prepare_request(struct request *rq) { + struct adios_data *ad = rq->q->elevator->elevator_data; + struct adios_rq_data *rd = get_rq_data(rq); + + rq->elv.priv[0] = NULL; + + /* Allocate adios_rq_data from the memory pool */ + rd = kmem_cache_zalloc(ad->rq_data_pool, GFP_ATOMIC); + if (WARN(!rd, "adios_prepare_request: " + "Failed to allocate memory from rq_data_pool. rd is NULL\n")) + return; + + rd->rq = rq; + rq->elv.priv[0] = rd; +} + +static struct adios_rq_data *get_dl_first_rd(struct adios_data *ad, bool idx) { + struct rb_root_cached *root = &ad->dl_tree[idx]; + struct rb_node *first = rb_first_cached(root); + struct dl_group *dl_group = rb_entry(first, struct dl_group, node); + + return list_first_entry(&dl_group->rqs, struct adios_rq_data, dl_node); +} + +// Comparison function for sorting requests by block address +static int cmp_rq_pos(void *priv, + const struct list_head *a, const struct list_head *b) { + struct request *rq_a = list_entry(a, struct request, queuelist); + struct request *rq_b = list_entry(b, struct request, queuelist); + u64 pos_a = blk_rq_pos(rq_a); + u64 pos_b = blk_rq_pos(rq_b); + + return (int)(pos_a > pos_b) - (int)(pos_a < pos_b); +} + +#ifndef list_last_entry_or_null +#define list_last_entry_or_null(ptr, type, member) \ + (!list_empty(ptr) ? list_last_entry(ptr, type, member) : NULL) +#endif + +// Update the elevator direction +static void update_elv_direction(struct adios_data *ad) { + if (!ad->is_rotational) + return; + + bool page = ad->bq_page; + struct list_head *q = &ad->batch_queue[page][1]; + if (ad->bq_batch_order[page] < ADIOS_BO_ELEVATOR || list_empty(q)) { + ad->elv_direction = 0; + return; + } + + // Get first and last request positions in the queue + struct request *rq_a = list_first_entry(q, struct request, queuelist); + struct request *rq_b = list_last_entry (q, struct request, queuelist); + u64 pos_a = blk_rq_pos(rq_a); + u64 pos_b = blk_rq_pos(rq_b); + u64 avg_rq_pos = (pos_a + pos_b) >> 1; + + ad->elv_direction = !!(ad->head_pos > avg_rq_pos); +} + +// Fill the batch queues with requests from the deadline-sorted red-black tree +static bool fill_batch_queues(struct adios_data *ad, u64 tpl) { + struct adios_rq_data *rd; + struct request *rq; + struct list_head *dest_q; + u8 dest_idx; + u64 added_lat = 0; + u32 optype_count[ADIOS_OPTYPES] = {0}; + u32 count = 0; + u8 optype; + bool page = !ad->bq_page, dl_idx, bias_idx, update_bias; + u32 dl_queued; + u8 bq_batch_order; + bool stop = false; + + // Reset batch queue counts for the back page + memset(&ad->batch_count[page], 0, sizeof(ad->batch_count[page])); + + ad->bq_batch_order[page] = + bq_batch_order = ad->batch_order; + + do { + scoped_guard(spinlock_irqsave, &ad->lock) + for (int i = 0; i < ADIOS_MAX_DELETES_PER_LOCK; i++) { + bool has_base = false; + + dl_queued = eval_adios_state(ad, ADIOS_STATE_DL); + // Check if there are any requests queued in the deadline tree + if (!dl_queued) { + stop = true; + break; + } + + // Reads if both queues have requests, otherwise pick the non-empty. + dl_idx = dl_queued >> 1; + + // Get the first request from the deadline-sorted tree + rd = get_dl_first_rd(ad, dl_idx); + + bias_idx = ad->dl_bias < 0; + // If read and write requests are queued, choose one based on bias + if (dl_queued == 0x3) { + struct adios_rq_data *trd[2] = {get_dl_first_rd(ad, 0), rd}; + rd = trd[bias_idx]; + + update_bias = (trd[bias_idx]->deadline > trd[!bias_idx]->deadline); + } else + update_bias = (bias_idx == dl_idx); + + rq = rd->rq; + optype = adios_optype(rq); + + rcu_read_lock(); + has_base = + !!rcu_dereference(ad->latency_model[optype].params)->base; + rcu_read_unlock(); + + // Check batch size and total predicted latency + if (count && (!has_base || + ad->batch_count[page][optype] >= ad->batch_limit[optype] || + (tpl + added_lat + rd->pred_lat) > ad->global_latency_window)) { + stop = true; + break; + } + + if (update_bias) { + s64 sign = ((s64)bias_idx << 1) - 1; + if (unlikely(!rd->pred_lat)) + ad->dl_bias = sign; + else + // Adjust the bias based on the predicted latency + ad->dl_bias += sign * (s64)((rd->pred_lat * + adios_prio_to_wmult[ad->dl_prio[bias_idx] + 20]) >> 10); + } + + remove_request(ad, rq); + + // Add request to the corresponding batch queue + dest_idx = (bq_batch_order == ADIOS_BO_OPTYPE || optype == ADIOS_OTHER)? + optype : !!(rd->deadline != rq->start_time_ns); + dest_q = &ad->batch_queue[page][dest_idx]; + list_add_tail(&rq->queuelist, dest_q); + ad->bq_state[page] |= 1U << dest_idx; + ad->batch_count[page][optype]++; + optype_count[optype]++; + added_lat += rd->pred_lat; + count++; + }} while (!stop); + + if (bq_batch_order == ADIOS_BO_ELEVATOR && ad->batch_count[page][1] > 1) + list_sort(NULL, &ad->batch_queue[page][1], cmp_rq_pos); + + if (count) { + /* We're sure that every request's rd->managed == true */ + union adios_in_flight_rqs ifr = { + .count = count, + .total_pred_lat = added_lat, + }; + atomic64_add(ifr.scalar, &ad->in_flight_rqs.atomic); + + set_adios_state(ad, ADIOS_STATE_BQ, page, true); + + for (optype = 0; optype < ADIOS_OPTYPES; optype++) + if (ad->batch_actual_max_size[optype] < optype_count[optype]) + ad->batch_actual_max_size[optype] = optype_count[optype]; + if (ad->batch_actual_max_total < count) + ad->batch_actual_max_total = count; + } + return count; +} + +// Flip to the next batch queue page +static void flip_bq_page(struct adios_data *ad) { + ad->bq_page = !ad->bq_page; + update_elv_direction(ad); +} + +// Pop a request from the specified index (optype or elevator tier) +static inline struct request *pop_bq_request( + struct adios_data *ad, u8 idx, bool direction) { + bool page = ad->bq_page; + struct list_head *q = &ad->batch_queue[page][idx]; + struct request *rq = direction ? + list_last_entry_or_null (q, struct request, queuelist): + list_first_entry_or_null(q, struct request, queuelist); + if (rq) { + list_del_init(&rq->queuelist); + if (list_empty(q)) + ad->bq_state[page] &= ~(1U << idx); + } + return rq; +} + +static struct request *pop_next_bq_request_optype(struct adios_data *ad) { + u32 bq_state = ad->bq_state[ad->bq_page]; + if (!bq_state) return NULL; + + struct request *rq; + u32 bq_idx = __builtin_ctz(bq_state); + + // Dispatch based on optype (FIFO within each) or single-queue elevator + rq = pop_bq_request(ad, bq_idx, false); + return rq; +} + +static struct request *pop_next_bq_request_elevator(struct adios_data *ad) { + u32 bq_state = ad->bq_state[ad->bq_page]; + if (!bq_state) return NULL; + + struct request *rq; + u32 bq_idx = __builtin_ctz(bq_state); + bool direction = (bq_idx == 1) & ad->elv_direction; + + // Tier-2 (sync) is always high priority + // Tier-3 (async) uses the pre-calculated elevator direction + rq = pop_bq_request(ad, bq_idx, direction); + + /* If batch queue for the sync requests just became empty */ + if (bq_idx == 0 && rq && !(bq_state & 0x1)) + update_elv_direction(ad); + + return rq; +} + +// Returns the state of the batch queue page +static inline bool bq_page_has_rq(u32 bq_state, bool page) { + return bq_state & (1U << page); +} + +// Dispatch a request from the batch queues +static struct request *dispatch_from_bq(struct adios_data *ad) { + struct request *rq; + + guard(spinlock_irqsave)(&ad->bq_lock); + + u32 state = get_adios_state(ad); + u32 bq_state = eval_this_adios_state(state, ADIOS_STATE_BQ); + u32 bq_curr_page_has_rq = bq_page_has_rq(bq_state, ad->bq_page); + union adios_in_flight_rqs ifr; + ifr.scalar = atomic64_read(&ad->in_flight_rqs.atomic); + u64 tpl = ifr.total_pred_lat; + + // Refill the batch queues if the back page is empty, dl_tree has work, and + // current page is empty or the total ongoing latency is below the threshold + if (!bq_page_has_rq(bq_state, !ad->bq_page) && + (!bq_curr_page_has_rq || (!tpl || tpl < div_u64( + ad->global_latency_window * ad->bq_refill_below_ratio, 100))) && + eval_this_adios_state(state, ADIOS_STATE_DL)) + fill_batch_queues(ad, tpl); + + // If current batch queue page is empty, and the other page has work, flip + if (!bq_curr_page_has_rq && + bq_page_has_rq(eval_adios_state(ad, ADIOS_STATE_BQ), !ad->bq_page)) + flip_bq_page(ad); + + // Use the per-page state to decide the dispatch logic, ensuring correctness + rq = (ad->bq_batch_order[ad->bq_page] == ADIOS_BO_ELEVATOR) ? + pop_next_bq_request_elevator(ad): + pop_next_bq_request_optype(ad); + + if (rq) { + bool page = ad->bq_page; + bool is_empty = !ad->bq_state[page]; + if (is_empty) + set_adios_state(ad, ADIOS_STATE_BQ, page, false); + return rq; + } + + return NULL; +} + +// Dispatch a request from the priority queue +static struct request *dispatch_from_pq(struct adios_data *ad) { + struct request *rq = NULL; + + guard(spinlock_irqsave)(&ad->pq_lock); + u32 pq_state = eval_adios_state(ad, ADIOS_STATE_PQ); + u8 pq_idx = pq_state >> 1; + struct list_head *q = &ad->prio_queue[pq_idx]; + + if (unlikely(list_empty(q))) return NULL; + + rq = list_first_entry(q, struct request, queuelist); + list_del_init(&rq->queuelist); + if (list_empty(q)) { + set_adios_state(ad, ADIOS_STATE_PQ, pq_idx, false); + update_elv_direction(ad); + } + return rq; +} + +static bool release_barrier_requests(struct adios_data *ad) { + u32 moved_count = 0; + LIST_HEAD(local_list); + + scoped_guard(spinlock_irqsave, &ad->barrier_lock) { + if (!list_empty(&ad->barrier_queue)) { + struct request *trq, *next; + bool first_barrier_moved = false; + + list_for_each_entry_safe(trq, next, &ad->barrier_queue, queuelist) { + if (!first_barrier_moved) { + list_del_init(&trq->queuelist); + insert_to_prio_queue(ad, trq, 1); + moved_count++; + first_barrier_moved = true; + continue; + } + + if (trq->cmd_flags & REQ_OP_FLUSH) + break; + + list_move_tail(&trq->queuelist, &local_list); + moved_count++; + } + + if (list_empty(&ad->barrier_queue)) + set_adios_state(ad, ADIOS_STATE_BP, 0, false); + } + } + + if (!moved_count) + return false; + + if (!list_empty(&local_list)) { + struct request *trq, *next; + LIST_HEAD(free_list); + + /* ad->lock is already held */ + list_for_each_entry_safe(trq, next, &local_list, queuelist) { + list_del_init(&trq->queuelist); + if (merge_or_insert_to_dl_tree(ad, trq, ad->queue, &free_list)) + continue; + } + + if (!list_empty(&free_list)) + blk_mq_free_requests(&free_list); + } + + return true; +} + +// Dispatch a request to the hardware queue +static struct request *adios_dispatch_request(struct blk_mq_hw_ctx *hctx) { + struct adios_data *ad = hctx->queue->elevator->elevator_data; + struct request *rq; + +retry: + rq = dispatch_from_pq(ad); + if (rq) + goto found; + + rq = dispatch_from_bq(ad); + if (rq) + goto found; + + /* + * If all active queues are empty, check if we need to process a barrier. + * This is the trigger to release requests that were held in barrier_queue + * due to a REQ_OP_FLUSH barrier. + */ + if (eval_adios_state(ad, ADIOS_STATE_BP)) { + bool barrier_released = false; + scoped_guard(spinlock_irqsave, &ad->lock) + barrier_released = release_barrier_requests(ad); + if (barrier_released) + goto retry; + } + + return NULL; +found: + if (ad->is_rotational) + ad->head_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); + + rq->rq_flags |= RQF_STARTED; + return rq; +} + +// Timer callback function to periodically update latency models +static void update_timer_callback(struct timer_list *t) { + struct adios_data *ad = timer_container_of(ad, t, update_timer); + + for (u8 optype = 0; optype < ADIOS_OPTYPES; optype++) + latency_model_update(ad, &ad->latency_model[optype]); +} + +// Handle the completion of a request +static void adios_completed_request(struct request *rq, u64 now) { + struct adios_data *ad = rq->q->elevator->elevator_data; + struct adios_rq_data *rd = get_rq_data(rq); + union adios_in_flight_rqs ifr = { .scalar = 0 }; + + if (rd->managed) { + union adios_in_flight_rqs ifr_to_sub = { + .count = 1, + .total_pred_lat = rd->pred_lat, + }; + ifr.scalar = atomic64_sub_return( + ifr_to_sub.scalar, &ad->in_flight_rqs.atomic); + } + u8 optype = adios_optype(rq); + + if (optype == ADIOS_OTHER) { + // Non-positional commands make the head position unpredictable. + // Invalidate our knowledge of the last completed position. + if (ad->is_rotational) + ad->last_completed_pos = 0; + return; + } + + u64 lct = ad->last_completed_time ?: rq->io_start_time_ns; + ad->last_completed_time = (ifr.count) ? now : 0; + + if (!rq->io_start_time_ns || !rd->block_size || unlikely(now < lct)) + return; + + u64 latency = now - lct; + if (latency > ad->lat_model_latency_limit) + return; + + u32 weight = 1; + if (ad->is_rotational) { + sector_t current_pos = blk_rq_pos(rq); + // Only calculate seek distance if we have a valid last position. + if (ad->last_completed_pos > 0) { + u64 seek_distance = abs( + (s64)current_pos - (s64)ad->last_completed_pos); + weight = 65 - __builtin_clzll(seek_distance); + } + // Update (or re-synchronize) our knowledge of the head position. + ad->last_completed_pos = current_pos + blk_rq_sectors(rq); + } + + latency_model_input(ad, &ad->latency_model[optype], + rd->block_size, latency, rd->pred_lat, weight); + timer_reduce(&ad->update_timer, jiffies + msecs_to_jiffies(100)); +} + +// Clean up after a request is finished +static void adios_finish_request(struct request *rq) { + struct adios_data *ad = rq->q->elevator->elevator_data; + + if (rq->elv.priv[0]) { + // Free adios_rq_data back to the memory pool + kmem_cache_free(ad->rq_data_pool, get_rq_data(rq)); + rq->elv.priv[0] = NULL; + } +} + +// Check if there are any requests available for dispatch +static bool adios_has_work(struct blk_mq_hw_ctx *hctx) { + struct adios_data *ad = hctx->queue->elevator->elevator_data; + + return atomic_read(&ad->state) != 0; +} + +// Initialize the scheduler-specific data when initializing the request queue +static int adios_init_sched(struct request_queue *q, struct elevator_queue *eq) { + struct adios_data *ad; + int ret = -ENOMEM; + u8 optype = 0; + + ad = kzalloc_node(sizeof(*ad), GFP_KERNEL, q->node); + if (!ad) { + pr_err("adios: Failed to create adios_data\n"); + goto put_eq; + } + + eq->elevator_data = ad; + + // Create a memory pool for adios_rq_data + ad->rq_data_pool = kmem_cache_create("rq_data_pool", + sizeof(struct adios_rq_data), + 0, SLAB_HWCACHE_ALIGN, NULL); + if (!ad->rq_data_pool) { + pr_err("adios: Failed to create rq_data_pool\n"); + goto free_ad; + } + + /* Create a memory pool for dl_group */ + ad->dl_group_pool = kmem_cache_create("dl_group_pool", + sizeof(struct dl_group), + 0, SLAB_HWCACHE_ALIGN, NULL); + if (!ad->dl_group_pool) { + pr_err("adios: Failed to create dl_group_pool\n"); + goto destroy_rq_data_pool; + } + + for (int i = 0; i < ADIOS_PQ_LEVELS; i++) + INIT_LIST_HEAD(&ad->prio_queue[i]); + + for (u8 i = 0; i < ADIOS_DL_TYPES; i++) { + ad->dl_tree[i] = RB_ROOT_CACHED; + ad->dl_prio[i] = default_dl_prio[i]; + } + ad->dl_bias = 0; + + for (u8 page = 0; page < ADIOS_BQ_PAGES; page++) + for (optype = 0; optype < ADIOS_OPTYPES; optype++) + INIT_LIST_HEAD(&ad->batch_queue[page][optype]); + + ad->aggr_buckets = kzalloc(sizeof(*ad->aggr_buckets), GFP_KERNEL); + if (!ad->aggr_buckets) { + pr_err("adios: Failed to allocate aggregation buckets\n"); + goto destroy_dl_group_pool; + } + + for (optype = 0; optype < ADIOS_OPTYPES; optype++) { + struct latency_model *model = &ad->latency_model[optype]; + struct latency_model_params *params; + + spin_lock_init(&model->update_lock); + params = kzalloc(sizeof(*params), GFP_KERNEL); + if (!params) { + pr_err("adios: Failed to allocate latency_model_params\n"); + goto free_buckets; + } + params->last_update_jiffies = jiffies; + RCU_INIT_POINTER(model->params, params); + + model->pcpu_buckets = alloc_percpu(struct lm_buckets); + if (!model->pcpu_buckets) { + pr_err("adios: Failed to allocate per-CPU buckets\n"); + kfree(params); + goto free_buckets; + } + + model->lm_shrink_at_kreqs = default_lm_shrink_at_kreqs; + model->lm_shrink_at_gbytes = default_lm_shrink_at_gbytes; + model->lm_shrink_resist = default_lm_shrink_resist; + } + + for (optype = 0; optype < ADIOS_OPTYPES; optype++) { + ad->latency_target[optype] = default_latency_target[optype]; + ad->batch_limit[optype] = default_batch_limit[optype]; + } + + eq->elevator_data = ad; + + ad->is_rotational = !!(q->limits.features & BLK_FEAT_ROTATIONAL); + ad->global_latency_window = (ad->is_rotational)? + default_global_latency_window_rotational: + default_global_latency_window; + ad->bq_refill_below_ratio = default_bq_refill_below_ratio; + ad->lat_model_latency_limit = default_lat_model_latency_limit; + ad->batch_order = default_batch_order; + ad->compliance_flags = default_compliance_flags; + + ad->models_stable = false; + + atomic_set(&ad->state, 0); + + spin_lock_init(&ad->lock); + spin_lock_init(&ad->pq_lock); + spin_lock_init(&ad->bq_lock); + spin_lock_init(&ad->barrier_lock); + INIT_LIST_HEAD(&ad->barrier_queue); + + timer_setup(&ad->update_timer, update_timer_callback, 0); + + /* We dispatch from request queue wide instead of hw queue */ + blk_queue_flag_set(QUEUE_FLAG_SQ_SCHED, q); + + ad->queue = q; + blk_stat_enable_accounting(q); + + q->elevator = eq; + adios_depth_updated(q); + return 0; + +free_buckets: + pr_err("adios: Failed to allocate per-cpu buckets\n"); + while (optype-- > 0) { + struct latency_model *prev_model = &ad->latency_model[optype]; + kfree(rcu_access_pointer(prev_model->params)); + free_percpu(prev_model->pcpu_buckets); + } + kfree(ad->aggr_buckets); +destroy_dl_group_pool: + kmem_cache_destroy(ad->dl_group_pool); +destroy_rq_data_pool: + kmem_cache_destroy(ad->rq_data_pool); +free_ad: + kfree(ad); +put_eq: + kobject_put(&eq->kobj); + return ret; +} + +// Clean up and free resources when exiting the scheduler +static void adios_exit_sched(struct elevator_queue *e) { + struct adios_data *ad = e->elevator_data; + + timer_shutdown_sync(&ad->update_timer); + + WARN_ON_ONCE(!list_empty(&ad->barrier_queue)); + for (int i = 0; i < 2; i++) + WARN_ON_ONCE(!list_empty(&ad->prio_queue[i])); + + for (u8 i = 0; i < ADIOS_OPTYPES; i++) { + struct latency_model *model = &ad->latency_model[i]; + struct latency_model_params *params = rcu_access_pointer(model->params); + + RCU_INIT_POINTER(model->params, NULL); + kfree_rcu(params, rcu); + + free_percpu(model->pcpu_buckets); + } + + synchronize_rcu(); + + kfree(ad->aggr_buckets); + + if (ad->rq_data_pool) + kmem_cache_destroy(ad->rq_data_pool); + + if (ad->dl_group_pool) + kmem_cache_destroy(ad->dl_group_pool); + + blk_stat_disable_accounting(ad->queue); + + kfree(ad); +} + +static void sideload_latency_model( + struct latency_model *model, u64 base, u64 slope) { + struct latency_model_params *old_params, *new_params; + unsigned long flags; + + new_params = kzalloc(sizeof(*new_params), GFP_KERNEL); + if (!new_params) + return; + + spin_lock_irqsave(&model->update_lock, flags); + + old_params = rcu_dereference_protected(model->params, + lockdep_is_held(&model->update_lock)); + + new_params->last_update_jiffies = jiffies; + + // Initialize base and its statistics as a single sample. + new_params->base = base; + new_params->small_sum_delay = base; + new_params->small_count = 1; + + // Initialize slope and its statistics as a single sample. + new_params->slope = slope; + new_params->large_sum_delay = slope; + new_params->large_sum_bsize = 1024; /* Corresponds to 1 KiB */ + + lm_reset_pcpu_buckets(model); + + rcu_assign_pointer(model->params, new_params); + spin_unlock_irqrestore(&model->update_lock, flags); + + kfree_rcu(old_params, rcu); +} + +// Define sysfs attributes for operation types +#define SYSFS_OPTYPE_DECL(name, optype) \ +static ssize_t adios_lat_model_##name##_show( \ + struct elevator_queue *e, char *page) { \ + struct adios_data *ad = e->elevator_data; \ + struct latency_model *model = &ad->latency_model[optype]; \ + struct latency_model_params *params; \ + ssize_t len = 0; \ + u64 base, slope; \ + rcu_read_lock(); \ + params = rcu_dereference(model->params); \ + base = params->base; \ + slope = params->slope; \ + rcu_read_unlock(); \ + len += sprintf(page, "base : %llu ns\n", base); \ + len += sprintf(page + len, "slope: %llu ns/KiB\n", slope); \ + return len; \ +} \ +static ssize_t adios_lat_model_##name##_store( \ + struct elevator_queue *e, const char *page, size_t count) { \ + struct adios_data *ad = e->elevator_data; \ + struct latency_model *model = &ad->latency_model[optype]; \ + u64 base, slope; \ + int ret; \ + ret = sscanf(page, "%llu %llu", &base, &slope); \ + if (ret != 2) \ + return -EINVAL; \ + sideload_latency_model(model, base, slope); \ + reset_buckets(ad->aggr_buckets); \ + return count; \ +} \ +static ssize_t adios_lat_target_##name##_show( \ + struct elevator_queue *e, char *page) { \ + struct adios_data *ad = e->elevator_data; \ + return sprintf(page, "%llu\n", ad->latency_target[optype]); \ +} \ +static ssize_t adios_lat_target_##name##_store( \ + struct elevator_queue *e, const char *page, size_t count) { \ + struct adios_data *ad = e->elevator_data; \ + unsigned long nsec; \ + int ret; \ + ret = kstrtoul(page, 10, &nsec); \ + if (ret) \ + return ret; \ + sideload_latency_model(&ad->latency_model[optype], 0, 0); \ + ad->latency_target[optype] = nsec; \ + return count; \ +} \ +static ssize_t adios_batch_limit_##name##_show( \ + struct elevator_queue *e, char *page) { \ + struct adios_data *ad = e->elevator_data; \ + return sprintf(page, "%u\n", ad->batch_limit[optype]); \ +} \ +static ssize_t adios_batch_limit_##name##_store( \ + struct elevator_queue *e, const char *page, size_t count) { \ + unsigned long max_batch; \ + int ret; \ + ret = kstrtoul(page, 10, &max_batch); \ + if (ret || max_batch == 0) \ + return -EINVAL; \ + struct adios_data *ad = e->elevator_data; \ + ad->batch_limit[optype] = max_batch; \ + return count; \ +} + +SYSFS_OPTYPE_DECL(read, ADIOS_READ); +SYSFS_OPTYPE_DECL(write, ADIOS_WRITE); +SYSFS_OPTYPE_DECL(discard, ADIOS_DISCARD); + +// Show the maximum batch size actually achieved for each operation type +static ssize_t adios_batch_actual_max_show( + struct elevator_queue *e, char *page) { + struct adios_data *ad = e->elevator_data; + u32 total_count, read_count, write_count, discard_count; + + total_count = ad->batch_actual_max_total; + read_count = ad->batch_actual_max_size[ADIOS_READ]; + write_count = ad->batch_actual_max_size[ADIOS_WRITE]; + discard_count = ad->batch_actual_max_size[ADIOS_DISCARD]; + + return sprintf(page, + "Total : %u\nDiscard: %u\nRead : %u\nWrite : %u\n", + total_count, discard_count, read_count, write_count); +} + +#define SYSFS_ULL_DECL(field, min_val, max_val) \ +static ssize_t adios_##field##_show( \ + struct elevator_queue *e, char *page) { \ + struct adios_data *ad = e->elevator_data; \ + return sprintf(page, "%llu\n", ad->field); \ +} \ +static ssize_t adios_##field##_store( \ + struct elevator_queue *e, const char *page, size_t count) { \ + struct adios_data *ad = e->elevator_data; \ + unsigned long val; \ + int ret; \ + ret = kstrtoul(page, 10, &val); \ + if (ret || val < (min_val) || val > (max_val)) \ + return -EINVAL; \ + ad->field = val; \ + return count; \ +} + +SYSFS_ULL_DECL(global_latency_window, 0, ULLONG_MAX) +SYSFS_ULL_DECL(compliance_flags, 0, ULLONG_MAX) + +#define SYSFS_INT_DECL(field, min_val, max_val) \ +static ssize_t adios_##field##_show( \ + struct elevator_queue *e, char *page) { \ + struct adios_data *ad = e->elevator_data; \ + return sprintf(page, "%d\n", ad->field); \ +} \ +static ssize_t adios_##field##_store( \ + struct elevator_queue *e, const char *page, size_t count) { \ + struct adios_data *ad = e->elevator_data; \ + int val; \ + int ret; \ + ret = kstrtoint(page, 10, &val); \ + if (ret || val < (min_val) || val > (max_val)) \ + return -EINVAL; \ + ad->field = val; \ + return count; \ +} + +SYSFS_INT_DECL(bq_refill_below_ratio, 0, 100) +SYSFS_INT_DECL(lat_model_latency_limit, 0, 2*NSEC_PER_SEC) +SYSFS_INT_DECL(batch_order, ADIOS_BO_OPTYPE, !!ad->is_rotational) + +// Show the read priority +static ssize_t adios_read_priority_show( + struct elevator_queue *e, char *page) { + struct adios_data *ad = e->elevator_data; + return sprintf(page, "%d\n", ad->dl_prio[0]); +} + +// Set the read priority +static ssize_t adios_read_priority_store( + struct elevator_queue *e, const char *page, size_t count) { + struct adios_data *ad = e->elevator_data; + int prio; + int ret; + + ret = kstrtoint(page, 10, &prio); + if (ret || prio < -20 || prio > 19) + return -EINVAL; + + guard(spinlock_irqsave)(&ad->lock); + ad->dl_prio[0] = prio; + ad->dl_bias = 0; + + return count; +} + +// Reset batch queue statistics +static ssize_t adios_reset_bq_stats_store( + struct elevator_queue *e, const char *page, size_t count) { + struct adios_data *ad = e->elevator_data; + unsigned long val; + int ret; + + ret = kstrtoul(page, 10, &val); + if (ret || val != 1) + return -EINVAL; + + for (u8 i = 0; i < ADIOS_OPTYPES; i++) + ad->batch_actual_max_size[i] = 0; + + ad->batch_actual_max_total = 0; + + return count; +} + +// Reset the latency model parameters or load them from user input +static ssize_t adios_reset_lat_model_store( + struct elevator_queue *e, const char *page, size_t count) +{ + struct adios_data *ad = e->elevator_data; + struct latency_model *model; + int ret; + + /* + * Differentiate between two modes based on input format: + * 1. "1": Fully reset the model (backward compatibility). + * 2. "R_base R_slope W_base W_slope D_base D_slope": Load values. + */ + if (!strchr(page, ' ')) { + // Mode 1: Full reset. + unsigned long val; + + ret = kstrtoul(page, 10, &val); + if (ret || val != 1) + return -EINVAL; + + for (u8 i = 0; i < ADIOS_OPTYPES; i++) { + model = &ad->latency_model[i]; + sideload_latency_model(model, 0, 0); + } + } else { + // Mode 2: Load initial values for all latency models. + u64 params[3][2]; /* 0:base, 1:slope for R, W, D */ + + ret = sscanf(page, "%llu %llu %llu %llu %llu %llu", + ¶ms[ADIOS_READ ][0], ¶ms[ADIOS_READ ][1], + ¶ms[ADIOS_WRITE ][0], ¶ms[ADIOS_WRITE ][1], + ¶ms[ADIOS_DISCARD][0], ¶ms[ADIOS_DISCARD][1]); + + if (ret != 6) + return -EINVAL; + + for (u8 i = ADIOS_READ; i <= ADIOS_DISCARD; i++) { + model = &ad->latency_model[i]; + sideload_latency_model(model, params[i][0], params[i][1]); + } + } + reset_buckets(ad->aggr_buckets); + + return count; +} + +// Show the ADIOS version +static ssize_t adios_version_show(struct elevator_queue *e, char *page) { + return sprintf(page, "%s\n", ADIOS_VERSION); +} + +// Define sysfs attributes for dynamic thresholds +#define SHRINK_THRESHOLD_ATTR_RW(name, model_field, min_value, max_value) \ +static ssize_t adios_shrink_##name##_store( \ + struct elevator_queue *e, const char *page, size_t count) { \ + struct adios_data *ad = e->elevator_data; \ + unsigned long val; \ + int ret; \ + ret = kstrtoul(page, 10, &val); \ + if (ret || val < min_value || val > max_value) \ + return -EINVAL; \ + for (u8 i = 0; i < ADIOS_OPTYPES; i++) { \ + struct latency_model *model = &ad->latency_model[i]; \ + unsigned long flags; \ + spin_lock_irqsave(&model->update_lock, flags); \ + model->model_field = val; \ + spin_unlock_irqrestore(&model->update_lock, flags); \ + } \ + return count; \ +} \ +static ssize_t adios_shrink_##name##_show( \ + struct elevator_queue *e, char *page) { \ + struct adios_data *ad = e->elevator_data; \ + u32 val = 0; \ + unsigned long flags; \ + struct latency_model *model = &ad->latency_model[0]; \ + spin_lock_irqsave(&model->update_lock, flags); \ + val = model->model_field; \ + spin_unlock_irqrestore(&model->update_lock, flags); \ + return sprintf(page, "%u\n", val); \ +} + +SHRINK_THRESHOLD_ATTR_RW(at_kreqs, lm_shrink_at_kreqs, 1, 100000) +SHRINK_THRESHOLD_ATTR_RW(at_gbytes, lm_shrink_at_gbytes, 1, 1000) +SHRINK_THRESHOLD_ATTR_RW(resist, lm_shrink_resist, 1, 3) + +// Define sysfs attributes +#define AD_ATTR(name, show_func, store_func) \ + __ATTR(name, 0644, show_func, store_func) +#define AD_ATTR_RW(name) \ + __ATTR(name, 0644, adios_##name##_show, adios_##name##_store) +#define AD_ATTR_RO(name) \ + __ATTR(name, 0444, adios_##name##_show, NULL) +#define AD_ATTR_WO(name) \ + __ATTR(name, 0200, NULL, adios_##name##_store) + +// Define sysfs attributes for ADIOS scheduler +static struct elv_fs_entry adios_sched_attrs[] = { + AD_ATTR_RO(batch_actual_max), + AD_ATTR_RW(bq_refill_below_ratio), + AD_ATTR_RW(global_latency_window), + AD_ATTR_RW(lat_model_latency_limit), + AD_ATTR_RW(batch_order), + AD_ATTR_RW(compliance_flags), + + AD_ATTR_RW(batch_limit_read), + AD_ATTR_RW(batch_limit_write), + AD_ATTR_RW(batch_limit_discard), + + AD_ATTR_RW(lat_model_read), + AD_ATTR_RW(lat_model_write), + AD_ATTR_RW(lat_model_discard), + + AD_ATTR_RW(lat_target_read), + AD_ATTR_RW(lat_target_write), + AD_ATTR_RW(lat_target_discard), + + AD_ATTR_RW(shrink_at_kreqs), + AD_ATTR_RW(shrink_at_gbytes), + AD_ATTR_RW(shrink_resist), + + AD_ATTR_RW(read_priority), + + AD_ATTR_WO(reset_bq_stats), + AD_ATTR_WO(reset_lat_model), + AD_ATTR(adios_version, adios_version_show, NULL), + + __ATTR_NULL +}; + +// Define the ADIOS scheduler type +static struct elevator_type mq_adios = { + .ops = { + .next_request = elv_rb_latter_request, + .former_request = elv_rb_former_request, + .limit_depth = adios_limit_depth, + .depth_updated = adios_depth_updated, + .request_merged = adios_request_merged, + .requests_merged = adios_merged_requests, + .bio_merge = adios_bio_merge, + .insert_requests = adios_insert_requests, + .prepare_request = adios_prepare_request, + .dispatch_request = adios_dispatch_request, + .completed_request = adios_completed_request, + .finish_request = adios_finish_request, + .has_work = adios_has_work, + .init_sched = adios_init_sched, + .exit_sched = adios_exit_sched, + }, + .elevator_attrs = adios_sched_attrs, + .elevator_name = "adios", + .elevator_owner = THIS_MODULE, +}; +MODULE_ALIAS("mq-adios-iosched"); + +#define ADIOS_PROGNAME "Adaptive Deadline I/O Scheduler" +#define ADIOS_AUTHOR "Masahito Suzuki" + +// Initialize the ADIOS scheduler module +static int __init adios_init(void) { + printk(KERN_INFO "%s %s by %s\n", + ADIOS_PROGNAME, ADIOS_VERSION, ADIOS_AUTHOR); + return elv_register(&mq_adios); +} + +// Exit the ADIOS scheduler module +static void __exit adios_exit(void) { + elv_unregister(&mq_adios); +} + +module_init(adios_init); +module_exit(adios_exit); + +MODULE_AUTHOR(ADIOS_AUTHOR); +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION(ADIOS_PROGNAME); \ No newline at end of file diff --git a/block/elevator.c b/block/elevator.c index a2f8b2251dc6..06cabf5b356d 100644 --- a/block/elevator.c +++ b/block/elevator.c @@ -739,6 +739,19 @@ void elevator_set_default(struct request_queue *q) if (q->tag_set->flags & BLK_MQ_F_NO_SCHED_BY_DEFAULT) return; +#ifdef CONFIG_MQ_IOSCHED_DEFAULT_ADIOS + ctx.name = "adios"; +#else // !CONFIG_MQ_IOSCHED_DEFAULT_ADIOS + bool is_sq = q->nr_hw_queues == 1 || blk_mq_is_shared_tags(q->tag_set->flags); +#if defined(CONFIG_CACHY) && defined(CONFIG_IOSCHED_BFQ) + if (is_sq) + ctx.name = "bfq"; +#else + if (!is_sq) + return; +#endif /* CONFIG_CACHY && CONFIG_IOSCHED_BFQ */ +#endif // CONFIG_MQ_IOSCHED_DEFAULT_ADIOS + /* * For single queue devices, default to using mq-deadline. If we * have multiple queues or mq-deadline is not available, default @@ -748,13 +761,10 @@ void elevator_set_default(struct request_queue *q) if (!ctx.type) return; - if ((q->nr_hw_queues == 1 || - blk_mq_is_shared_tags(q->tag_set->flags))) { - err = elevator_change(q, &ctx); - if (err < 0) - pr_warn("\"%s\" elevator initialization, failed %d, falling back to \"none\"\n", - ctx.name, err); - } + err = elevator_change(q, &ctx); + if (err < 0) + pr_warn("\"%s\" elevator initialization, failed %d, falling back to \"none\"\n", + ctx.name, err); elevator_put(ctx.type); } diff --git a/drivers/Makefile b/drivers/Makefile index ccc05f1eae3e..83bb1e476a73 100644 --- a/drivers/Makefile +++ b/drivers/Makefile @@ -64,14 +64,8 @@ obj-y += char/ # iommu/ comes before gpu as gpu are using iommu controllers obj-y += iommu/ -# gpu/ comes after char for AGP vs DRM startup and after iommu -obj-y += gpu/ - obj-$(CONFIG_CONNECTOR) += connector/ -# i810fb depends on char/agp/ -obj-$(CONFIG_FB_I810) += video/fbdev/i810/ - obj-$(CONFIG_PARPORT) += parport/ obj-y += base/ block/ misc/ mfd/ nfc/ obj-$(CONFIG_LIBNVDIMM) += nvdimm/ @@ -83,6 +77,13 @@ obj-y += macintosh/ obj-y += scsi/ obj-y += nvme/ obj-$(CONFIG_ATA) += ata/ + +# gpu/ comes after char for AGP vs DRM startup and after iommu +obj-y += gpu/ + +# i810fb depends on char/agp/ +obj-$(CONFIG_FB_I810) += video/fbdev/i810/ + obj-$(CONFIG_TARGET_CORE) += target/ obj-$(CONFIG_MTD) += mtd/ obj-$(CONFIG_SPI) += spi/ diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c index 931d0081169b..02ee4c4fa5b7 100644 --- a/drivers/ata/ahci.c +++ b/drivers/ata/ahci.c @@ -1672,7 +1672,7 @@ static irqreturn_t ahci_thunderx_irq_handler(int irq, void *dev_instance) } #endif -static void ahci_remap_check(struct pci_dev *pdev, int bar, +static int ahci_remap_check(struct pci_dev *pdev, int bar, struct ahci_host_priv *hpriv) { int i; @@ -1685,7 +1685,7 @@ static void ahci_remap_check(struct pci_dev *pdev, int bar, pci_resource_len(pdev, bar) < SZ_512K || bar != AHCI_PCI_BAR_STANDARD || !(readl(hpriv->mmio + AHCI_VSCAP) & 1)) - return; + return 0; cap = readq(hpriv->mmio + AHCI_REMAP_CAP); for (i = 0; i < AHCI_MAX_REMAP; i++) { @@ -1700,18 +1700,11 @@ static void ahci_remap_check(struct pci_dev *pdev, int bar, } if (!hpriv->remapped_nvme) - return; - - dev_warn(&pdev->dev, "Found %u remapped NVMe devices.\n", - hpriv->remapped_nvme); - dev_warn(&pdev->dev, - "Switch your BIOS from RAID to AHCI mode to use them.\n"); + return 0; - /* - * Don't rely on the msi-x capability in the remap case, - * share the legacy interrupt across ahci and remapped devices. - */ - hpriv->flags |= AHCI_HFLAG_NO_MSI; + /* Abort probe, allowing intel-nvme-remap to step in when available */ + dev_info(&pdev->dev, "Device will be handled by intel-nvme-remap.\n"); + return -ENODEV; } static int ahci_get_irq_vector(struct ata_host *host, int port) @@ -1975,7 +1968,9 @@ static int ahci_init_one(struct pci_dev *pdev, const struct pci_device_id *ent) return -ENOMEM; /* detect remapped nvme devices */ - ahci_remap_check(pdev, ahci_pci_bar, hpriv); + rc = ahci_remap_check(pdev, ahci_pci_bar, hpriv); + if (rc) + return rc; sysfs_add_file_to_group(&pdev->dev.kobj, &dev_attr_remapped_nvme.attr, diff --git a/drivers/cpufreq/Kconfig.x86 b/drivers/cpufreq/Kconfig.x86 index 2c5c228408bf..918e2bebfe78 100644 --- a/drivers/cpufreq/Kconfig.x86 +++ b/drivers/cpufreq/Kconfig.x86 @@ -9,7 +9,6 @@ config X86_INTEL_PSTATE select ACPI_PROCESSOR if ACPI select ACPI_CPPC_LIB if X86_64 && ACPI && SCHED_MC_PRIO select CPU_FREQ_GOV_PERFORMANCE - select CPU_FREQ_GOV_SCHEDUTIL if SMP help This driver provides a P state for Intel core processors. The driver implements an internal governor and will become @@ -39,7 +38,6 @@ config X86_AMD_PSTATE depends on X86 && ACPI select ACPI_PROCESSOR select ACPI_CPPC_LIB if X86_64 - select CPU_FREQ_GOV_SCHEDUTIL if SMP help This driver adds a CPUFreq driver which utilizes a fine grain processor performance frequency control range instead of legacy diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c index ec4abe374573..00c78f252e03 100644 --- a/drivers/cpufreq/intel_pstate.c +++ b/drivers/cpufreq/intel_pstate.c @@ -3914,6 +3914,8 @@ static int __init intel_pstate_setup(char *str) if (!strcmp(str, "disable")) no_load = 1; + else if (!strcmp(str, "enable")) + no_load = 0; else if (!strcmp(str, "active")) default_driver = &intel_pstate; else if (!strcmp(str, "passive")) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index b20a06abb65d..4dd9214873c1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -164,6 +164,7 @@ struct amdgpu_watchdog_timer { */ extern int amdgpu_modeset; extern unsigned int amdgpu_vram_limit; +extern int amdgpu_ignore_min_pcap; extern int amdgpu_vis_vram_limit; extern int amdgpu_gart_size; extern int amdgpu_gtt_size; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 39387da8586b..26b4dcc58471 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -150,6 +150,7 @@ enum AMDGPU_DEBUG_MASK { }; unsigned int amdgpu_vram_limit = UINT_MAX; +int amdgpu_ignore_min_pcap = 0; /* do not ignore by default */ int amdgpu_vis_vram_limit; int amdgpu_gart_size = -1; /* auto */ int amdgpu_gtt_size = -1; /* auto */ @@ -272,6 +273,15 @@ struct amdgpu_watchdog_timer amdgpu_watchdog_timer = { .period = 0x0, /* default to 0x0 (timeout disable) */ }; +/** + * DOC: ignore_min_pcap (int) + * Ignore the minimum power cap. + * Useful on graphics cards where the minimum power cap is very high. + * The default is 0 (Do not ignore). + */ +MODULE_PARM_DESC(ignore_min_pcap, "Ignore the minimum power cap"); +module_param_named(ignore_min_pcap, amdgpu_ignore_min_pcap, int, 0600); + /** * DOC: vramlimit (int) * Restrict the total amount of VRAM in MiB for testing. The default is 0 (Use full VRAM). diff --git a/drivers/gpu/drm/amd/display/Kconfig b/drivers/gpu/drm/amd/display/Kconfig index abd3b6564373..46937e6fa78d 100644 --- a/drivers/gpu/drm/amd/display/Kconfig +++ b/drivers/gpu/drm/amd/display/Kconfig @@ -56,4 +56,10 @@ config DRM_AMD_SECURE_DISPLAY This option enables the calculation of crc of specific region via debugfs. Cooperate with specific DMCU FW. +config AMD_PRIVATE_COLOR + bool "Enable KMS color management by AMD for AMD" + default n + help + This option extends the KMS color management API with AMD driver-specific properties to enhance the color management support on AMD Steam Deck. + endmenu diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c index a8a59126b2d2..75fa46f18588 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -4931,7 +4931,7 @@ static int amdgpu_dm_mode_config_init(struct amdgpu_device *adev) return r; } -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR if (amdgpu_dm_create_color_properties(adev)) { dc_state_release(state->context); kfree(state); diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c index 1dcc79b35225..be01e61bd039 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c @@ -181,7 +181,7 @@ static inline struct fixed31_32 amdgpu_dm_fixpt_from_s3132(__u64 x) return val; } -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR /* Pre-defined Transfer Functions (TF) * * AMD driver supports pre-defined mathematical functions for transferring diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c index 9fcd72d87d25..6137439a7009 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c @@ -503,7 +503,7 @@ static int amdgpu_dm_crtc_late_register(struct drm_crtc *crtc) } #endif -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR /** * dm_crtc_additional_color_mgmt - enable additional color properties * @crtc: DRM CRTC @@ -585,7 +585,7 @@ static const struct drm_crtc_funcs amdgpu_dm_crtc_funcs = { #if defined(CONFIG_DEBUG_FS) .late_register = amdgpu_dm_crtc_late_register, #endif -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR .atomic_set_property = amdgpu_dm_atomic_crtc_set_property, .atomic_get_property = amdgpu_dm_atomic_crtc_get_property, #endif @@ -780,7 +780,7 @@ int amdgpu_dm_crtc_init(struct amdgpu_display_manager *dm, drm_mode_crtc_set_gamma_size(&acrtc->base, MAX_COLOR_LEGACY_LUT_ENTRIES); -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR dm_crtc_additional_color_mgmt(&acrtc->base); #endif return 0; diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c index 7c4496fb4b9d..df86a28eda58 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c @@ -1602,7 +1602,7 @@ static void amdgpu_dm_plane_drm_plane_destroy_state(struct drm_plane *plane, drm_atomic_helper_plane_destroy_state(plane, state); } -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR static void dm_atomic_plane_attach_color_mgmt_properties(struct amdgpu_display_manager *dm, struct drm_plane *plane) @@ -1831,7 +1831,7 @@ static const struct drm_plane_funcs dm_plane_funcs = { .atomic_duplicate_state = amdgpu_dm_plane_drm_plane_duplicate_state, .atomic_destroy_state = amdgpu_dm_plane_drm_plane_destroy_state, .format_mod_supported = amdgpu_dm_plane_format_mod_supported, -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR .atomic_set_property = dm_atomic_plane_set_property, .atomic_get_property = dm_atomic_plane_get_property, #endif @@ -1927,7 +1927,7 @@ int amdgpu_dm_plane_init(struct amdgpu_display_manager *dm, else drm_plane_helper_add(plane, &dm_plane_helper_funcs); -#ifdef AMD_PRIVATE_COLOR +#ifdef CONFIG_AMD_PRIVATE_COLOR dm_atomic_plane_attach_color_mgmt_properties(dm, plane); #else res = dm_plane_init_colorops(plane); diff --git a/drivers/gpu/drm/amd/pm/amdgpu_pm.c b/drivers/gpu/drm/amd/pm/amdgpu_pm.c index 65296a819e6a..3d37cbefe4b8 100644 --- a/drivers/gpu/drm/amd/pm/amdgpu_pm.c +++ b/drivers/gpu/drm/amd/pm/amdgpu_pm.c @@ -3342,6 +3342,9 @@ static ssize_t amdgpu_hwmon_show_power_cap_min(struct device *dev, struct device_attribute *attr, char *buf) { + if (amdgpu_ignore_min_pcap) + return sysfs_emit(buf, "%i\n", 0); + return amdgpu_hwmon_show_power_cap_generic(dev, attr, buf, PP_PWR_LIMIT_MIN); } diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c index f51fa265230b..82aec5c85d41 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c +++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c @@ -2974,7 +2974,10 @@ int smu_get_power_limit(void *handle, *limit = smu->max_power_limit; break; case SMU_PPT_LIMIT_MIN: - *limit = smu->min_power_limit; + if (amdgpu_ignore_min_pcap) + *limit = 0; + else + *limit = smu->min_power_limit; break; default: return -EINVAL; @@ -2995,7 +2998,15 @@ static int smu_set_power_limit(void *handle, uint32_t limit_type, uint32_t limit if (limit_type == SMU_DEFAULT_PPT_LIMIT) { if (!limit) limit = smu->current_power_limit; - if ((limit > smu->max_power_limit) || (limit < smu->min_power_limit)) { + + if (amdgpu_ignore_min_pcap) { + if (limit > smu->max_power_limit) { + dev_err(smu->adev->dev, + "New power limit (%d) is over the max allowed %d\n", + limit, smu->max_power_limit); + return -EINVAL; + } + } else if ((limit > smu->max_power_limit) || (limit < smu->min_power_limit)) { dev_err(smu->adev->dev, "New power limit (%d) is out of range [%d,%d]\n", limit, smu->min_power_limit, smu->max_power_limit); diff --git a/drivers/input/evdev.c b/drivers/input/evdev.c index 90ff6be85cf4..15159c1cf6e1 100644 --- a/drivers/input/evdev.c +++ b/drivers/input/evdev.c @@ -46,6 +46,7 @@ struct evdev_client { struct fasync_struct *fasync; struct evdev *evdev; struct list_head node; + struct rcu_head rcu; enum input_clock_type clk_type; bool revoked; unsigned long *evmasks[EV_CNT]; @@ -368,13 +369,22 @@ static void evdev_attach_client(struct evdev *evdev, spin_unlock(&evdev->client_lock); } +static void evdev_reclaim_client(struct rcu_head *rp) +{ + struct evdev_client *client = container_of(rp, struct evdev_client, rcu); + unsigned int i; + for (i = 0; i < EV_CNT; ++i) + bitmap_free(client->evmasks[i]); + kvfree(client); +} + static void evdev_detach_client(struct evdev *evdev, struct evdev_client *client) { spin_lock(&evdev->client_lock); list_del_rcu(&client->node); spin_unlock(&evdev->client_lock); - synchronize_rcu(); + call_rcu(&client->rcu, evdev_reclaim_client); } static int evdev_open_device(struct evdev *evdev) @@ -427,7 +437,6 @@ static int evdev_release(struct inode *inode, struct file *file) { struct evdev_client *client = file->private_data; struct evdev *evdev = client->evdev; - unsigned int i; mutex_lock(&evdev->mutex); @@ -439,11 +448,6 @@ static int evdev_release(struct inode *inode, struct file *file) evdev_detach_client(evdev, client); - for (i = 0; i < EV_CNT; ++i) - bitmap_free(client->evmasks[i]); - - kvfree(client); - evdev_close_device(evdev); return 0; @@ -486,7 +490,6 @@ static int evdev_open(struct inode *inode, struct file *file) err_free_client: evdev_detach_client(evdev, client); - kvfree(client); return error; } diff --git a/drivers/media/v4l2-core/Kconfig b/drivers/media/v4l2-core/Kconfig index d50ccac9733c..bd785d6a24ba 100644 --- a/drivers/media/v4l2-core/Kconfig +++ b/drivers/media/v4l2-core/Kconfig @@ -40,6 +40,11 @@ config VIDEO_TUNER config V4L2_JPEG_HELPER tristate +config V4L2_LOOPBACK + tristate "V4L2 loopback device" + help + V4L2 loopback device + # Used by drivers that need v4l2-h264.ko config V4L2_H264 tristate diff --git a/drivers/media/v4l2-core/Makefile b/drivers/media/v4l2-core/Makefile index 329f0eadce99..dd34e32edb02 100644 --- a/drivers/media/v4l2-core/Makefile +++ b/drivers/media/v4l2-core/Makefile @@ -34,5 +34,7 @@ obj-$(CONFIG_V4L2_JPEG_HELPER) += v4l2-jpeg.o obj-$(CONFIG_V4L2_MEM2MEM_DEV) += v4l2-mem2mem.o obj-$(CONFIG_V4L2_VP9) += v4l2-vp9.o +obj-$(CONFIG_V4L2_LOOPBACK) += v4l2loopback.o + obj-$(CONFIG_VIDEO_TUNER) += tuner.o obj-$(CONFIG_VIDEO_DEV) += v4l2-dv-timings.o videodev.o diff --git a/drivers/media/v4l2-core/v4l2loopback.c b/drivers/media/v4l2-core/v4l2loopback.c new file mode 100644 index 000000000000..22e270717191 --- /dev/null +++ b/drivers/media/v4l2-core/v4l2loopback.c @@ -0,0 +1,3331 @@ +/* -*- c-file-style: "linux" -*- */ +/* + * v4l2loopback.c -- video4linux2 loopback driver + * + * Copyright (C) 2005-2009 Vasily Levin (vasaka@gmail.com) + * Copyright (C) 2010-2023 IOhannes m zmoelnig (zmoelnig@iem.at) + * Copyright (C) 2011 Stefan Diewald (stefan.diewald@mytum.de) + * Copyright (C) 2012 Anton Novikov (random.plant@gmail.com) + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include "v4l2loopback.h" + +#define V4L2LOOPBACK_CTL_ADD_legacy 0x4C80 +#define V4L2LOOPBACK_CTL_REMOVE_legacy 0x4C81 +#define V4L2LOOPBACK_CTL_QUERY_legacy 0x4C82 + +#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 0, 0) +#error This module is not supported on kernels before 4.0.0. +#endif + +#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 3, 0) +#define strscpy strlcpy +#endif + +#if defined(timer_setup) +#define HAVE_TIMER_SETUP +#endif + +#if LINUX_VERSION_CODE < KERNEL_VERSION(5, 7, 0) +#define VFL_TYPE_VIDEO VFL_TYPE_GRABBER +#endif + +#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 2, 0) +#define timer_delete_sync del_timer_sync +#endif + +#define fh_to_opener(ptr) container_of((ptr), struct v4l2_loopback_opener, fh) +#define file_to_opener(ptr) \ + container_of(file_to_v4l2_fh(ptr), struct v4l2_loopback_opener, fh) + +#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 18, 0) +#define v4l2l_f_to_opener(_file, _fh) fh_to_opener(_fh) +#define v4l2_fh_add(fh, filp) v4l2_fh_add(fh) +#define v4l2_fh_del(fh, filp) v4l2_fh_del(fh) +#else +#define v4l2l_f_to_opener(_file, _fh) file_to_opener(_file) +#endif + +#define V4L2LOOPBACK_VERSION_CODE \ + KERNEL_VERSION(V4L2LOOPBACK_VERSION_MAJOR, V4L2LOOPBACK_VERSION_MINOR, \ + V4L2LOOPBACK_VERSION_BUGFIX) + +MODULE_DESCRIPTION("V4L2 loopback video device"); +MODULE_AUTHOR("Vasily Levin, " + "IOhannes m zmoelnig ," + "Stefan Diewald," + "Anton Novikov" + "et al."); +#ifdef SNAPSHOT_VERSION +MODULE_VERSION(__stringify(SNAPSHOT_VERSION)); +#else +MODULE_VERSION("" __stringify(V4L2LOOPBACK_VERSION_MAJOR) "." __stringify( + V4L2LOOPBACK_VERSION_MINOR) "." __stringify(V4L2LOOPBACK_VERSION_BUGFIX)); +#endif +MODULE_LICENSE("GPL"); + +/* + * helpers + */ +#define dprintk(fmt, args...) \ + do { \ + if (debug > 0) { \ + printk(KERN_INFO "v4l2-loopback[" __stringify( \ + __LINE__) "], pid(%d): " fmt, \ + task_pid_nr(current), ##args); \ + } \ + } while (0) + +#define MARK() \ + do { \ + if (debug > 1) { \ + printk(KERN_INFO "%s:%d[%s], pid(%d)\n", __FILE__, \ + __LINE__, __func__, task_pid_nr(current)); \ + } \ + } while (0) + +#define dprintkrw(fmt, args...) \ + do { \ + if (debug > 2) { \ + printk(KERN_INFO "v4l2-loopback[" __stringify( \ + __LINE__) "], pid(%d): " fmt, \ + task_pid_nr(current), ##args); \ + } \ + } while (0) + +static inline void v4l2l_get_timestamp(struct v4l2_buffer *b) +{ + struct timespec64 ts; + ktime_get_ts64(&ts); + + b->timestamp.tv_sec = ts.tv_sec; + b->timestamp.tv_usec = (ts.tv_nsec / NSEC_PER_USEC); + b->flags |= V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC; + b->flags &= ~V4L2_BUF_FLAG_TIMESTAMP_COPY; +} + +#if BITS_PER_LONG == 32 +#include /* do_div() for 64bit division */ +static inline int v4l2l_mod64(const s64 A, const u32 B) +{ + u64 a = (u64)A; + u32 b = B; + + if (A > 0) + return do_div(a, b); + a = -A; + return -do_div(a, b); +} +#else +static inline int v4l2l_mod64(const s64 A, const u32 B) +{ + return A % B; +} +#endif + +#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 16, 0) +typedef unsigned __poll_t; +#endif + +/* module constants + * can be overridden during he build process using something like + * make KCPPFLAGS="-DMAX_DEVICES=100" + */ + +/* maximum number of v4l2loopback devices that can be created */ +#ifndef MAX_DEVICES +#define MAX_DEVICES 8 +#endif + +/* whether the default is to announce capabilities exclusively or not */ +#ifndef V4L2LOOPBACK_DEFAULT_EXCLUSIVECAPS +#define V4L2LOOPBACK_DEFAULT_EXCLUSIVECAPS 0 +#endif + +/* when a producer is considered to have gone stale */ +#ifndef MAX_TIMEOUT +#define MAX_TIMEOUT (100 * 1000) /* in msecs */ +#endif + +/* max buffers that can be mapped, actually they + * are all mapped to max_buffers buffers */ +#ifndef MAX_BUFFERS +#define MAX_BUFFERS 32 +#endif + +/* module parameters */ +static int debug = 0; +module_param(debug, int, S_IRUGO | S_IWUSR); +MODULE_PARM_DESC(debug, "debugging level (higher values == more verbose)"); + +#define V4L2LOOPBACK_DEFAULT_MAX_BUFFERS 2 +static int max_buffers = V4L2LOOPBACK_DEFAULT_MAX_BUFFERS; +module_param(max_buffers, int, S_IRUGO); +MODULE_PARM_DESC(max_buffers, + "how many buffers should be allocated [DEFAULT: " __stringify( + V4L2LOOPBACK_DEFAULT_MAX_BUFFERS) "]"); + +/* how many times a device can be opened + * the per-module default value can be overridden on a per-device basis using + * the /sys/devices interface + * + * note that max_openers should be at least 2 in order to get a working system: + * one opener for the producer and one opener for the consumer + * however, we leave that to the user + */ +#define V4L2LOOPBACK_DEFAULT_MAX_OPENERS 10 +static int max_openers = V4L2LOOPBACK_DEFAULT_MAX_OPENERS; +module_param(max_openers, int, S_IRUGO | S_IWUSR); +MODULE_PARM_DESC( + max_openers, + "how many users can open the loopback device [DEFAULT: " __stringify( + V4L2LOOPBACK_DEFAULT_MAX_OPENERS) "]"); + +static int devices = -1; +module_param(devices, int, 0); +MODULE_PARM_DESC(devices, "how many devices should be created"); + +static int video_nr[MAX_DEVICES] = { [0 ...(MAX_DEVICES - 1)] = -1 }; +module_param_array(video_nr, int, NULL, 0444); +MODULE_PARM_DESC(video_nr, + "video device numbers (-1=auto, 0=/dev/video0, etc.)"); + +static char *card_label[MAX_DEVICES]; +module_param_array(card_label, charp, NULL, 0000); +MODULE_PARM_DESC(card_label, "card labels for each device"); + +static bool exclusive_caps[MAX_DEVICES] = { + [0 ...(MAX_DEVICES - 1)] = V4L2LOOPBACK_DEFAULT_EXCLUSIVECAPS +}; +module_param_array(exclusive_caps, bool, NULL, 0444); +/* FIXXME: wording */ +MODULE_PARM_DESC( + exclusive_caps, + "whether to announce OUTPUT/CAPTURE capabilities exclusively or not [DEFAULT: " __stringify( + V4L2LOOPBACK_DEFAULT_EXCLUSIVECAPS) "]"); + +/* format specifications */ +#define V4L2LOOPBACK_SIZE_MIN_WIDTH 2 +#define V4L2LOOPBACK_SIZE_MIN_HEIGHT 1 +#define V4L2LOOPBACK_SIZE_DEFAULT_MAX_WIDTH 8192 +#define V4L2LOOPBACK_SIZE_DEFAULT_MAX_HEIGHT 8192 + +#define V4L2LOOPBACK_SIZE_DEFAULT_WIDTH 640 +#define V4L2LOOPBACK_SIZE_DEFAULT_HEIGHT 480 + +static int max_width = V4L2LOOPBACK_SIZE_DEFAULT_MAX_WIDTH; +module_param(max_width, int, S_IRUGO); +MODULE_PARM_DESC(max_width, + "maximum allowed frame width [DEFAULT: " __stringify( + V4L2LOOPBACK_SIZE_DEFAULT_MAX_WIDTH) "]"); +static int max_height = V4L2LOOPBACK_SIZE_DEFAULT_MAX_HEIGHT; +module_param(max_height, int, S_IRUGO); +MODULE_PARM_DESC(max_height, + "maximum allowed frame height [DEFAULT: " __stringify( + V4L2LOOPBACK_SIZE_DEFAULT_MAX_HEIGHT) "]"); + +static DEFINE_IDR(v4l2loopback_index_idr); +static DEFINE_MUTEX(v4l2loopback_ctl_mutex); + +/* frame intervals */ +#define V4L2LOOPBACK_FRAME_INTERVAL_MAX __UINT32_MAX__ +#define V4L2LOOPBACK_FPS_DEFAULT 30 +#define V4L2LOOPBACK_FPS_MAX 1000 + +/* control IDs */ +#define V4L2LOOPBACK_CID_BASE (V4L2_CID_USER_BASE | 0xf000) +#define CID_KEEP_FORMAT (V4L2LOOPBACK_CID_BASE + 0) +#define CID_SUSTAIN_FRAMERATE (V4L2LOOPBACK_CID_BASE + 1) +#define CID_TIMEOUT (V4L2LOOPBACK_CID_BASE + 2) +#define CID_TIMEOUT_IMAGE_IO (V4L2LOOPBACK_CID_BASE + 3) + +static int v4l2loopback_s_ctrl(struct v4l2_ctrl *ctrl); +static const struct v4l2_ctrl_ops v4l2loopback_ctrl_ops = { + .s_ctrl = v4l2loopback_s_ctrl, +}; +static const struct v4l2_ctrl_config v4l2loopback_ctrl_keepformat = { + // clang-format off + .ops = &v4l2loopback_ctrl_ops, + .id = CID_KEEP_FORMAT, + .name = "keep_format", + .type = V4L2_CTRL_TYPE_BOOLEAN, + .min = 0, + .max = 1, + .step = 1, + .def = 0, + // clang-format on +}; +static const struct v4l2_ctrl_config v4l2loopback_ctrl_sustainframerate = { + // clang-format off + .ops = &v4l2loopback_ctrl_ops, + .id = CID_SUSTAIN_FRAMERATE, + .name = "sustain_framerate", + .type = V4L2_CTRL_TYPE_BOOLEAN, + .min = 0, + .max = 1, + .step = 1, + .def = 0, + // clang-format on +}; +static const struct v4l2_ctrl_config v4l2loopback_ctrl_timeout = { + // clang-format off + .ops = &v4l2loopback_ctrl_ops, + .id = CID_TIMEOUT, + .name = "timeout", + .type = V4L2_CTRL_TYPE_INTEGER, + .min = 0, + .max = MAX_TIMEOUT, + .step = 1, + .def = 0, + // clang-format on +}; +static const struct v4l2_ctrl_config v4l2loopback_ctrl_timeoutimageio = { + // clang-format off + .ops = &v4l2loopback_ctrl_ops, + .id = CID_TIMEOUT_IMAGE_IO, + .name = "timeout_image_io", + .type = V4L2_CTRL_TYPE_BUTTON, + .min = 0, + .max = 0, + .step = 0, + .def = 0, + // clang-format on +}; + +/* module structures */ +struct v4l2loopback_private { + int device_nr; +}; + +/* TODO(vasaka) use typenames which are common to kernel, but first find out if + * it is needed */ +/* struct keeping state and settings of loopback device */ + +struct v4l2l_buffer { + struct v4l2_buffer buffer; + struct list_head list_head; + atomic_t use_count; +}; + +struct v4l2_loopback_device { + struct v4l2_device v4l2_dev; + struct v4l2_ctrl_handler ctrl_handler; + struct video_device *vdev; + + /* loopback device-specific parameters */ + char card_label[32]; + bool announce_all_caps; /* announce both OUTPUT and CAPTURE capabilities + * when true; else announce OUTPUT when no + * writer is streaming, otherwise CAPTURE. */ + int max_openers; /* how many times can this device be opened */ + int min_width, max_width; + int min_height, max_height; + + /* pixel and stream format */ + struct v4l2_pix_format pix_format; + bool pix_format_has_valid_sizeimage; + struct v4l2_captureparm capture_param; + unsigned long frame_jiffies; + + /* ctrls */ + int keep_format; /* CID_KEEP_FORMAT; lock the format, do not free + * on close(), and when `!announce_all_caps` do NOT + * fall back to OUTPUT when no writers attached (clear + * `keep_format` to attach a new writer) */ + int sustain_framerate; /* CID_SUSTAIN_FRAMERATE; duplicate frames to maintain + (close to) nominal framerate */ + unsigned long timeout_jiffies; /* CID_TIMEOUT; 0 means disabled */ + int timeout_image_io; /* CID_TIMEOUT_IMAGE_IO; next opener will + * queue/dequeue the timeout image buffer */ + + /* buffers for OUTPUT and CAPTURE */ + u8 *image; /* pointer to actual buffers data */ + unsigned long image_size; /* number of bytes alloc'd for all buffers */ + struct v4l2l_buffer buffers[MAX_BUFFERS]; /* inner driver buffers */ + u32 buffer_count; /* should not be big, 4 is a good choice */ + u32 buffer_size; /* number of bytes alloc'd per buffer */ + u32 used_buffer_count; /* number of buffers allocated to openers */ + struct list_head outbufs_list; /* FIFO queue for OUTPUT buffers */ + u32 bufpos2index[MAX_BUFFERS]; /* mapping of `(position % used_buffers)` + * to `buffers[index]` */ + s64 write_position; /* sequence number of last 'displayed' buffer plus + * one */ + + /* synchronization between openers */ + atomic_t open_count; + struct mutex image_mutex; /* mutex for allocating image(s) and + * exchanging format tokens */ + spinlock_t lock; /* lock for the timeout and framerate timers */ + spinlock_t list_lock; /* lock for the OUTPUT buffer queue */ + wait_queue_head_t read_event; + u32 format_tokens; /* tokens to 'set format' for OUTPUT, CAPTURE, or + * timeout buffers */ + u32 stream_tokens; /* tokens to 'start' OUTPUT, CAPTURE, or timeout + * stream */ + + /* sustain framerate */ + struct timer_list sustain_timer; + unsigned int reread_count; + + /* timeout */ + u8 *timeout_image; /* copied to outgoing buffers when timeout passes */ + struct v4l2l_buffer timeout_buffer; + u32 timeout_buffer_size; /* number bytes alloc'd for timeout buffer */ + struct timer_list timeout_timer; + int timeout_happened; +}; + +enum v4l2l_io_method { + V4L2L_IO_NONE = 0, + V4L2L_IO_MMAP = 1, + V4L2L_IO_FILE = 2, + V4L2L_IO_TIMEOUT = 3, +}; + +/* struct keeping state and type of opener */ +struct v4l2_loopback_opener { + u32 format_token; /* token (if any) for type used in call to S_FMT or + * REQBUFS */ + u32 stream_token; /* token (if any) for type used in call to STREAMON */ + u32 buffer_count; /* number of buffers (if any) that opener acquired via + * REQBUFS */ + s64 read_position; /* sequence number of the next 'captured' frame */ + unsigned int reread_count; + enum v4l2l_io_method io_method; + + struct v4l2_fh fh; +}; + +/* this is heavily inspired by the bttv driver found in the linux kernel */ +struct v4l2l_format { + char *name; + int fourcc; /* video4linux 2 */ + int depth; /* bit/pixel */ + int flags; +}; +/* set the v4l2l_format.flags to PLANAR for non-packed formats */ +#define FORMAT_FLAGS_PLANAR 0x01 +#define FORMAT_FLAGS_COMPRESSED 0x02 + +#include "v4l2loopback_formats.h" + +#ifndef V4L2_TYPE_IS_CAPTURE +#define V4L2_TYPE_IS_CAPTURE(type) \ + ((type) == V4L2_BUF_TYPE_VIDEO_CAPTURE || \ + (type) == V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE) +#endif /* V4L2_TYPE_IS_CAPTURE */ +#ifndef V4L2_TYPE_IS_OUTPUT +#define V4L2_TYPE_IS_OUTPUT(type) \ + ((type) == V4L2_BUF_TYPE_VIDEO_OUTPUT || \ + (type) == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) +#endif /* V4L2_TYPE_IS_OUTPUT */ + +/* token values for privilege to set format or start/stop stream */ +#define V4L2L_TOKEN_CAPTURE 0x01 +#define V4L2L_TOKEN_OUTPUT 0x02 +#define V4L2L_TOKEN_TIMEOUT 0x04 +#define V4L2L_TOKEN_MASK \ + (V4L2L_TOKEN_CAPTURE | V4L2L_TOKEN_OUTPUT | V4L2L_TOKEN_TIMEOUT) + +/* helpers for token exchange and token status */ +#define token_from_type(type) \ + (V4L2_TYPE_IS_CAPTURE(type) ? V4L2L_TOKEN_CAPTURE : V4L2L_TOKEN_OUTPUT) +#define acquire_token(dev, opener, label, token) \ + do { \ + (opener)->label##_token = token; \ + (dev)->label##_tokens &= ~token; \ + } while (0) +#define release_token(dev, opener, label) \ + do { \ + (dev)->label##_tokens |= (opener)->label##_token; \ + (opener)->label##_token = 0; \ + } while (0) +#define has_output_token(token) (token & V4L2L_TOKEN_OUTPUT) +#define has_capture_token(token) (token & V4L2L_TOKEN_CAPTURE) +#define has_no_owners(dev) ((~((dev)->format_tokens) & V4L2L_TOKEN_MASK) == 0) +#define has_other_owners(opener, dev) \ + (~((dev)->format_tokens ^ (opener)->format_token) & V4L2L_TOKEN_MASK) +#define need_timeout_buffer(dev, token) \ + ((dev)->timeout_jiffies > 0 || (token) & V4L2L_TOKEN_TIMEOUT) + +static const unsigned int FORMATS = ARRAY_SIZE(formats); + +static char *fourcc2str(unsigned int fourcc, char buf[5]) +{ + buf[0] = (fourcc >> 0) & 0xFF; + buf[1] = (fourcc >> 8) & 0xFF; + buf[2] = (fourcc >> 16) & 0xFF; + buf[3] = (fourcc >> 24) & 0xFF; + buf[4] = 0; + + return buf; +} + +static const struct v4l2l_format *format_by_fourcc(int fourcc) +{ + unsigned int i; + char buf[5]; + + for (i = 0; i < FORMATS; i++) { + if (formats[i].fourcc == fourcc) + return formats + i; + } + + dprintk("unsupported format '%4s'\n", fourcc2str(fourcc, buf)); + return NULL; +} + +static void pix_format_set_size(struct v4l2_pix_format *f, + const struct v4l2l_format *fmt, + unsigned int width, unsigned int height) +{ + f->width = width; + f->height = height; + + if (fmt->flags & FORMAT_FLAGS_PLANAR) { + f->bytesperline = width; /* Y plane */ + f->sizeimage = (width * height * fmt->depth) >> 3; + } else if (fmt->flags & FORMAT_FLAGS_COMPRESSED) { + /* doesn't make sense for compressed formats */ + f->bytesperline = 0; + f->sizeimage = (width * height * fmt->depth) >> 3; + } else { + f->bytesperline = (width * fmt->depth) >> 3; + f->sizeimage = height * f->bytesperline; + } +} + +static int v4l2l_fill_format(struct v4l2_format *fmt, const u32 minwidth, + const u32 maxwidth, const u32 minheight, + const u32 maxheight) +{ + u32 width = fmt->fmt.pix.width, height = fmt->fmt.pix.height; + u32 pixelformat = fmt->fmt.pix.pixelformat; + struct v4l2_format fmt0 = *fmt; + u32 bytesperline = 0, sizeimage = 0; + + if (!width) + width = V4L2LOOPBACK_SIZE_DEFAULT_WIDTH; + if (!height) + height = V4L2LOOPBACK_SIZE_DEFAULT_HEIGHT; + width = clamp_val(width, minwidth, maxwidth); + height = clamp_val(height, minheight, maxheight); + + /* sets: width,height,pixelformat,bytesperline,sizeimage */ + if (!(V4L2_TYPE_IS_MULTIPLANAR(fmt0.type))) { + fmt0.fmt.pix.bytesperline = 0; + fmt0.fmt.pix.sizeimage = 0; + } + + if (0) { + ; +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 2, 0) + } else if (!v4l2_fill_pixfmt(&fmt0.fmt.pix, pixelformat, width, + height)) { + ; + } else if (!v4l2_fill_pixfmt_mp(&fmt0.fmt.pix_mp, pixelformat, width, + height)) { + ; +#endif + } else { + const struct v4l2l_format *format = + format_by_fourcc(pixelformat); + if (!format) + return -EINVAL; + pix_format_set_size(&fmt0.fmt.pix, format, width, height); + fmt0.fmt.pix.pixelformat = format->fourcc; + } + + if (V4L2_TYPE_IS_MULTIPLANAR(fmt0.type)) { + *fmt = fmt0; + + if ((fmt->fmt.pix_mp.colorspace == V4L2_COLORSPACE_DEFAULT) || + (fmt->fmt.pix_mp.colorspace > V4L2_COLORSPACE_DCI_P3)) + fmt->fmt.pix_mp.colorspace = V4L2_COLORSPACE_SRGB; + if (V4L2_FIELD_ANY == fmt->fmt.pix_mp.field) + fmt->fmt.pix_mp.field = V4L2_FIELD_NONE; + } else { + bytesperline = fmt->fmt.pix.bytesperline; + sizeimage = fmt->fmt.pix.sizeimage; + + *fmt = fmt0; + + if (!fmt->fmt.pix.bytesperline) + fmt->fmt.pix.bytesperline = bytesperline; + if (!fmt->fmt.pix.sizeimage) + fmt->fmt.pix.sizeimage = sizeimage; + + if ((fmt->fmt.pix.colorspace == V4L2_COLORSPACE_DEFAULT) || + (fmt->fmt.pix.colorspace > V4L2_COLORSPACE_DCI_P3)) + fmt->fmt.pix.colorspace = V4L2_COLORSPACE_SRGB; + if (V4L2_FIELD_ANY == fmt->fmt.pix.field) + fmt->fmt.pix.field = V4L2_FIELD_NONE; + } + + return 0; +} + +/* Checks if v4l2l_fill_format() has set a valid, fixed sizeimage val. */ +static bool v4l2l_pix_format_has_valid_sizeimage(struct v4l2_format *fmt) +{ +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 2, 0) + const struct v4l2_format_info *info; + + info = v4l2_format_info(fmt->fmt.pix.pixelformat); + if (info && info->mem_planes == 1) + return true; +#endif + + return false; +} + +static int pix_format_eq(const struct v4l2_pix_format *ref, + const struct v4l2_pix_format *tgt, int strict) +{ + /* check if the two formats are equivalent. + * ANY fields are handled gracefully + */ +#define _pix_format_eq0(x) \ + if (ref->x != tgt->x) \ + result = 0 +#define _pix_format_eq1(x, def) \ + do { \ + if ((def != tgt->x) && (ref->x != tgt->x)) { \ + printk(KERN_INFO #x " failed"); \ + result = 0; \ + } \ + } while (0) + int result = 1; + _pix_format_eq0(width); + _pix_format_eq0(height); + _pix_format_eq0(pixelformat); + if (!strict) + return result; + _pix_format_eq1(field, V4L2_FIELD_ANY); + _pix_format_eq0(bytesperline); + _pix_format_eq0(sizeimage); + _pix_format_eq1(colorspace, V4L2_COLORSPACE_DEFAULT); + return result; +} + +static void set_timeperframe(struct v4l2_loopback_device *dev, + struct v4l2_fract *tpf) +{ + if (!tpf->denominator && !tpf->numerator) { + tpf->numerator = 1; + tpf->denominator = V4L2LOOPBACK_FPS_DEFAULT; + } else if (tpf->numerator > + V4L2LOOPBACK_FRAME_INTERVAL_MAX * tpf->denominator) { + /* divide-by-zero or greater than maximum interval => min FPS */ + tpf->numerator = V4L2LOOPBACK_FRAME_INTERVAL_MAX; + tpf->denominator = 1; + } else if (tpf->numerator * V4L2LOOPBACK_FPS_MAX < tpf->denominator) { + /* zero or lower than minimum interval => max FPS */ + tpf->numerator = 1; + tpf->denominator = V4L2LOOPBACK_FPS_MAX; + } + + dev->capture_param.timeperframe = *tpf; + dev->frame_jiffies = + max(1UL, (msecs_to_jiffies(1000) * tpf->numerator) / + tpf->denominator); +} + +static struct v4l2_loopback_device *v4l2loopback_cd2dev(struct device *cd); + +/* device attributes */ +/* available via sysfs: /sys/devices/virtual/video4linux/video* */ + +static ssize_t attr_show_format(struct device *cd, + struct device_attribute *attr, char *buf) +{ + /* gets the current format as "FOURCC:WxH@f/s", e.g. "YUYV:320x240@1000/30" */ + struct v4l2_loopback_device *dev = v4l2loopback_cd2dev(cd); + const struct v4l2_fract *tpf; + char buf4cc[5], buf_fps[32]; + + if (!dev || (has_no_owners(dev) && !dev->keep_format)) + return 0; + tpf = &dev->capture_param.timeperframe; + + fourcc2str(dev->pix_format.pixelformat, buf4cc); + if (tpf->numerator == 1) + snprintf(buf_fps, sizeof(buf_fps), "%u", tpf->denominator); + else + snprintf(buf_fps, sizeof(buf_fps), "%u/%u", tpf->denominator, + tpf->numerator); + return sprintf(buf, "%4s:%ux%u@%s\n", buf4cc, dev->pix_format.width, + dev->pix_format.height, buf_fps); +} + +static ssize_t attr_store_format(struct device *cd, + struct device_attribute *attr, const char *buf, + size_t len) +{ + struct v4l2_loopback_device *dev = v4l2loopback_cd2dev(cd); + int fps_num = 0, fps_den = 1; + + if (!dev) + return -ENODEV; + + /* only fps changing is supported */ + if (sscanf(buf, "@%u/%u", &fps_num, &fps_den) > 0) { + struct v4l2_fract f = { .numerator = fps_den, + .denominator = fps_num }; + set_timeperframe(dev, &f); + return len; + } + return -EINVAL; +} + +static DEVICE_ATTR(format, S_IRUGO | S_IWUSR, attr_show_format, + attr_store_format); + +static ssize_t attr_show_buffers(struct device *cd, + struct device_attribute *attr, char *buf) +{ + struct v4l2_loopback_device *dev = v4l2loopback_cd2dev(cd); + + if (!dev) + return -ENODEV; + + return sprintf(buf, "%u\n", dev->used_buffer_count); +} + +static DEVICE_ATTR(buffers, S_IRUGO, attr_show_buffers, NULL); + +static ssize_t attr_show_maxopeners(struct device *cd, + struct device_attribute *attr, char *buf) +{ + struct v4l2_loopback_device *dev = v4l2loopback_cd2dev(cd); + + if (!dev) + return -ENODEV; + + return sprintf(buf, "%d\n", dev->max_openers); +} + +static ssize_t attr_store_maxopeners(struct device *cd, + struct device_attribute *attr, + const char *buf, size_t len) +{ + struct v4l2_loopback_device *dev = NULL; + unsigned long curr = 0; + + if (kstrtoul(buf, 0, &curr)) + return -EINVAL; + + dev = v4l2loopback_cd2dev(cd); + if (!dev) + return -ENODEV; + + if (dev->max_openers == curr) + return len; + + if (curr > __INT_MAX__ || dev->open_count.counter > curr) { + /* request to limit to less openers as are currently attached to us */ + return -EINVAL; + } + + dev->max_openers = (int)curr; + + return len; +} + +static DEVICE_ATTR(max_openers, S_IRUGO | S_IWUSR, attr_show_maxopeners, + attr_store_maxopeners); + +static ssize_t attr_show_state(struct device *cd, struct device_attribute *attr, + char *buf) +{ + struct v4l2_loopback_device *dev = v4l2loopback_cd2dev(cd); + + if (!dev) + return -ENODEV; + + if (!has_output_token(dev->stream_tokens) || dev->keep_format) { + return sprintf(buf, "capture\n"); + } else + return sprintf(buf, "output\n"); + + return -EAGAIN; +} + +static DEVICE_ATTR(state, S_IRUGO, attr_show_state, NULL); + +static void v4l2loopback_remove_sysfs(struct video_device *vdev) +{ +#define V4L2_SYSFS_DESTROY(x) device_remove_file(&vdev->dev, &dev_attr_##x) + + if (vdev) { + V4L2_SYSFS_DESTROY(format); + V4L2_SYSFS_DESTROY(buffers); + V4L2_SYSFS_DESTROY(max_openers); + V4L2_SYSFS_DESTROY(state); + /* ... */ + } +} + +static void v4l2loopback_create_sysfs(struct video_device *vdev) +{ + int res = 0; + +#define V4L2_SYSFS_CREATE(x) \ + res = device_create_file(&vdev->dev, &dev_attr_##x); \ + if (res < 0) \ + break + if (!vdev) + return; + do { + V4L2_SYSFS_CREATE(format); + V4L2_SYSFS_CREATE(buffers); + V4L2_SYSFS_CREATE(max_openers); + V4L2_SYSFS_CREATE(state); + /* ... */ + } while (0); + + if (res >= 0) + return; + dev_err(&vdev->dev, "%s error: %d\n", __func__, res); +} + +/* Event APIs */ + +#define V4L2LOOPBACK_EVENT_BASE (V4L2_EVENT_PRIVATE_START) +#define V4L2LOOPBACK_EVENT_OFFSET 0x08E00000 +#define V4L2_EVENT_PRI_CLIENT_USAGE \ + (V4L2LOOPBACK_EVENT_BASE + V4L2LOOPBACK_EVENT_OFFSET + 1) + +struct v4l2_event_client_usage { + __u32 count; +}; + +/* global module data */ +/* find a device based on it's device-number (e.g. '3' for /dev/video3) */ +struct v4l2loopback_lookup_cb_data { + int device_nr; + struct v4l2_loopback_device *device; +}; +static int v4l2loopback_lookup_cb(int id, void *ptr, void *data) +{ + struct v4l2_loopback_device *device = ptr; + struct v4l2loopback_lookup_cb_data *cbdata = data; + if (cbdata && device && device->vdev) { + if (device->vdev->num == cbdata->device_nr) { + cbdata->device = device; + cbdata->device_nr = id; + return 1; + } + } + return 0; +} +static int v4l2loopback_lookup(int device_nr, + struct v4l2_loopback_device **device) +{ + struct v4l2loopback_lookup_cb_data data = { + .device_nr = device_nr, + .device = NULL, + }; + int err = idr_for_each(&v4l2loopback_index_idr, &v4l2loopback_lookup_cb, + &data); + if (1 == err) { + if (device) + *device = data.device; + return data.device_nr; + } + return -ENODEV; +} +#define v4l2loopback_get_vdev_nr(vdev) \ + ((struct v4l2loopback_private *)video_get_drvdata(vdev))->device_nr +static struct v4l2_loopback_device *v4l2loopback_cd2dev(struct device *cd) +{ + struct video_device *loopdev = to_video_device(cd); + int device_nr = v4l2loopback_get_vdev_nr(loopdev); + + return idr_find(&v4l2loopback_index_idr, device_nr); +} + +static struct v4l2_loopback_device *v4l2loopback_getdevice(struct file *f) +{ + struct v4l2loopback_private *ptr = video_drvdata(f); + int nr = ptr->device_nr; + + return idr_find(&v4l2loopback_index_idr, nr); +} + +/* forward declarations */ +static void client_usage_queue_event(struct video_device *vdev); +static bool any_buffers_mapped(struct v4l2_loopback_device *dev); +static int allocate_buffers(struct v4l2_loopback_device *dev, + struct v4l2_pix_format *pix_format); +static void init_buffers(struct v4l2_loopback_device *dev, u32 bytes_used, + u32 buffer_size); +static void free_buffers(struct v4l2_loopback_device *dev); +static int allocate_timeout_buffer(struct v4l2_loopback_device *dev); +static void free_timeout_buffer(struct v4l2_loopback_device *dev); +static void check_timers(struct v4l2_loopback_device *dev); +static const struct v4l2_file_operations v4l2_loopback_fops; +static const struct v4l2_ioctl_ops v4l2_loopback_ioctl_ops; + +/* V4L2 ioctl caps and params calls */ +/* returns device capabilities + * called on VIDIOC_QUERYCAP + */ +static int vidioc_querycap(struct file *file, void *fh, + struct v4l2_capability *cap) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + int device_nr = v4l2loopback_get_vdev_nr(dev->vdev); + __u32 capabilities = V4L2_CAP_STREAMING | V4L2_CAP_READWRITE; + + strscpy(cap->driver, "v4l2 loopback", sizeof(cap->driver)); + snprintf(cap->card, sizeof(cap->card), "%s", dev->card_label); + snprintf(cap->bus_info, sizeof(cap->bus_info), + "platform:v4l2loopback-%03d", device_nr); + + if (dev->announce_all_caps) { + capabilities |= V4L2_CAP_VIDEO_CAPTURE | V4L2_CAP_VIDEO_OUTPUT; + } else { + if (opener->io_method == V4L2L_IO_TIMEOUT || + (has_output_token(dev->stream_tokens) && + !dev->keep_format)) { + capabilities |= V4L2_CAP_VIDEO_OUTPUT; + } else + capabilities |= V4L2_CAP_VIDEO_CAPTURE; + } + +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 7, 0) + dev->vdev->device_caps = +#endif /* >=linux-4.7.0 */ + cap->device_caps = cap->capabilities = capabilities; + + cap->capabilities |= V4L2_CAP_DEVICE_CAPS; + + memset(cap->reserved, 0, sizeof(cap->reserved)); + return 0; +} + +static int vidioc_enum_framesizes(struct file *file, void *fh, + struct v4l2_frmsizeenum *argp) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + + /* there can be only one... */ + if (argp->index) + return -EINVAL; + + if (dev->keep_format || has_other_owners(opener, dev)) { + /* only current frame size supported */ + if (argp->pixel_format != dev->pix_format.pixelformat) + return -EINVAL; + + argp->type = V4L2_FRMSIZE_TYPE_DISCRETE; + + argp->discrete.width = dev->pix_format.width; + argp->discrete.height = dev->pix_format.height; + } else { + /* return continuous sizes if pixel format is supported */ + if (NULL == format_by_fourcc(argp->pixel_format)) + return -EINVAL; + + if (dev->min_width == dev->max_width && + dev->min_height == dev->max_height) { + argp->type = V4L2_FRMSIZE_TYPE_DISCRETE; + + argp->discrete.width = dev->min_width; + argp->discrete.height = dev->min_height; + } else { + argp->type = V4L2_FRMSIZE_TYPE_CONTINUOUS; + + argp->stepwise.min_width = dev->min_width; + argp->stepwise.min_height = dev->min_height; + + argp->stepwise.max_width = dev->max_width; + argp->stepwise.max_height = dev->max_height; + + argp->stepwise.step_width = 1; + argp->stepwise.step_height = 1; + } + } + return 0; +} + +/* Test if the device is currently 'capable' of the buffer (stream) type when + * the `exclusive_caps` parameter is set. `keep_format` should lock the format + * and prevent free of buffers */ +static int check_buffer_capability(struct v4l2_loopback_device *dev, + struct v4l2_loopback_opener *opener, + enum v4l2_buf_type type) +{ + /* short-circuit for (non-compliant) timeout image mode */ + if (opener->io_method == V4L2L_IO_TIMEOUT) + return 0; + if (dev->announce_all_caps) + return (type == V4L2_BUF_TYPE_VIDEO_CAPTURE || + type == V4L2_BUF_TYPE_VIDEO_OUTPUT) ? + 0 : + -EINVAL; + /* CAPTURE if opener has a capture format or a writer is streaming; + * else OUTPUT. */ + switch (type) { + case V4L2_BUF_TYPE_VIDEO_CAPTURE: + if (!(has_capture_token(opener->format_token) || + !has_output_token(dev->stream_tokens))) + return -EINVAL; + break; + case V4L2_BUF_TYPE_VIDEO_OUTPUT: + if (!(has_output_token(opener->format_token) || + has_output_token(dev->stream_tokens))) + return -EINVAL; + break; + default: + return -EINVAL; + } + return 0; +} +/* returns frameinterval (fps) for the set resolution + * called on VIDIOC_ENUM_FRAMEINTERVALS + */ +static int vidioc_enum_frameintervals(struct file *file, void *fh, + struct v4l2_frmivalenum *argp) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + + /* there can be only one... */ + if (argp->index) + return -EINVAL; + + if (dev->keep_format || has_other_owners(opener, dev)) { + /* keep_format also locks the frame rate */ + if (argp->width != dev->pix_format.width || + argp->height != dev->pix_format.height || + argp->pixel_format != dev->pix_format.pixelformat) + return -EINVAL; + + argp->type = V4L2_FRMIVAL_TYPE_DISCRETE; + argp->discrete = dev->capture_param.timeperframe; + } else { + if (argp->width < dev->min_width || + argp->width > dev->max_width || + argp->height < dev->min_height || + argp->height > dev->max_height || + !format_by_fourcc(argp->pixel_format)) + return -EINVAL; + + argp->type = V4L2_FRMIVAL_TYPE_CONTINUOUS; + argp->stepwise.min.numerator = 1; + argp->stepwise.min.denominator = V4L2LOOPBACK_FPS_MAX; + argp->stepwise.max.numerator = V4L2LOOPBACK_FRAME_INTERVAL_MAX; + argp->stepwise.max.denominator = 1; + argp->stepwise.step.numerator = 1; + argp->stepwise.step.denominator = 1; + } + + return 0; +} + +/* Enumerate device formats + * Returns: + * - EINVAL the index is out of bounds; or if non-zero when format is fixed + * - EFAULT unexpected null pointer */ +static int vidioc_enum_fmt_vid(struct file *file, void *fh, + struct v4l2_fmtdesc *f) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + int fixed = dev->keep_format || has_other_owners(opener, dev); + const struct v4l2l_format *fmt; + + if (check_buffer_capability(dev, opener, f->type) < 0) + return -EINVAL; + + if (!(f->index < FORMATS)) + return -EINVAL; + /* TODO: Support 6.14 V4L2_FMTDESC_FLAG_ENUM_ALL */ + if (fixed && f->index) + return -EINVAL; + + fmt = fixed ? format_by_fourcc(dev->pix_format.pixelformat) : + &formats[f->index]; + if (!fmt) + return -EFAULT; + + f->flags = 0; + if (fmt->flags & FORMAT_FLAGS_COMPRESSED) + f->flags |= V4L2_FMT_FLAG_COMPRESSED; + snprintf(f->description, sizeof(f->description), fmt->name); + f->pixelformat = fmt->fourcc; + return 0; +} + +/* Tests (or tries) the format. + * Returns: + * - EINVAL if the buffer type or format is not supported + */ +static int vidioc_try_fmt_vid(struct file *file, void *fh, + struct v4l2_format *f) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + + if (check_buffer_capability(dev, opener, f->type) < 0) + return -EINVAL; + if (v4l2l_fill_format(f, dev->min_width, dev->max_width, + dev->min_height, dev->max_height) != 0) + return -EINVAL; + if (dev->keep_format || has_other_owners(opener, dev)) + /* use existing format - including colorspace info */ + f->fmt.pix = dev->pix_format; + + return 0; +} + +/* Sets new format. Fills 'f' argument with the requested or existing format. + * Side-effect: buffers are allocated for the (returned) format. + * Returns: + * - EINVAL if the type is not supported + * - EBUSY if buffers are already allocated + * TODO: (vasaka) set subregions of input + */ +static int vidioc_s_fmt_vid(struct file *file, void *fh, struct v4l2_format *f) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + u32 token = opener->io_method == V4L2L_IO_TIMEOUT ? + V4L2L_TOKEN_TIMEOUT : + token_from_type(f->type); + int changed, result; + char buf[5]; + + result = vidioc_try_fmt_vid(file, fh, f); + if (result < 0) + return result; + + if (opener->buffer_count > 0) + /* must free buffers before format can be set */ + return -EBUSY; + + result = mutex_lock_killable(&dev->image_mutex); + if (result < 0) + return result; + + if (opener->format_token) + release_token(dev, opener, format); + if (!(dev->format_tokens & token)) { + result = -EBUSY; + goto exit_s_fmt_unlock; + } + + dprintk("S_FMT[%s] %4s:%ux%u size=%u\n", + V4L2_TYPE_IS_CAPTURE(f->type) ? "CAPTURE" : "OUTPUT", + fourcc2str(f->fmt.pix.pixelformat, buf), f->fmt.pix.width, + f->fmt.pix.height, f->fmt.pix.sizeimage); + changed = !pix_format_eq(&dev->pix_format, &f->fmt.pix, 0); + if (changed || has_no_owners(dev)) { + result = allocate_buffers(dev, &f->fmt.pix); + if (result < 0) + goto exit_s_fmt_unlock; + } + if ((dev->timeout_image && changed) || + (!dev->timeout_image && need_timeout_buffer(dev, token))) { + result = allocate_timeout_buffer(dev); + if (result < 0) + goto exit_s_fmt_free; + } + if (changed) { + dev->pix_format = f->fmt.pix; + dev->pix_format_has_valid_sizeimage = + v4l2l_pix_format_has_valid_sizeimage(f); + } + acquire_token(dev, opener, format, token); + if (opener->io_method == V4L2L_IO_TIMEOUT) + dev->timeout_image_io = 0; + goto exit_s_fmt_unlock; +exit_s_fmt_free: + free_buffers(dev); +exit_s_fmt_unlock: + mutex_unlock(&dev->image_mutex); + return result; +} + +/* ------------------ CAPTURE ----------------------- */ +/* ioctl for VIDIOC_ENUM_FMT, _G_FMT, _S_FMT, and _TRY_FMT when buffer type + * is V4L2_BUF_TYPE_VIDEO_CAPTURE */ + +static int vidioc_enum_fmt_cap(struct file *file, void *fh, + struct v4l2_fmtdesc *f) +{ + return vidioc_enum_fmt_vid(file, fh, f); +} + +static int vidioc_g_fmt_cap(struct file *file, void *fh, struct v4l2_format *f) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + if (check_buffer_capability(dev, opener, f->type) < 0) + return -EINVAL; + f->fmt.pix = dev->pix_format; + return 0; +} + +static int vidioc_try_fmt_cap(struct file *file, void *fh, + struct v4l2_format *f) +{ + return vidioc_try_fmt_vid(file, fh, f); +} + +static int vidioc_s_fmt_cap(struct file *file, void *fh, struct v4l2_format *f) +{ + return vidioc_s_fmt_vid(file, fh, f); +} + +/* ------------------ OUTPUT ----------------------- */ +/* ioctl for VIDIOC_ENUM_FMT, _G_FMT, _S_FMT, and _TRY_FMT when buffer type + * is V4L2_BUF_TYPE_VIDEO_OUTPUT */ + +static int vidioc_enum_fmt_out(struct file *file, void *fh, + struct v4l2_fmtdesc *f) +{ + return vidioc_enum_fmt_vid(file, fh, f); +} + +static int vidioc_g_fmt_out(struct file *file, void *fh, struct v4l2_format *f) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + if (check_buffer_capability(dev, opener, f->type) < 0) + return -EINVAL; + /* + * LATER: this should return the currently valid format + * gstreamer doesn't like it, if this returns -EINVAL, as it + * then concludes that there is _no_ valid format + * CHECK whether this assumption is wrong, + * or whether we have to always provide a valid format + */ + f->fmt.pix = dev->pix_format; + return 0; +} + +static int vidioc_try_fmt_out(struct file *file, void *fh, + struct v4l2_format *f) +{ + return vidioc_try_fmt_vid(file, fh, f); +} + +static int vidioc_s_fmt_out(struct file *file, void *fh, struct v4l2_format *f) +{ + return vidioc_s_fmt_vid(file, fh, f); +} + +// #define V4L2L_OVERLAY +#ifdef V4L2L_OVERLAY +/* ------------------ OVERLAY ----------------------- */ +/* currently unsupported */ +/* GSTreamer's v4l2sink is buggy, as it requires the overlay to work + * while it should only require it, if overlay is requested + * once the gstreamer element is fixed, remove the overlay dummies + */ +#warning OVERLAY dummies +static int vidioc_g_fmt_overlay(struct file *file, void *priv, + struct v4l2_format *fmt) +{ + return 0; +} + +static int vidioc_s_fmt_overlay(struct file *file, void *priv, + struct v4l2_format *fmt) +{ + return 0; +} +#endif /* V4L2L_OVERLAY */ + +/* ------------------ PARAMs ----------------------- */ + +/* get some data flow parameters, only capability, fps and readbuffers has + * effect on this driver + * called on VIDIOC_G_PARM + */ +static int vidioc_g_parm(struct file *file, void *fh, + struct v4l2_streamparm *parm) +{ + /* do not care about type of opener, hope these enums would always be + * compatible */ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + if (check_buffer_capability(dev, opener, parm->type) < 0) + return -EINVAL; + parm->parm.capture = dev->capture_param; + return 0; +} + +/* get some data flow parameters, only capability, fps and readbuffers has + * effect on this driver + * called on VIDIOC_S_PARM + */ +static int vidioc_s_parm(struct file *file, void *fh, + struct v4l2_streamparm *parm) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + + dprintk("S_PARM(frame-time=%u/%u)\n", + parm->parm.capture.timeperframe.numerator, + parm->parm.capture.timeperframe.denominator); + if (check_buffer_capability(dev, opener, parm->type) < 0) + return -EINVAL; + + switch (parm->type) { + case V4L2_BUF_TYPE_VIDEO_CAPTURE: + set_timeperframe(dev, &parm->parm.capture.timeperframe); + break; + case V4L2_BUF_TYPE_VIDEO_OUTPUT: + set_timeperframe(dev, &parm->parm.output.timeperframe); + break; + default: + return -EINVAL; + } + + parm->parm.capture = dev->capture_param; + return 0; +} + +#ifdef V4L2LOOPBACK_WITH_STD +/* sets a tv standard, actually we do not need to handle this any special way + * added to support effecttv + * called on VIDIOC_S_STD + */ +static int vidioc_s_std(struct file *file, void *fh, v4l2_std_id *_std) +{ + v4l2_std_id req_std = 0, supported_std = 0; + const v4l2_std_id all_std = V4L2_STD_ALL, no_std = 0; + + if (_std) { + req_std = *_std; + *_std = all_std; + } + + /* we support everything in V4L2_STD_ALL, but not more... */ + supported_std = (all_std & req_std); + if (no_std == supported_std) + return -EINVAL; + + return 0; +} + +/* gets a fake video standard + * called on VIDIOC_G_STD + */ +static int vidioc_g_std(struct file *file, void *fh, v4l2_std_id *norm) +{ + if (norm) + *norm = V4L2_STD_ALL; + return 0; +} +/* gets a fake video standard + * called on VIDIOC_QUERYSTD + */ +static int vidioc_querystd(struct file *file, void *fh, v4l2_std_id *norm) +{ + if (norm) + *norm = V4L2_STD_ALL; + return 0; +} +#endif /* V4L2LOOPBACK_WITH_STD */ + +static int v4l2loopback_set_ctrl(struct v4l2_loopback_device *dev, u32 id, + s64 val) +{ + int result = 0; + switch (id) { + case CID_KEEP_FORMAT: + if (val < 0 || val > 1) + return -EINVAL; + dev->keep_format = val; + result = mutex_lock_killable(&dev->image_mutex); + if (result < 0) + return result; + if (!dev->keep_format) { + if (has_no_owners(dev) && !any_buffers_mapped(dev)) + free_buffers(dev); + } + mutex_unlock(&dev->image_mutex); + break; + case CID_SUSTAIN_FRAMERATE: + if (val < 0 || val > 1) + return -EINVAL; + spin_lock_bh(&dev->lock); + dev->sustain_framerate = val; + check_timers(dev); + spin_unlock_bh(&dev->lock); + break; + case CID_TIMEOUT: + if (val < 0 || val > MAX_TIMEOUT) + return -EINVAL; + if (val > 0) { + result = mutex_lock_killable(&dev->image_mutex); + if (result < 0) + return result; + /* on-the-fly allocate if device is owned; else + * allocate occurs on next S_FMT or REQBUFS */ + if (!has_no_owners(dev)) + result = allocate_timeout_buffer(dev); + mutex_unlock(&dev->image_mutex); + if (result < 0) { + /* disable timeout as buffer not alloc'd */ + spin_lock_bh(&dev->lock); + dev->timeout_jiffies = 0; + spin_unlock_bh(&dev->lock); + return result; + } + } + spin_lock_bh(&dev->lock); + dev->timeout_jiffies = msecs_to_jiffies(val); + check_timers(dev); + spin_unlock_bh(&dev->lock); + break; + case CID_TIMEOUT_IMAGE_IO: + dev->timeout_image_io = 1; + break; + default: + return -EINVAL; + } + return 0; +} + +static int v4l2loopback_s_ctrl(struct v4l2_ctrl *ctrl) +{ + struct v4l2_loopback_device *dev = container_of( + ctrl->handler, struct v4l2_loopback_device, ctrl_handler); + return v4l2loopback_set_ctrl(dev, ctrl->id, ctrl->val); +} + +/* returns set of device outputs, in our case there is only one + * called on VIDIOC_ENUMOUTPUT + */ +static int vidioc_enum_output(struct file *file, void *fh, + struct v4l2_output *outp) +{ + __u32 index = outp->index; + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + + if (check_buffer_capability(dev, opener, V4L2_BUF_TYPE_VIDEO_OUTPUT)) + return -ENOTTY; + if (index) + return -EINVAL; + + /* clear all data (including the reserved fields) */ + memset(outp, 0, sizeof(*outp)); + + outp->index = index; + strscpy(outp->name, "loopback in", sizeof(outp->name)); + outp->type = V4L2_OUTPUT_TYPE_ANALOG; + outp->audioset = 0; + outp->modulator = 0; +#ifdef V4L2LOOPBACK_WITH_STD + outp->std = V4L2_STD_ALL; +#ifdef V4L2_OUT_CAP_STD + outp->capabilities |= V4L2_OUT_CAP_STD; +#endif /* V4L2_OUT_CAP_STD */ +#endif /* V4L2LOOPBACK_WITH_STD */ + + return 0; +} + +/* which output is currently active, + * called on VIDIOC_G_OUTPUT + */ +static int vidioc_g_output(struct file *file, void *fh, unsigned int *index) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + if (check_buffer_capability(dev, opener, V4L2_BUF_TYPE_VIDEO_OUTPUT)) + return -ENOTTY; + if (index) + *index = 0; + return 0; +} + +/* set output, can make sense if we have more than one video src, + * called on VIDIOC_S_OUTPUT + */ +static int vidioc_s_output(struct file *file, void *fh, unsigned int index) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + if (check_buffer_capability(dev, opener, V4L2_BUF_TYPE_VIDEO_OUTPUT)) + return -ENOTTY; + return index == 0 ? index : -EINVAL; +} + +/* returns set of device inputs, in our case there is only one, + * but later I may add more + * called on VIDIOC_ENUMINPUT + */ +static int vidioc_enum_input(struct file *file, void *fh, + struct v4l2_input *inp) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + __u32 index = inp->index; + + if (check_buffer_capability(dev, opener, V4L2_BUF_TYPE_VIDEO_CAPTURE)) + return -ENOTTY; + if (index) + return -EINVAL; + + /* clear all data (including the reserved fields) */ + memset(inp, 0, sizeof(*inp)); + + inp->index = index; + strscpy(inp->name, "loopback", sizeof(inp->name)); + inp->type = V4L2_INPUT_TYPE_CAMERA; + inp->audioset = 0; + inp->tuner = 0; + inp->status = 0; + +#ifdef V4L2LOOPBACK_WITH_STD + inp->std = V4L2_STD_ALL; +#ifdef V4L2_IN_CAP_STD + inp->capabilities |= V4L2_IN_CAP_STD; +#endif +#endif /* V4L2LOOPBACK_WITH_STD */ + + if (has_output_token(dev->stream_tokens) && !dev->keep_format) + /* if no outputs attached; pretend device is powered off */ + inp->status |= V4L2_IN_ST_NO_SIGNAL; + + return 0; +} + +/* which input is currently active, + * called on VIDIOC_G_INPUT + */ +static int vidioc_g_input(struct file *file, void *fh, unsigned int *index) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + if (check_buffer_capability(dev, opener, V4L2_BUF_TYPE_VIDEO_CAPTURE)) + return -ENOTTY; /* NOTE: -EAGAIN might be more informative */ + if (index) + *index = 0; + return 0; +} + +/* set input, can make sense if we have more than one video src, + * called on VIDIOC_S_INPUT + */ +static int vidioc_s_input(struct file *file, void *fh, unsigned int index) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + if (index != 0) + return -EINVAL; + if (check_buffer_capability(dev, opener, V4L2_BUF_TYPE_VIDEO_CAPTURE)) + return -ENOTTY; /* NOTE: -EAGAIN might be more informative */ + return 0; +} + +/* --------------- V4L2 ioctl buffer related calls ----------------- */ + +#define is_allocated(opener, type, index) \ + (opener->format_token & (opener->io_method == V4L2L_IO_TIMEOUT ? \ + V4L2L_TOKEN_TIMEOUT : \ + token_from_type(type)) && \ + (index) < (opener)->buffer_count) +#define BUFFER_DEBUG_FMT_STR \ + "buffer#%u @ %p type=%u bytesused=%u length=%u flags=%x " \ + "field=%u timestamp= %lld.%06lldsequence=%u\n" +#define BUFFER_DEBUG_FMT_ARGS(buf) \ + (buf)->index, (buf), (buf)->type, (buf)->bytesused, (buf)->length, \ + (buf)->flags, (buf)->field, \ + (long long)(buf)->timestamp.tv_sec, \ + (long long)(buf)->timestamp.tv_usec, (buf)->sequence +/* Buffer flag helpers */ +#define unset_flags(flags) \ + do { \ + flags &= ~V4L2_BUF_FLAG_QUEUED; \ + flags &= ~V4L2_BUF_FLAG_DONE; \ + } while (0) +#define set_queued(flags) \ + do { \ + flags |= V4L2_BUF_FLAG_QUEUED; \ + flags &= ~V4L2_BUF_FLAG_DONE; \ + } while (0) +#define set_done(flags) \ + do { \ + flags &= ~V4L2_BUF_FLAG_QUEUED; \ + flags |= V4L2_BUF_FLAG_DONE; \ + } while (0) + +static bool any_buffers_mapped(struct v4l2_loopback_device *dev) +{ + u32 index; + for (index = 0; index < dev->buffer_count; ++index) + if (dev->buffers[index].buffer.flags & V4L2_BUF_FLAG_MAPPED) + return true; + return false; +} + +static void prepare_buffer_queue(struct v4l2_loopback_device *dev, int count) +{ + struct v4l2l_buffer *bufd, *n; + u32 pos; + + spin_lock_bh(&dev->list_lock); + + /* ensure sufficient number of buffers in queue */ + for (pos = 0; pos < count; ++pos) { + bufd = &dev->buffers[pos]; + if (list_empty(&bufd->list_head)) + list_add_tail(&bufd->list_head, &dev->outbufs_list); + } + if (list_empty(&dev->outbufs_list)) + goto exit_prepare_queue_unlock; + + /* remove any excess buffers */ + list_for_each_entry_safe(bufd, n, &dev->outbufs_list, list_head) { + if (bufd->buffer.index >= count) + list_del_init(&bufd->list_head); + } + + /* buffers are no longer queued; and `write_position` will correspond + * to the first item of `outbufs_list`. */ + pos = v4l2l_mod64(dev->write_position, count); + list_for_each_entry(bufd, &dev->outbufs_list, list_head) { + unset_flags(bufd->buffer.flags); + dev->bufpos2index[pos % count] = bufd->buffer.index; + ++pos; + } +exit_prepare_queue_unlock: + spin_unlock_bh(&dev->list_lock); +} + +/* forward declaration */ +static int vidioc_streamoff(struct file *file, void *fh, + enum v4l2_buf_type type); +/* negotiate buffer type + * only mmap streaming supported + * called on VIDIOC_REQBUFS + */ +static int vidioc_reqbufs(struct file *file, void *fh, + struct v4l2_requestbuffers *reqbuf) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + u32 token = opener->io_method == V4L2L_IO_TIMEOUT ? + V4L2L_TOKEN_TIMEOUT : + token_from_type(reqbuf->type); + u32 req_count = reqbuf->count; + int result = 0; + + dprintk("REQBUFS(memory=%u, req_count=%u) and device-bufs=%u/%u " + "[used/max]\n", + reqbuf->memory, req_count, dev->used_buffer_count, + dev->buffer_count); + + switch (reqbuf->memory) { + case V4L2_MEMORY_MMAP: +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 20, 0) + reqbuf->capabilities = 0; /* only guarantee MMAP support */ +#endif +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 16, 0) + reqbuf->flags = 0; /* no memory consistency support */ +#endif + break; + default: + return -EINVAL; + } + + if (opener->format_token & ~token) + /* different (buffer) type already assigned to descriptor by + * S_FMT or REQBUFS */ + return -EINVAL; + + MARK(); + result = mutex_lock_killable(&dev->image_mutex); + if (result < 0) + return result; /* -EINTR */ + + /* CASE queue/dequeue timeout-buffer only: */ + if (opener->format_token & V4L2L_TOKEN_TIMEOUT) { + opener->buffer_count = req_count; + if (req_count == 0) + release_token(dev, opener, format); + goto exit_reqbufs_unlock; + } + + MARK(); + /* CASE count is zero: streamoff, free buffers, release their token */ + if (req_count == 0) { + if (dev->format_tokens & token) { + acquire_token(dev, opener, format, token); + opener->io_method = V4L2L_IO_MMAP; + } + result = vidioc_streamoff(file, fh, reqbuf->type); + opener->buffer_count = 0; + /* undocumented requirement - REQBUFS with count zero should + * ALSO release lock on logical stream */ + if (opener->format_token) + release_token(dev, opener, format); + if (has_no_owners(dev)) + dev->used_buffer_count = 0; + goto exit_reqbufs_unlock; + } + + /* CASE count non-zero: allocate buffers and acquire token for them */ + MARK(); + switch (reqbuf->type) { + case V4L2_BUF_TYPE_VIDEO_CAPTURE: + case V4L2_BUF_TYPE_VIDEO_OUTPUT: + if (!(dev->format_tokens & token || + opener->format_token & token)) + /* only exclusive ownership for each stream */ + result = -EBUSY; + break; + default: + result = -EINVAL; + } + if (result < 0) + goto exit_reqbufs_unlock; + + if (has_other_owners(opener, dev) && dev->used_buffer_count > 0) { + /* allow 'allocation' of existing number of buffers */ + req_count = dev->used_buffer_count; + } else if (any_buffers_mapped(dev)) { + /* do not allow re-allocation if buffers are mapped */ + result = -EBUSY; + goto exit_reqbufs_unlock; + } + + MARK(); + opener->buffer_count = 0; + + if (req_count > dev->buffer_count) + req_count = dev->buffer_count; + + if (has_no_owners(dev)) { + result = allocate_buffers(dev, &dev->pix_format); + if (result < 0) + goto exit_reqbufs_unlock; + } + if (!dev->timeout_image && need_timeout_buffer(dev, token)) { + result = allocate_timeout_buffer(dev); + if (result < 0) + goto exit_reqbufs_unlock; + } + acquire_token(dev, opener, format, token); + + MARK(); + switch (opener->io_method) { + case V4L2L_IO_TIMEOUT: + dev->timeout_image_io = 0; + opener->buffer_count = req_count; + break; + default: + opener->io_method = V4L2L_IO_MMAP; + prepare_buffer_queue(dev, req_count); + dev->used_buffer_count = opener->buffer_count = req_count; + } +exit_reqbufs_unlock: + mutex_unlock(&dev->image_mutex); + reqbuf->count = opener->buffer_count; + return result; +} + +/* returns buffer asked for; + * give app as many buffers as it wants, if it less than MAX, + * but map them in our inner buffers + * called on VIDIOC_QUERYBUF + */ +static int vidioc_querybuf(struct file *file, void *fh, struct v4l2_buffer *buf) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + u32 type = buf->type; + u32 index = buf->index; + + if ((type != V4L2_BUF_TYPE_VIDEO_CAPTURE) && + (type != V4L2_BUF_TYPE_VIDEO_OUTPUT)) + return -EINVAL; + if (!is_allocated(opener, type, index)) + return -EINVAL; + + if (opener->format_token & V4L2L_TOKEN_TIMEOUT) { + *buf = dev->timeout_buffer.buffer; + buf->index = index; + } else + *buf = dev->buffers[index].buffer; + + buf->type = type; + + if (!(buf->flags & (V4L2_BUF_FLAG_DONE | V4L2_BUF_FLAG_QUEUED))) { + /* v4l2-compliance requires these to be zero */ + buf->sequence = 0; + buf->timestamp.tv_sec = buf->timestamp.tv_usec = 0; + } else if (V4L2_TYPE_IS_CAPTURE(type)) { + /* guess flags based on sequence values */ + if (buf->sequence >= opener->read_position) { + set_done(buf->flags); + } else if (buf->flags & V4L2_BUF_FLAG_DONE) { + set_queued(buf->flags); + } + } + dprintkrw("QUERYBUF(%s, index=%u) -> " BUFFER_DEBUG_FMT_STR, + V4L2_TYPE_IS_CAPTURE(type) ? "CAPTURE" : "OUTPUT", index, + BUFFER_DEBUG_FMT_ARGS(buf)); + return 0; +} + +static void buffer_written(struct v4l2_loopback_device *dev, + struct v4l2l_buffer *buf) +{ + timer_delete_sync(&dev->sustain_timer); + timer_delete_sync(&dev->timeout_timer); + + spin_lock_bh(&dev->list_lock); + list_move_tail(&buf->list_head, &dev->outbufs_list); + spin_unlock_bh(&dev->list_lock); + + spin_lock_bh(&dev->lock); + dev->bufpos2index[v4l2l_mod64(dev->write_position, + dev->used_buffer_count)] = + buf->buffer.index; + ++dev->write_position; + dev->reread_count = 0; + + check_timers(dev); + spin_unlock_bh(&dev->lock); +} + +/* put buffer to queue + * called on VIDIOC_QBUF + */ +static int vidioc_qbuf(struct file *file, void *fh, struct v4l2_buffer *buf) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + struct v4l2l_buffer *bufd; + u32 index = buf->index; + u32 type = buf->type; + + if (!is_allocated(opener, type, index)) + return -EINVAL; + bufd = &dev->buffers[index]; + + switch (buf->memory) { + case V4L2_MEMORY_MMAP: + if (!(bufd->buffer.flags & V4L2_BUF_FLAG_MAPPED)) + dprintkrw("QBUF() unmapped buffer [index=%u]\n", index); + break; + default: + return -EINVAL; + } + + if (opener->format_token & V4L2L_TOKEN_TIMEOUT) { + set_queued(buf->flags); + return 0; + } + + switch (type) { + case V4L2_BUF_TYPE_VIDEO_CAPTURE: + dprintkrw("QBUF(CAPTURE, index=%u) -> " BUFFER_DEBUG_FMT_STR, + index, BUFFER_DEBUG_FMT_ARGS(buf)); + set_queued(buf->flags); + break; + case V4L2_BUF_TYPE_VIDEO_OUTPUT: + dprintkrw("QBUF(OUTPUT, index=%u) -> " BUFFER_DEBUG_FMT_STR, + index, BUFFER_DEBUG_FMT_ARGS(buf)); + if (!(bufd->buffer.flags & V4L2_BUF_FLAG_TIMESTAMP_COPY) && + (buf->timestamp.tv_sec == 0 && + buf->timestamp.tv_usec == 0)) { + v4l2l_get_timestamp(&bufd->buffer); + } else { + bufd->buffer.timestamp = buf->timestamp; + bufd->buffer.flags |= V4L2_BUF_FLAG_TIMESTAMP_COPY; + bufd->buffer.flags &= + ~V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC; + } + if (dev->pix_format_has_valid_sizeimage) { + if (buf->bytesused >= dev->pix_format.sizeimage) { + bufd->buffer.bytesused = + dev->pix_format.sizeimage; + } else { +#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0) + dev_warn_ratelimited( + &dev->vdev->dev, +#else + dprintkrw( +#endif + "warning queued output buffer bytesused too small %u < %u\n", + buf->bytesused, + dev->pix_format.sizeimage); + bufd->buffer.bytesused = buf->bytesused; + } + } else { + bufd->buffer.bytesused = buf->bytesused; + } + bufd->buffer.sequence = dev->write_position; + set_queued(bufd->buffer.flags); + *buf = bufd->buffer; + buffer_written(dev, bufd); + set_done(bufd->buffer.flags); + wake_up_all(&dev->read_event); + break; + default: + return -EINVAL; + } + buf->type = type; + return 0; +} + +static int can_read(struct v4l2_loopback_device *dev, + struct v4l2_loopback_opener *opener) +{ + int ret; + + spin_lock_bh(&dev->lock); + check_timers(dev); + ret = dev->write_position > opener->read_position || + dev->reread_count > opener->reread_count || dev->timeout_happened; + spin_unlock_bh(&dev->lock); + return ret; +} + +static int get_capture_buffer(struct file *file) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = + v4l2l_f_to_opener(file, file->private_data); + int pos, timeout_happened; + u32 index; + + if ((file->f_flags & O_NONBLOCK) && + (dev->write_position <= opener->read_position && + dev->reread_count <= opener->reread_count && + !dev->timeout_happened)) + return -EAGAIN; + wait_event_interruptible(dev->read_event, can_read(dev, opener)); + + spin_lock_bh(&dev->lock); + if (dev->write_position == opener->read_position) { + if (dev->reread_count > opener->reread_count + 2) + opener->reread_count = dev->reread_count - 1; + ++opener->reread_count; + pos = v4l2l_mod64(opener->read_position + + dev->used_buffer_count - 1, + dev->used_buffer_count); + } else { + opener->reread_count = 0; + if (dev->write_position > + opener->read_position + dev->used_buffer_count) + opener->read_position = dev->write_position - 1; + pos = v4l2l_mod64(opener->read_position, + dev->used_buffer_count); + ++opener->read_position; + } + timeout_happened = dev->timeout_happened && (dev->timeout_jiffies > 0); + dev->timeout_happened = 0; + spin_unlock_bh(&dev->lock); + + index = dev->bufpos2index[pos]; + if (timeout_happened) { + if (index >= dev->used_buffer_count) { + dprintkrw("get_capture_buffer() read position is at " + "an unallocated buffer [index=%u]\n", + index); + return -EFAULT; + } + /* although allocated on-demand, timeout_image is freed only + * in free_buffers(), so we don't need to worry about it being + * deallocated suddenly */ + memcpy(dev->image + dev->buffers[index].buffer.m.offset, + dev->timeout_image, dev->buffer_size); + } + return (int)index; +} + +/* put buffer to dequeue + * called on VIDIOC_DQBUF + */ +static int vidioc_dqbuf(struct file *file, void *fh, struct v4l2_buffer *buf) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + u32 type = buf->type; + int index; + struct v4l2l_buffer *bufd; + + if (buf->memory != V4L2_MEMORY_MMAP) + return -EINVAL; + if (opener->format_token & V4L2L_TOKEN_TIMEOUT) { + *buf = dev->timeout_buffer.buffer; + buf->type = type; + unset_flags(buf->flags); + return 0; + } + if ((opener->buffer_count == 0) || + !(opener->format_token & token_from_type(type))) + return -EINVAL; + + switch (type) { + case V4L2_BUF_TYPE_VIDEO_CAPTURE: + index = get_capture_buffer(file); + if (index < 0) + return index; + *buf = dev->buffers[index].buffer; + unset_flags(buf->flags); + break; + case V4L2_BUF_TYPE_VIDEO_OUTPUT: + spin_lock_bh(&dev->list_lock); + + bufd = list_first_entry_or_null(&dev->outbufs_list, + struct v4l2l_buffer, list_head); + if (bufd) + list_move_tail(&bufd->list_head, &dev->outbufs_list); + + spin_unlock_bh(&dev->list_lock); + if (!bufd) + return -EFAULT; + unset_flags(bufd->buffer.flags); + *buf = bufd->buffer; + break; + default: + return -EINVAL; + } + + buf->type = type; + dprintkrw("DQBUF(%s, index=%u) -> " BUFFER_DEBUG_FMT_STR, + V4L2_TYPE_IS_CAPTURE(type) ? "CAPTURE" : "OUTPUT", index, + BUFFER_DEBUG_FMT_ARGS(buf)); + return 0; +} + +/* ------------- STREAMING ------------------- */ + +/* start streaming + * called on VIDIOC_STREAMON + */ +static int vidioc_streamon(struct file *file, void *fh, enum v4l2_buf_type type) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + u32 token = token_from_type(type); + + /* short-circuit when using timeout buffer set */ + if (opener->format_token & V4L2L_TOKEN_TIMEOUT) + return 0; + /* opener must have claimed (same) buffer set via REQBUFS */ + if (!opener->buffer_count || !(opener->format_token & token)) + return -EINVAL; + + switch (type) { + case V4L2_BUF_TYPE_VIDEO_CAPTURE: + if (has_output_token(dev->stream_tokens) && !dev->keep_format) + return -EIO; + if (dev->stream_tokens & token) { + acquire_token(dev, opener, stream, token); + client_usage_queue_event(dev->vdev); + } + return 0; + case V4L2_BUF_TYPE_VIDEO_OUTPUT: + if (dev->stream_tokens & token) + acquire_token(dev, opener, stream, token); + return 0; + default: + return -EINVAL; + } +} + +/* stop streaming + * called on VIDIOC_STREAMOFF + */ +static int vidioc_streamoff(struct file *file, void *fh, + enum v4l2_buf_type type) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + u32 token = token_from_type(type); + + /* short-circuit when using timeout buffer set */ + if (opener->format_token & V4L2L_TOKEN_TIMEOUT) + return 0; + /* short-circuit when buffer set has no owner */ + if (dev->format_tokens & token) + return 0; + /* opener needs a claim to buffer set */ + if (!opener->format_token) + return -EBUSY; + if (opener->format_token & ~token) + return -EINVAL; + + switch (type) { + case V4L2_BUF_TYPE_VIDEO_OUTPUT: + if (opener->stream_token & token) + release_token(dev, opener, stream); + /* reset output queue */ + if (dev->used_buffer_count > 0) + prepare_buffer_queue(dev, dev->used_buffer_count); + return 0; + case V4L2_BUF_TYPE_VIDEO_CAPTURE: + if (opener->stream_token & token) { + release_token(dev, opener, stream); + client_usage_queue_event(dev->vdev); + } + return 0; + default: + return -EINVAL; + } +} + +#ifdef CONFIG_VIDEO_V4L1_COMPAT +static int vidiocgmbuf(struct file *file, void *fh, struct video_mbuf *p) +{ + struct v4l2_loopback_device *dev; + MARK(); + + dev = v4l2loopback_getdevice(file); + p->frames = dev->buffer_count; + p->offsets[0] = 0; + p->offsets[1] = 0; + p->size = dev->buffer_size; + return 0; +} +#endif + +static void client_usage_queue_event(struct video_device *vdev) +{ + struct v4l2_event ev; + struct v4l2_loopback_device *dev; + + dev = container_of(vdev->v4l2_dev, struct v4l2_loopback_device, + v4l2_dev); + + memset(&ev, 0, sizeof(ev)); + ev.type = V4L2_EVENT_PRI_CLIENT_USAGE; + ((struct v4l2_event_client_usage *)&ev.u)->count = + !has_capture_token(dev->stream_tokens); + + v4l2_event_queue(vdev, &ev); +} + +static int client_usage_ops_add(struct v4l2_subscribed_event *sev, + unsigned elems) +{ + if (!(sev->flags & V4L2_EVENT_SUB_FL_SEND_INITIAL)) + return 0; + + client_usage_queue_event(sev->fh->vdev); + return 0; +} + +static void client_usage_ops_replace(struct v4l2_event *old, + const struct v4l2_event *new) +{ + *((struct v4l2_event_client_usage *)&old->u) = + *((struct v4l2_event_client_usage *)&new->u); +} + +static void client_usage_ops_merge(const struct v4l2_event *old, + struct v4l2_event *new) +{ + *((struct v4l2_event_client_usage *)&new->u) = + *((struct v4l2_event_client_usage *)&old->u); +} + +const struct v4l2_subscribed_event_ops client_usage_ops = { + .add = client_usage_ops_add, + .replace = client_usage_ops_replace, + .merge = client_usage_ops_merge, +}; + +static int vidioc_subscribe_event(struct v4l2_fh *fh, + const struct v4l2_event_subscription *sub) +{ + switch (sub->type) { + case V4L2_EVENT_CTRL: + return v4l2_ctrl_subscribe_event(fh, sub); + case V4L2_EVENT_PRI_CLIENT_USAGE: + return v4l2_event_subscribe(fh, sub, 0, &client_usage_ops); + } + + return -EINVAL; +} + +/* file operations */ +static void vm_open(struct vm_area_struct *vma) +{ + struct v4l2l_buffer *buf; + MARK(); + + buf = vma->vm_private_data; + atomic_inc(&buf->use_count); + buf->buffer.flags |= V4L2_BUF_FLAG_MAPPED; +} + +static void vm_close(struct vm_area_struct *vma) +{ + struct v4l2l_buffer *buf; + MARK(); + + buf = vma->vm_private_data; + if (atomic_dec_and_test(&buf->use_count)) + buf->buffer.flags &= ~V4L2_BUF_FLAG_MAPPED; +} + +static struct vm_operations_struct vm_ops = { + .open = vm_open, + .close = vm_close, +}; + +static int v4l2_loopback_mmap(struct file *file, struct vm_area_struct *vma) +{ + u8 *addr; + unsigned long start, size, offset; + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = + v4l2l_f_to_opener(file, file->private_data); + struct v4l2l_buffer *buffer = NULL; + int result = 0; + MARK(); + + offset = (unsigned long)vma->vm_pgoff << PAGE_SHIFT; + start = (unsigned long)vma->vm_start; + size = (unsigned long)(vma->vm_end - vma->vm_start); /* always != 0 */ + + /* ensure buffer size, count, and allocated image(s) are not altered by + * other file descriptors */ + result = mutex_lock_killable(&dev->image_mutex); + if (result < 0) + return result; + + if (size > dev->buffer_size) { + dprintk("mmap() attempt to map %lubytes when %ubytes are " + "allocated to buffers\n", + size, dev->buffer_size); + result = -EINVAL; + goto exit_mmap_unlock; + } + if (offset % dev->buffer_size != 0) { + dprintk("mmap() offset does not match start of any buffer\n"); + result = -EINVAL; + goto exit_mmap_unlock; + } + switch (opener->format_token) { + case V4L2L_TOKEN_TIMEOUT: + if (offset != (unsigned long)dev->buffer_size * MAX_BUFFERS) { + dprintk("mmap() incorrect offset for timeout image\n"); + result = -EINVAL; + goto exit_mmap_unlock; + } + buffer = &dev->timeout_buffer; + addr = dev->timeout_image; + break; + default: + if (offset >= dev->image_size) { + dprintk("mmap() attempt to map beyond all buffers\n"); + result = -EINVAL; + goto exit_mmap_unlock; + } + u32 index = offset / dev->buffer_size; + buffer = &dev->buffers[index]; + addr = dev->image + offset; + break; + } + + while (size > 0) { + struct page *page = vmalloc_to_page(addr); + + result = vm_insert_page(vma, start, page); + if (result < 0) + goto exit_mmap_unlock; + + start += PAGE_SIZE; + addr += PAGE_SIZE; + size -= PAGE_SIZE; + } + + vma->vm_ops = &vm_ops; + vma->vm_private_data = buffer; + + vm_open(vma); +exit_mmap_unlock: + mutex_unlock(&dev->image_mutex); + return result; +} + +static unsigned int v4l2_loopback_poll(struct file *file, + struct poll_table_struct *pts) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = + v4l2l_f_to_opener(file, file->private_data); + __poll_t req_events = poll_requested_events(pts); + int ret_mask = 0; + + /* call poll_wait in first call, regardless, to ensure that the + * wait-queue is not null */ + poll_wait(file, &dev->read_event, pts); + poll_wait(file, &opener->fh.wait, pts); + + if (req_events & POLLPRI) { + if (v4l2_event_pending(&opener->fh)) { + ret_mask |= POLLPRI; + if (!(req_events & DEFAULT_POLLMASK)) + return ret_mask; + } + } + + switch (opener->format_token) { + case V4L2L_TOKEN_OUTPUT: + if (opener->stream_token != 0 || + opener->io_method == V4L2L_IO_NONE) + ret_mask |= POLLOUT | POLLWRNORM; + break; + case V4L2L_TOKEN_CAPTURE: + if ((opener->io_method == V4L2L_IO_NONE || + opener->stream_token != 0) && + can_read(dev, opener)) + ret_mask |= POLLIN | POLLWRNORM; + break; + case V4L2L_TOKEN_TIMEOUT: + ret_mask |= POLLOUT | POLLWRNORM; + break; + default: + break; + } + + return ret_mask; +} + +/* do not want to limit device opens, it can be as many readers as user want, + * writers are limited by means of setting writer field */ +static int v4l2_loopback_open(struct file *file) +{ + struct v4l2_loopback_device *dev; + struct v4l2_loopback_opener *opener; + + dev = v4l2loopback_getdevice(file); + if (dev->open_count.counter >= dev->max_openers) + return -EBUSY; + /* kfree on close */ + opener = kzalloc(sizeof(*opener), GFP_KERNEL); + if (opener == NULL) + return -ENOMEM; + + atomic_inc(&dev->open_count); + if (dev->timeout_image_io && dev->format_tokens & V4L2L_TOKEN_TIMEOUT) + /* will clear timeout_image_io once buffer set acquired */ + opener->io_method = V4L2L_IO_TIMEOUT; + + v4l2_fh_init(&opener->fh, video_devdata(file)); + file->private_data = &opener->fh; + + v4l2_fh_add(&opener->fh, file); + dprintk("open() -> dev@%p with image@%p\n", dev, + dev ? dev->image : NULL); + return 0; +} + +static int v4l2_loopback_close(struct file *file) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = + v4l2l_f_to_opener(file, file->private_data); + int result = 0; + dprintk("close() -> dev@%p with image@%p\n", dev, + dev ? dev->image : NULL); + + if (opener->format_token) { + struct v4l2_requestbuffers reqbuf = { + .count = 0, .memory = V4L2_MEMORY_MMAP, .type = 0 + }; + switch (opener->format_token) { + case V4L2L_TOKEN_CAPTURE: + reqbuf.type = V4L2_BUF_TYPE_VIDEO_CAPTURE; + break; + case V4L2L_TOKEN_OUTPUT: + case V4L2L_TOKEN_TIMEOUT: + reqbuf.type = V4L2_BUF_TYPE_VIDEO_OUTPUT; + break; + } + if (reqbuf.type) + result = vidioc_reqbufs(file, file->private_data, + &reqbuf); + if (result < 0) + dprintk("failed to free buffers REQBUFS(count=0) " + " returned %d\n", + result); + mutex_lock(&dev->image_mutex); + release_token(dev, opener, format); + mutex_unlock(&dev->image_mutex); + } + + if (atomic_dec_and_test(&dev->open_count)) { + timer_delete_sync(&dev->sustain_timer); + timer_delete_sync(&dev->timeout_timer); + if (!dev->keep_format) { + mutex_lock(&dev->image_mutex); + free_buffers(dev); + mutex_unlock(&dev->image_mutex); + } + } + + v4l2_fh_del(&opener->fh, file); + v4l2_fh_exit(&opener->fh); + + kfree(opener); + return 0; +} + +static int start_fileio(struct file *file, void *fh, enum v4l2_buf_type type) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_loopback_opener *opener = v4l2l_f_to_opener(file, fh); + struct v4l2_requestbuffers reqbuf = { .count = dev->buffer_count, + .memory = V4L2_MEMORY_MMAP, + .type = type }; + int token = token_from_type(type); + int result; + + if (opener->format_token & V4L2L_TOKEN_TIMEOUT || + opener->format_token & ~token) + return -EBUSY; /* NOTE: -EBADF might be more informative */ + + /* short-circuit if already have stream token */ + if (opener->stream_token && opener->io_method == V4L2L_IO_FILE) + return 0; + + /* otherwise attempt to acquire stream token and assign IO method */ + if (!(dev->stream_tokens & token) || opener->io_method != V4L2L_IO_NONE) + return -EBUSY; + + result = vidioc_reqbufs(file, fh, &reqbuf); + if (result < 0) + return result; + result = vidioc_streamon(file, fh, type); + if (result < 0) + return result; + + opener->io_method = V4L2L_IO_FILE; + return 0; +} + +static ssize_t v4l2_loopback_read(struct file *file, char __user *buf, + size_t count, loff_t *ppos) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_buffer *b; + int index, result; + + dprintkrw("read() %zu bytes\n", count); + result = start_fileio(file, file->private_data, + V4L2_BUF_TYPE_VIDEO_CAPTURE); + if (result < 0) + return result; + + index = get_capture_buffer(file); + if (index < 0) + return index; + b = &dev->buffers[index].buffer; + if (count > b->bytesused) + count = b->bytesused; + if (copy_to_user((void *)buf, (void *)(dev->image + b->m.offset), + count)) { + printk(KERN_ERR "v4l2-loopback read() failed copy_to_user()\n"); + return -EFAULT; + } + return count; +} + +static ssize_t v4l2_loopback_write(struct file *file, const char __user *buf, + size_t count, loff_t *ppos) +{ + struct v4l2_loopback_device *dev = v4l2loopback_getdevice(file); + struct v4l2_buffer *b; + int index, result; + + dprintkrw("write() %zu bytes\n", count); + result = start_fileio(file, file->private_data, + V4L2_BUF_TYPE_VIDEO_OUTPUT); + if (result < 0) + return result; + + if (count > dev->buffer_size) + count = dev->buffer_size; + index = v4l2l_mod64(dev->write_position, dev->used_buffer_count); + b = &dev->buffers[index].buffer; + + if (copy_from_user((void *)(dev->image + b->m.offset), (void *)buf, + count)) { + printk(KERN_ERR + "v4l2-loopback write() failed copy_from_user()\n"); + return -EFAULT; + } + b->bytesused = count; + + v4l2l_get_timestamp(b); + b->sequence = dev->write_position; + set_queued(b->flags); + buffer_written(dev, &dev->buffers[index]); + set_done(b->flags); + wake_up_all(&dev->read_event); + + return count; +} + +/* init functions */ +/* frees buffers, if allocated */ +static void free_buffers(struct v4l2_loopback_device *dev) +{ + dprintk("free_buffers() with image@%p\n", dev->image); + if (!dev->image) + return; + if (!has_no_owners(dev) || any_buffers_mapped(dev)) + /* maybe an opener snuck in before image_mutex was acquired */ + printk(KERN_WARNING + "v4l2-loopback free_buffers() buffers of video device " + "#%u freed while still mapped to userspace\n", + dev->vdev->num); + vfree(dev->image); + dev->image = NULL; + dev->image_size = 0; + dev->buffer_size = 0; +} + +static void free_timeout_buffer(struct v4l2_loopback_device *dev) +{ + dprintk("free_timeout_buffer() with timeout_image@%p\n", + dev->timeout_image); + if (!dev->timeout_image) + return; + + if ((dev->timeout_jiffies > 0 && !has_no_owners(dev)) || + dev->timeout_buffer.buffer.flags & V4L2_BUF_FLAG_MAPPED) + printk(KERN_WARNING + "v4l2-loopback free_timeout_buffer() timeout image " + "of device #%u freed while still mapped to userspace\n", + dev->vdev->num); + + vfree(dev->timeout_image); + dev->timeout_image = NULL; + dev->timeout_buffer_size = 0; +} +/* allocates buffers if no (other) openers are already using them */ +static int allocate_buffers(struct v4l2_loopback_device *dev, + struct v4l2_pix_format *pix_format) +{ + u32 buffer_size = PAGE_ALIGN(pix_format->sizeimage); + unsigned long image_size = + (unsigned long)buffer_size * (unsigned long)dev->buffer_count; + /* vfree on close file operation in case no open handles left */ + + if (buffer_size == 0 || dev->buffer_count == 0 || + buffer_size < pix_format->sizeimage) + return -EINVAL; + + if ((__LONG_MAX__ / buffer_size) < dev->buffer_count) + return -ENOSPC; + + dprintk("allocate_buffers() size %lubytes = %ubytes x %ubuffers\n", + image_size, buffer_size, dev->buffer_count); + if (dev->image) { + /* check that no buffers are expected in user-space */ + if (!has_no_owners(dev) || any_buffers_mapped(dev)) + return -EBUSY; + dprintk("allocate_buffers() existing size=%lubytes\n", + dev->image_size); + /* FIXME: prevent double allocation more intelligently! */ + if (image_size == dev->image_size) { + dprintk("allocate_buffers() keep existing\n"); + return 0; + } + free_buffers(dev); + } + + /* FIXME: set buffers to 0 */ + dev->image = vmalloc(image_size); + if (dev->image == NULL) { + dev->buffer_size = dev->image_size = 0; + return -ENOMEM; + } + init_buffers(dev, pix_format->sizeimage, buffer_size); + dev->buffer_size = buffer_size; + dev->image_size = image_size; + dprintk("allocate_buffers() -> vmalloc'd %lubytes\n", dev->image_size); + return 0; +} +static int allocate_timeout_buffer(struct v4l2_loopback_device *dev) +{ + /* device's `buffer_size` and `buffers` must be initialised in + * allocate_buffers() */ + + dprintk("allocate_timeout_buffer() size %ubytes\n", dev->buffer_size); + if (dev->buffer_size == 0) + return -EINVAL; + + if (dev->timeout_image) { + if (dev->timeout_buffer.buffer.flags & V4L2_BUF_FLAG_MAPPED) + return -EBUSY; + if (dev->buffer_size == dev->timeout_buffer_size) + return 0; + free_timeout_buffer(dev); + } + + dev->timeout_image = vzalloc(dev->buffer_size); + if (!dev->timeout_image) { + dev->timeout_buffer_size = 0; + return -ENOMEM; + } + dev->timeout_buffer_size = dev->buffer_size; + return 0; +} +/* init inner buffers, they are capture mode and flags are set as for capture + * mode buffers */ +static void init_buffers(struct v4l2_loopback_device *dev, u32 bytes_used, + u32 buffer_size) +{ + u32 i; + + for (i = 0; i < dev->buffer_count; ++i) { + struct v4l2_buffer *b = &dev->buffers[i].buffer; + b->index = i; + b->bytesused = bytes_used; + b->length = buffer_size; + b->field = V4L2_FIELD_NONE; + b->flags = 0; + b->m.offset = i * buffer_size; + b->memory = V4L2_MEMORY_MMAP; + b->sequence = 0; + b->timestamp.tv_sec = 0; + b->timestamp.tv_usec = 0; + b->type = V4L2_BUF_TYPE_VIDEO_CAPTURE; + + v4l2l_get_timestamp(b); + } + dev->timeout_buffer = dev->buffers[0]; + dev->timeout_buffer.buffer.m.offset = MAX_BUFFERS * buffer_size; +} + +/* fills and register video device */ +static void init_vdev(struct video_device *vdev, int nr) +{ +#ifdef V4L2LOOPBACK_WITH_STD + vdev->tvnorms = V4L2_STD_ALL; +#endif /* V4L2LOOPBACK_WITH_STD */ + + vdev->vfl_type = VFL_TYPE_VIDEO; + vdev->fops = &v4l2_loopback_fops; + vdev->ioctl_ops = &v4l2_loopback_ioctl_ops; + vdev->release = &video_device_release; + vdev->minor = -1; +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 7, 0) + vdev->device_caps = V4L2_CAP_DEVICE_CAPS | V4L2_CAP_VIDEO_CAPTURE | + V4L2_CAP_VIDEO_OUTPUT | V4L2_CAP_READWRITE | + V4L2_CAP_STREAMING; +#endif + + if (debug > 1) + vdev->dev_debug = V4L2_DEV_DEBUG_IOCTL | + V4L2_DEV_DEBUG_IOCTL_ARG; + + vdev->vfl_dir = VFL_DIR_M2M; +} + +/* init default capture parameters, only fps may be changed in future */ +static void init_capture_param(struct v4l2_captureparm *capture_param) +{ + capture_param->capability = V4L2_CAP_TIMEPERFRAME; /* since 2.16 */ + capture_param->capturemode = 0; + capture_param->extendedmode = 0; + capture_param->readbuffers = max_buffers; + capture_param->timeperframe.numerator = 1; + capture_param->timeperframe.denominator = V4L2LOOPBACK_FPS_DEFAULT; +} + +static void check_timers(struct v4l2_loopback_device *dev) +{ + if (has_output_token(dev->stream_tokens)) + return; + + if (dev->timeout_jiffies > 0 && !timer_pending(&dev->timeout_timer)) + mod_timer(&dev->timeout_timer, jiffies + dev->timeout_jiffies); + if (dev->sustain_framerate && !timer_pending(&dev->sustain_timer)) + mod_timer(&dev->sustain_timer, + jiffies + dev->frame_jiffies * 3 / 2); +} +#ifdef HAVE_TIMER_SETUP +static void sustain_timer_clb(struct timer_list *t) +{ + struct v4l2_loopback_device *dev = + container_of(t, struct v4l2_loopback_device, sustain_timer); +#else +static void sustain_timer_clb(unsigned long nr) +{ + struct v4l2_loopback_device *dev = + idr_find(&v4l2loopback_index_idr, nr); +#endif + spin_lock(&dev->lock); + if (dev->sustain_framerate) { + dev->reread_count++; + dprintkrw("sustain_timer_clb() write_pos=%lld reread=%u\n", + (long long)dev->write_position, dev->reread_count); + if (dev->reread_count == 1) + mod_timer(&dev->sustain_timer, + jiffies + max(1UL, dev->frame_jiffies / 2)); + else + mod_timer(&dev->sustain_timer, + jiffies + dev->frame_jiffies); + wake_up_all(&dev->read_event); + } + spin_unlock(&dev->lock); +} +#ifdef HAVE_TIMER_SETUP +static void timeout_timer_clb(struct timer_list *t) +{ + struct v4l2_loopback_device *dev = + container_of(t, struct v4l2_loopback_device, timeout_timer); +#else +static void timeout_timer_clb(unsigned long nr) +{ + struct v4l2_loopback_device *dev = + idr_find(&v4l2loopback_index_idr, nr); +#endif + spin_lock(&dev->lock); + if (dev->timeout_jiffies > 0) { + dev->timeout_happened = 1; + mod_timer(&dev->timeout_timer, jiffies + dev->timeout_jiffies); + wake_up_all(&dev->read_event); + } + spin_unlock(&dev->lock); +} + +/* init loopback main structure */ +#define DEFAULT_FROM_CONF(confmember, default_condition, default_value) \ + ((conf) ? \ + ((conf->confmember default_condition) ? (default_value) : \ + (conf->confmember)) : \ + default_value) + +static int v4l2_loopback_add(struct v4l2_loopback_config *conf, int *ret_nr) +{ + struct v4l2_loopback_device *dev; + struct v4l2_ctrl_handler *hdl; + struct v4l2loopback_private *vdev_priv = NULL; + int err; + + u32 _width = V4L2LOOPBACK_SIZE_DEFAULT_WIDTH; + u32 _height = V4L2LOOPBACK_SIZE_DEFAULT_HEIGHT; + + u32 _min_width = DEFAULT_FROM_CONF(min_width, + < V4L2LOOPBACK_SIZE_MIN_WIDTH, + V4L2LOOPBACK_SIZE_MIN_WIDTH); + u32 _min_height = DEFAULT_FROM_CONF(min_height, + < V4L2LOOPBACK_SIZE_MIN_HEIGHT, + V4L2LOOPBACK_SIZE_MIN_HEIGHT); + u32 _max_width = DEFAULT_FROM_CONF(max_width, < _min_width, max_width); + u32 _max_height = + DEFAULT_FROM_CONF(max_height, < _min_height, max_height); + bool _announce_all_caps = (conf && conf->announce_all_caps >= 0) ? + (bool)(conf->announce_all_caps) : + !(V4L2LOOPBACK_DEFAULT_EXCLUSIVECAPS); + int _max_buffers = DEFAULT_FROM_CONF(max_buffers, <= 0, max_buffers); + int _max_openers = DEFAULT_FROM_CONF(max_openers, <= 0, max_openers); + struct v4l2_format _fmt; + + int nr = -1; + + if (conf) { + const int output_nr = conf->output_nr; +#ifdef SPLIT_DEVICES + const int capture_nr = conf->capture_nr; +#else + const int capture_nr = output_nr; +#endif + if (capture_nr >= 0 && output_nr == capture_nr) { + nr = output_nr; + } else if (capture_nr < 0 && output_nr < 0) { + nr = -1; + } else if (capture_nr < 0) { + nr = output_nr; + } else if (output_nr < 0) { + nr = capture_nr; + } else { + printk(KERN_ERR + "v4l2-loopback add() split OUTPUT and CAPTURE " + "devices not yet supported.\n"); + printk(KERN_INFO + "v4l2-loopback add() both devices must have the " + "same number (%d != %d).\n", + output_nr, capture_nr); + return -EINVAL; + } + } + + if (idr_find(&v4l2loopback_index_idr, nr)) + return -EEXIST; + + /* initialisation of a new device */ + dprintk("add() creating device #%d\n", nr); + dev = kzalloc(sizeof(*dev), GFP_KERNEL); + if (!dev) + return -ENOMEM; + + /* allocate id, if @id >= 0, we're requesting that specific id */ + if (nr >= 0) { + err = idr_alloc(&v4l2loopback_index_idr, dev, nr, nr + 1, + GFP_KERNEL); + if (err == -ENOSPC) + err = -EEXIST; + } else { + err = idr_alloc(&v4l2loopback_index_idr, dev, 0, 0, GFP_KERNEL); + } + if (err < 0) + goto out_free_dev; + + /* register new device */ + MARK(); + nr = err; + + if (conf && conf->card_label[0]) { + snprintf(dev->card_label, sizeof(dev->card_label), "%s", + conf->card_label); + } else { + snprintf(dev->card_label, sizeof(dev->card_label), + "Dummy video device (0x%04X)", nr); + } + snprintf(dev->v4l2_dev.name, sizeof(dev->v4l2_dev.name), + "v4l2loopback-%03d", nr); + + err = v4l2_device_register(NULL, &dev->v4l2_dev); + if (err) + goto out_free_idr; + + /* initialise the _video_ device */ + MARK(); + err = -ENOMEM; + dev->vdev = video_device_alloc(); + if (dev->vdev == NULL) + goto out_unregister; + + vdev_priv = kzalloc(sizeof(struct v4l2loopback_private), GFP_KERNEL); + if (vdev_priv == NULL) + goto out_unregister; + + video_set_drvdata(dev->vdev, vdev_priv); + if (video_get_drvdata(dev->vdev) == NULL) + goto out_unregister; + + snprintf(dev->vdev->name, sizeof(dev->vdev->name), "%s", + dev->card_label); + vdev_priv->device_nr = nr; + init_vdev(dev->vdev, nr); + dev->vdev->v4l2_dev = &dev->v4l2_dev; + + /* initialise v4l2-loopback specific parameters */ + MARK(); + dev->announce_all_caps = _announce_all_caps; + dev->min_width = _min_width; + dev->min_height = _min_height; + dev->max_width = _max_width; + dev->max_height = _max_height; + dev->max_openers = _max_openers; + + /* set (initial) pixel and stream format */ + _width = clamp_val(_width, _min_width, _max_width); + _height = clamp_val(_height, _min_height, _max_height); + _fmt = (struct v4l2_format){ + .type = V4L2_BUF_TYPE_VIDEO_CAPTURE, + .fmt.pix = { .width = _width, + .height = _height, + .pixelformat = formats[0].fourcc, + .colorspace = V4L2_COLORSPACE_DEFAULT, + .field = V4L2_FIELD_NONE } + }; + + err = v4l2l_fill_format(&_fmt, _min_width, _max_width, _min_height, + _max_height); + if (err) + /* highly unexpected failure to assign default format */ + goto out_unregister; + dev->pix_format = _fmt.fmt.pix; + init_capture_param(&dev->capture_param); + set_timeperframe(dev, &dev->capture_param.timeperframe); + + /* ctrls parameters */ + dev->keep_format = 0; + dev->sustain_framerate = 0; + dev->timeout_jiffies = 0; + dev->timeout_image_io = 0; + + /* initialise OUTPUT and CAPTURE buffer values */ + dev->image = NULL; + dev->image_size = 0; + dev->buffer_count = _max_buffers; + dev->buffer_size = 0; + dev->used_buffer_count = 0; + INIT_LIST_HEAD(&dev->outbufs_list); + do { + u32 index; + for (index = 0; index < dev->buffer_count; ++index) + INIT_LIST_HEAD(&dev->buffers[index].list_head); + + } while (0); + memset(dev->bufpos2index, 0, sizeof(dev->bufpos2index)); + dev->write_position = 0; + + /* initialise synchronisation data */ + atomic_set(&dev->open_count, 0); + mutex_init(&dev->image_mutex); + spin_lock_init(&dev->lock); + spin_lock_init(&dev->list_lock); + init_waitqueue_head(&dev->read_event); + dev->format_tokens = V4L2L_TOKEN_MASK; + dev->stream_tokens = V4L2L_TOKEN_MASK; + + /* initialise sustain frame rate and timeout parameters, and timers */ + dev->reread_count = 0; + dev->timeout_image = NULL; + dev->timeout_happened = 0; +#ifdef HAVE_TIMER_SETUP + timer_setup(&dev->sustain_timer, sustain_timer_clb, 0); + timer_setup(&dev->timeout_timer, timeout_timer_clb, 0); +#else + setup_timer(&dev->sustain_timer, sustain_timer_clb, nr); + setup_timer(&dev->timeout_timer, timeout_timer_clb, nr); +#endif + + /* initialise the control handler and add controls */ + MARK(); + hdl = &dev->ctrl_handler; + err = v4l2_ctrl_handler_init(hdl, 4); + if (err) + goto out_unregister; + v4l2_ctrl_new_custom(hdl, &v4l2loopback_ctrl_keepformat, NULL); + v4l2_ctrl_new_custom(hdl, &v4l2loopback_ctrl_sustainframerate, NULL); + v4l2_ctrl_new_custom(hdl, &v4l2loopback_ctrl_timeout, NULL); + v4l2_ctrl_new_custom(hdl, &v4l2loopback_ctrl_timeoutimageio, NULL); + if (hdl->error) { + err = hdl->error; + goto out_free_handler; + } + dev->v4l2_dev.ctrl_handler = hdl; + + err = v4l2_ctrl_handler_setup(hdl); + if (err) + goto out_free_handler; + + /* register the device (creates /dev/video*) */ + MARK(); + if (video_register_device(dev->vdev, VFL_TYPE_VIDEO, nr) < 0) { + printk(KERN_ERR + "v4l2-loopback add() failed video_register_device()\n"); + err = -EFAULT; + goto out_free_device; + } + v4l2loopback_create_sysfs(dev->vdev); + /* NOTE: ambivalent if sysfs entries fail */ + + if (ret_nr) + *ret_nr = dev->vdev->num; + return 0; + +out_free_device: +out_free_handler: + v4l2_ctrl_handler_free(&dev->ctrl_handler); +out_unregister: + if (dev->vdev) + video_set_drvdata(dev->vdev, NULL); + video_device_release(dev->vdev); + if (vdev_priv != NULL) + kfree(vdev_priv); + v4l2_device_unregister(&dev->v4l2_dev); +out_free_idr: + idr_remove(&v4l2loopback_index_idr, nr); +out_free_dev: + kfree(dev); + return err; +} + +static void v4l2_loopback_remove(struct v4l2_loopback_device *dev) +{ + int device_nr = v4l2loopback_get_vdev_nr(dev->vdev); + mutex_lock(&dev->image_mutex); + free_buffers(dev); + free_timeout_buffer(dev); + mutex_unlock(&dev->image_mutex); + v4l2loopback_remove_sysfs(dev->vdev); + v4l2_ctrl_handler_free(&dev->ctrl_handler); + kfree(video_get_drvdata(dev->vdev)); + video_unregister_device(dev->vdev); + v4l2_device_unregister(&dev->v4l2_dev); + idr_remove(&v4l2loopback_index_idr, device_nr); + kfree(dev); +} + +static long v4l2loopback_control_ioctl(struct file *file, unsigned int cmd, + unsigned long parm) +{ + struct v4l2_loopback_device *dev; + struct v4l2_loopback_config conf; + struct v4l2_loopback_config *confptr = &conf; + int device_nr, capture_nr, output_nr; + int ret; + const __u32 version = V4L2LOOPBACK_VERSION_CODE; + + ret = mutex_lock_killable(&v4l2loopback_ctl_mutex); + if (ret) + return ret; + + ret = -EINVAL; + switch (cmd) { + default: + ret = -ENOSYS; + break; + /* add a v4l2loopback device (pair), based on the user-provided specs */ + case V4L2LOOPBACK_CTL_ADD: + case V4L2LOOPBACK_CTL_ADD_legacy: + if (parm) { + if ((ret = copy_from_user(&conf, (void *)parm, + sizeof(conf))) < 0) + break; + } else + confptr = NULL; + ret = v4l2_loopback_add(confptr, &device_nr); + if (ret >= 0) + ret = device_nr; + break; + /* remove a v4l2loopback device (both capture and output) */ + case V4L2LOOPBACK_CTL_REMOVE: + case V4L2LOOPBACK_CTL_REMOVE_legacy: + ret = v4l2loopback_lookup((__u32)parm, &dev); + if (ret >= 0 && dev) { + ret = -EBUSY; + if (dev->open_count.counter > 0) + break; + v4l2_loopback_remove(dev); + ret = 0; + }; + break; + /* get information for a loopback device. + * this is mostly about limits (which cannot be queried directly with VIDIOC_G_FMT and friends + */ + case V4L2LOOPBACK_CTL_QUERY: + case V4L2LOOPBACK_CTL_QUERY_legacy: + if (!parm) + break; + if ((ret = copy_from_user(&conf, (void *)parm, sizeof(conf))) < + 0) + break; + capture_nr = output_nr = conf.output_nr; +#ifdef SPLIT_DEVICES + capture_nr = conf.capture_nr; +#endif + device_nr = (output_nr < 0) ? capture_nr : output_nr; + MARK(); + /* get the device from either capture_nr or output_nr (whatever is valid) */ + if ((ret = v4l2loopback_lookup(device_nr, &dev)) < 0) + break; + MARK(); + /* if we got the device from output_nr and there is a valid capture_nr, + * make sure that both refer to the same device (or bail out) + */ + if ((device_nr != capture_nr) && (capture_nr >= 0) && + ((ret = v4l2loopback_lookup(capture_nr, 0)) < 0)) + break; + MARK(); + /* if otoh, we got the device from capture_nr and there is a valid output_nr, + * make sure that both refer to the same device (or bail out) + */ + if ((device_nr != output_nr) && (output_nr >= 0) && + ((ret = v4l2loopback_lookup(output_nr, 0)) < 0)) + break; + + /* v4l2_loopback_config identified a single device, so fetch the data */ + snprintf(conf.card_label, sizeof(conf.card_label), "%s", + dev->card_label); + + conf.output_nr = dev->vdev->num; +#ifdef SPLIT_DEVICES + conf.capture_nr = dev->vdev->num; +#endif + conf.min_width = dev->min_width; + conf.min_height = dev->min_height; + conf.max_width = dev->max_width; + conf.max_height = dev->max_height; + conf.announce_all_caps = dev->announce_all_caps; + conf.max_buffers = dev->buffer_count; + conf.max_openers = dev->max_openers; + conf.debug = debug; + MARK(); + if (copy_to_user((void *)parm, &conf, sizeof(conf))) { + ret = -EFAULT; + break; + } + ret = 0; + break; + case V4L2LOOPBACK_CTL_VERSION: + if (!parm) + break; + if (copy_to_user((void *)parm, &version, sizeof(version))) { + ret = -EFAULT; + break; + } + ret = 0; + break; + } + + mutex_unlock(&v4l2loopback_ctl_mutex); + MARK(); + return ret; +} + +/* LINUX KERNEL */ + +static const struct file_operations v4l2loopback_ctl_fops = { + // clang-format off + .owner = THIS_MODULE, + .open = nonseekable_open, + .unlocked_ioctl = v4l2loopback_control_ioctl, + .compat_ioctl = v4l2loopback_control_ioctl, + .llseek = noop_llseek, + // clang-format on +}; + +static struct miscdevice v4l2loopback_misc = { + // clang-format off + .minor = MISC_DYNAMIC_MINOR, + .name = "v4l2loopback", + .fops = &v4l2loopback_ctl_fops, + // clang-format on +}; + +static const struct v4l2_file_operations v4l2_loopback_fops = { + // clang-format off + .owner = THIS_MODULE, + .open = v4l2_loopback_open, + .release = v4l2_loopback_close, + .read = v4l2_loopback_read, + .write = v4l2_loopback_write, + .poll = v4l2_loopback_poll, + .mmap = v4l2_loopback_mmap, + .unlocked_ioctl = video_ioctl2, + // clang-format on +}; + +static const struct v4l2_ioctl_ops v4l2_loopback_ioctl_ops = { + // clang-format off + .vidioc_querycap = &vidioc_querycap, + .vidioc_enum_framesizes = &vidioc_enum_framesizes, + .vidioc_enum_frameintervals = &vidioc_enum_frameintervals, + + .vidioc_enum_output = &vidioc_enum_output, + .vidioc_g_output = &vidioc_g_output, + .vidioc_s_output = &vidioc_s_output, + + .vidioc_enum_input = &vidioc_enum_input, + .vidioc_g_input = &vidioc_g_input, + .vidioc_s_input = &vidioc_s_input, + + .vidioc_enum_fmt_vid_cap = &vidioc_enum_fmt_cap, + .vidioc_g_fmt_vid_cap = &vidioc_g_fmt_cap, + .vidioc_s_fmt_vid_cap = &vidioc_s_fmt_cap, + .vidioc_try_fmt_vid_cap = &vidioc_try_fmt_cap, + + .vidioc_enum_fmt_vid_out = &vidioc_enum_fmt_out, + .vidioc_s_fmt_vid_out = &vidioc_s_fmt_out, + .vidioc_g_fmt_vid_out = &vidioc_g_fmt_out, + .vidioc_try_fmt_vid_out = &vidioc_try_fmt_out, + +#ifdef V4L2L_OVERLAY + .vidioc_s_fmt_vid_overlay = &vidioc_s_fmt_overlay, + .vidioc_g_fmt_vid_overlay = &vidioc_g_fmt_overlay, +#endif + +#ifdef V4L2LOOPBACK_WITH_STD + .vidioc_s_std = &vidioc_s_std, + .vidioc_g_std = &vidioc_g_std, + .vidioc_querystd = &vidioc_querystd, +#endif /* V4L2LOOPBACK_WITH_STD */ + + .vidioc_g_parm = &vidioc_g_parm, + .vidioc_s_parm = &vidioc_s_parm, + + .vidioc_reqbufs = &vidioc_reqbufs, + .vidioc_querybuf = &vidioc_querybuf, + .vidioc_qbuf = &vidioc_qbuf, + .vidioc_dqbuf = &vidioc_dqbuf, + + .vidioc_streamon = &vidioc_streamon, + .vidioc_streamoff = &vidioc_streamoff, + +#ifdef CONFIG_VIDEO_V4L1_COMPAT + .vidiocgmbuf = &vidiocgmbuf, +#endif + + .vidioc_subscribe_event = &vidioc_subscribe_event, + .vidioc_unsubscribe_event = &v4l2_event_unsubscribe, + // clang-format on +}; + +static int free_device_cb(int id, void *ptr, void *data) +{ + struct v4l2_loopback_device *dev = ptr; + v4l2_loopback_remove(dev); + return 0; +} +static void free_devices(void) +{ + idr_for_each(&v4l2loopback_index_idr, &free_device_cb, NULL); + idr_destroy(&v4l2loopback_index_idr); +} + +static int __init v4l2loopback_init_module(void) +{ + const u32 min_width = V4L2LOOPBACK_SIZE_MIN_WIDTH; + const u32 min_height = V4L2LOOPBACK_SIZE_MIN_HEIGHT; + int err; + int i; + MARK(); + + err = misc_register(&v4l2loopback_misc); + if (err < 0) + return err; + + if (devices < 0) { + devices = 1; + + /* try guessing the devices from the "video_nr" parameter */ + for (i = MAX_DEVICES - 1; i >= 0; i--) { + if (video_nr[i] >= 0) { + devices = i + 1; + break; + } + } + } + + if (devices > MAX_DEVICES) { + devices = MAX_DEVICES; + printk(KERN_INFO + "v4l2-loopback init() number of initial devices is " + "limited to: %d\n", + MAX_DEVICES); + } + + if (max_buffers > MAX_BUFFERS) { + max_buffers = MAX_BUFFERS; + printk(KERN_INFO + "v4l2-loopback init() number of buffers is limited " + "to: %d\n", + MAX_BUFFERS); + } + + if (max_openers < 0) { + printk(KERN_INFO + "v4l2-loopback init() allowing %d openers rather " + "than %d\n", + 2, max_openers); + max_openers = 2; + } + + if (max_width < min_width) { + max_width = V4L2LOOPBACK_SIZE_DEFAULT_MAX_WIDTH; + printk(KERN_INFO "v4l2-loopback init() using max_width %d\n", + max_width); + } + if (max_height < min_height) { + max_height = V4L2LOOPBACK_SIZE_DEFAULT_MAX_HEIGHT; + printk(KERN_INFO "v4l2-loopback init() using max_height %d\n", + max_height); + } + + for (i = 0; i < devices; i++) { + struct v4l2_loopback_config cfg = { + // clang-format off + .output_nr = video_nr[i], +#ifdef SPLIT_DEVICES + .capture_nr = video_nr[i], +#endif + .min_width = min_width, + .min_height = min_height, + .max_width = max_width, + .max_height = max_height, + .announce_all_caps = (!exclusive_caps[i]), + .max_buffers = max_buffers, + .max_openers = max_openers, + .debug = debug, + // clang-format on + }; + cfg.card_label[0] = 0; + if (card_label[i]) + snprintf(cfg.card_label, sizeof(cfg.card_label), "%s", + card_label[i]); + err = v4l2_loopback_add(&cfg, 0); + if (err) { + free_devices(); + goto error; + } + } + + dprintk("module installed\n"); + + printk(KERN_INFO "v4l2-loopback driver version %d.%d.%d%s loaded\n", + // clang-format off + (V4L2LOOPBACK_VERSION_CODE >> 16) & 0xff, + (V4L2LOOPBACK_VERSION_CODE >> 8) & 0xff, + (V4L2LOOPBACK_VERSION_CODE ) & 0xff, +#ifdef SNAPSHOT_VERSION + " (" __stringify(SNAPSHOT_VERSION) ")" +#else + "" +#endif + ); + // clang-format on + + return 0; +error: + misc_deregister(&v4l2loopback_misc); + return err; +} + +static void v4l2loopback_cleanup_module(void) +{ + MARK(); + /* unregister the device -> it deletes /dev/video* */ + free_devices(); + /* and get rid of /dev/v4l2loopback */ + misc_deregister(&v4l2loopback_misc); + dprintk("module removed\n"); +} + +MODULE_ALIAS_MISCDEV(MISC_DYNAMIC_MINOR); + +module_init(v4l2loopback_init_module); +module_exit(v4l2loopback_cleanup_module); diff --git a/drivers/media/v4l2-core/v4l2loopback.h b/drivers/media/v4l2-core/v4l2loopback.h new file mode 100644 index 000000000000..8961a98eff58 --- /dev/null +++ b/drivers/media/v4l2-core/v4l2loopback.h @@ -0,0 +1,108 @@ +/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */ +/* + * v4l2loopback.h + * + * Written by IOhannes m zmölnig, 7/1/20. + * + * Copyright 2020 by IOhannes m zmölnig. Redistribution of this file is + * permitted under the GNU General Public License. + */ +#ifndef _V4L2LOOPBACK_H +#define _V4L2LOOPBACK_H + +#define V4L2LOOPBACK_VERSION_MAJOR 0 +#define V4L2LOOPBACK_VERSION_MINOR 15 +#define V4L2LOOPBACK_VERSION_BUGFIX 3 + +/* /dev/v4l2loopback interface */ + +struct v4l2_loopback_config { + /** + * the device-number (/dev/video) + * V4L2LOOPBACK_CTL_ADD: + * setting this to a value<0, will allocate an available one + * if nr>=0 and the device already exists, the ioctl will EEXIST + * if output_nr and capture_nr are the same, only a single device will be created + * NOTE: currently split-devices (where output_nr and capture_nr differ) + * are not implemented yet. + * until then, requesting different device-IDs will result in EINVAL. + * + * V4L2LOOPBACK_CTL_QUERY: + * either both output_nr and capture_nr must refer to the same loopback, + * or one (and only one) of them must be -1 + * + */ + __s32 output_nr; + __s32 unused; /*capture_nr;*/ + + /** + * a nice name for your device + * if (*card_label)==0, an automatic name is assigned + */ + char card_label[32]; + + /** + * allowed frame size + * if too low, default values are used + */ + __u32 min_width; + __u32 max_width; + __u32 min_height; + __u32 max_height; + + /** + * number of buffers to allocate for the queue + * if set to <=0, default values are used + */ + __s32 max_buffers; + + /** + * how many consumers are allowed to open this device concurrently + * if set to <=0, default values are used + */ + __s32 max_openers; + + /** + * set the debugging level for this device + */ + __s32 debug; + + /** + * whether to announce OUTPUT/CAPTURE capabilities exclusively + * for this device or not + * (!exclusive_caps) + * NOTE: this is going to be removed once separate output/capture + * devices are implemented + */ + __s32 announce_all_caps; +}; + +#define V4L2LOOPBACK_CTL_IOCTLMAGIC '~' + +/* a pointer to an (unsigned int) that - on success - will hold + * the version code of the v4l2loopback module + * as returned by KERNEL_VERSION(MAJOR, MINOR, BUGFIX) + */ +#define V4L2LOOPBACK_CTL_VERSION _IOR(V4L2LOOPBACK_CTL_IOCTLMAGIC, 0, __u32) + +/* a pointer to a (struct v4l2_loopback_config) that has all values you wish to impose on the + * to-be-created device set. + * if the ptr is NULL, a new device is created with default values at the driver's discretion. + * + * returns the device_nr of the OUTPUT device (which can be used with V4L2LOOPBACK_CTL_QUERY, + * to get more information on the device) + */ +#define V4L2LOOPBACK_CTL_ADD \ + _IOW(V4L2LOOPBACK_CTL_IOCTLMAGIC, 1, struct v4l2_loopback_config) + +/* the device-number (either CAPTURE or OUTPUT) associated with the loopback-device */ +#define V4L2LOOPBACK_CTL_REMOVE _IOW(V4L2LOOPBACK_CTL_IOCTLMAGIC, 2, __u32) + +/* a pointer to a (struct v4l2_loopback_config) that has output_nr and/or capture_nr set + * (the two values must either refer to video-devices associated with the same loopback device + * or exactly one of them must be <0 + */ +#define V4L2LOOPBACK_CTL_QUERY \ + _IOWR(V4L2LOOPBACK_CTL_IOCTLMAGIC, 3, struct v4l2_loopback_config) + +#endif /* _V4L2LOOPBACK_H */ diff --git a/drivers/media/v4l2-core/v4l2loopback_formats.h b/drivers/media/v4l2-core/v4l2loopback_formats.h new file mode 100644 index 000000000000..d855a3796554 --- /dev/null +++ b/drivers/media/v4l2-core/v4l2loopback_formats.h @@ -0,0 +1,445 @@ +static const struct v4l2l_format formats[] = { +#ifndef V4L2_PIX_FMT_VP9 +#define V4L2_PIX_FMT_VP9 v4l2_fourcc('V', 'P', '9', '0') +#endif +#ifndef V4L2_PIX_FMT_HEVC +#define V4L2_PIX_FMT_HEVC v4l2_fourcc('H', 'E', 'V', 'C') +#endif + + /* here come the packed formats */ + { + .name = "32 bpp RGB, le", + .fourcc = V4L2_PIX_FMT_BGR32, + .depth = 32, + .flags = 0, + }, + { + .name = "32 bpp RGB, be", + .fourcc = V4L2_PIX_FMT_RGB32, + .depth = 32, + .flags = 0, + }, + { + .name = "24 bpp RGB, le", + .fourcc = V4L2_PIX_FMT_BGR24, + .depth = 24, + .flags = 0, + }, + { + .name = "24 bpp RGB, be", + .fourcc = V4L2_PIX_FMT_RGB24, + .depth = 24, + .flags = 0, + }, +#ifdef V4L2_PIX_FMT_ABGR32 + { + .name = "32 bpp RGBA, le", + .fourcc = V4L2_PIX_FMT_ABGR32, + .depth = 32, + .flags = 0, + }, +#endif +#ifdef V4L2_PIX_FMT_RGBA32 + { + .name = "32 bpp RGBA", + .fourcc = V4L2_PIX_FMT_RGBA32, + .depth = 32, + .flags = 0, + }, +#endif +#ifdef V4L2_PIX_FMT_RGB332 + { + .name = "8 bpp RGB-3-3-2", + .fourcc = V4L2_PIX_FMT_RGB332, + .depth = 8, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_RGB332 */ +#ifdef V4L2_PIX_FMT_RGB444 + { + .name = "16 bpp RGB (xxxxrrrr ggggbbbb)", + .fourcc = V4L2_PIX_FMT_RGB444, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_RGB444 */ +#ifdef V4L2_PIX_FMT_RGB555 + { + .name = "16 bpp RGB-5-5-5", + .fourcc = V4L2_PIX_FMT_RGB555, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_RGB555 */ +#ifdef V4L2_PIX_FMT_RGB565 + { + .name = "16 bpp RGB-5-6-5", + .fourcc = V4L2_PIX_FMT_RGB565, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_RGB565 */ +#ifdef V4L2_PIX_FMT_RGB555X + { + .name = "16 bpp RGB-5-5-5 BE", + .fourcc = V4L2_PIX_FMT_RGB555X, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_RGB555X */ +#ifdef V4L2_PIX_FMT_RGB565X + { + .name = "16 bpp RGB-5-6-5 BE", + .fourcc = V4L2_PIX_FMT_RGB565X, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_RGB565X */ +#ifdef V4L2_PIX_FMT_BGR666 + { + .name = "18 bpp BGR-6-6-6", + .fourcc = V4L2_PIX_FMT_BGR666, + .depth = 18, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_BGR666 */ + { + .name = "4:2:2, packed, YUYV", + .fourcc = V4L2_PIX_FMT_YUYV, + .depth = 16, + .flags = 0, + }, + { + .name = "4:2:2, packed, UYVY", + .fourcc = V4L2_PIX_FMT_UYVY, + .depth = 16, + .flags = 0, + }, +#ifdef V4L2_PIX_FMT_YVYU + { + .name = "4:2:2, packed YVYU", + .fourcc = V4L2_PIX_FMT_YVYU, + .depth = 16, + .flags = 0, + }, +#endif +#ifdef V4L2_PIX_FMT_VYUY + { + .name = "4:2:2, packed VYUY", + .fourcc = V4L2_PIX_FMT_VYUY, + .depth = 16, + .flags = 0, + }, +#endif + { + .name = "4:2:2, packed YYUV", + .fourcc = V4L2_PIX_FMT_YYUV, + .depth = 16, + .flags = 0, + }, + { + .name = "YUV-8-8-8-8", + .fourcc = V4L2_PIX_FMT_YUV32, + .depth = 32, + .flags = 0, + }, + { + .name = "8 bpp, Greyscale", + .fourcc = V4L2_PIX_FMT_GREY, + .depth = 8, + .flags = 0, + }, +#ifdef V4L2_PIX_FMT_Y4 + { + .name = "4 bpp Greyscale", + .fourcc = V4L2_PIX_FMT_Y4, + .depth = 4, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_Y4 */ +#ifdef V4L2_PIX_FMT_Y6 + { + .name = "6 bpp Greyscale", + .fourcc = V4L2_PIX_FMT_Y6, + .depth = 6, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_Y6 */ +#ifdef V4L2_PIX_FMT_Y10 + { + .name = "10 bpp Greyscale", + .fourcc = V4L2_PIX_FMT_Y10, + .depth = 10, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_Y10 */ +#ifdef V4L2_PIX_FMT_Y12 + { + .name = "12 bpp Greyscale", + .fourcc = V4L2_PIX_FMT_Y12, + .depth = 12, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_Y12 */ + { + .name = "16 bpp, Greyscale", + .fourcc = V4L2_PIX_FMT_Y16, + .depth = 16, + .flags = 0, + }, +#ifdef V4L2_PIX_FMT_YUV444 + { + .name = "16 bpp xxxxyyyy uuuuvvvv", + .fourcc = V4L2_PIX_FMT_YUV444, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_YUV444 */ +#ifdef V4L2_PIX_FMT_YUV555 + { + .name = "16 bpp YUV-5-5-5", + .fourcc = V4L2_PIX_FMT_YUV555, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_YUV555 */ +#ifdef V4L2_PIX_FMT_YUV565 + { + .name = "16 bpp YUV-5-6-5", + .fourcc = V4L2_PIX_FMT_YUV565, + .depth = 16, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_YUV565 */ + +/* bayer formats */ +#ifdef V4L2_PIX_FMT_SRGGB8 + { + .name = "Bayer RGGB 8bit", + .fourcc = V4L2_PIX_FMT_SRGGB8, + .depth = 8, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_SRGGB8 */ +#ifdef V4L2_PIX_FMT_SGRBG8 + { + .name = "Bayer GRBG 8bit", + .fourcc = V4L2_PIX_FMT_SGRBG8, + .depth = 8, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_SGRBG8 */ +#ifdef V4L2_PIX_FMT_SGBRG8 + { + .name = "Bayer GBRG 8bit", + .fourcc = V4L2_PIX_FMT_SGBRG8, + .depth = 8, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_SGBRG8 */ +#ifdef V4L2_PIX_FMT_SBGGR8 + { + .name = "Bayer BA81 8bit", + .fourcc = V4L2_PIX_FMT_SBGGR8, + .depth = 8, + .flags = 0, + }, +#endif /* V4L2_PIX_FMT_SBGGR8 */ + + /* here come the planar formats */ + { + .name = "4:1:0, planar, Y-Cr-Cb", + .fourcc = V4L2_PIX_FMT_YVU410, + .depth = 9, + .flags = FORMAT_FLAGS_PLANAR, + }, + { + .name = "4:2:0, planar, Y-Cr-Cb", + .fourcc = V4L2_PIX_FMT_YVU420, + .depth = 12, + .flags = FORMAT_FLAGS_PLANAR, + }, + { + .name = "4:1:0, planar, Y-Cb-Cr", + .fourcc = V4L2_PIX_FMT_YUV410, + .depth = 9, + .flags = FORMAT_FLAGS_PLANAR, + }, + { + .name = "4:2:0, planar, Y-Cb-Cr", + .fourcc = V4L2_PIX_FMT_YUV420, + .depth = 12, + .flags = FORMAT_FLAGS_PLANAR, + }, +#ifdef V4L2_PIX_FMT_YUV422P + { + .name = "16 bpp YVU422 planar", + .fourcc = V4L2_PIX_FMT_YUV422P, + .depth = 16, + .flags = FORMAT_FLAGS_PLANAR, + }, +#endif /* V4L2_PIX_FMT_YUV422P */ +#ifdef V4L2_PIX_FMT_YUV411P + { + .name = "16 bpp YVU411 planar", + .fourcc = V4L2_PIX_FMT_YUV411P, + .depth = 16, + .flags = FORMAT_FLAGS_PLANAR, + }, +#endif /* V4L2_PIX_FMT_YUV411P */ +#ifdef V4L2_PIX_FMT_Y41P + { + .name = "12 bpp YUV 4:1:1", + .fourcc = V4L2_PIX_FMT_Y41P, + .depth = 12, + .flags = FORMAT_FLAGS_PLANAR, + }, +#endif /* V4L2_PIX_FMT_Y41P */ +#ifdef V4L2_PIX_FMT_NV12 + { + .name = "12 bpp Y/CbCr 4:2:0 ", + .fourcc = V4L2_PIX_FMT_NV12, + .depth = 12, + .flags = FORMAT_FLAGS_PLANAR, + }, +#endif /* V4L2_PIX_FMT_NV12 */ + +/* here come the compressed formats */ + +#ifdef V4L2_PIX_FMT_MJPEG + { + .name = "Motion-JPEG", + .fourcc = V4L2_PIX_FMT_MJPEG, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_MJPEG */ +#ifdef V4L2_PIX_FMT_JPEG + { + .name = "JFIF JPEG", + .fourcc = V4L2_PIX_FMT_JPEG, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_JPEG */ +#ifdef V4L2_PIX_FMT_DV + { + .name = "DV1394", + .fourcc = V4L2_PIX_FMT_DV, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_DV */ +#ifdef V4L2_PIX_FMT_MPEG + { + .name = "MPEG-1/2/4 Multiplexed", + .fourcc = V4L2_PIX_FMT_MPEG, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_MPEG */ +#ifdef V4L2_PIX_FMT_H264 + { + .name = "H264 with start codes", + .fourcc = V4L2_PIX_FMT_H264, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_H264 */ +#ifdef V4L2_PIX_FMT_H264_NO_SC + { + .name = "H264 without start codes", + .fourcc = V4L2_PIX_FMT_H264_NO_SC, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_H264_NO_SC */ +#ifdef V4L2_PIX_FMT_H264_MVC + { + .name = "H264 MVC", + .fourcc = V4L2_PIX_FMT_H264_MVC, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_H264_MVC */ +#ifdef V4L2_PIX_FMT_H263 + { + .name = "H263", + .fourcc = V4L2_PIX_FMT_H263, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_H263 */ +#ifdef V4L2_PIX_FMT_MPEG1 + { + .name = "MPEG-1 ES", + .fourcc = V4L2_PIX_FMT_MPEG1, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_MPEG1 */ +#ifdef V4L2_PIX_FMT_MPEG2 + { + .name = "MPEG-2 ES", + .fourcc = V4L2_PIX_FMT_MPEG2, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_MPEG2 */ +#ifdef V4L2_PIX_FMT_MPEG4 + { + .name = "MPEG-4 part 2 ES", + .fourcc = V4L2_PIX_FMT_MPEG4, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_MPEG4 */ +#ifdef V4L2_PIX_FMT_XVID + { + .name = "Xvid", + .fourcc = V4L2_PIX_FMT_XVID, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_XVID */ +#ifdef V4L2_PIX_FMT_VC1_ANNEX_G + { + .name = "SMPTE 421M Annex G compliant stream", + .fourcc = V4L2_PIX_FMT_VC1_ANNEX_G, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_VC1_ANNEX_G */ +#ifdef V4L2_PIX_FMT_VC1_ANNEX_L + { + .name = "SMPTE 421M Annex L compliant stream", + .fourcc = V4L2_PIX_FMT_VC1_ANNEX_L, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_VC1_ANNEX_L */ +#ifdef V4L2_PIX_FMT_VP8 + { + .name = "VP8", + .fourcc = V4L2_PIX_FMT_VP8, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_VP8 */ +#ifdef V4L2_PIX_FMT_VP9 + { + .name = "VP9", + .fourcc = V4L2_PIX_FMT_VP9, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_VP9 */ +#ifdef V4L2_PIX_FMT_HEVC + { + .name = "HEVC", + .fourcc = V4L2_PIX_FMT_HEVC, + .depth = 32, + .flags = FORMAT_FLAGS_COMPRESSED, + }, +#endif /* V4L2_PIX_FMT_HEVC */ +}; diff --git a/drivers/pci/controller/Makefile b/drivers/pci/controller/Makefile index 229929a945c2..79811c68a278 100644 --- a/drivers/pci/controller/Makefile +++ b/drivers/pci/controller/Makefile @@ -1,4 +1,10 @@ # SPDX-License-Identifier: GPL-2.0 +ifdef CONFIG_X86_64 +ifdef CONFIG_SATA_AHCI +obj-y += intel-nvme-remap.o +endif +endif + obj-$(CONFIG_PCIE_CADENCE) += cadence/ obj-$(CONFIG_PCI_FTPCI100) += pci-ftpci100.o obj-$(CONFIG_PCI_IXP4XX) += pci-ixp4xx.o diff --git a/drivers/pci/controller/intel-nvme-remap.c b/drivers/pci/controller/intel-nvme-remap.c new file mode 100644 index 000000000000..e105e6f5cc91 --- /dev/null +++ b/drivers/pci/controller/intel-nvme-remap.c @@ -0,0 +1,462 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Intel remapped NVMe device support. + * + * Copyright (c) 2019 Endless Mobile, Inc. + * Author: Daniel Drake + * + * Some products ship by default with the SATA controller in "RAID" or + * "Intel RST Premium With Intel Optane System Acceleration" mode. Under this + * mode, which we refer to as "remapped NVMe" mode, any installed NVMe + * devices disappear from the PCI bus, and instead their I/O memory becomes + * available within the AHCI device BARs. + * + * This scheme is understood to be a way of avoiding usage of the standard + * Windows NVMe driver under that OS, instead mandating usage of Intel's + * driver instead, which has better power management, and presumably offers + * some RAID/disk-caching solutions too. + * + * Here in this driver, we support the remapped NVMe mode by claiming the + * AHCI device and creating a fake PCIe root port. On the new bus, the + * original AHCI device is exposed with only minor tweaks. Then, fake PCI + * devices corresponding to the remapped NVMe devices are created. The usual + * ahci and nvme drivers are then expected to bind to these devices and + * operate as normal. + * + * The PCI configuration space for the NVMe devices is completely + * unavailable, so we fake a minimal one and hope for the best. + * + * Interrupts are shared between the AHCI and NVMe devices. For simplicity, + * we only support the legacy interrupt here, although MSI support + * could potentially be added later. + */ + +#define MODULE_NAME "intel-nvme-remap" + +#include +#include +#include +#include +#include + +#define AHCI_PCI_BAR_STANDARD 5 + +struct nvme_remap_dev { + struct pci_dev *dev; /* AHCI device */ + struct pci_bus *bus; /* our fake PCI bus */ + struct pci_sysdata sysdata; + int irq_base; /* our fake interrupts */ + + /* + * When we detect an all-ones write to a BAR register, this flag + * is set, so that we return the BAR size on the next read (a + * standard PCI behaviour). + * This includes the assumption that an all-ones BAR write is + * immediately followed by a read of the same register. + */ + bool bar_sizing; + + /* + * Resources copied from the AHCI device, to be regarded as + * resources on our fake bus. + */ + struct resource ahci_resources[PCI_NUM_RESOURCES]; + + /* Resources corresponding to the NVMe devices. */ + struct resource remapped_dev_mem[AHCI_MAX_REMAP]; + + /* Number of remapped NVMe devices found. */ + int num_remapped_devices; +}; + +static inline struct nvme_remap_dev *nrdev_from_bus(struct pci_bus *bus) +{ + return container_of(bus->sysdata, struct nvme_remap_dev, sysdata); +} + + +/******** PCI configuration space **********/ + +/* + * Helper macros for tweaking returned contents of PCI configuration space. + * + * value contains len bytes of data read from reg. + * If fixup_reg is included in that range, fix up the contents of that + * register to fixed_value. + */ +#define NR_FIX8(fixup_reg, fixed_value) do { \ + if (reg <= fixup_reg && fixup_reg < reg + len) \ + ((u8 *) value)[fixup_reg - reg] = (u8) (fixed_value); \ + } while (0) + +#define NR_FIX16(fixup_reg, fixed_value) do { \ + NR_FIX8(fixup_reg, fixed_value); \ + NR_FIX8(fixup_reg + 1, fixed_value >> 8); \ + } while (0) + +#define NR_FIX24(fixup_reg, fixed_value) do { \ + NR_FIX8(fixup_reg, fixed_value); \ + NR_FIX8(fixup_reg + 1, fixed_value >> 8); \ + NR_FIX8(fixup_reg + 2, fixed_value >> 16); \ + } while (0) + +#define NR_FIX32(fixup_reg, fixed_value) do { \ + NR_FIX16(fixup_reg, (u16) fixed_value); \ + NR_FIX16(fixup_reg + 2, fixed_value >> 16); \ + } while (0) + +/* + * Read PCI config space of the slot 0 (AHCI) device. + * We pass through the read request to the underlying device, but + * tweak the results in some cases. + */ +static int nvme_remap_pci_read_slot0(struct pci_bus *bus, int reg, + int len, u32 *value) +{ + struct nvme_remap_dev *nrdev = nrdev_from_bus(bus); + struct pci_bus *ahci_dev_bus = nrdev->dev->bus; + int ret; + + ret = ahci_dev_bus->ops->read(ahci_dev_bus, nrdev->dev->devfn, + reg, len, value); + if (ret) + return ret; + + /* + * Adjust the device class, to prevent this driver from attempting to + * additionally probe the device we're simulating here. + */ + NR_FIX24(PCI_CLASS_PROG, PCI_CLASS_STORAGE_SATA_AHCI); + + /* + * Unset interrupt pin, otherwise ACPI tries to find routing + * info for our virtual IRQ, fails, and complains. + */ + NR_FIX8(PCI_INTERRUPT_PIN, 0); + + /* + * Truncate the AHCI BAR to not include the region that covers the + * hidden devices. This will cause the ahci driver to successfully + * probe th new device (instead of handing it over to this driver). + */ + if (nrdev->bar_sizing) { + NR_FIX32(PCI_BASE_ADDRESS_5, ~(SZ_16K - 1)); + nrdev->bar_sizing = false; + } + + return PCIBIOS_SUCCESSFUL; +} + +/* + * Read PCI config space of a remapped device. + * Since the original PCI config space is inaccessible, we provide a minimal, + * fake config space instead. + */ +static int nvme_remap_pci_read_remapped(struct pci_bus *bus, unsigned int port, + int reg, int len, u32 *value) +{ + struct nvme_remap_dev *nrdev = nrdev_from_bus(bus); + struct resource *remapped_mem; + + if (port > nrdev->num_remapped_devices) + return PCIBIOS_DEVICE_NOT_FOUND; + + *value = 0; + remapped_mem = &nrdev->remapped_dev_mem[port - 1]; + + /* Set a Vendor ID, otherwise Linux assumes no device is present */ + NR_FIX16(PCI_VENDOR_ID, PCI_VENDOR_ID_INTEL); + + /* Always appear on & bus mastering */ + NR_FIX16(PCI_COMMAND, PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER); + + /* Set class so that nvme driver probes us */ + NR_FIX24(PCI_CLASS_PROG, PCI_CLASS_STORAGE_EXPRESS); + + if (nrdev->bar_sizing) { + NR_FIX32(PCI_BASE_ADDRESS_0, + ~(resource_size(remapped_mem) - 1)); + nrdev->bar_sizing = false; + } else { + resource_size_t mem_start = remapped_mem->start; + + mem_start |= PCI_BASE_ADDRESS_MEM_TYPE_64; + NR_FIX32(PCI_BASE_ADDRESS_0, mem_start); + mem_start >>= 32; + NR_FIX32(PCI_BASE_ADDRESS_1, mem_start); + } + + return PCIBIOS_SUCCESSFUL; +} + +/* Read PCI configuration space. */ +static int nvme_remap_pci_read(struct pci_bus *bus, unsigned int devfn, + int reg, int len, u32 *value) +{ + if (PCI_SLOT(devfn) == 0) + return nvme_remap_pci_read_slot0(bus, reg, len, value); + else + return nvme_remap_pci_read_remapped(bus, PCI_SLOT(devfn), + reg, len, value); +} + +/* + * Write PCI config space of the slot 0 (AHCI) device. + * Apart from the special case of BAR sizing, we disable all writes. + * Otherwise, the ahci driver could make changes (e.g. unset PCI bus master) + * that would affect the operation of the NVMe devices. + */ +static int nvme_remap_pci_write_slot0(struct pci_bus *bus, int reg, + int len, u32 value) +{ + struct nvme_remap_dev *nrdev = nrdev_from_bus(bus); + struct pci_bus *ahci_dev_bus = nrdev->dev->bus; + + if (reg >= PCI_BASE_ADDRESS_0 && reg <= PCI_BASE_ADDRESS_5) { + /* + * Writing all-ones to a BAR means that the size of the + * memory region is being checked. Flag this so that we can + * reply with an appropriate size on the next read. + */ + if (value == ~0) + nrdev->bar_sizing = true; + + return ahci_dev_bus->ops->write(ahci_dev_bus, + nrdev->dev->devfn, + reg, len, value); + } + + return PCIBIOS_SET_FAILED; +} + +/* + * Write PCI config space of a remapped device. + * Since the original PCI config space is inaccessible, we reject all + * writes, except for the special case of BAR probing. + */ +static int nvme_remap_pci_write_remapped(struct pci_bus *bus, + unsigned int port, + int reg, int len, u32 value) +{ + struct nvme_remap_dev *nrdev = nrdev_from_bus(bus); + + if (port > nrdev->num_remapped_devices) + return PCIBIOS_DEVICE_NOT_FOUND; + + /* + * Writing all-ones to a BAR means that the size of the memory + * region is being checked. Flag this so that we can reply with + * an appropriate size on the next read. + */ + if (value == ~0 && reg >= PCI_BASE_ADDRESS_0 + && reg <= PCI_BASE_ADDRESS_5) { + nrdev->bar_sizing = true; + return PCIBIOS_SUCCESSFUL; + } + + return PCIBIOS_SET_FAILED; +} + +/* Write PCI configuration space. */ +static int nvme_remap_pci_write(struct pci_bus *bus, unsigned int devfn, + int reg, int len, u32 value) +{ + if (PCI_SLOT(devfn) == 0) + return nvme_remap_pci_write_slot0(bus, reg, len, value); + else + return nvme_remap_pci_write_remapped(bus, PCI_SLOT(devfn), + reg, len, value); +} + +static struct pci_ops nvme_remap_pci_ops = { + .read = nvme_remap_pci_read, + .write = nvme_remap_pci_write, +}; + + +/******** Initialization & exit **********/ + +/* + * Find a PCI domain ID to use for our fake bus. + * Start at 0x10000 to not clash with ACPI _SEG domains (16 bits). + */ +static int find_free_domain(void) +{ + int domain = 0xffff; + struct pci_bus *bus = NULL; + + while ((bus = pci_find_next_bus(bus)) != NULL) + domain = max_t(int, domain, pci_domain_nr(bus)); + + return domain + 1; +} + +static int find_remapped_devices(struct nvme_remap_dev *nrdev, + struct list_head *resources) +{ + void __iomem *mmio; + int i, count = 0; + u32 cap; + + mmio = pcim_iomap(nrdev->dev, AHCI_PCI_BAR_STANDARD, + pci_resource_len(nrdev->dev, + AHCI_PCI_BAR_STANDARD)); + if (!mmio) + return -ENODEV; + + /* Check if this device might have remapped nvme devices. */ + if (pci_resource_len(nrdev->dev, AHCI_PCI_BAR_STANDARD) < SZ_512K || + !(readl(mmio + AHCI_VSCAP) & 1)) + return -ENODEV; + + cap = readq(mmio + AHCI_REMAP_CAP); + for (i = AHCI_MAX_REMAP-1; i >= 0; i--) { + struct resource *remapped_mem; + + if ((cap & (1 << i)) == 0) + continue; + if (readl(mmio + ahci_remap_dcc(i)) + != PCI_CLASS_STORAGE_EXPRESS) + continue; + + /* We've found a remapped device */ + remapped_mem = &nrdev->remapped_dev_mem[count++]; + remapped_mem->start = + pci_resource_start(nrdev->dev, AHCI_PCI_BAR_STANDARD) + + ahci_remap_base(i); + remapped_mem->end = remapped_mem->start + + AHCI_REMAP_N_SIZE - 1; + remapped_mem->flags = IORESOURCE_MEM | IORESOURCE_PCI_FIXED; + pci_add_resource(resources, remapped_mem); + } + + pcim_iounmap(nrdev->dev, mmio); + + if (count == 0) + return -ENODEV; + + nrdev->num_remapped_devices = count; + dev_info(&nrdev->dev->dev, "Found %d remapped NVMe devices\n", + nrdev->num_remapped_devices); + return 0; +} + +static void nvme_remap_remove_root_bus(void *data) +{ + struct pci_bus *bus = data; + + pci_stop_root_bus(bus); + pci_remove_root_bus(bus); +} + +static int nvme_remap_probe(struct pci_dev *dev, + const struct pci_device_id *id) +{ + struct nvme_remap_dev *nrdev; + LIST_HEAD(resources); + int i; + int ret; + struct pci_dev *child; + + nrdev = devm_kzalloc(&dev->dev, sizeof(*nrdev), GFP_KERNEL); + nrdev->sysdata.domain = find_free_domain(); + nrdev->sysdata.nvme_remap_dev = dev; + nrdev->dev = dev; + pci_set_drvdata(dev, nrdev); + + ret = pcim_enable_device(dev); + if (ret < 0) + return ret; + + pci_set_master(dev); + + ret = find_remapped_devices(nrdev, &resources); + if (ret) + return ret; + + /* Add resources from the original AHCI device */ + for (i = 0; i < PCI_NUM_RESOURCES; i++) { + struct resource *res = &dev->resource[i]; + + if (res->start) { + struct resource *nr_res = &nrdev->ahci_resources[i]; + + nr_res->start = res->start; + nr_res->end = res->end; + nr_res->flags = res->flags; + pci_add_resource(&resources, nr_res); + } + } + + /* Create virtual interrupts */ + nrdev->irq_base = devm_irq_alloc_descs(&dev->dev, -1, 0, + nrdev->num_remapped_devices + 1, + 0); + if (nrdev->irq_base < 0) + return nrdev->irq_base; + + /* Create and populate PCI bus */ + nrdev->bus = pci_create_root_bus(&dev->dev, 0, &nvme_remap_pci_ops, + &nrdev->sysdata, &resources); + if (!nrdev->bus) + return -ENODEV; + + if (devm_add_action_or_reset(&dev->dev, nvme_remap_remove_root_bus, + nrdev->bus)) + return -ENOMEM; + + /* We don't support sharing MSI interrupts between these devices */ + nrdev->bus->bus_flags |= PCI_BUS_FLAGS_NO_MSI; + + pci_scan_child_bus(nrdev->bus); + + list_for_each_entry(child, &nrdev->bus->devices, bus_list) { + /* + * Prevent PCI core from trying to move memory BARs around. + * The hidden NVMe devices are at fixed locations. + */ + for (i = 0; i < PCI_NUM_RESOURCES; i++) { + struct resource *res = &child->resource[i]; + + if (res->flags & IORESOURCE_MEM) + res->flags |= IORESOURCE_PCI_FIXED; + } + + /* Share the legacy IRQ between all devices */ + child->irq = dev->irq; + } + + pci_assign_unassigned_bus_resources(nrdev->bus); + pci_bus_add_devices(nrdev->bus); + + return 0; +} + +static const struct pci_device_id nvme_remap_ids[] = { + /* + * Match all Intel RAID controllers. + * + * There's overlap here with the set of devices detected by the ahci + * driver, but ahci will only successfully probe when there + * *aren't* any remapped NVMe devices, and this driver will only + * successfully probe when there *are* remapped NVMe devices that + * need handling. + */ + { + PCI_VDEVICE(INTEL, PCI_ANY_ID), + .class = PCI_CLASS_STORAGE_RAID << 8, + .class_mask = 0xffffff00, + }, + {0,} +}; +MODULE_DEVICE_TABLE(pci, nvme_remap_ids); + +static struct pci_driver nvme_remap_drv = { + .name = MODULE_NAME, + .id_table = nvme_remap_ids, + .probe = nvme_remap_probe, +}; +module_pci_driver(nvme_remap_drv); + +MODULE_AUTHOR("Daniel Drake "); +MODULE_LICENSE("GPL v2"); diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c index 280cd50d693b..465888670244 100644 --- a/drivers/pci/quirks.c +++ b/drivers/pci/quirks.c @@ -3748,6 +3748,106 @@ static void quirk_no_bus_reset(struct pci_dev *dev) dev->dev_flags |= PCI_DEV_FLAGS_NO_BUS_RESET; } +static bool acs_on_downstream; +static bool acs_on_multifunction; + +#define NUM_ACS_IDS 16 +struct acs_on_id { + unsigned short vendor; + unsigned short device; +}; +static struct acs_on_id acs_on_ids[NUM_ACS_IDS]; +static u8 max_acs_id; + +static __init int pcie_acs_override_setup(char *p) +{ + if (!p) + return -EINVAL; + + while (*p) { + if (!strncmp(p, "downstream", 10)) + acs_on_downstream = true; + if (!strncmp(p, "multifunction", 13)) + acs_on_multifunction = true; + if (!strncmp(p, "id:", 3)) { + char opt[5]; + int ret; + long val; + + if (max_acs_id >= NUM_ACS_IDS - 1) { + pr_warn("Out of PCIe ACS override slots (%d)\n", + NUM_ACS_IDS); + goto next; + } + + p += 3; + snprintf(opt, 5, "%s", p); + ret = kstrtol(opt, 16, &val); + if (ret) { + pr_warn("PCIe ACS ID parse error %d\n", ret); + goto next; + } + acs_on_ids[max_acs_id].vendor = val; + + p += strcspn(p, ":"); + if (*p != ':') { + pr_warn("PCIe ACS invalid ID\n"); + goto next; + } + + p++; + snprintf(opt, 5, "%s", p); + ret = kstrtol(opt, 16, &val); + if (ret) { + pr_warn("PCIe ACS ID parse error %d\n", ret); + goto next; + } + acs_on_ids[max_acs_id].device = val; + max_acs_id++; + } +next: + p += strcspn(p, ","); + if (*p == ',') + p++; + } + + if (acs_on_downstream || acs_on_multifunction || max_acs_id) + pr_warn("Warning: PCIe ACS overrides enabled; This may allow non-IOMMU protected peer-to-peer DMA\n"); + + return 0; +} +early_param("pcie_acs_override", pcie_acs_override_setup); + +static int pcie_acs_overrides(struct pci_dev *dev, u16 acs_flags) +{ + int i; + + /* Never override ACS for legacy devices or devices with ACS caps */ + if (!pci_is_pcie(dev) || + pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS)) + return -ENOTTY; + + for (i = 0; i < max_acs_id; i++) + if (acs_on_ids[i].vendor == dev->vendor && + acs_on_ids[i].device == dev->device) + return 1; + + switch (pci_pcie_type(dev)) { + case PCI_EXP_TYPE_DOWNSTREAM: + case PCI_EXP_TYPE_ROOT_PORT: + if (acs_on_downstream) + return 1; + break; + case PCI_EXP_TYPE_ENDPOINT: + case PCI_EXP_TYPE_UPSTREAM: + case PCI_EXP_TYPE_LEG_END: + case PCI_EXP_TYPE_RC_END: + if (acs_on_multifunction && dev->multifunction) + return 1; + } + + return -ENOTTY; +} /* * Some NVIDIA GPU devices do not work with bus reset, SBR needs to be * prevented for those affected devices. @@ -5195,6 +5295,7 @@ static const struct pci_dev_acs_enabled { { PCI_VENDOR_ID_ZHAOXIN, PCI_ANY_ID, pci_quirk_zhaoxin_pcie_ports_acs }, /* Wangxun nics */ { PCI_VENDOR_ID_WANGXUN, PCI_ANY_ID, pci_quirk_wangxun_nic_acs }, + { PCI_ANY_ID, PCI_ANY_ID, pcie_acs_overrides }, { 0 } }; diff --git a/drivers/scsi/Kconfig b/drivers/scsi/Kconfig index 19d0884479a2..b37d8fcc1205 100644 --- a/drivers/scsi/Kconfig +++ b/drivers/scsi/Kconfig @@ -1524,4 +1524,6 @@ endif # SCSI_LOWLEVEL source "drivers/scsi/device_handler/Kconfig" +source "drivers/scsi/vhba/Kconfig" + endmenu diff --git a/drivers/scsi/Makefile b/drivers/scsi/Makefile index 16de3e41f94c..4e88f6e3e67b 100644 --- a/drivers/scsi/Makefile +++ b/drivers/scsi/Makefile @@ -152,6 +152,7 @@ obj-$(CONFIG_CHR_DEV_SCH) += ch.o obj-$(CONFIG_SCSI_ENCLOSURE) += ses.o obj-$(CONFIG_SCSI_HISI_SAS) += hisi_sas/ +obj-$(CONFIG_VHBA) += vhba/ # This goes last, so that "real" scsi devices probe earlier obj-$(CONFIG_SCSI_DEBUG) += scsi_debug.o diff --git a/drivers/scsi/vhba/Kconfig b/drivers/scsi/vhba/Kconfig new file mode 100644 index 000000000000..e70a381fe3df --- /dev/null +++ b/drivers/scsi/vhba/Kconfig @@ -0,0 +1,9 @@ +config VHBA + tristate "Virtual (SCSI) Host Bus Adapter" + depends on SCSI + help + This is the in-kernel part of CDEmu, a CD/DVD-ROM device + emulator. + + This driver can also be built as a module. If so, the module + will be called vhba. diff --git a/drivers/scsi/vhba/Makefile b/drivers/scsi/vhba/Makefile new file mode 100644 index 000000000000..560d24689b75 --- /dev/null +++ b/drivers/scsi/vhba/Makefile @@ -0,0 +1,4 @@ +VHBA_VERSION := 20250329 + +obj-$(CONFIG_VHBA) += vhba.o +ccflags-y := -DVHBA_VERSION=\"$(VHBA_VERSION)\" -Werror diff --git a/drivers/scsi/vhba/vhba.c b/drivers/scsi/vhba/vhba.c new file mode 100644 index 000000000000..64b09ece2e5a --- /dev/null +++ b/drivers/scsi/vhba/vhba.c @@ -0,0 +1,1132 @@ +/* + * vhba.c + * + * Copyright (C) 2007-2012 Chia-I Wu + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + */ + +#define pr_fmt(fmt) "vhba: " fmt + +#include + +#include +#include +#include +#include +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0) +#include +#else +#include +#endif +#include +#include +#include +#include +#include +#ifdef CONFIG_COMPAT +#include +#endif +#include +#include +#include +#include +#include +#include + + +MODULE_AUTHOR("Chia-I Wu"); +MODULE_VERSION(VHBA_VERSION); +MODULE_DESCRIPTION("Virtual SCSI HBA"); +MODULE_LICENSE("GPL"); + + +#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 15, 0) +#define sdev_dbg(sdev, fmt, a...) \ + dev_dbg(&(sdev)->sdev_gendev, fmt, ##a) +#define scmd_dbg(scmd, fmt, a...) \ + dev_dbg(&(scmd)->device->sdev_gendev, fmt, ##a) +#endif + +#define VHBA_MAX_SECTORS_PER_IO 256 +#define VHBA_MAX_BUS 16 +#define VHBA_MAX_ID 16 +#define VHBA_MAX_DEVICES (VHBA_MAX_BUS * (VHBA_MAX_ID-1)) +#define VHBA_KBUF_SIZE PAGE_SIZE + +#define DATA_TO_DEVICE(dir) ((dir) == DMA_TO_DEVICE || (dir) == DMA_BIDIRECTIONAL) +#define DATA_FROM_DEVICE(dir) ((dir) == DMA_FROM_DEVICE || (dir) == DMA_BIDIRECTIONAL) + + +static int vhba_can_queue = 32; +module_param_named(can_queue, vhba_can_queue, int, 0); + + +enum vhba_req_state { + VHBA_REQ_FREE, + VHBA_REQ_PENDING, + VHBA_REQ_READING, + VHBA_REQ_SENT, + VHBA_REQ_WRITING, +}; + +struct vhba_command { + struct scsi_cmnd *cmd; + /* metatags are per-host. not to be confused with + queue tags that are usually per-lun */ + unsigned long metatag; + int status; + struct list_head entry; +}; + +struct vhba_device { + unsigned int num; + spinlock_t cmd_lock; + struct list_head cmd_list; + wait_queue_head_t cmd_wq; + atomic_t refcnt; + + unsigned char *kbuf; + size_t kbuf_size; +}; + +struct vhba_host { + struct Scsi_Host *shost; + spinlock_t cmd_lock; + int cmd_next; + struct vhba_command *commands; + spinlock_t dev_lock; + struct vhba_device *devices[VHBA_MAX_DEVICES]; + int num_devices; + DECLARE_BITMAP(chgmap, VHBA_MAX_DEVICES); + int chgtype[VHBA_MAX_DEVICES]; + struct work_struct scan_devices; +}; + +#define MAX_COMMAND_SIZE 16 + +struct vhba_request { + __u32 metatag; + __u32 lun; + __u8 cdb[MAX_COMMAND_SIZE]; + __u8 cdb_len; + __u32 data_len; +}; + +struct vhba_response { + __u32 metatag; + __u32 status; + __u32 data_len; +}; + + + +static struct vhba_command *vhba_alloc_command (void); +static void vhba_free_command (struct vhba_command *vcmd); + +static struct platform_device vhba_platform_device; + + + +/* These functions define a symmetric 1:1 mapping between device numbers and + the bus and id. We have reserved the last id per bus for the host itself. */ +static void devnum_to_bus_and_id(unsigned int devnum, unsigned int *bus, unsigned int *id) +{ + *bus = devnum / (VHBA_MAX_ID-1); + *id = devnum % (VHBA_MAX_ID-1); +} + +static unsigned int bus_and_id_to_devnum(unsigned int bus, unsigned int id) +{ + return (bus * (VHBA_MAX_ID-1)) + id; +} + +static struct vhba_device *vhba_device_alloc (void) +{ + struct vhba_device *vdev; + + vdev = kzalloc(sizeof(struct vhba_device), GFP_KERNEL); + if (!vdev) { + return NULL; + } + + spin_lock_init(&vdev->cmd_lock); + INIT_LIST_HEAD(&vdev->cmd_list); + init_waitqueue_head(&vdev->cmd_wq); + atomic_set(&vdev->refcnt, 1); + + vdev->kbuf = NULL; + vdev->kbuf_size = 0; + + return vdev; +} + +static void vhba_device_put (struct vhba_device *vdev) +{ + if (atomic_dec_and_test(&vdev->refcnt)) { + kfree(vdev); + } +} + +static struct vhba_device *vhba_device_get (struct vhba_device *vdev) +{ + atomic_inc(&vdev->refcnt); + + return vdev; +} + +static int vhba_device_queue (struct vhba_device *vdev, struct scsi_cmnd *cmd) +{ + struct vhba_host *vhost; + struct vhba_command *vcmd; + unsigned long flags; + + vhost = platform_get_drvdata(&vhba_platform_device); + + vcmd = vhba_alloc_command(); + if (!vcmd) { + return SCSI_MLQUEUE_HOST_BUSY; + } + + vcmd->cmd = cmd; + + spin_lock_irqsave(&vdev->cmd_lock, flags); +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 15, 0) + vcmd->metatag = scsi_cmd_to_rq(vcmd->cmd)->tag; +#else + vcmd->metatag = vcmd->cmd->request->tag; +#endif + list_add_tail(&vcmd->entry, &vdev->cmd_list); + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + wake_up_interruptible(&vdev->cmd_wq); + + return 0; +} + +static int vhba_device_dequeue (struct vhba_device *vdev, struct scsi_cmnd *cmd) +{ + struct vhba_command *vcmd; + int retval; + unsigned long flags; + + spin_lock_irqsave(&vdev->cmd_lock, flags); + list_for_each_entry(vcmd, &vdev->cmd_list, entry) { + if (vcmd->cmd == cmd) { + list_del_init(&vcmd->entry); + break; + } + } + + /* command not found */ + if (&vcmd->entry == &vdev->cmd_list) { + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + return SUCCESS; + } + + while (vcmd->status == VHBA_REQ_READING || vcmd->status == VHBA_REQ_WRITING) { + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + scmd_dbg(cmd, "wait for I/O before aborting\n"); + schedule_timeout(1); + spin_lock_irqsave(&vdev->cmd_lock, flags); + } + + retval = (vcmd->status == VHBA_REQ_SENT) ? FAILED : SUCCESS; + + vhba_free_command(vcmd); + + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + return retval; +} + +#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 19, 0) +static int vhba_slave_alloc(struct scsi_device *sdev) +{ + struct Scsi_Host *shost = sdev->host; + + sdev_dbg(sdev, "enabling tagging (queue depth: %i).\n", sdev->queue_depth); +#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 17, 0) + if (!shost_use_blk_mq(shost) && shost->bqt) { +#else + if (shost->bqt) { +#endif + blk_queue_init_tags(sdev->request_queue, sdev->queue_depth, shost->bqt); + } + scsi_adjust_queue_depth(sdev, 0, sdev->queue_depth); + + return 0; +} +#endif + +static void vhba_scan_devices_add (struct vhba_host *vhost, int bus, int id) +{ + struct scsi_device *sdev; + + sdev = scsi_device_lookup(vhost->shost, bus, id, 0); + if (!sdev) { + scsi_add_device(vhost->shost, bus, id, 0); + } else { + dev_warn(&vhost->shost->shost_gendev, "tried to add an already-existing device %d:%d:0!\n", bus, id); + scsi_device_put(sdev); + } +} + +static void vhba_scan_devices_remove (struct vhba_host *vhost, int bus, int id) +{ + struct scsi_device *sdev; + + sdev = scsi_device_lookup(vhost->shost, bus, id, 0); + if (sdev) { + scsi_remove_device(sdev); + scsi_device_put(sdev); + } else { + dev_warn(&vhost->shost->shost_gendev, "tried to remove non-existing device %d:%d:0!\n", bus, id); + } +} + +static void vhba_scan_devices (struct work_struct *work) +{ + struct vhba_host *vhost = container_of(work, struct vhba_host, scan_devices); + unsigned long flags; + int change, exists; + unsigned int devnum; + unsigned int bus, id; + + for (;;) { + spin_lock_irqsave(&vhost->dev_lock, flags); + + devnum = find_first_bit(vhost->chgmap, VHBA_MAX_DEVICES); + if (devnum >= VHBA_MAX_DEVICES) { + spin_unlock_irqrestore(&vhost->dev_lock, flags); + break; + } + change = vhost->chgtype[devnum]; + exists = vhost->devices[devnum] != NULL; + + vhost->chgtype[devnum] = 0; + clear_bit(devnum, vhost->chgmap); + + spin_unlock_irqrestore(&vhost->dev_lock, flags); + + devnum_to_bus_and_id(devnum, &bus, &id); + + if (change < 0) { + dev_dbg(&vhost->shost->shost_gendev, "trying to remove target %d:%d:0\n", bus, id); + vhba_scan_devices_remove(vhost, bus, id); + } else if (change > 0) { + dev_dbg(&vhost->shost->shost_gendev, "trying to add target %d:%d:0\n", bus, id); + vhba_scan_devices_add(vhost, bus, id); + } else { + /* quick sequence of add/remove or remove/add; we determine + which one it was by checking if device structure exists */ + if (exists) { + /* remove followed by add: remove and (re)add */ + dev_dbg(&vhost->shost->shost_gendev, "trying to (re)add target %d:%d:0\n", bus, id); + vhba_scan_devices_remove(vhost, bus, id); + vhba_scan_devices_add(vhost, bus, id); + } else { + /* add followed by remove: no-op */ + dev_dbg(&vhost->shost->shost_gendev, "no-op for target %d:%d:0\n", bus, id); + } + } + } +} + +static int vhba_add_device (struct vhba_device *vdev) +{ + struct vhba_host *vhost; + unsigned int devnum; + unsigned long flags; + + vhost = platform_get_drvdata(&vhba_platform_device); + + vhba_device_get(vdev); + + spin_lock_irqsave(&vhost->dev_lock, flags); + if (vhost->num_devices >= VHBA_MAX_DEVICES) { + spin_unlock_irqrestore(&vhost->dev_lock, flags); + vhba_device_put(vdev); + return -EBUSY; + } + + for (devnum = 0; devnum < VHBA_MAX_DEVICES; devnum++) { + if (vhost->devices[devnum] == NULL) { + vdev->num = devnum; + vhost->devices[devnum] = vdev; + vhost->num_devices++; + set_bit(devnum, vhost->chgmap); + vhost->chgtype[devnum]++; + break; + } + } + spin_unlock_irqrestore(&vhost->dev_lock, flags); + + schedule_work(&vhost->scan_devices); + + return 0; +} + +static int vhba_remove_device (struct vhba_device *vdev) +{ + struct vhba_host *vhost; + unsigned long flags; + + vhost = platform_get_drvdata(&vhba_platform_device); + + spin_lock_irqsave(&vhost->dev_lock, flags); + set_bit(vdev->num, vhost->chgmap); + vhost->chgtype[vdev->num]--; + vhost->devices[vdev->num] = NULL; + vhost->num_devices--; + spin_unlock_irqrestore(&vhost->dev_lock, flags); + + vhba_device_put(vdev); + + schedule_work(&vhost->scan_devices); + + return 0; +} + +static struct vhba_device *vhba_lookup_device (int devnum) +{ + struct vhba_host *vhost; + struct vhba_device *vdev = NULL; + unsigned long flags; + + vhost = platform_get_drvdata(&vhba_platform_device); + + if (likely(devnum < VHBA_MAX_DEVICES)) { + spin_lock_irqsave(&vhost->dev_lock, flags); + vdev = vhost->devices[devnum]; + if (vdev) { + vdev = vhba_device_get(vdev); + } + + spin_unlock_irqrestore(&vhost->dev_lock, flags); + } + + return vdev; +} + +static struct vhba_command *vhba_alloc_command (void) +{ + struct vhba_host *vhost; + struct vhba_command *vcmd; + unsigned long flags; + int i; + + vhost = platform_get_drvdata(&vhba_platform_device); + + spin_lock_irqsave(&vhost->cmd_lock, flags); + + vcmd = vhost->commands + vhost->cmd_next++; + if (vcmd->status != VHBA_REQ_FREE) { + for (i = 0; i < vhba_can_queue; i++) { + vcmd = vhost->commands + i; + + if (vcmd->status == VHBA_REQ_FREE) { + vhost->cmd_next = i + 1; + break; + } + } + + if (i == vhba_can_queue) { + vcmd = NULL; + } + } + + if (vcmd) { + vcmd->status = VHBA_REQ_PENDING; + } + + vhost->cmd_next %= vhba_can_queue; + + spin_unlock_irqrestore(&vhost->cmd_lock, flags); + + return vcmd; +} + +static void vhba_free_command (struct vhba_command *vcmd) +{ + struct vhba_host *vhost; + unsigned long flags; + + vhost = platform_get_drvdata(&vhba_platform_device); + + spin_lock_irqsave(&vhost->cmd_lock, flags); + vcmd->status = VHBA_REQ_FREE; + spin_unlock_irqrestore(&vhost->cmd_lock, flags); +} + +static int vhba_queuecommand (struct Scsi_Host *shost, struct scsi_cmnd *cmd) +{ + struct vhba_device *vdev; + int retval; + unsigned int devnum; + +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 15, 0) + scmd_dbg(cmd, "queue %p tag %i\n", cmd, scsi_cmd_to_rq(cmd)->tag); +#else + scmd_dbg(cmd, "queue %p tag %i\n", cmd, cmd->request->tag); +#endif + + devnum = bus_and_id_to_devnum(cmd->device->channel, cmd->device->id); + vdev = vhba_lookup_device(devnum); + if (!vdev) { + scmd_dbg(cmd, "no such device\n"); + + cmd->result = DID_NO_CONNECT << 16; +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 16, 0) + scsi_done(cmd); +#else + cmd->scsi_done(cmd); +#endif + + return 0; + } + + retval = vhba_device_queue(vdev, cmd); + + vhba_device_put(vdev); + + return retval; +} + +static int vhba_abort (struct scsi_cmnd *cmd) +{ + struct vhba_device *vdev; + int retval = SUCCESS; + unsigned int devnum; + + scmd_dbg(cmd, "abort %p\n", cmd); + + devnum = bus_and_id_to_devnum(cmd->device->channel, cmd->device->id); + vdev = vhba_lookup_device(devnum); + if (vdev) { + retval = vhba_device_dequeue(vdev, cmd); + vhba_device_put(vdev); + } else { + cmd->result = DID_NO_CONNECT << 16; + } + + return retval; +} + +static struct scsi_host_template vhba_template = { + .module = THIS_MODULE, + .name = "vhba", + .proc_name = "vhba", + .queuecommand = vhba_queuecommand, + .eh_abort_handler = vhba_abort, + .this_id = -1, + .max_sectors = VHBA_MAX_SECTORS_PER_IO, + .sg_tablesize = 256, +#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 19, 0) + .slave_alloc = vhba_slave_alloc, +#endif +#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 14, 0) + .tag_alloc_policy_rr = true, +#elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 0, 0) + .tag_alloc_policy = BLK_TAG_ALLOC_RR, +#endif +#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 19, 0) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 4, 0) + .use_blk_tags = 1, +#endif +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 0, 0) + .max_segment_size = VHBA_KBUF_SIZE, +#endif +}; + +static ssize_t do_request (struct vhba_device *vdev, unsigned long metatag, struct scsi_cmnd *cmd, char __user *buf, size_t buf_len) +{ + struct vhba_request vreq; + ssize_t ret; + + scmd_dbg(cmd, "request %lu (%p), cdb 0x%x, bufflen %d, sg count %d\n", + metatag, cmd, cmd->cmnd[0], scsi_bufflen(cmd), scsi_sg_count(cmd)); + + ret = sizeof(vreq); + if (DATA_TO_DEVICE(cmd->sc_data_direction)) { + ret += scsi_bufflen(cmd); + } + + if (ret > buf_len) { + scmd_dbg(cmd, "buffer too small (%zd < %zd) for a request\n", buf_len, ret); + return -EIO; + } + + vreq.metatag = metatag; + vreq.lun = cmd->device->lun; + memcpy(vreq.cdb, cmd->cmnd, MAX_COMMAND_SIZE); + vreq.cdb_len = cmd->cmd_len; + vreq.data_len = scsi_bufflen(cmd); + + if (copy_to_user(buf, &vreq, sizeof(vreq))) { + return -EFAULT; + } + + if (DATA_TO_DEVICE(cmd->sc_data_direction) && vreq.data_len) { + buf += sizeof(vreq); + + if (scsi_sg_count(cmd)) { + unsigned char *kaddr, *uaddr; + struct scatterlist *sglist = scsi_sglist(cmd); + struct scatterlist *sg; + int i; + + uaddr = (unsigned char *) buf; + + for_each_sg(sglist, sg, scsi_sg_count(cmd), i) { + size_t len = sg->length; + + if (len > vdev->kbuf_size) { + scmd_dbg(cmd, "segment size (%zu) exceeds kbuf size (%zu)!", len, vdev->kbuf_size); + len = vdev->kbuf_size; + } + + kaddr = kmap_atomic(sg_page(sg)); + memcpy(vdev->kbuf, kaddr + sg->offset, len); + kunmap_atomic(kaddr); + + if (copy_to_user(uaddr, vdev->kbuf, len)) { + return -EFAULT; + } + uaddr += len; + } + } else { + if (copy_to_user(buf, scsi_sglist(cmd), vreq.data_len)) { + return -EFAULT; + } + } + } + + return ret; +} + +static ssize_t do_response (struct vhba_device *vdev, unsigned long metatag, struct scsi_cmnd *cmd, const char __user *buf, size_t buf_len, struct vhba_response *res) +{ + ssize_t ret = 0; + + scmd_dbg(cmd, "response %lu (%p), status %x, data len %d, sg count %d\n", + metatag, cmd, res->status, res->data_len, scsi_sg_count(cmd)); + + if (res->status) { + if (res->data_len > SCSI_SENSE_BUFFERSIZE) { + scmd_dbg(cmd, "truncate sense (%d < %d)", SCSI_SENSE_BUFFERSIZE, res->data_len); + res->data_len = SCSI_SENSE_BUFFERSIZE; + } + + if (copy_from_user(cmd->sense_buffer, buf, res->data_len)) { + return -EFAULT; + } + + cmd->result = res->status; + + ret += res->data_len; + } else if (DATA_FROM_DEVICE(cmd->sc_data_direction) && scsi_bufflen(cmd)) { + size_t to_read; + + if (res->data_len > scsi_bufflen(cmd)) { + scmd_dbg(cmd, "truncate data (%d < %d)\n", scsi_bufflen(cmd), res->data_len); + res->data_len = scsi_bufflen(cmd); + } + + to_read = res->data_len; + + if (scsi_sg_count(cmd)) { + unsigned char *kaddr, *uaddr; + struct scatterlist *sglist = scsi_sglist(cmd); + struct scatterlist *sg; + int i; + + uaddr = (unsigned char *)buf; + + for_each_sg(sglist, sg, scsi_sg_count(cmd), i) { + size_t len = (sg->length < to_read) ? sg->length : to_read; + + if (len > vdev->kbuf_size) { + scmd_dbg(cmd, "segment size (%zu) exceeds kbuf size (%zu)!", len, vdev->kbuf_size); + len = vdev->kbuf_size; + } + + if (copy_from_user(vdev->kbuf, uaddr, len)) { + return -EFAULT; + } + uaddr += len; + + kaddr = kmap_atomic(sg_page(sg)); + memcpy(kaddr + sg->offset, vdev->kbuf, len); + kunmap_atomic(kaddr); + + to_read -= len; + if (to_read == 0) { + break; + } + } + } else { + if (copy_from_user(scsi_sglist(cmd), buf, res->data_len)) { + return -EFAULT; + } + + to_read -= res->data_len; + } + + scsi_set_resid(cmd, to_read); + + ret += res->data_len - to_read; + } + + return ret; +} + +static struct vhba_command *next_command (struct vhba_device *vdev) +{ + struct vhba_command *vcmd; + + list_for_each_entry(vcmd, &vdev->cmd_list, entry) { + if (vcmd->status == VHBA_REQ_PENDING) { + break; + } + } + + if (&vcmd->entry == &vdev->cmd_list) { + vcmd = NULL; + } + + return vcmd; +} + +static struct vhba_command *match_command (struct vhba_device *vdev, __u32 metatag) +{ + struct vhba_command *vcmd; + + list_for_each_entry(vcmd, &vdev->cmd_list, entry) { + if (vcmd->metatag == metatag) { + break; + } + } + + if (&vcmd->entry == &vdev->cmd_list) { + vcmd = NULL; + } + + return vcmd; +} + +static struct vhba_command *wait_command (struct vhba_device *vdev, unsigned long flags) +{ + struct vhba_command *vcmd; + DEFINE_WAIT(wait); + + while (!(vcmd = next_command(vdev))) { + if (signal_pending(current)) { + break; + } + + prepare_to_wait(&vdev->cmd_wq, &wait, TASK_INTERRUPTIBLE); + + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + schedule(); + + spin_lock_irqsave(&vdev->cmd_lock, flags); + } + + finish_wait(&vdev->cmd_wq, &wait); + if (vcmd) { + vcmd->status = VHBA_REQ_READING; + } + + return vcmd; +} + +static ssize_t vhba_ctl_read (struct file *file, char __user *buf, size_t buf_len, loff_t *offset) +{ + struct vhba_device *vdev; + struct vhba_command *vcmd; + ssize_t ret; + unsigned long flags; + + vdev = file->private_data; + + /* Get next command */ + if (file->f_flags & O_NONBLOCK) { + /* Non-blocking variant */ + spin_lock_irqsave(&vdev->cmd_lock, flags); + vcmd = next_command(vdev); + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + if (!vcmd) { + return -EWOULDBLOCK; + } + } else { + /* Blocking variant */ + spin_lock_irqsave(&vdev->cmd_lock, flags); + vcmd = wait_command(vdev, flags); + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + if (!vcmd) { + return -ERESTARTSYS; + } + } + + ret = do_request(vdev, vcmd->metatag, vcmd->cmd, buf, buf_len); + + spin_lock_irqsave(&vdev->cmd_lock, flags); + if (ret >= 0) { + vcmd->status = VHBA_REQ_SENT; + *offset += ret; + } else { + vcmd->status = VHBA_REQ_PENDING; + } + + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + return ret; +} + +static ssize_t vhba_ctl_write (struct file *file, const char __user *buf, size_t buf_len, loff_t *offset) +{ + struct vhba_device *vdev; + struct vhba_command *vcmd; + struct vhba_response res; + ssize_t ret; + unsigned long flags; + + if (buf_len < sizeof(res)) { + return -EIO; + } + + if (copy_from_user(&res, buf, sizeof(res))) { + return -EFAULT; + } + + vdev = file->private_data; + + spin_lock_irqsave(&vdev->cmd_lock, flags); + vcmd = match_command(vdev, res.metatag); + if (!vcmd || vcmd->status != VHBA_REQ_SENT) { + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + pr_debug("ctl dev #%u not expecting response\n", vdev->num); + return -EIO; + } + vcmd->status = VHBA_REQ_WRITING; + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + ret = do_response(vdev, vcmd->metatag, vcmd->cmd, buf + sizeof(res), buf_len - sizeof(res), &res); + + spin_lock_irqsave(&vdev->cmd_lock, flags); + if (ret >= 0) { +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 16, 0) + scsi_done(vcmd->cmd); +#else + vcmd->cmd->scsi_done(vcmd->cmd); +#endif + ret += sizeof(res); + + /* don't compete with vhba_device_dequeue */ + if (!list_empty(&vcmd->entry)) { + list_del_init(&vcmd->entry); + vhba_free_command(vcmd); + } + } else { + vcmd->status = VHBA_REQ_SENT; + } + + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + return ret; +} + +static long vhba_ctl_ioctl (struct file *file, unsigned int cmd, unsigned long arg) +{ + struct vhba_device *vdev = file->private_data; + struct vhba_host *vhost = platform_get_drvdata(&vhba_platform_device); + + switch (cmd) { + case 0xBEEF001: { + unsigned int ident[4]; /* host, channel, id, lun */ + + ident[0] = vhost->shost->host_no; + devnum_to_bus_and_id(vdev->num, &ident[1], &ident[2]); + ident[3] = 0; /* lun */ + + if (copy_to_user((void *) arg, ident, sizeof(ident))) { + return -EFAULT; + } + + return 0; + } + case 0xBEEF002: { + unsigned int devnum = vdev->num; + + if (copy_to_user((void *) arg, &devnum, sizeof(devnum))) { + return -EFAULT; + } + + return 0; + } + } + + return -ENOTTY; +} + +#ifdef CONFIG_COMPAT +static long vhba_ctl_compat_ioctl (struct file *file, unsigned int cmd, unsigned long arg) +{ + unsigned long compat_arg = (unsigned long)compat_ptr(arg); + return vhba_ctl_ioctl(file, cmd, compat_arg); +} +#endif + +static unsigned int vhba_ctl_poll (struct file *file, poll_table *wait) +{ + struct vhba_device *vdev = file->private_data; + unsigned int mask = 0; + unsigned long flags; + + poll_wait(file, &vdev->cmd_wq, wait); + + spin_lock_irqsave(&vdev->cmd_lock, flags); + if (next_command(vdev)) { + mask |= POLLIN | POLLRDNORM; + } + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + return mask; +} + +static int vhba_ctl_open (struct inode *inode, struct file *file) +{ + struct vhba_device *vdev; + int retval; + + pr_debug("ctl dev open\n"); + + /* check if vhba is probed */ + if (!platform_get_drvdata(&vhba_platform_device)) { + return -ENODEV; + } + + vdev = vhba_device_alloc(); + if (!vdev) { + return -ENOMEM; + } + + vdev->kbuf_size = VHBA_KBUF_SIZE; + vdev->kbuf = kzalloc(vdev->kbuf_size, GFP_KERNEL); + if (!vdev->kbuf) { + return -ENOMEM; + } + + if (!(retval = vhba_add_device(vdev))) { + file->private_data = vdev; + } + + vhba_device_put(vdev); + + return retval; +} + +static int vhba_ctl_release (struct inode *inode, struct file *file) +{ + struct vhba_device *vdev; + struct vhba_command *vcmd; + unsigned long flags; + + vdev = file->private_data; + + pr_debug("ctl dev release\n"); + + vhba_device_get(vdev); + vhba_remove_device(vdev); + + spin_lock_irqsave(&vdev->cmd_lock, flags); + list_for_each_entry(vcmd, &vdev->cmd_list, entry) { + WARN_ON(vcmd->status == VHBA_REQ_READING || vcmd->status == VHBA_REQ_WRITING); + + scmd_dbg(vcmd->cmd, "device released with command %lu (%p)\n", vcmd->metatag, vcmd->cmd); + vcmd->cmd->result = DID_NO_CONNECT << 16; +#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 16, 0) + scsi_done(vcmd->cmd); +#else + vcmd->cmd->scsi_done(vcmd->cmd); +#endif + vhba_free_command(vcmd); + } + INIT_LIST_HEAD(&vdev->cmd_list); + spin_unlock_irqrestore(&vdev->cmd_lock, flags); + + kfree(vdev->kbuf); + vdev->kbuf = NULL; + + vhba_device_put(vdev); + + return 0; +} + +static struct file_operations vhba_ctl_fops = { + .owner = THIS_MODULE, + .open = vhba_ctl_open, + .release = vhba_ctl_release, + .read = vhba_ctl_read, + .write = vhba_ctl_write, + .poll = vhba_ctl_poll, + .unlocked_ioctl = vhba_ctl_ioctl, +#ifdef CONFIG_COMPAT + .compat_ioctl = vhba_ctl_compat_ioctl, +#endif +}; + +static struct miscdevice vhba_miscdev = { + .minor = MISC_DYNAMIC_MINOR, + .name = "vhba_ctl", + .fops = &vhba_ctl_fops, +}; + +static int vhba_probe (struct platform_device *pdev) +{ + struct Scsi_Host *shost; + struct vhba_host *vhost; + int i; + + vhba_can_queue = clamp(vhba_can_queue, 1, 256); + + shost = scsi_host_alloc(&vhba_template, sizeof(struct vhba_host)); + if (!shost) { + return -ENOMEM; + } + + shost->max_channel = VHBA_MAX_BUS-1; + shost->max_id = VHBA_MAX_ID; + /* we don't support lun > 0 */ + shost->max_lun = 1; + shost->max_cmd_len = MAX_COMMAND_SIZE; + shost->can_queue = vhba_can_queue; + shost->cmd_per_lun = vhba_can_queue; + + vhost = (struct vhba_host *)shost->hostdata; + memset(vhost, 0, sizeof(struct vhba_host)); + + vhost->shost = shost; + vhost->num_devices = 0; + spin_lock_init(&vhost->dev_lock); + spin_lock_init(&vhost->cmd_lock); + INIT_WORK(&vhost->scan_devices, vhba_scan_devices); + vhost->cmd_next = 0; + vhost->commands = kzalloc(vhba_can_queue * sizeof(struct vhba_command), GFP_KERNEL); + if (!vhost->commands) { + return -ENOMEM; + } + + for (i = 0; i < vhba_can_queue; i++) { + vhost->commands[i].status = VHBA_REQ_FREE; + } + + platform_set_drvdata(pdev, vhost); + +#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 4, 0) + i = scsi_init_shared_tag_map(shost, vhba_can_queue); + if (i) return i; +#endif + + if (scsi_add_host(shost, &pdev->dev)) { + scsi_host_put(shost); + return -ENOMEM; + } + + return 0; +} + +#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 11, 0) +static int vhba_remove (struct platform_device *pdev) +#else +static void vhba_remove (struct platform_device *pdev) +#endif +{ + struct vhba_host *vhost; + struct Scsi_Host *shost; + + vhost = platform_get_drvdata(pdev); + shost = vhost->shost; + + scsi_remove_host(shost); + scsi_host_put(shost); + + kfree(vhost->commands); + +#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 11, 0) + return 0; +#endif +} + +static void vhba_release (struct device * dev) +{ + return; +} + +static struct platform_device vhba_platform_device = { + .name = "vhba", + .id = -1, + .dev = { + .release = vhba_release, + }, +}; + +static struct platform_driver vhba_platform_driver = { + .driver = { + .owner = THIS_MODULE, + .name = "vhba", + }, + .probe = vhba_probe, + .remove = vhba_remove, +}; + +static int __init vhba_init (void) +{ + int ret; + + ret = platform_device_register(&vhba_platform_device); + if (ret < 0) { + return ret; + } + + ret = platform_driver_register(&vhba_platform_driver); + if (ret < 0) { + platform_device_unregister(&vhba_platform_device); + return ret; + } + + ret = misc_register(&vhba_miscdev); + if (ret < 0) { + platform_driver_unregister(&vhba_platform_driver); + platform_device_unregister(&vhba_platform_device); + return ret; + } + + return 0; +} + +static void __exit vhba_exit(void) +{ + misc_deregister(&vhba_miscdev); + platform_driver_unregister(&vhba_platform_driver); + platform_device_unregister(&vhba_platform_device); +} + +module_init(vhba_init); +module_exit(vhba_exit); + diff --git a/include/linux/mm.h b/include/linux/mm.h index f0d5be9dc736..f25889adfa43 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -207,6 +207,14 @@ static inline void __mm_zero_struct_page(struct page *page) extern int sysctl_max_map_count; +extern bool sysctl_workingset_protection; +extern u8 sysctl_anon_min_ratio; +extern u8 sysctl_clean_low_ratio; +extern u8 sysctl_clean_min_ratio; +int vm_workingset_protection_update_handler( + const struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, loff_t *ppos); + extern unsigned long sysctl_user_reserve_kbytes; extern unsigned long sysctl_admin_reserve_kbytes; diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h index ec442af3f886..9a0364eb5ee1 100644 --- a/include/linux/pagemap.h +++ b/include/linux/pagemap.h @@ -1376,7 +1376,7 @@ struct readahead_control { ._index = i, \ } -#define VM_READAHEAD_PAGES (SZ_128K / PAGE_SIZE) +#define VM_READAHEAD_PAGES (SZ_8M / PAGE_SIZE) void page_cache_ra_unbounded(struct readahead_control *, unsigned long nr_to_read, unsigned long lookahead_count); diff --git a/init/Kconfig b/init/Kconfig index fa79feb8fe57..6c8ba1a1a8b9 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -194,6 +194,10 @@ config THREAD_INFO_IN_TASK menu "General setup" +config CACHY + bool "Some kernel tweaks by CachyOS" + default y + config BROKEN bool help @@ -1398,6 +1402,22 @@ config USER_NS If unsure, say N. +config USER_NS_UNPRIVILEGED + bool "Allow unprivileged users to create namespaces" + default y + depends on USER_NS + help + When disabled, unprivileged users will not be able to create + new namespaces. Allowing users to create their own namespaces + has been part of several recent local privilege escalation + exploits, so if you need user namespaces but are + paranoid^Wsecurity-conscious you want to disable this. + + This setting can be overridden at runtime via the + kernel.unprivileged_userns_clone sysctl. + + If unsure, say Y. + config PID_NS bool "PID Namespaces" default y @@ -1566,6 +1586,12 @@ config CC_OPTIMIZE_FOR_PERFORMANCE with the "-O2" compiler flag for best performance and most helpful compile-time warnings. +config CC_OPTIMIZE_FOR_PERFORMANCE_O3 + bool "Optimize more for performance (-O3)" + help + Choosing this option will pass "-O3" to your compiler to optimize + the kernel yet more for performance. + config CC_OPTIMIZE_FOR_SIZE bool "Optimize for size (-Os)" help diff --git a/kernel/Kconfig.hz b/kernel/Kconfig.hz index ce1435cb08b1..e1359db5561e 100644 --- a/kernel/Kconfig.hz +++ b/kernel/Kconfig.hz @@ -40,6 +40,27 @@ choice on SMP and NUMA systems and exactly dividing by both PAL and NTSC frame rates for video and multimedia work. + config HZ_500 + bool "500 HZ" + help + 500 Hz is a balanced timer frequency. Provides fast interactivity + on desktops with good smoothness without increasing CPU power + consumption and sacrificing the battery life on laptops. + + config HZ_600 + bool "600 HZ" + help + 600 Hz is a balanced timer frequency. Provides fast interactivity + on desktops with good smoothness without increasing CPU power + consumption and sacrificing the battery life on laptops. + + config HZ_750 + bool "750 HZ" + help + 750 Hz is a balanced timer frequency. Provides fast interactivity + on desktops with good smoothness without increasing CPU power + consumption and sacrificing the battery life on laptops. + config HZ_1000 bool "1000 HZ" help @@ -53,6 +74,9 @@ config HZ default 100 if HZ_100 default 250 if HZ_250 default 300 if HZ_300 + default 500 if HZ_500 + default 600 if HZ_600 + default 750 if HZ_750 default 1000 if HZ_1000 config SCHED_HRTICK diff --git a/kernel/Kconfig.preempt b/kernel/Kconfig.preempt index da326800c1c9..cf5267cac7bb 100644 --- a/kernel/Kconfig.preempt +++ b/kernel/Kconfig.preempt @@ -88,7 +88,7 @@ endchoice config PREEMPT_RT bool "Fully Preemptible Kernel (Real-Time)" - depends on EXPERT && ARCH_SUPPORTS_RT && !COMPILE_TEST + depends on ARCH_SUPPORTS_RT && !COMPILE_TEST select PREEMPTION help This option turns the kernel into a real-time kernel by replacing diff --git a/kernel/fork.c b/kernel/fork.c index b1f3915d5f8e..f779ad112643 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -123,6 +123,16 @@ #include +#ifdef CONFIG_USER_NS +# ifdef CONFIG_USER_NS_UNPRIVILEGED +static int unprivileged_userns_clone = 1; +# else +static int unprivileged_userns_clone = 0; +# endif +#else +#define unprivileged_userns_clone 1 +#endif + /* * Minimum number of threads to boot the kernel */ @@ -2030,6 +2040,11 @@ __latent_entropy struct task_struct *copy_process( return ERR_PTR(-EINVAL); } + if ((clone_flags & CLONE_NEWUSER) && !unprivileged_userns_clone) { + if (!capable(CAP_SYS_ADMIN)) + return ERR_PTR(-EPERM); + } + /* * Force any signals received before this point to be delivered * before the fork happens. Collect up signals sent to multiple @@ -3067,6 +3082,10 @@ static int check_unshare_flags(unsigned long unshare_flags) if (!current_is_single_threaded()) return -EINVAL; } + if ((unshare_flags & CLONE_NEWUSER) && !unprivileged_userns_clone) { + if (!capable(CAP_SYS_ADMIN)) + return -EPERM; + } return 0; } @@ -3297,6 +3316,15 @@ static const struct ctl_table fork_sysctl_table[] = { .mode = 0644, .proc_handler = sysctl_max_threads, }, +#ifdef CONFIG_USER_NS + { + .procname = "unprivileged_userns_clone", + .data = &unprivileged_userns_clone, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = proc_dointvec, + }, +#endif }; static int __init init_fork_sysctl(void) diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c index 24df4d98f7d2..1d5923996fa5 100644 --- a/kernel/locking/rwsem.c +++ b/kernel/locking/rwsem.c @@ -746,6 +746,7 @@ rwsem_spin_on_owner(struct rw_semaphore *sem) struct task_struct *new, *owner; unsigned long flags, new_flags; enum owner_state state; + int i = 0; lockdep_assert_preemption_disabled(); @@ -782,7 +783,8 @@ rwsem_spin_on_owner(struct rw_semaphore *sem) break; } - cpu_relax(); + if (i++ > 1000) + cpu_relax(); } return state; diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 3eaeceda71b0..66289244f71e 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -76,10 +76,19 @@ unsigned int sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG; * * (default: 0.70 msec * (1 + ilog(ncpus)), units: nanoseconds) */ +#ifdef CONFIG_CACHY +unsigned int sysctl_sched_base_slice = 350000ULL; +static unsigned int normalized_sysctl_sched_base_slice = 350000ULL; +#else unsigned int sysctl_sched_base_slice = 700000ULL; static unsigned int normalized_sysctl_sched_base_slice = 700000ULL; +#endif /* CONFIG_CACHY */ +#ifdef CONFIG_CACHY +__read_mostly unsigned int sysctl_sched_migration_cost = 300000UL; +#else __read_mostly unsigned int sysctl_sched_migration_cost = 500000UL; +#endif static int __init setup_sched_thermal_decay_shift(char *str) { @@ -122,8 +131,12 @@ int __weak arch_asym_cpu_priority(int cpu) * * (default: 5 msec, units: microseconds) */ +#ifdef CONFIG_CACHY +static unsigned int sysctl_sched_cfs_bandwidth_slice = 3000UL; +#else static unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL; #endif +#endif #ifdef CONFIG_NUMA_BALANCING /* Restrict the NUMA promotion throughput (MB/s) for each target node. */ @@ -1048,7 +1061,7 @@ static bool update_deadline(struct cfs_rq *cfs_rq, struct sched_entity *se) #include "pelt.h" -static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu); +static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu, int sync); static unsigned long task_h_load(struct task_struct *p); static unsigned long capacity_of(int cpu); @@ -7502,6 +7515,24 @@ static inline int __select_idle_cpu(int cpu, struct task_struct *p) DEFINE_STATIC_KEY_FALSE(sched_smt_present); EXPORT_SYMBOL_GPL(sched_smt_present); +/* + * Return true if all the CPUs in the SMT core where @cpu belongs are idle, + * false otherwise. + */ +static bool is_idle_core(int cpu) +{ + int sibling; + + if (!sched_smt_active()) + return (available_idle_cpu(cpu) || sched_idle_cpu(cpu)); + + for_each_cpu(sibling, cpu_smt_mask(cpu)) + if (!available_idle_cpu(sibling) && !sched_idle_cpu(sibling)) + return false; + + return true; +} + static inline void set_idle_cores(int cpu, int val) { struct sched_domain_shared *sds; @@ -7770,13 +7801,26 @@ static inline bool asym_fits_cpu(unsigned long util, /* * Try and locate an idle core/thread in the LLC cache domain. */ -static int select_idle_sibling(struct task_struct *p, int prev, int target) +static int select_idle_sibling(struct task_struct *p, int prev, int target, int sync) { bool has_idle_core = false; struct sched_domain *sd; unsigned long task_util, util_min, util_max; int i, recent_used_cpu, prev_aff = -1; + /* Check a recently used CPU as a potential idle candidate: */ + recent_used_cpu = p->recent_used_cpu; + p->recent_used_cpu = prev; + if (recent_used_cpu != prev && + recent_used_cpu != target && + cpus_share_cache(recent_used_cpu, target) && + is_idle_core(recent_used_cpu) && + cpumask_test_cpu(recent_used_cpu, p->cpus_ptr)) { + return recent_used_cpu; + } else { + recent_used_cpu = -1; + } + /* * On asymmetric system, update task utilization because we will check * that the task fits with CPU's capacity. @@ -7793,7 +7837,7 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target) */ lockdep_assert_irqs_disabled(); - if ((available_idle_cpu(target) || sched_idle_cpu(target)) && + if (sync && is_idle_core(target) && asym_fits_cpu(task_util, util_min, util_max, target)) return target; @@ -7827,24 +7871,6 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target) return prev; } - /* Check a recently used CPU as a potential idle candidate: */ - recent_used_cpu = p->recent_used_cpu; - p->recent_used_cpu = prev; - if (recent_used_cpu != prev && - recent_used_cpu != target && - cpus_share_cache(recent_used_cpu, target) && - (available_idle_cpu(recent_used_cpu) || sched_idle_cpu(recent_used_cpu)) && - cpumask_test_cpu(recent_used_cpu, p->cpus_ptr) && - asym_fits_cpu(task_util, util_min, util_max, recent_used_cpu)) { - - if (!static_branch_unlikely(&sched_cluster_active) || - cpus_share_resources(recent_used_cpu, target)) - return recent_used_cpu; - - } else { - recent_used_cpu = -1; - } - /* * For asymmetric CPU capacity systems, our domain of interest is * sd_asym_cpucapacity rather than sd_llc. @@ -8578,7 +8604,14 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags) new_cpu = sched_balance_find_dst_cpu(sd, p, cpu, prev_cpu, sd_flag); } else if (wake_flags & WF_TTWU) { /* XXX always ? */ /* Fast path */ - new_cpu = select_idle_sibling(p, prev_cpu, new_cpu); + /* + * If the previous CPU is an idle core, retain the same for + * cache locality. Otherwise, search for an idle sibling. + */ + if (is_idle_core(prev_cpu)) + new_cpu = prev_cpu; + else + new_cpu = select_idle_sibling(p, prev_cpu, new_cpu, sync); } rcu_read_unlock(); @@ -10262,7 +10295,7 @@ static bool sched_use_asym_prio(struct sched_domain *sd, int cpu) if (!sched_smt_active()) return true; - return sd->flags & SD_SHARE_CPUCAPACITY || is_core_idle(cpu); + return sd->flags & SD_SHARE_CPUCAPACITY || is_idle_core(cpu); } static inline bool sched_asym(struct sched_domain *sd, int dst_cpu, int src_cpu) diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 1b4283e9edc3..7294157bfadc 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2965,7 +2965,7 @@ extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags); -#ifdef CONFIG_PREEMPT_RT +#if defined(CONFIG_PREEMPT_RT) || defined(CONFIG_CACHY) # define SCHED_NR_MIGRATE_BREAK 8 #else # define SCHED_NR_MIGRATE_BREAK 32 diff --git a/mm/Kconfig b/mm/Kconfig index a992f2203eb9..df5f168a3f30 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -435,6 +435,69 @@ config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP config ARCH_WANT_HUGETLB_VMEMMAP_PREINIT bool +config ANON_MIN_RATIO + int "Default value for vm.anon_min_ratio" + depends on SYSCTL + range 0 100 + default 1 + help + This option sets the default value for vm.anon_min_ratio sysctl knob. + + The vm.anon_min_ratio sysctl knob provides *hard* protection of + anonymous pages. The anonymous pages on the current node won't be + reclaimed under any conditions when their amount is below + vm.anon_min_ratio. This knob may be used to prevent excessive swap + thrashing when anonymous memory is low (for example, when memory is + going to be overfilled by compressed data of zram module). + + Setting this value too high (close to MemTotal) can result in + inability to swap and can lead to early OOM under memory pressure. + +config CLEAN_LOW_RATIO + int "Default value for vm.clean_low_ratio" + depends on SYSCTL + range 0 100 + default 15 + help + This option sets the default value for vm.clean_low_ratio sysctl knob. + + The vm.clean_low_ratio sysctl knob provides *best-effort* + protection of clean file pages. The file pages on the current node + won't be reclaimed under memory pressure when the amount of clean file + pages is below vm.clean_low_ratio *unless* we threaten to OOM. + Protection of clean file pages using this knob may be used when + swapping is still possible to + - prevent disk I/O thrashing under memory pressure; + - improve performance in disk cache-bound tasks under memory + pressure. + + Setting it to a high value may result in a early eviction of anonymous + pages into the swap space by attempting to hold the protected amount + of clean file pages in memory. + +config CLEAN_MIN_RATIO + int "Default value for vm.clean_min_ratio" + depends on SYSCTL + range 0 100 + default 4 + help + This option sets the default value for vm.clean_min_ratio sysctl knob. + + The vm.clean_min_ratio sysctl knob provides *hard* protection of + clean file pages. The file pages on the current node won't be + reclaimed under memory pressure when the amount of clean file pages is + below vm.clean_min_ratio. Hard protection of clean file pages using + this knob may be used to + - prevent disk I/O thrashing under memory pressure even with no free + swap space; + - improve performance in disk cache-bound tasks under memory + pressure; + - avoid high latency and prevent livelock in near-OOM conditions. + + Setting it to a high value may result in a early out-of-memory condition + due to the inability to reclaim the protected amount of clean file pages + when other types of pages cannot be reclaimed. + config HAVE_MEMBLOCK_PHYS_MAP bool @@ -631,7 +694,7 @@ config COMPACTION config COMPACT_UNEVICTABLE_DEFAULT int depends on COMPACTION - default 0 if PREEMPT_RT + default 0 if PREEMPT_RT || CACHY default 1 # diff --git a/mm/compaction.c b/mm/compaction.c index 1e8f8eca318c..32326c48162b 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -1887,7 +1887,11 @@ static int sysctl_compact_unevictable_allowed __read_mostly = CONFIG_COMPACT_UNE * aggressively the kernel should compact memory in the * background. It takes values in the range [0, 100]. */ +#ifdef CONFIG_CACHY +static unsigned int __read_mostly sysctl_compaction_proactiveness; +#else static unsigned int __read_mostly sysctl_compaction_proactiveness = 20; +#endif static int sysctl_extfrag_threshold = 500; static int __read_mostly sysctl_compact_memory; diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 40cf59301c21..50a3a2165bcf 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -63,7 +63,11 @@ unsigned long transparent_hugepage_flags __read_mostly = #ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE (1<may_unmap && folio_mapped(folio)) goto keep_locked; + if (folio_is_file_lru(folio) ? sc->clean_below_min : + (sc->anon_below_min && !sc->clean_below_min)) + goto keep_locked; + /* * The number of dirty pages determines if a node is marked * reclaim_congested. kswapd will stall and start writing @@ -2572,6 +2598,15 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, goto out; } + /* + * Force-scan anon if clean file pages is under vm.clean_low_ratio + * or vm.clean_min_ratio. + */ + if (sc->clean_below_low || sc->clean_below_min) { + scan_balance = SCAN_ANON; + goto out; + } + /* * If there is enough inactive page cache, we do not reclaim * anything from the anonymous working right now to make sure @@ -2630,6 +2665,14 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, BUG(); } + /* + * Hard protection of the working set. + * Don't reclaim anon/file pages when the amount is + * below the watermark of the same type. + */ + if (file ? sc->clean_below_min : sc->anon_below_min) + scan = 0; + nr[lru] = scan; } } @@ -2650,6 +2693,96 @@ static bool can_age_anon_pages(struct lruvec *lruvec, lruvec_memcg(lruvec)); } +int vm_workingset_protection_update_handler(const struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, loff_t *ppos) +{ + int ret = proc_dou8vec_minmax(table, write, buffer, lenp, ppos); + if (ret || !write) + return ret; + + workingset_protection_prev_totalram = 0; + + return 0; +} + +static void prepare_workingset_protection(pg_data_t *pgdat, struct scan_control *sc) +{ + unsigned long node_mem_total; + struct sysinfo i; + + if (!(sysctl_workingset_protection)) { + sc->anon_below_min = 0; + sc->clean_below_low = 0; + sc->clean_below_min = 0; + return; + } + + if (likely(sysctl_anon_min_ratio || + sysctl_clean_low_ratio || + sysctl_clean_min_ratio)) { +#ifdef CONFIG_NUMA + si_meminfo_node(&i, pgdat->node_id); +#else //CONFIG_NUMA + si_meminfo(&i); +#endif //CONFIG_NUMA + node_mem_total = i.totalram; + + if (unlikely(workingset_protection_prev_totalram != node_mem_total)) { + sysctl_anon_min_ratio_kb = + node_mem_total * sysctl_anon_min_ratio / 100; + sysctl_clean_low_ratio_kb = + node_mem_total * sysctl_clean_low_ratio / 100; + sysctl_clean_min_ratio_kb = + node_mem_total * sysctl_clean_min_ratio / 100; + workingset_protection_prev_totalram = node_mem_total; + } + } + + /* + * Check the number of anonymous pages to protect them from + * reclaiming if their amount is below the specified. + */ + if (sysctl_anon_min_ratio) { + unsigned long reclaimable_anon; + + reclaimable_anon = + node_page_state(pgdat, NR_ACTIVE_ANON) + + node_page_state(pgdat, NR_INACTIVE_ANON) + + node_page_state(pgdat, NR_ISOLATED_ANON); + + sc->anon_below_min = reclaimable_anon < sysctl_anon_min_ratio_kb; + } else + sc->anon_below_min = 0; + + /* + * Check the number of clean file pages to protect them from + * reclaiming if their amount is below the specified. + */ + if (sysctl_clean_low_ratio || sysctl_clean_min_ratio) { + unsigned long reclaimable_file, dirty, clean; + + reclaimable_file = + node_page_state(pgdat, NR_ACTIVE_FILE) + + node_page_state(pgdat, NR_INACTIVE_FILE) + + node_page_state(pgdat, NR_ISOLATED_FILE); + dirty = node_page_state(pgdat, NR_FILE_DIRTY); + /* + * node_page_state() sum can go out of sync since + * all the values are not read at once. + */ + if (likely(reclaimable_file > dirty)) + clean = reclaimable_file - dirty; + else + clean = 0; + + sc->clean_below_low = clean < sysctl_clean_low_ratio_kb; + sc->clean_below_min = clean < sysctl_clean_min_ratio_kb; + } else { + sc->clean_below_low = 0; + sc->clean_below_min = 0; + } +} + #ifdef CONFIG_LRU_GEN #ifdef CONFIG_LRU_GEN_ENABLED @@ -4634,11 +4767,21 @@ static int get_tier_idx(struct lruvec *lruvec, int type) return tier - 1; } -static int get_type_to_scan(struct lruvec *lruvec, int swappiness) +static int get_type_to_scan(struct lruvec *lruvec, struct scan_control *sc, int swappiness) { struct ctrl_pos sp, pv; - if (swappiness <= MIN_SWAPPINESS + 1) + if (swappiness == MIN_SWAPPINESS) + return LRU_GEN_FILE; + + if (sc->clean_below_min) + return LRU_GEN_ANON; + if (sc->anon_below_min) + return LRU_GEN_FILE; + if (sc->clean_below_low) + return LRU_GEN_ANON; + + if (swappiness == MIN_SWAPPINESS + 1) return LRU_GEN_FILE; if (swappiness >= MAX_SWAPPINESS) @@ -4658,7 +4801,7 @@ static int isolate_folios(unsigned long nr_to_scan, struct lruvec *lruvec, int *type_scanned, struct list_head *list) { int i; - int type = get_type_to_scan(lruvec, swappiness); + int type = get_type_to_scan(lruvec, sc, swappiness); for_each_evictable_type(i, swappiness) { int scanned; @@ -4904,6 +5047,12 @@ static int shrink_one(struct lruvec *lruvec, struct scan_control *sc) struct mem_cgroup *memcg = lruvec_memcg(lruvec); struct pglist_data *pgdat = lruvec_pgdat(lruvec); + prepare_workingset_protection(pgdat, sc); + + if (sysctl_workingset_protection && sc->clean_below_min && + !can_reclaim_anon_pages(memcg, pgdat->node_id, sc)) + return 0; + /* lru_gen_age_node() called mem_cgroup_calculate_protection() */ if (mem_cgroup_below_min(NULL, memcg)) return MEMCG_LRU_YOUNG; @@ -6058,6 +6207,8 @@ static void shrink_node(pg_data_t *pgdat, struct scan_control *sc) prepare_scan_control(pgdat, sc); + prepare_workingset_protection(pgdat, sc); + shrink_node_memcgs(pgdat, sc); flush_reclaim_state(sc); diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib index 28a1c08e3b22..13f100a3381d 100644 --- a/scripts/Makefile.lib +++ b/scripts/Makefile.lib @@ -239,6 +239,12 @@ ifdef CONFIG_LTO_CLANG cmd_ld_single = $(if $(objtool-enabled)$(is-single-obj-m), ; $(LD) $(ld_flags) -r -o $(tmp-target) $@; mv $(tmp-target) $@) endif +ifdef CONFIG_LTO_CLANG_THIN_DIST +# Save the _c_flags, sliently. +quiet_cmd_save_c_flags = + cmd_save_c_flags = printf '\n%s\n' 'saved_c_flags_$@ := $(call escsq,$(_c_flags))' >> $(dot-target).cmd +endif + quiet_cmd_cc_o_c = CC $(quiet_modtag) $@ cmd_cc_o_c = $(CC) $(c_flags) -c -o $@ $< \ $(cmd_ld_single) \ @@ -246,6 +252,7 @@ quiet_cmd_cc_o_c = CC $(quiet_modtag) $@ define rule_cc_o_c $(call cmd_and_fixdep,cc_o_c) + $(call cmd,save_c_flags) $(call cmd,checksrc) $(call cmd,checkdoc) $(call cmd,gen_objtooldep) diff --git a/scripts/Makefile.thinlto b/scripts/Makefile.thinlto new file mode 100644 index 000000000000..03349ac69de5 --- /dev/null +++ b/scripts/Makefile.thinlto @@ -0,0 +1,40 @@ +PHONY := __default +__default: + +include include/config/auto.conf +include $(srctree)/scripts/Kbuild.include +include $(srctree)/scripts/Makefile.lib + +native-objs := $(patsubst %.o,%.thinlto-native.o,$(call read-file, vmlinux.thinlto-index)) + +__default: $(native-objs) + +# Generate .thinlto-native.o (obj) from .o (bitcode) and .thinlto.bc (summary) files +# --------------------------------------------------------------------------- +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@ + be_flags = $(shell sed -n '/saved_c_flags_/s/.*:= //p' \ + $(dir $(<)).$(notdir $(<)).cmd) + cmd_cc_o_bc = \ + $(CC) $(be_flags) -x ir -fno-lto -Wno-unused-command-line-argument \ + -fthinlto-index=$(word 2, $^) -c -o $@ $< + +targets += $(native-objs) +$(native-objs): %.thinlto-native.o: %.o %.o.thinlto.bc FORCE + $(call if_changed,cc_o_bc) + +# Add FORCE to the prerequisites of a target to force it to be always rebuilt. +# --------------------------------------------------------------------------- + +PHONY += FORCE +FORCE: + +# Read all saved command lines and dependencies for the $(targets) we +# may be building above, using $(if_changed{,_dep}). As an +# optimization, we don't need to read them if the target does not +# exist, we will rebuild anyway in that case. + +existing-targets := $(wildcard $(sort $(targets))) + +-include $(foreach f,$(existing-targets),$(dir $(f)).$(notdir $(f)).cmd) + +.PHONY: $(PHONY) diff --git a/scripts/Makefile.vmlinux_a b/scripts/Makefile.vmlinux_a new file mode 100644 index 000000000000..382cebe1ee57 --- /dev/null +++ b/scripts/Makefile.vmlinux_a @@ -0,0 +1,83 @@ +# SPDX-License-Identifier: GPL-2.0-only + +PHONY := __default +__default: vmlinux.a + +include include/config/auto.conf +include $(srctree)/scripts/Kbuild.include +include $(srctree)/scripts/Makefile.lib + +# Link of built-in-fixup.a +# --------------------------------------------------------------------------- + +# '$(AR) mPi' needs --thin to workaround the bug of llvm-ar <= 14 +quiet_cmd_ar_builtin_fixup = AR $@ + cmd_ar_builtin_fixup = \ + rm -f $@; \ + $(AR) cDPrS --thin $@ $(KBUILD_VMLINUX_OBJS); \ + $(AR) mPi --thin $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt) + +targets += built-in-fixup.a +built-in-fixup.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE + $(call if_changed,ar_builtin_fixup) + +ifdef CONFIG_LTO_CLANG_THIN_DIST + +quiet_cmd_builtin.order = GEN $@ + cmd_builtin.order = $(AR) t $< > $@ + +targets += builtin.order +builtin.order: built-in-fixup.a FORCE + $(call if_changed,builtin.order) + +quiet_cmd_ld_thinlto_index = LD $@ + cmd_ld_thinlto_index = \ + $(LD) $(KBUILD_LDFLAGS) -r --thinlto-index-only=$@ @$< + +targets += vmlinux.thinlto-index +vmlinux.thinlto-index: builtin.order FORCE + $(call if_changed,ld_thinlto_index) + +quiet_cmd_ar_vmlinux.a = GEN $@ + cmd_ar_vmlinux.a = \ + rm -f $@; \ + while read -r obj; do \ + if grep -q $${obj} $(word 2, $^); then \ + echo $${obj%.o}.thinlto-native.o; \ + else \ + echo $${obj}; \ + fi; \ + done < $< | xargs $(AR) cDPrS --thin $@ + +targets += vmlinux.a +vmlinux.a: builtin.order vmlinux.thinlto-index FORCE + $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.thinlto + $(call if_changed,ar_vmlinux.a) + +else + +# vmlinux.a +# --------------------------------------------------------------------------- + +targets += vmlinux.a +vmlinux.a: built-in-fixup.a FORCE + $(call if_changed,copy) + +endif + +# Add FORCE to the prerequisites of a target to force it to be always rebuilt. +# --------------------------------------------------------------------------- + +PHONY += FORCE +FORCE: + +# Read all saved command lines and dependencies for the $(targets) we +# may be building above, using $(if_changed{,_dep}). As an +# optimization, we don't need to read them if the target does not +# exist, we will rebuild anyway in that case. + +existing-targets := $(wildcard $(sort $(targets))) + +-include $(foreach f,$(existing-targets),$(dir $(f)).$(notdir $(f)).cmd) + +.PHONY: $(PHONY) diff --git a/scripts/mod/modpost.c b/scripts/mod/modpost.c index 755b842f1f9b..7d9881a2243c 100644 --- a/scripts/mod/modpost.c +++ b/scripts/mod/modpost.c @@ -1478,13 +1478,22 @@ static void extract_crcs_for_object(const char *object, struct module *mod) char cmd_file[PATH_MAX]; char *buf, *p; const char *base; - int dirlen, ret; + int dirlen, baselen_without_suffix, ret; base = get_basename(object); dirlen = base - object; - ret = snprintf(cmd_file, sizeof(cmd_file), "%.*s.%s.cmd", - dirlen, object, base); + baselen_without_suffix = strlen(object) - dirlen - strlen(".o"); + + /* + * When CONFIG_LTO_CLANG_THIN_DIST=y, the ELF is *.thinlto-native.o + * but the symbol CRCs are recorded in *.o.cmd file. + */ + if (strends(object, ".thinlto-native.o")) + baselen_without_suffix -= strlen(".thinlto-native"); + + ret = snprintf(cmd_file, sizeof(cmd_file), "%.*s.%.*s.o.cmd", + dirlen, object, baselen_without_suffix, base); if (ret >= sizeof(cmd_file)) { error("%s: too long path was truncated\n", cmd_file); return; diff --git a/scripts/package/PKGBUILD b/scripts/package/PKGBUILD index 452374d63c24..08f80d7c5df0 100644 --- a/scripts/package/PKGBUILD +++ b/scripts/package/PKGBUILD @@ -90,6 +90,11 @@ _package-headers() { "${srctree}/scripts/package/install-extmod-build" "${builddir}" fi + # required when DEBUG_INFO_BTF_MODULES is enabled + if [ -f tools/bpf/resolve_btfids/resolve_btfids ]; then + install -Dt "$builddir/tools/bpf/resolve_btfids" tools/bpf/resolve_btfids/resolve_btfids + fi + echo "Installing System.map and config..." mkdir -p "${builddir}" cp System.map "${builddir}/System.map" -- 2.53.0 From 4f273466ec22c1d4609e03270d5a8473ebfa8781 Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Mon, 16 Feb 2026 13:51:09 +0100 Subject: [PATCH 4/7] fixes Signed-off-by: Peter Jung --- arch/arm/include/asm/mmu_context.h | 2 +- arch/riscv/include/asm/sync_core.h | 2 +- arch/s390/include/asm/mmu_context.h | 2 +- arch/sparc/include/asm/mmu_context_64.h | 2 +- arch/x86/include/asm/mmu_context.h | 23 +++++++- arch/x86/include/asm/sync_core.h | 2 +- arch/x86/kernel/cpu/amd.c | 6 ++ arch/x86/mm/tlb.c | 21 ------- drivers/bluetooth/btusb.c | 4 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 +- .../amd/display/amdgpu_dm/amdgpu_dm_helpers.c | 1 + .../amd/display/include/ddc_service_types.h | 1 + drivers/gpu/drm/i915/gt/intel_rc6.c | 13 ++++- drivers/usb/core/quirks.c | 9 +++ include/linux/perf_event.h | 2 +- include/linux/sched/mm.h | 10 ++-- include/linux/sched/topology.h | 3 - include/linux/tick.h | 4 +- include/linux/vtime.h | 8 +-- kernel/rcu/tree.c | 12 ++++ kernel/sched/core.c | 22 +++----- kernel/sched/ext.c | 4 +- kernel/sched/fair.c | 55 +++---------------- kernel/sched/features.h | 5 -- kernel/sched/sched.h | 31 +++++------ kernel/sched/topology.c | 6 -- sound/hda/codecs/realtek/alc269.c | 1 + 27 files changed, 117 insertions(+), 138 deletions(-) diff --git a/arch/arm/include/asm/mmu_context.h b/arch/arm/include/asm/mmu_context.h index db2cb06aa8cf..bebde469f81a 100644 --- a/arch/arm/include/asm/mmu_context.h +++ b/arch/arm/include/asm/mmu_context.h @@ -80,7 +80,7 @@ static inline void check_and_switch_context(struct mm_struct *mm, #ifndef MODULE #define finish_arch_post_lock_switch \ finish_arch_post_lock_switch -static inline void finish_arch_post_lock_switch(void) +static __always_inline void finish_arch_post_lock_switch(void) { struct mm_struct *mm = current->mm; diff --git a/arch/riscv/include/asm/sync_core.h b/arch/riscv/include/asm/sync_core.h index 9153016da8f1..2fe6b7fe6b12 100644 --- a/arch/riscv/include/asm/sync_core.h +++ b/arch/riscv/include/asm/sync_core.h @@ -6,7 +6,7 @@ * RISC-V implements return to user-space through an xRET instruction, * which is not core serializing. */ -static inline void sync_core_before_usermode(void) +static __always_inline void sync_core_before_usermode(void) { asm volatile ("fence.i" ::: "memory"); } diff --git a/arch/s390/include/asm/mmu_context.h b/arch/s390/include/asm/mmu_context.h index d9b8501bc93d..c124ef6a01b3 100644 --- a/arch/s390/include/asm/mmu_context.h +++ b/arch/s390/include/asm/mmu_context.h @@ -97,7 +97,7 @@ static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next, } #define finish_arch_post_lock_switch finish_arch_post_lock_switch -static inline void finish_arch_post_lock_switch(void) +static __always_inline void finish_arch_post_lock_switch(void) { struct task_struct *tsk = current; struct mm_struct *mm = tsk->mm; diff --git a/arch/sparc/include/asm/mmu_context_64.h b/arch/sparc/include/asm/mmu_context_64.h index 78bbacc14d2d..d1967214ef25 100644 --- a/arch/sparc/include/asm/mmu_context_64.h +++ b/arch/sparc/include/asm/mmu_context_64.h @@ -160,7 +160,7 @@ static inline void arch_start_context_switch(struct task_struct *prev) } #define finish_arch_post_lock_switch finish_arch_post_lock_switch -static inline void finish_arch_post_lock_switch(void) +static __always_inline void finish_arch_post_lock_switch(void) { /* Restore the state of MCDPER register for the new process * just switched to. diff --git a/arch/x86/include/asm/mmu_context.h b/arch/x86/include/asm/mmu_context.h index 73bf3b1b44e8..ecd134dcfb34 100644 --- a/arch/x86/include/asm/mmu_context.h +++ b/arch/x86/include/asm/mmu_context.h @@ -136,8 +136,29 @@ static inline void mm_reset_untag_mask(struct mm_struct *mm) } #endif +/* + * Please ignore the name of this function. It should be called + * switch_to_kernel_thread(). + * + * enter_lazy_tlb() is a hint from the scheduler that we are entering a + * kernel thread or other context without an mm. Acceptable implementations + * include doing nothing whatsoever, switching to init_mm, or various clever + * lazy tricks to try to minimize TLB flushes. + * + * The scheduler reserves the right to call enter_lazy_tlb() several times + * in a row. It will notify us that we're going back to a real mm by + * calling switch_mm_irqs_off(). + */ #define enter_lazy_tlb enter_lazy_tlb -extern void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk); +#ifndef MODULE +static __always_inline void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk) +{ + if (this_cpu_read(cpu_tlbstate.loaded_mm) == &init_mm) + return; + + this_cpu_write(cpu_tlbstate_shared.is_lazy, true); +} +#endif #define mm_init_global_asid mm_init_global_asid extern void mm_init_global_asid(struct mm_struct *mm); diff --git a/arch/x86/include/asm/sync_core.h b/arch/x86/include/asm/sync_core.h index 96bda43538ee..4b55fa353bb5 100644 --- a/arch/x86/include/asm/sync_core.h +++ b/arch/x86/include/asm/sync_core.h @@ -93,7 +93,7 @@ static __always_inline void sync_core(void) * to user-mode. x86 implements return to user-space through sysexit, * sysrel, and sysretq, which are not core serializing. */ -static inline void sync_core_before_usermode(void) +static __always_inline void sync_core_before_usermode(void) { /* With PTI, we unconditionally serialize before running user code. */ if (static_cpu_has(X86_FEATURE_PTI)) diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c index bc94ff1e250a..fcc1a954ccb3 100644 --- a/arch/x86/kernel/cpu/amd.c +++ b/arch/x86/kernel/cpu/amd.c @@ -1035,6 +1035,7 @@ static void init_amd_zen4(struct cpuinfo_x86 *c) } } +#ifndef CONFIG_CACHY static const struct x86_cpu_id zen5_rdseed_microcode[] = { ZEN_MODEL_STEP_UCODE(0x1a, 0x02, 0x1, 0x0b00215a), ZEN_MODEL_STEP_UCODE(0x1a, 0x08, 0x1, 0x0b008121), @@ -1056,6 +1057,11 @@ static void init_amd_zen5(struct cpuinfo_x86 *c) pr_emerg_once("RDSEED32 is broken. Disabling the corresponding CPUID bit.\n"); } } +#else +static void init_amd_zen5(struct cpuinfo_x86 *c) +{ +} +#endif /* !CONFIG_CACHY */ static void init_amd(struct cpuinfo_x86 *c) { diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c index f5b93e01e347..71abaf0bdb91 100644 --- a/arch/x86/mm/tlb.c +++ b/arch/x86/mm/tlb.c @@ -971,27 +971,6 @@ void switch_mm_irqs_off(struct mm_struct *unused, struct mm_struct *next, } } -/* - * Please ignore the name of this function. It should be called - * switch_to_kernel_thread(). - * - * enter_lazy_tlb() is a hint from the scheduler that we are entering a - * kernel thread or other context without an mm. Acceptable implementations - * include doing nothing whatsoever, switching to init_mm, or various clever - * lazy tricks to try to minimize TLB flushes. - * - * The scheduler reserves the right to call enter_lazy_tlb() several times - * in a row. It will notify us that we're going back to a real mm by - * calling switch_mm_irqs_off(). - */ -void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk) -{ - if (this_cpu_read(cpu_tlbstate.loaded_mm) == &init_mm) - return; - - this_cpu_write(cpu_tlbstate_shared.is_lazy, true); -} - /* * Using a temporary mm allows to set temporary mappings that are not accessible * by other CPUs. Such mappings are needed to perform sensitive memory writes diff --git a/drivers/bluetooth/btusb.c b/drivers/bluetooth/btusb.c index 80ccfa8fd982..0d5eae8295d4 100644 --- a/drivers/bluetooth/btusb.c +++ b/drivers/bluetooth/btusb.c @@ -559,6 +559,8 @@ static const struct usb_device_id quirks_table[] = { BTUSB_WIDEBAND_SPEECH }, { USB_DEVICE(0x13d3, 0x3592), .driver_info = BTUSB_REALTEK | BTUSB_WIDEBAND_SPEECH }, + { USB_DEVICE(0x0489, 0xe112), .driver_info = BTUSB_REALTEK | + BTUSB_WIDEBAND_SPEECH }, { USB_DEVICE(0x0489, 0xe122), .driver_info = BTUSB_REALTEK | BTUSB_WIDEBAND_SPEECH }, @@ -669,6 +671,8 @@ static const struct usb_device_id quirks_table[] = { BTUSB_WIDEBAND_SPEECH }, /* MediaTek MT7922 Bluetooth devices */ + { USB_DEVICE(0x13d3, 0x3625), .driver_info = BTUSB_MEDIATEK | + BTUSB_WIDEBAND_SPEECH }, { USB_DEVICE(0x13d3, 0x3585), .driver_info = BTUSB_MEDIATEK | BTUSB_WIDEBAND_SPEECH }, { USB_DEVICE(0x13d3, 0x3610), .driver_info = BTUSB_MEDIATEK | diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index d2c3885de711..8900e0dc8a61 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5068,7 +5068,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) * before ip_fini_early to prevent kfd locking refcount issues by calling * amdgpu_amdkfd_suspend() */ - if (drm_dev_is_unplugged(adev_to_drm(adev))) + if (pci_dev_is_disconnected(adev->pdev)) amdgpu_amdkfd_device_fini_sw(adev); amdgpu_device_ip_fini_early(adev); @@ -5080,7 +5080,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) amdgpu_gart_dummy_page_fini(adev); - if (drm_dev_is_unplugged(adev_to_drm(adev))) + if (pci_dev_is_disconnected(adev->pdev)) amdgpu_device_unmap_mmio(adev); } diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c index e5e993d3ef74..45b34ba0927e 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c @@ -1383,6 +1383,7 @@ static bool dm_is_freesync_pcon_whitelist(const uint32_t branch_dev_id) case DP_BRANCH_DEVICE_ID_0060AD: case DP_BRANCH_DEVICE_ID_00E04C: case DP_BRANCH_DEVICE_ID_90CC24: + case DP_BRANCH_DEVICE_ID_2B02F0: ret_val = true; break; default: diff --git a/drivers/gpu/drm/amd/display/include/ddc_service_types.h b/drivers/gpu/drm/amd/display/include/ddc_service_types.h index 1c603b12957f..e838f7c1269c 100644 --- a/drivers/gpu/drm/amd/display/include/ddc_service_types.h +++ b/drivers/gpu/drm/amd/display/include/ddc_service_types.h @@ -36,6 +36,7 @@ #define DP_BRANCH_DEVICE_ID_006037 0x006037 #define DP_BRANCH_DEVICE_ID_001CF8 0x001CF8 #define DP_BRANCH_DEVICE_ID_0060AD 0x0060AD +#define DP_BRANCH_DEVICE_ID_2B02F0 0x2B02F0 /* Chrontel CH7218 */ #define DP_BRANCH_HW_REV_10 0x10 #define DP_BRANCH_HW_REV_20 0x20 diff --git a/drivers/gpu/drm/i915/gt/intel_rc6.c b/drivers/gpu/drm/i915/gt/intel_rc6.c index 286d49ecc449..472c68b32516 100644 --- a/drivers/gpu/drm/i915/gt/intel_rc6.c +++ b/drivers/gpu/drm/i915/gt/intel_rc6.c @@ -2,9 +2,9 @@ /* * Copyright © 2019 Intel Corporation */ - #include #include +#include #include @@ -497,6 +497,17 @@ static bool rc6_supported(struct intel_rc6 *rc6) if (!HAS_RC6(i915)) return false; + if (dmi_match(DMI_PRODUCT_SKU, "IBP1XI08MK1") || + dmi_match(DMI_PRODUCT_SKU, "IBP14I08MK2") || + dmi_match(DMI_PRODUCT_SKU, "IBP1XI08MK2") || + dmi_match(DMI_PRODUCT_SKU, "IBP16I08MK2") || + dmi_match(DMI_PRODUCT_SKU, "OMNIA08IMK1") || + dmi_match(DMI_PRODUCT_SKU, "OMNIA08IMK2")) { + drm_notice(&i915->drm, + "RC6 disabled by quirk\n"); + return false; + } + if (intel_vgpu_active(i915)) return false; diff --git a/drivers/usb/core/quirks.c b/drivers/usb/core/quirks.c index c4d85089d19b..f4c80063864f 100644 --- a/drivers/usb/core/quirks.c +++ b/drivers/usb/core/quirks.c @@ -377,6 +377,9 @@ static const struct usb_device_id usb_quirk_list[] = { /* SanDisk Extreme 55AE */ { USB_DEVICE(0x0781, 0x55ae), .driver_info = USB_QUIRK_NO_LPM }, + /* Avermedia Live Gamer Ultra 2.1 (GC553G2) - BOS descriptor fetch hangs at SuperSpeed Plus */ + { USB_DEVICE(0x07ca, 0x2553), .driver_info = USB_QUIRK_NO_BOS }, + /* Realforce 87U Keyboard */ { USB_DEVICE(0x0853, 0x011b), .driver_info = USB_QUIRK_NO_LPM }, @@ -437,6 +440,9 @@ static const struct usb_device_id usb_quirk_list[] = { { USB_DEVICE(0x0b05, 0x17e0), .driver_info = USB_QUIRK_IGNORE_REMOTE_WAKEUP }, + /* ASUS TUF 4K PRO - BOS descriptor fetch hangs at SuperSpeed Plus */ + { USB_DEVICE(0x0b05, 0x1ab9), .driver_info = USB_QUIRK_NO_BOS }, + /* Realtek Semiconductor Corp. Mass Storage Device (Multicard Reader)*/ { USB_DEVICE(0x0bda, 0x0151), .driver_info = USB_QUIRK_CONFIG_INTF_STRINGS }, @@ -565,6 +571,9 @@ static const struct usb_device_id usb_quirk_list[] = { { USB_DEVICE(0x2386, 0x350e), .driver_info = USB_QUIRK_NO_LPM }, + /* UGREEN 35871 - BOS descriptor fetch hangs at SuperSpeed Plus */ + { USB_DEVICE(0x2b89, 0x5871), .driver_info = USB_QUIRK_NO_BOS }, + /* APTIV AUTOMOTIVE HUB */ { USB_DEVICE(0x2c48, 0x0132), .driver_info = USB_QUIRK_SHORT_SET_ADDRESS_REQ_TIMEOUT }, diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 9ded2e582c60..fbe94b65b30a 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -1624,7 +1624,7 @@ static inline void perf_event_task_migrate(struct task_struct *task) task->sched_migrated = 1; } -static inline void perf_event_task_sched_in(struct task_struct *prev, +static __always_inline void perf_event_task_sched_in(struct task_struct *prev, struct task_struct *task) { if (static_branch_unlikely(&perf_sched_events)) diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index 95d0040df584..4a279ee2d026 100644 --- a/include/linux/sched/mm.h +++ b/include/linux/sched/mm.h @@ -44,7 +44,7 @@ static inline void smp_mb__after_mmgrab(void) extern void __mmdrop(struct mm_struct *mm); -static inline void mmdrop(struct mm_struct *mm) +static __always_inline void mmdrop(struct mm_struct *mm) { /* * The implicit full barrier implied by atomic_dec_and_test() is @@ -71,14 +71,14 @@ static inline void __mmdrop_delayed(struct rcu_head *rhp) * Invoked from finish_task_switch(). Delegates the heavy lifting on RT * kernels via RCU. */ -static inline void mmdrop_sched(struct mm_struct *mm) +static __always_inline void mmdrop_sched(struct mm_struct *mm) { /* Provides a full memory barrier. See mmdrop() */ if (atomic_dec_and_test(&mm->mm_count)) call_rcu(&mm->delayed_drop, __mmdrop_delayed); } #else -static inline void mmdrop_sched(struct mm_struct *mm) +static __always_inline void mmdrop_sched(struct mm_struct *mm) { mmdrop(mm); } @@ -104,7 +104,7 @@ static inline void mmdrop_lazy_tlb(struct mm_struct *mm) } } -static inline void mmdrop_lazy_tlb_sched(struct mm_struct *mm) +static __always_inline void mmdrop_lazy_tlb_sched(struct mm_struct *mm) { if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) mmdrop_sched(mm); @@ -532,7 +532,7 @@ enum { #include #endif -static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm) +static __always_inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm) { /* * The atomic_read() below prevents CSE. The following should diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h index 45c0022b91ce..bbcfdf12aa6e 100644 --- a/include/linux/sched/topology.h +++ b/include/linux/sched/topology.h @@ -92,9 +92,6 @@ struct sched_domain { unsigned int nr_balance_failed; /* initialise to 0 */ /* idle_balance() stats */ - unsigned int newidle_call; - unsigned int newidle_success; - unsigned int newidle_ratio; u64 max_newidle_lb_cost; unsigned long last_decay_max_lb_cost; diff --git a/include/linux/tick.h b/include/linux/tick.h index ac76ae9fa36d..fce16aa10ba2 100644 --- a/include/linux/tick.h +++ b/include/linux/tick.h @@ -175,7 +175,7 @@ extern cpumask_var_t tick_nohz_full_mask; #ifdef CONFIG_NO_HZ_FULL extern bool tick_nohz_full_running; -static inline bool tick_nohz_full_enabled(void) +static __always_inline bool tick_nohz_full_enabled(void) { if (!context_tracking_enabled()) return false; @@ -299,7 +299,7 @@ static inline void __tick_nohz_task_switch(void) { } static inline void tick_nohz_full_setup(cpumask_var_t cpumask) { } #endif -static inline void tick_nohz_task_switch(void) +static __always_inline void tick_nohz_task_switch(void) { if (tick_nohz_full_enabled()) __tick_nohz_task_switch(); diff --git a/include/linux/vtime.h b/include/linux/vtime.h index 29dd5b91dd7d..428464bb81b3 100644 --- a/include/linux/vtime.h +++ b/include/linux/vtime.h @@ -67,24 +67,24 @@ static __always_inline void vtime_account_guest_exit(void) * For now vtime state is tied to context tracking. We might want to decouple * those later if necessary. */ -static inline bool vtime_accounting_enabled(void) +static __always_inline bool vtime_accounting_enabled(void) { return context_tracking_enabled(); } -static inline bool vtime_accounting_enabled_cpu(int cpu) +static __always_inline bool vtime_accounting_enabled_cpu(int cpu) { return context_tracking_enabled_cpu(cpu); } -static inline bool vtime_accounting_enabled_this_cpu(void) +static __always_inline bool vtime_accounting_enabled_this_cpu(void) { return context_tracking_enabled_this_cpu(); } extern void vtime_task_switch_generic(struct task_struct *prev); -static inline void vtime_task_switch(struct task_struct *prev) +static __always_inline void vtime_task_switch(struct task_struct *prev) { if (vtime_accounting_enabled_this_cpu()) vtime_task_switch_generic(prev); diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index 293bbd9ac3f4..2c1c9759e278 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -160,6 +160,7 @@ static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, unsigned long gps, unsigned long flags); static void invoke_rcu_core(void); static void rcu_report_exp_rdp(struct rcu_data *rdp); +static void rcu_report_qs_rdp(struct rcu_data *rdp); static void check_cb_ovld_locked(struct rcu_data *rdp, struct rcu_node *rnp); static bool rcu_rdp_is_offloaded(struct rcu_data *rdp); static bool rcu_rdp_cpu_online(struct rcu_data *rdp); @@ -1983,6 +1984,17 @@ static noinline_for_stack bool rcu_gp_init(void) if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) on_each_cpu(rcu_strict_gp_boundary, NULL, 0); + /* + * Immediately report QS for the GP kthread's CPU. The GP kthread + * cannot be in an RCU read-side critical section while running + * the FQS scan. This eliminates the need for a second FQS wait + * when all CPUs are idle. + */ + preempt_disable(); + rcu_qs(); + rcu_report_qs_rdp(this_cpu_ptr(&rcu_data)); + preempt_enable(); + return true; } diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 61c2d65156b5..7224d5fe3355 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -121,7 +121,6 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); EXPORT_TRACEPOINT_SYMBOL_GPL(sched_compute_energy_tp); DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); -DEFINE_PER_CPU(struct rnd_state, sched_rnd_state); #ifdef CONFIG_SCHED_PROXY_EXEC DEFINE_STATIC_KEY_TRUE(__sched_proxy_exec); @@ -678,11 +677,6 @@ bool raw_spin_rq_trylock(struct rq *rq) } } -void raw_spin_rq_unlock(struct rq *rq) -{ - raw_spin_unlock(rq_lockp(rq)); -} - /* * double_rq_lock - safely lock two runqueues */ @@ -4869,7 +4863,7 @@ static inline void prepare_task(struct task_struct *next) WRITE_ONCE(next->on_cpu, 1); } -static inline void finish_task(struct task_struct *prev) +static __always_inline void finish_task(struct task_struct *prev) { /* * This must be the very last reference to @prev from this CPU. After @@ -4885,7 +4879,7 @@ static inline void finish_task(struct task_struct *prev) smp_store_release(&prev->on_cpu, 0); } -static void do_balance_callbacks(struct rq *rq, struct balance_callback *head) +static __always_inline void do_balance_callbacks(struct rq *rq, struct balance_callback *head) { void (*func)(struct rq *rq); struct balance_callback *next; @@ -4920,7 +4914,7 @@ struct balance_callback balance_push_callback = { .func = balance_push, }; -static inline struct balance_callback * +static __always_inline struct balance_callback * __splice_balance_callbacks(struct rq *rq, bool split) { struct balance_callback *head = rq->balance_callback; @@ -4950,7 +4944,7 @@ struct balance_callback *splice_balance_callbacks(struct rq *rq) return __splice_balance_callbacks(rq, true); } -void __balance_callbacks(struct rq *rq, struct rq_flags *rf) +__always_inline void __balance_callbacks(struct rq *rq, struct rq_flags *rf) { if (rf) rq_unpin_lock(rq, rf); @@ -4987,7 +4981,7 @@ prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf #endif } -static inline void finish_lock_switch(struct rq *rq) +static __always_inline void finish_lock_switch(struct rq *rq) { /* * If we are tracking spinlock dependencies then we have to @@ -5019,7 +5013,7 @@ static inline void kmap_local_sched_out(void) #endif } -static inline void kmap_local_sched_in(void) +static __always_inline void kmap_local_sched_in(void) { #ifdef CONFIG_KMAP_LOCAL if (unlikely(current->kmap_ctrl.idx)) @@ -5072,7 +5066,7 @@ prepare_task_switch(struct rq *rq, struct task_struct *prev, * past. 'prev == current' is still correct but we need to recalculate this_rq * because prev may have moved to another CPU. */ -static struct rq *finish_task_switch(struct task_struct *prev) +static __always_inline struct rq *finish_task_switch(struct task_struct *prev) __releases(rq->lock) { struct rq *rq = this_rq(); @@ -8491,8 +8485,6 @@ void __init sched_init_smp(void) { sched_init_numa(NUMA_NO_NODE); - prandom_init_once(&sched_rnd_state); - /* * There's no userspace yet to cause hotplug operations; hence all the * CPU masks are stable and all blatant races in the below code cannot diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c index 0bb8fa927e9e..94d75bf0c119 100644 --- a/kernel/sched/ext.c +++ b/kernel/sched/ext.c @@ -3488,8 +3488,8 @@ static void destroy_dsq(struct scx_sched *sch, u64 dsq_id) * operations inside scheduler locks. */ dsq->id = SCX_DSQ_INVALID; - llist_add(&dsq->free_node, &dsqs_to_free); - irq_work_queue(&free_dsq_irq_work); + if (llist_add(&dsq->free_node, &dsqs_to_free)) + irq_work_queue(&free_dsq_irq_work); out_unlock_dsq: raw_spin_unlock_irqrestore(&dsq->lock, flags); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 66289244f71e..b859f1f314cf 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -2471,11 +2471,8 @@ static void task_numa_find_cpu(struct task_numa_env *env, maymove = !load_too_imbalanced(src_load, dst_load, env); } - for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) { - /* Skip this CPU if the source task cannot migrate */ - if (!cpumask_test_cpu(cpu, env->p->cpus_ptr)) - continue; - + /* Skip CPUs if the source task cannot migrate */ + for_each_cpu_and(cpu, cpumask_of_node(env->dst_nid), env->p->cpus_ptr) { env->dst_cpu = cpu; if (task_numa_compare(env, taskimp, groupimp, maymove)) break; @@ -8391,9 +8388,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu) int max_spare_cap_cpu = -1; int fits, max_fits = -1; - cpumask_and(cpus, perf_domain_span(pd), cpu_online_mask); - - if (cpumask_empty(cpus)) + if (!cpumask_and(cpus, perf_domain_span(pd), cpu_online_mask)) continue; /* Account external pressure for the energy estimation */ @@ -12241,27 +12236,11 @@ void update_max_interval(void) max_load_balance_interval = HZ*num_online_cpus()/10; } -static inline void update_newidle_stats(struct sched_domain *sd, unsigned int success) -{ - sd->newidle_call++; - sd->newidle_success += success; - - if (sd->newidle_call >= 1024) { - sd->newidle_ratio = sd->newidle_success; - sd->newidle_call /= 2; - sd->newidle_success /= 2; - } -} - -static inline bool -update_newidle_cost(struct sched_domain *sd, u64 cost, unsigned int success) +static inline bool update_newidle_cost(struct sched_domain *sd, u64 cost) { unsigned long next_decay = sd->last_decay_max_lb_cost + HZ; unsigned long now = jiffies; - if (cost) - update_newidle_stats(sd, success); - if (cost > sd->max_newidle_lb_cost) { /* * Track max cost of a domain to make sure to not delay the @@ -12309,7 +12288,7 @@ static void sched_balance_domains(struct rq *rq, enum cpu_idle_type idle) * Decay the newidle max times here because this is a regular * visit to all the domains. */ - need_decay = update_newidle_cost(sd, 0, 0); + need_decay = update_newidle_cost(sd, 0); max_cost += sd->max_newidle_lb_cost; /* @@ -12945,22 +12924,6 @@ static int sched_balance_newidle(struct rq *this_rq, struct rq_flags *rf) break; if (sd->flags & SD_BALANCE_NEWIDLE) { - unsigned int weight = 1; - - if (sched_feat(NI_RANDOM)) { - /* - * Throw a 1k sided dice; and only run - * newidle_balance according to the success - * rate. - */ - u32 d1k = sched_rng() % 1024; - weight = 1 + sd->newidle_ratio; - if (d1k > weight) { - update_newidle_stats(sd, 0); - continue; - } - weight = (1024 + weight/2) / weight; - } pulled_task = sched_balance_rq(this_cpu, this_rq, sd, CPU_NEWLY_IDLE, @@ -12968,14 +12931,10 @@ static int sched_balance_newidle(struct rq *this_rq, struct rq_flags *rf) t1 = sched_clock_cpu(this_cpu); domain_cost = t1 - t0; + update_newidle_cost(sd, domain_cost); + curr_cost += domain_cost; t0 = t1; - - /* - * Track max cost of a domain to make sure to not delay the - * next wakeup on the CPU. - */ - update_newidle_cost(sd, domain_cost, weight * !!pulled_task); } /* diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 136a6584be79..3c12d9f93331 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -121,8 +121,3 @@ SCHED_FEAT(WA_BIAS, true) SCHED_FEAT(UTIL_EST, true) SCHED_FEAT(LATENCY_WARN, false) - -/* - * Do newidle balancing proportional to its success rate using randomization. - */ -SCHED_FEAT(NI_RANDOM, true) diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 7294157bfadc..18503287e561 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -5,7 +5,6 @@ #ifndef _KERNEL_SCHED_SCHED_H #define _KERNEL_SCHED_SCHED_H -#include #include #include #include @@ -1351,12 +1350,6 @@ static inline bool is_migration_disabled(struct task_struct *p) } DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); -DECLARE_PER_CPU(struct rnd_state, sched_rnd_state); - -static inline u32 sched_rng(void) -{ - return prandom_u32_state(this_cpu_ptr(&sched_rnd_state)); -} #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) #define this_rq() this_cpu_ptr(&runqueues) @@ -1403,12 +1396,12 @@ static inline struct cpumask *sched_group_span(struct sched_group *sg); DECLARE_STATIC_KEY_FALSE(__sched_core_enabled); -static inline bool sched_core_enabled(struct rq *rq) +static __always_inline bool sched_core_enabled(struct rq *rq) { return static_branch_unlikely(&__sched_core_enabled) && rq->core_enabled; } -static inline bool sched_core_disabled(void) +static __always_inline bool sched_core_disabled(void) { return !static_branch_unlikely(&__sched_core_enabled); } @@ -1417,7 +1410,7 @@ static inline bool sched_core_disabled(void) * Be careful with this function; not for general use. The return value isn't * stable unless you actually hold a relevant rq->__lock. */ -static inline raw_spinlock_t *rq_lockp(struct rq *rq) +static __always_inline raw_spinlock_t *rq_lockp(struct rq *rq) { if (sched_core_enabled(rq)) return &rq->core->__lock; @@ -1425,7 +1418,7 @@ static inline raw_spinlock_t *rq_lockp(struct rq *rq) return &rq->__lock; } -static inline raw_spinlock_t *__rq_lockp(struct rq *rq) +static __always_inline raw_spinlock_t *__rq_lockp(struct rq *rq) { if (rq->core_enabled) return &rq->core->__lock; @@ -1519,12 +1512,12 @@ static inline bool sched_core_disabled(void) return true; } -static inline raw_spinlock_t *rq_lockp(struct rq *rq) +static __always_inline raw_spinlock_t *rq_lockp(struct rq *rq) { return &rq->__lock; } -static inline raw_spinlock_t *__rq_lockp(struct rq *rq) +static __always_inline raw_spinlock_t *__rq_lockp(struct rq *rq) { return &rq->__lock; } @@ -1573,20 +1566,24 @@ static inline void lockdep_assert_rq_held(struct rq *rq) extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass); extern bool raw_spin_rq_trylock(struct rq *rq); -extern void raw_spin_rq_unlock(struct rq *rq); -static inline void raw_spin_rq_lock(struct rq *rq) +static __always_inline void raw_spin_rq_lock(struct rq *rq) { raw_spin_rq_lock_nested(rq, 0); } -static inline void raw_spin_rq_lock_irq(struct rq *rq) +static __always_inline void raw_spin_rq_unlock(struct rq *rq) +{ + raw_spin_unlock(rq_lockp(rq)); +} + +static __always_inline void raw_spin_rq_lock_irq(struct rq *rq) { local_irq_disable(); raw_spin_rq_lock(rq); } -static inline void raw_spin_rq_unlock_irq(struct rq *rq) +static __always_inline void raw_spin_rq_unlock_irq(struct rq *rq) { raw_spin_rq_unlock(rq); local_irq_enable(); diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index cf643a5ddedd..711076aa4980 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -1669,12 +1669,6 @@ sd_init(struct sched_domain_topology_level *tl, .last_balance = jiffies, .balance_interval = sd_weight, - - /* 50% success rate */ - .newidle_call = 512, - .newidle_success = 256, - .newidle_ratio = 512, - .max_newidle_lb_cost = 0, .last_decay_max_lb_cost = jiffies, .child = child, diff --git a/sound/hda/codecs/realtek/alc269.c b/sound/hda/codecs/realtek/alc269.c index b66965a52107..9afa52ff5e0c 100644 --- a/sound/hda/codecs/realtek/alc269.c +++ b/sound/hda/codecs/realtek/alc269.c @@ -7057,6 +7057,7 @@ static const struct hda_quirk alc269_fixup_tbl[] = { SND_PCI_QUIRK(0x144d, 0xc886, "Samsung Galaxy Book3 Pro (NP964XFG)", ALC298_FIXUP_SAMSUNG_AMP_V2_4_AMPS), SND_PCI_QUIRK(0x144d, 0xc1ca, "Samsung Galaxy Book3 Pro 360 (NP960QFG)", ALC298_FIXUP_SAMSUNG_AMP_V2_4_AMPS), SND_PCI_QUIRK(0x144d, 0xc1cc, "Samsung Galaxy Book3 Ultra (NT960XFH)", ALC298_FIXUP_SAMSUNG_AMP_V2_4_AMPS), + SND_PCI_QUIRK(0x1458, 0x900e, "Gigabyte G5 KF5 (2023)", ALC2XX_FIXUP_HEADSET_MIC), SND_PCI_QUIRK(0x1458, 0xfa53, "Gigabyte BXBT-2807", ALC283_FIXUP_HEADSET_MIC), SND_PCI_QUIRK(0x1462, 0xb120, "MSI Cubi MS-B120", ALC283_FIXUP_HEADSET_MIC), SND_PCI_QUIRK(0x1462, 0xb171, "Cubi N 8GL (MS-B171)", ALC283_FIXUP_HEADSET_MIC), -- 2.53.0 From e5bfbb3a8c17d42592f27cb4505b36fb09eb4857 Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Mon, 16 Feb 2026 13:51:26 +0100 Subject: [PATCH 5/7] t2 Signed-off-by: Peter Jung --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 + drivers/gpu/drm/i915/display/intel_ddi.c | 3 + drivers/gpu/drm/i915/display/intel_fbdev.c | 6 +- drivers/gpu/drm/i915/display/intel_quirks.c | 15 + drivers/gpu/drm/i915/display/intel_quirks.h | 1 + drivers/gpu/vga/vga_switcheroo.c | 7 +- drivers/hid/Kconfig | 10 +- drivers/hid/Makefile | 4 + drivers/hid/dockchannel-hid/Kconfig | 14 + drivers/hid/dockchannel-hid/Makefile | 6 + drivers/hid/dockchannel-hid/dockchannel-hid.c | 1213 +++++++++++++++++ drivers/hid/hid-apple.c | 45 +- drivers/hid/hid-core.c | 11 +- drivers/hid/hid-ids.h | 10 + drivers/hid/hid-magicmouse.c | 770 ++++++++++- drivers/hid/spi-hid/Kconfig | 26 + drivers/hid/spi-hid/Makefile | 10 + drivers/hid/spi-hid/spi-hid-apple-core.c | 1194 ++++++++++++++++ drivers/hid/spi-hid/spi-hid-apple-of.c | 153 +++ drivers/hid/spi-hid/spi-hid-apple.h | 35 + drivers/hwmon/applesmc.c | 1138 ++++++++++++---- drivers/pci/vgaarb.c | 1 + drivers/platform/x86/apple-gmux.c | 18 + drivers/soc/apple/Kconfig | 24 + drivers/soc/apple/Makefile | 6 + drivers/soc/apple/dockchannel.c | 406 ++++++ drivers/soc/apple/rtkit-helper.c | 151 ++ drivers/staging/Kconfig | 2 + drivers/staging/Makefile | 1 + drivers/staging/apple-bce/Kconfig | 18 + drivers/staging/apple-bce/Makefile | 28 + drivers/staging/apple-bce/apple_bce.c | 445 ++++++ drivers/staging/apple-bce/apple_bce.h | 41 + drivers/staging/apple-bce/audio/audio.c | 711 ++++++++++ drivers/staging/apple-bce/audio/audio.h | 125 ++ drivers/staging/apple-bce/audio/description.h | 42 + drivers/staging/apple-bce/audio/pcm.c | 308 +++++ drivers/staging/apple-bce/audio/pcm.h | 16 + drivers/staging/apple-bce/audio/protocol.c | 347 +++++ drivers/staging/apple-bce/audio/protocol.h | 147 ++ .../staging/apple-bce/audio/protocol_bce.c | 226 +++ .../staging/apple-bce/audio/protocol_bce.h | 72 + drivers/staging/apple-bce/mailbox.c | 155 +++ drivers/staging/apple-bce/mailbox.h | 53 + drivers/staging/apple-bce/queue.c | 415 ++++++ drivers/staging/apple-bce/queue.h | 177 +++ drivers/staging/apple-bce/queue_dma.c | 220 +++ drivers/staging/apple-bce/queue_dma.h | 50 + drivers/staging/apple-bce/vhci/command.h | 204 +++ drivers/staging/apple-bce/vhci/queue.c | 268 ++++ drivers/staging/apple-bce/vhci/queue.h | 76 ++ drivers/staging/apple-bce/vhci/transfer.c | 673 +++++++++ drivers/staging/apple-bce/vhci/transfer.h | 75 + drivers/staging/apple-bce/vhci/vhci.c | 763 +++++++++++ drivers/staging/apple-bce/vhci/vhci.h | 52 + include/linux/hid.h | 6 +- include/linux/soc/apple/dockchannel.h | 26 + 57 files changed, 10671 insertions(+), 351 deletions(-) create mode 100644 drivers/hid/dockchannel-hid/Kconfig create mode 100644 drivers/hid/dockchannel-hid/Makefile create mode 100644 drivers/hid/dockchannel-hid/dockchannel-hid.c create mode 100644 drivers/hid/spi-hid/Kconfig create mode 100644 drivers/hid/spi-hid/Makefile create mode 100644 drivers/hid/spi-hid/spi-hid-apple-core.c create mode 100644 drivers/hid/spi-hid/spi-hid-apple-of.c create mode 100644 drivers/hid/spi-hid/spi-hid-apple.h create mode 100644 drivers/soc/apple/dockchannel.c create mode 100644 drivers/soc/apple/rtkit-helper.c create mode 100644 drivers/staging/apple-bce/Kconfig create mode 100644 drivers/staging/apple-bce/Makefile create mode 100644 drivers/staging/apple-bce/apple_bce.c create mode 100644 drivers/staging/apple-bce/apple_bce.h create mode 100644 drivers/staging/apple-bce/audio/audio.c create mode 100644 drivers/staging/apple-bce/audio/audio.h create mode 100644 drivers/staging/apple-bce/audio/description.h create mode 100644 drivers/staging/apple-bce/audio/pcm.c create mode 100644 drivers/staging/apple-bce/audio/pcm.h create mode 100644 drivers/staging/apple-bce/audio/protocol.c create mode 100644 drivers/staging/apple-bce/audio/protocol.h create mode 100644 drivers/staging/apple-bce/audio/protocol_bce.c create mode 100644 drivers/staging/apple-bce/audio/protocol_bce.h create mode 100644 drivers/staging/apple-bce/mailbox.c create mode 100644 drivers/staging/apple-bce/mailbox.h create mode 100644 drivers/staging/apple-bce/queue.c create mode 100644 drivers/staging/apple-bce/queue.h create mode 100644 drivers/staging/apple-bce/queue_dma.c create mode 100644 drivers/staging/apple-bce/queue_dma.h create mode 100644 drivers/staging/apple-bce/vhci/command.h create mode 100644 drivers/staging/apple-bce/vhci/queue.c create mode 100644 drivers/staging/apple-bce/vhci/queue.h create mode 100644 drivers/staging/apple-bce/vhci/transfer.c create mode 100644 drivers/staging/apple-bce/vhci/transfer.h create mode 100644 drivers/staging/apple-bce/vhci/vhci.c create mode 100644 drivers/staging/apple-bce/vhci/vhci.h create mode 100644 include/linux/soc/apple/dockchannel.h diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 26b4dcc58471..9c3dc6af33e0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -2403,6 +2403,9 @@ static int amdgpu_pci_probe(struct pci_dev *pdev, int ret, retry = 0, i; bool supports_atomic = false; + if (vga_switcheroo_client_probe_defer(pdev)) + return -EPROBE_DEFER; + if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA || (pdev->class >> 8) == PCI_CLASS_DISPLAY_OTHER) { if (drm_firmware_drivers_only() && amdgpu_modeset == -1) diff --git a/drivers/gpu/drm/i915/display/intel_ddi.c b/drivers/gpu/drm/i915/display/intel_ddi.c index 002ccd47856d..c5d2e908f4a7 100644 --- a/drivers/gpu/drm/i915/display/intel_ddi.c +++ b/drivers/gpu/drm/i915/display/intel_ddi.c @@ -4890,6 +4890,9 @@ static bool intel_ddi_a_force_4_lanes(struct intel_digital_port *dig_port) if (dig_port->ddi_a_4_lanes) return false; + if (intel_has_quirk(display, QUIRK_DDI_A_FORCE_4_LANES)) + return true; + /* Broxton/Geminilake: Bspec says that DDI_A_4_LANES is the only * supported configuration */ diff --git a/drivers/gpu/drm/i915/display/intel_fbdev.c b/drivers/gpu/drm/i915/display/intel_fbdev.c index 44f4fcce526e..695eadc88eab 100644 --- a/drivers/gpu/drm/i915/display/intel_fbdev.c +++ b/drivers/gpu/drm/i915/display/intel_fbdev.c @@ -278,10 +278,10 @@ int intel_fbdev_driver_fbdev_probe(struct drm_fb_helper *helper, ifbdev->fb = NULL; if (fb && - (sizes->fb_width > fb->base.width || - sizes->fb_height > fb->base.height)) { + (sizes->fb_width != fb->base.width || + sizes->fb_height != fb->base.height)) { drm_dbg_kms(display->drm, - "BIOS fb too small (%dx%d), we require (%dx%d)," + "BIOS fb not valid (%dx%d), we require (%dx%d)," " releasing it\n", fb->base.width, fb->base.height, sizes->fb_width, sizes->fb_height); diff --git a/drivers/gpu/drm/i915/display/intel_quirks.c b/drivers/gpu/drm/i915/display/intel_quirks.c index d2e16b79d6be..85f26517771a 100644 --- a/drivers/gpu/drm/i915/display/intel_quirks.c +++ b/drivers/gpu/drm/i915/display/intel_quirks.c @@ -66,6 +66,18 @@ static void quirk_increase_ddi_disabled_time(struct intel_display *display) drm_info(display->drm, "Applying Increase DDI Disabled quirk\n"); } +/* + * In some cases, the firmware might not set the lane count to 4 (for example, + * when booting in some dual GPU Macs with the dGPU as the default GPU), this + * quirk is used to force it as otherwise it might not be possible to compute a + * valid link configuration. + */ +static void quirk_ddi_a_force_4_lanes(struct intel_display *display) +{ + intel_set_quirk(display, QUIRK_DDI_A_FORCE_4_LANES); + drm_info(display->drm, "Applying DDI A Forced 4 Lanes quirk\n"); +} + static void quirk_no_pps_backlight_power_hook(struct intel_display *display) { intel_set_quirk(display, QUIRK_NO_PPS_BACKLIGHT_POWER_HOOK); @@ -240,6 +252,9 @@ static struct intel_quirk intel_quirks[] = { /* Dell XPS 13 7390 2-in-1 */ { 0x8a12, 0x1028, 0x08b0, quirk_edp_limit_rate_hbr2 }, + + /* Apple MacBookPro15,1 */ + { 0x3e9b, 0x106b, 0x0176, quirk_ddi_a_force_4_lanes }, }; static const struct intel_dpcd_quirk intel_dpcd_quirks[] = { diff --git a/drivers/gpu/drm/i915/display/intel_quirks.h b/drivers/gpu/drm/i915/display/intel_quirks.h index 06da0e286c67..71b89e52ebe7 100644 --- a/drivers/gpu/drm/i915/display/intel_quirks.h +++ b/drivers/gpu/drm/i915/display/intel_quirks.h @@ -21,6 +21,7 @@ enum intel_quirk_id { QUIRK_NO_PPS_BACKLIGHT_POWER_HOOK, QUIRK_FW_SYNC_LEN, QUIRK_EDP_LIMIT_RATE_HBR2, + QUIRK_DDI_A_FORCE_4_LANES, }; void intel_init_quirks(struct intel_display *display); diff --git a/drivers/gpu/vga/vga_switcheroo.c b/drivers/gpu/vga/vga_switcheroo.c index 68e45a26e85f..7580172daade 100644 --- a/drivers/gpu/vga/vga_switcheroo.c +++ b/drivers/gpu/vga/vga_switcheroo.c @@ -438,12 +438,7 @@ find_active_client(struct list_head *head) bool vga_switcheroo_client_probe_defer(struct pci_dev *pdev) { if (pci_is_display(pdev)) { - /* - * apple-gmux is needed on pre-retina MacBook Pro - * to probe the panel if pdev is the inactive GPU. - */ - if (apple_gmux_present() && pdev != vga_default_device() && - !vgasr_priv.handler_flags) + if (apple_gmux_present() && !vgasr_priv.handler_flags) return true; } diff --git a/drivers/hid/Kconfig b/drivers/hid/Kconfig index 920a64b66b25..ecb3059c68fb 100644 --- a/drivers/hid/Kconfig +++ b/drivers/hid/Kconfig @@ -140,7 +140,7 @@ config HID_APPLE tristate "Apple {i,Power,Mac}Books" depends on LEDS_CLASS depends on NEW_LEDS - default !EXPERT + default !EXPERT || SPI_HID_APPLE help Support for some Apple devices which less or more break HID specification. @@ -727,11 +727,13 @@ config LOGIWHEELS_FF config HID_MAGICMOUSE tristate "Apple Magic Mouse/Trackpad multi-touch support" + default SPI_HID_APPLE help Support for the Apple Magic Mouse/Trackpad multi-touch. Say Y here if you want support for the multi-touch features of the - Apple Wireless "Magic" Mouse and the Apple Wireless "Magic" Trackpad. + Apple Wireless "Magic" Mouse, the Apple Wireless "Magic" Trackpad and + force touch Trackpads in Macbooks starting from 2015. config HID_MALTRON tristate "Maltron L90 keyboard" @@ -1448,4 +1450,8 @@ endif # HID source "drivers/hid/usbhid/Kconfig" +source "drivers/hid/spi-hid/Kconfig" + +source "drivers/hid/dockchannel-hid/Kconfig" + endif # HID_SUPPORT diff --git a/drivers/hid/Makefile b/drivers/hid/Makefile index 361a7daedeb8..397b6cf5b6c3 100644 --- a/drivers/hid/Makefile +++ b/drivers/hid/Makefile @@ -173,6 +173,10 @@ obj-$(CONFIG_INTEL_ISH_HID) += intel-ish-hid/ obj-$(CONFIG_AMD_SFH_HID) += amd-sfh-hid/ +obj-$(CONFIG_SPI_HID_APPLE_CORE) += spi-hid/ + +obj-$(CONFIG_HID_DOCKCHANNEL) += dockchannel-hid/ + obj-$(CONFIG_SURFACE_HID_CORE) += surface-hid/ obj-$(CONFIG_INTEL_THC_HID) += intel-thc-hid/ diff --git a/drivers/hid/dockchannel-hid/Kconfig b/drivers/hid/dockchannel-hid/Kconfig new file mode 100644 index 000000000000..8a81d551a83d --- /dev/null +++ b/drivers/hid/dockchannel-hid/Kconfig @@ -0,0 +1,14 @@ +# SPDX-License-Identifier: GPL-2.0-only OR MIT +menu "DockChannel HID support" + depends on APPLE_DOCKCHANNEL + +config HID_DOCKCHANNEL + tristate "HID over DockChannel transport layer for Apple Silicon SoCs" + default ARCH_APPLE + depends on APPLE_DOCKCHANNEL && INPUT && OF && HID + help + Say Y here if you use an M2 or later Apple Silicon based laptop. + The keyboard and touchpad are HID based devices connected via the + proprietary DockChannel interface. + +endmenu diff --git a/drivers/hid/dockchannel-hid/Makefile b/drivers/hid/dockchannel-hid/Makefile new file mode 100644 index 000000000000..7dba766b047f --- /dev/null +++ b/drivers/hid/dockchannel-hid/Makefile @@ -0,0 +1,6 @@ +# SPDX-License-Identifier: GPL-2.0-only OR MIT +# +# Makefile for DockChannel HID transport drivers +# + +obj-$(CONFIG_HID_DOCKCHANNEL) += dockchannel-hid.o diff --git a/drivers/hid/dockchannel-hid/dockchannel-hid.c b/drivers/hid/dockchannel-hid/dockchannel-hid.c new file mode 100644 index 000000000000..a712a724ded3 --- /dev/null +++ b/drivers/hid/dockchannel-hid/dockchannel-hid.c @@ -0,0 +1,1213 @@ +/* + * SPDX-License-Identifier: GPL-2.0 OR MIT + * + * Apple DockChannel HID transport driver + * + * Copyright The Asahi Linux Contributors + */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "../hid-ids.h" + +#define COMMAND_TIMEOUT_MS 1000 +#define START_TIMEOUT_MS 2000 + +#define MAX_INTERFACES 16 + +/* Data + checksum */ +#define MAX_PKT_SIZE (0xffff + 4) + +#define DCHID_CHANNEL_CMD 0x11 +#define DCHID_CHANNEL_REPORT 0x12 + +struct dchid_hdr { + u8 hdr_len; + u8 channel; + u16 length; + u8 seq; + u8 iface; + u16 pad; +} __packed; + +#define IFACE_COMM 0 + +#define FLAGS_GROUP GENMASK(7, 6) +#define FLAGS_REQ GENMASK(5, 0) + +#define REQ_SET_REPORT 0 +#define REQ_GET_REPORT 1 + +struct dchid_subhdr { + u8 flags; + u8 unk; + u16 length; + u32 retcode; +} __packed; + +#define EVENT_GPIO_CMD 0xa0 +#define EVENT_INIT 0xf0 +#define EVENT_READY 0xf1 + +struct dchid_init_hdr { + u8 type; + u8 unk1; + u8 unk2; + u8 iface; + char name[16]; + u8 more_packets; + u8 unkpad; +} __packed; + +#define INIT_HID_DESCRIPTOR 0 +#define INIT_GPIO_REQUEST 1 +#define INIT_TERMINATOR 2 +#define INIT_PRODUCT_NAME 7 + +#define CMD_RESET_INTERFACE 0x40 +#define CMD_SEND_FIRMWARE 0x95 +#define CMD_ENABLE_INTERFACE 0xb4 +#define CMD_ACK_GPIO_CMD 0xa1 + +struct dchid_init_block_hdr { + u16 type; + u16 length; +} __packed; + +#define MAX_GPIO_NAME 32 + +struct dchid_gpio_request { + u16 unk; + u16 id; + char name[MAX_GPIO_NAME]; +} __packed; + +struct dchid_gpio_cmd { + u8 type; + u8 iface; + u8 gpio; + u8 unk; + u8 cmd; +} __packed; + +struct dchid_gpio_ack { + u8 type; + u32 retcode; + u8 cmd[]; +} __packed; + +#define STM_REPORT_ID 0x10 +#define STM_REPORT_SERIAL 0x11 +#define STM_REPORT_KEYBTYPE 0x14 + +struct dchid_stm_id { + u8 unk; + u16 vendor_id; + u16 product_id; + u16 version_number; + u8 unk2; + u8 unk3; + u8 keyboard_type; + u8 serial_length; + /* Serial follows, but we grab it with a different report. */ +} __packed; + +#define FW_MAGIC 0x46444948 +#define FW_VER 1 + +struct fw_header { + u32 magic; + u32 version; + u32 hdr_length; + u32 data_length; + u32 iface_offset; +} __packed; + +struct dchid_work { + struct work_struct work; + struct dchid_iface *iface; + + struct dchid_hdr hdr; + u8 data[]; +}; + +struct dchid_iface { + struct dockchannel_hid *dchid; + struct hid_device *hid; + struct workqueue_struct *wq; + + bool creating; + struct work_struct create_work; + + int index; + const char *name; + const struct device_node *of_node; + + uint8_t tx_seq; + bool deferred; + bool starting; + bool open; + struct completion ready; + + void *hid_desc; + size_t hid_desc_len; + + struct gpio_desc *gpio; + char gpio_name[MAX_GPIO_NAME]; + int gpio_id; + + struct mutex out_mutex; + u32 out_flags; + int out_report; + u32 retcode; + void *resp_buf; + size_t resp_size; + struct completion out_complete; + + u32 keyboard_layout_id; +}; + +struct dockchannel_hid { + struct device *dev; + struct dockchannel *dc; + struct device_link *helper_link; + + bool id_ready; + struct dchid_stm_id device_id; + char serial[64]; + + struct dchid_iface *comm; + struct dchid_iface *ifaces[MAX_INTERFACES]; + + u8 pkt_buf[MAX_PKT_SIZE]; + + /* Workqueue to asynchronously create HID devices */ + struct workqueue_struct *new_iface_wq; +}; + +static ssize_t apple_layout_id_show(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct hid_device *hdev = to_hid_device(dev); + struct dchid_iface *iface = hdev->driver_data; + + return scnprintf(buf, PAGE_SIZE, "%d\n", iface->keyboard_layout_id); +} + +static DEVICE_ATTR_RO(apple_layout_id); + +static struct dchid_iface * +dchid_get_interface(struct dockchannel_hid *dchid, int index, const char *name) +{ + struct dchid_iface *iface; + + if (index >= MAX_INTERFACES) { + dev_err(dchid->dev, "Interface index %d out of range\n", index); + return NULL; + } + + if (dchid->ifaces[index]) + return dchid->ifaces[index]; + + iface = devm_kzalloc(dchid->dev, sizeof(struct dchid_iface), GFP_KERNEL); + if (!iface) + return NULL; + + iface->index = index; + iface->name = devm_kstrdup(dchid->dev, name, GFP_KERNEL); + iface->dchid = dchid; + iface->out_report= -1; + init_completion(&iface->out_complete); + init_completion(&iface->ready); + mutex_init(&iface->out_mutex); + iface->wq = alloc_ordered_workqueue("dchid-%s", WQ_MEM_RECLAIM, iface->name); + if (!iface->wq) + return NULL; + + /* Comm is not a HID subdevice */ + if (!strcmp(name, "comm")) { + dchid->ifaces[index] = iface; + return iface; + } + + iface->of_node = of_get_child_by_name(dchid->dev->of_node, name); + if (!iface->of_node) { + dev_warn(dchid->dev, "No OF node for subdevice %s, ignoring.", name); + return NULL; + } + + dchid->ifaces[index] = iface; + return iface; +} + +static u32 dchid_checksum(void *p, size_t length) +{ + u32 sum = 0; + + while (length >= 4) { + sum += get_unaligned_le32(p); + p += 4; + length -= 4; + } + + WARN_ON_ONCE(length); + return sum; +} + +static int dchid_send(struct dchid_iface *iface, u32 flags, void *msg, size_t size) +{ + u32 checksum = 0xffffffff; + size_t wsize = round_down(size, 4); + size_t tsize = size - wsize; + int ret; + struct { + struct dchid_hdr hdr; + struct dchid_subhdr sub; + } __packed h; + + memset(&h, 0, sizeof(h)); + h.hdr.hdr_len = sizeof(h.hdr); + h.hdr.channel = DCHID_CHANNEL_CMD; + h.hdr.length = round_up(size, 4) + sizeof(h.sub); + h.hdr.seq = iface->tx_seq; + h.hdr.iface = iface->index; + h.sub.flags = flags; + h.sub.length = size; + + ret = dockchannel_send(iface->dchid->dc, &h, sizeof(h)); + if (ret < 0) + return ret; + checksum -= dchid_checksum(&h, sizeof(h)); + + ret = dockchannel_send(iface->dchid->dc, msg, wsize); + if (ret < 0) + return ret; + checksum -= dchid_checksum(msg, wsize); + + if (tsize) { + u8 tail[4] = {0, 0, 0, 0}; + + memcpy(tail, msg + wsize, tsize); + ret = dockchannel_send(iface->dchid->dc, tail, sizeof(tail)); + if (ret < 0) + return ret; + checksum -= dchid_checksum(tail, sizeof(tail)); + } + + ret = dockchannel_send(iface->dchid->dc, &checksum, sizeof(checksum)); + if (ret < 0) + return ret; + + return 0; +} + +static int dchid_cmd(struct dchid_iface *iface, u32 type, u32 req, + void *data, size_t size, void *resp_buf, size_t resp_size) +{ + int ret; + int report_id = *(u8*)data; + + mutex_lock(&iface->out_mutex); + + WARN_ON(iface->out_report != -1); + iface->out_report = report_id; + iface->out_flags = FIELD_PREP(FLAGS_GROUP, type) | FIELD_PREP(FLAGS_REQ, req); + iface->resp_buf = resp_buf; + iface->resp_size = resp_size; + reinit_completion(&iface->out_complete); + + ret = dchid_send(iface, iface->out_flags, data, size); + if (ret < 0) + goto done; + + if (!wait_for_completion_timeout(&iface->out_complete, msecs_to_jiffies(COMMAND_TIMEOUT_MS))) { + dev_err(iface->dchid->dev, "output report 0x%x to iface %d (%s) timed out\n", + report_id, iface->index, iface->name); + ret = -ETIMEDOUT; + goto done; + } + + ret = iface->resp_size; + if (iface->retcode) { + dev_err(iface->dchid->dev, + "output report 0x%x to iface %d (%s) failed with err 0x%x\n", + report_id, iface->index, iface->name, iface->retcode); + ret = -EIO; + } + +done: + iface->tx_seq++; + iface->out_report = -1; + iface->out_flags = 0; + iface->resp_buf = NULL; + iface->resp_size = 0; + mutex_unlock(&iface->out_mutex); + return ret; +} + +static int dchid_comm_cmd(struct dockchannel_hid *dchid, void *cmd, size_t size) +{ + return dchid_cmd(dchid->comm, HID_FEATURE_REPORT, REQ_SET_REPORT, cmd, size, NULL, 0); +} + +static int dchid_enable_interface(struct dchid_iface *iface) +{ + u8 msg[] = { CMD_ENABLE_INTERFACE, iface->index }; + + return dchid_comm_cmd(iface->dchid, msg, sizeof(msg)); +} + +static int dchid_reset_interface(struct dchid_iface *iface, int state) +{ + u8 msg[] = { CMD_RESET_INTERFACE, 1, iface->index, state }; + + return dchid_comm_cmd(iface->dchid, msg, sizeof(msg)); +} + +static int dchid_send_firmware(struct dchid_iface *iface, void *firmware, size_t size) +{ + struct { + u8 cmd; + u8 unk1; + u8 unk2; + u8 iface; + u64 addr; + u32 size; + } __packed msg = { + .cmd = CMD_SEND_FIRMWARE, + .unk1 = 2, + .unk2 = 0, + .iface = iface->index, + .size = size, + }; + dma_addr_t addr; + void *buf = dmam_alloc_coherent(iface->dchid->dev, size, &addr, GFP_KERNEL); + + if (IS_ERR_OR_NULL(buf)) + return buf ? PTR_ERR(buf) : -ENOMEM; + + msg.addr = addr; + memcpy(buf, firmware, size); + wmb(); + + return dchid_comm_cmd(iface->dchid, &msg, sizeof(msg)); +} + +static int dchid_get_firmware(struct dchid_iface *iface, void **firmware, size_t *size) +{ + int ret; + const char *fw_name; + const struct firmware *fw; + struct fw_header *hdr; + u8 *fw_data; + + ret = of_property_read_string(iface->of_node, "firmware-name", &fw_name); + if (ret) { + /* Firmware is only for some devices */ + *firmware = NULL; + *size = 0; + return 0; + } + + ret = request_firmware(&fw, fw_name, iface->dchid->dev); + if (ret) + return ret; + + hdr = (struct fw_header *)fw->data; + + if (hdr->magic != FW_MAGIC || hdr->version != FW_VER || + hdr->hdr_length < sizeof(*hdr) || hdr->hdr_length > fw->size || + (hdr->hdr_length + (size_t)hdr->data_length) > fw->size || + hdr->iface_offset >= hdr->data_length) { + dev_warn(iface->dchid->dev, "%s: invalid firmware header\n", + fw_name); + ret = -EINVAL; + goto done; + } + + fw_data = devm_kmemdup(iface->dchid->dev, fw->data + hdr->hdr_length, + hdr->data_length, GFP_KERNEL); + if (!fw_data) { + ret = -ENOMEM; + goto done; + } + + if (hdr->iface_offset) + fw_data[hdr->iface_offset] = iface->index; + + *firmware = fw_data; + *size = hdr->data_length; + +done: + release_firmware(fw); + return ret; +} + +static int dchid_request_gpio(struct dchid_iface *iface) +{ + char prop_name[MAX_GPIO_NAME + 16]; + + if (iface->gpio) + return 0; + + dev_info(iface->dchid->dev, "Requesting GPIO %s#%d: %s\n", + iface->name, iface->gpio_id, iface->gpio_name); + + snprintf(prop_name, sizeof(prop_name), "apple,%s", iface->gpio_name); + + iface->gpio = devm_gpiod_get_index(iface->dchid->dev, prop_name, 0, GPIOD_OUT_LOW); + + if (IS_ERR_OR_NULL(iface->gpio)) { + dev_err(iface->dchid->dev, "Failed to request GPIO %s-gpios\n", prop_name); + iface->gpio = NULL; + return -1; + } + + return 0; +} + +static int dchid_start_interface(struct dchid_iface *iface) +{ + void *fw; + size_t size; + int ret; + + if (iface->starting) { + dev_warn(iface->dchid->dev, "Interface %s is already starting", iface->name); + return -EINPROGRESS; + } + + dev_info(iface->dchid->dev, "Starting interface %s\n", iface->name); + + iface->starting = true; + + /* Look to see if we need firmware */ + ret = dchid_get_firmware(iface, &fw, &size); + if (ret < 0) + goto err; + + /* If we need a GPIO, make sure we have it. */ + if (iface->gpio_id) { + ret = dchid_request_gpio(iface); + if (ret < 0) + goto err; + } + + /* Only multi-touch has firmware */ + if (fw && size) { + + /* Send firmware to the device */ + dev_info(iface->dchid->dev, "Sending firmware for %s\n", iface->name); + ret = dchid_send_firmware(iface, fw, size); + if (ret < 0) { + dev_err(iface->dchid->dev, "Failed to send %s firmwareS", iface->name); + goto err; + } + + /* After loading firmware, multi-touch needs a reset */ + dev_info(iface->dchid->dev, "Resetting %s\n", iface->name); + dchid_reset_interface(iface, 0); + dchid_reset_interface(iface, 2); + } + + return 0; + +err: + iface->starting = false; + return ret; +} + +static int dchid_start(struct hid_device *hdev) +{ + struct dchid_iface *iface = hdev->driver_data; + + if (iface->keyboard_layout_id) { + int ret = device_create_file(&hdev->dev, &dev_attr_apple_layout_id); + if (ret) { + dev_warn(iface->dchid->dev, "Failed to create apple_layout_id: %d", ret); + iface->keyboard_layout_id = 0; + } + } + + return 0; +}; + +static void dchid_stop(struct hid_device *hdev) +{ + struct dchid_iface *iface = hdev->driver_data; + + if (iface->keyboard_layout_id) + device_remove_file(&hdev->dev, &dev_attr_apple_layout_id); +} + +static int dchid_open(struct hid_device *hdev) +{ + struct dchid_iface *iface = hdev->driver_data; + int ret; + + if (!completion_done(&iface->ready)) { + ret = dchid_start_interface(iface); + if (ret < 0) + return ret; + + if (!wait_for_completion_timeout(&iface->ready, msecs_to_jiffies(START_TIMEOUT_MS))) { + dev_err(iface->dchid->dev, "iface %s start timed out\n", iface->name); + return -ETIMEDOUT; + } + } + + iface->open = true; + return 0; +} + +static void dchid_close(struct hid_device *hdev) +{ + struct dchid_iface *iface = hdev->driver_data; + + iface->open = false; +} + +static int dchid_parse(struct hid_device *hdev) +{ + struct dchid_iface *iface = hdev->driver_data; + + return hid_parse_report(hdev, iface->hid_desc, iface->hid_desc_len); +} + +/* Note: buf excludes report number! For ease of fetching strings/etc. */ +static int dchid_get_report_cmd(struct dchid_iface *iface, u8 reportnum, void *buf, size_t len) +{ + int ret = dchid_cmd(iface, HID_FEATURE_REPORT, REQ_GET_REPORT, &reportnum, 1, buf, len); + + return ret <= 0 ? ret : ret - 1; +} + +/* Note: buf includes report number! */ +static int dchid_set_report(struct dchid_iface *iface, void *buf, size_t len) +{ + return dchid_cmd(iface, HID_OUTPUT_REPORT, REQ_SET_REPORT, buf, len, NULL, 0); +} + +static int dchid_raw_request(struct hid_device *hdev, + unsigned char reportnum, __u8 *buf, size_t len, + unsigned char rtype, int reqtype) +{ + struct dchid_iface *iface = hdev->driver_data; + + switch (reqtype) { + case HID_REQ_GET_REPORT: + buf[0] = reportnum; + return dchid_cmd(iface, rtype, REQ_GET_REPORT, &reportnum, 1, buf + 1, len - 1); + case HID_REQ_SET_REPORT: + return dchid_set_report(iface, buf, len); + default: + return -EIO; + } + + return 0; +} + +static struct hid_ll_driver dchid_ll = { + .start = &dchid_start, + .stop = &dchid_stop, + .open = &dchid_open, + .close = &dchid_close, + .parse = &dchid_parse, + .raw_request = &dchid_raw_request, +}; + +static void dchid_create_interface_work(struct work_struct *ws) +{ + struct dchid_iface *iface = container_of(ws, struct dchid_iface, create_work); + struct dockchannel_hid *dchid = iface->dchid; + struct hid_device *hid; + int ret; + + if (iface->hid) { + dev_warn(dchid->dev, "Interface %s already created!\n", + iface->name); + return; + } + + dev_info(dchid->dev, "New interface %s\n", iface->name); + + /* Start the interface. This is not the entire init process, as firmware is loaded later on device open. */ + ret = dchid_enable_interface(iface); + if (ret < 0) { + dev_warn(dchid->dev, "Failed to enable %s: %d\n", iface->name, ret); + return; + } + + iface->deferred = false; + + hid = hid_allocate_device(); + if (IS_ERR(hid)) + return; + + snprintf(hid->name, sizeof(hid->name), "Apple MTP %s", iface->name); + snprintf(hid->phys, sizeof(hid->phys), "%s.%d (%s)", + dev_name(dchid->dev), iface->index, iface->name); + strscpy(hid->uniq, dchid->serial, sizeof(hid->uniq)); + + hid->ll_driver = &dchid_ll; + hid->bus = BUS_HOST; + hid->vendor = dchid->device_id.vendor_id; + hid->product = dchid->device_id.product_id; + hid->version = dchid->device_id.version_number; + hid->type = HID_TYPE_OTHER; + if (!strcmp(iface->name, "multi-touch")) { + hid->type = HID_TYPE_SPI_MOUSE; + } else if (!strcmp(iface->name, "keyboard")) { + u32 country_code = 0; + + hid->type = HID_TYPE_SPI_KEYBOARD; + + /* + * We have to get the country code from the device tree, since the + * device provides no reliable way to get this info. + */ + if (!of_property_read_u32(iface->of_node, "hid-country-code", &country_code)) + hid->country = country_code; + + of_property_read_u32(iface->of_node, "apple,keyboard-layout-id", + &iface->keyboard_layout_id); + } + + hid->dev.parent = iface->dchid->dev; + hid->driver_data = iface; + + iface->hid = hid; + + ret = hid_add_device(hid); + if (ret < 0) { + iface->hid = NULL; + hid_destroy_device(hid); + dev_warn(iface->dchid->dev, "Failed to register hid device %s", iface->name); + } +} + +static int dchid_create_interface(struct dchid_iface *iface) +{ + if (iface->creating) + return -EBUSY; + + iface->creating = true; + INIT_WORK(&iface->create_work, dchid_create_interface_work); + return queue_work(iface->dchid->new_iface_wq, &iface->create_work); +} + +static void dchid_handle_descriptor(struct dchid_iface *iface, void *hid_desc, size_t desc_len) +{ + if (iface->hid) { + dev_warn(iface->dchid->dev, "Tried to initialize already started interface %s!\n", + iface->name); + return; + } + + iface->hid_desc = devm_kmemdup(iface->dchid->dev, hid_desc, desc_len, GFP_KERNEL); + if (!iface->hid_desc) + return; + + iface->hid_desc_len = desc_len; +} + +static void dchid_handle_ready(struct dockchannel_hid *dchid, void *data, size_t length) +{ + struct dchid_iface *iface; + u8 *pkt = data; + u8 index; + int i, ret; + + if (length < 2) { + dev_err(dchid->dev, "Bad length for ready message: %zu\n", length); + return; + } + + index = pkt[1]; + + if (index >= MAX_INTERFACES) { + dev_err(dchid->dev, "Got ready notification for bad iface %d\n", index); + return; + } + + iface = dchid->ifaces[index]; + if (!iface) { + dev_err(dchid->dev, "Got ready notification for unknown iface %d\n", index); + return; + } + + dev_info(dchid->dev, "Interface %s is now ready\n", iface->name); + complete_all(&iface->ready); + + /* When STM is ready, grab global device info */ + if (!strcmp(iface->name, "stm")) { + ret = dchid_get_report_cmd(iface, STM_REPORT_ID, &dchid->device_id, + sizeof(dchid->device_id)); + if (ret < sizeof(dchid->device_id)) { + dev_warn(iface->dchid->dev, "Failed to get device ID from STM!\n"); + /* Fake it and keep going. Things might still work... */ + memset(&dchid->device_id, 0, sizeof(dchid->device_id)); + dchid->device_id.vendor_id = HOST_VENDOR_ID_APPLE; + } + ret = dchid_get_report_cmd(iface, STM_REPORT_SERIAL, dchid->serial, + sizeof(dchid->serial) - 1); + if (ret < 0) { + dev_warn(iface->dchid->dev, "Failed to get serial from STM!\n"); + dchid->serial[0] = 0; + } + + dchid->id_ready = true; + for (i = 0; i < MAX_INTERFACES; i++) { + if (!dchid->ifaces[i] || !dchid->ifaces[i]->deferred) + continue; + dchid_create_interface(dchid->ifaces[i]); + } + } +} + +static void dchid_handle_init(struct dockchannel_hid *dchid, void *data, size_t length) +{ + struct dchid_init_hdr *hdr = data; + struct dchid_iface *iface; + struct dchid_init_block_hdr *blk; + + if (length < sizeof(*hdr)) + return; + + iface = dchid_get_interface(dchid, hdr->iface, hdr->name); + if (!iface) + return; + + data += sizeof(*hdr); + length -= sizeof(*hdr); + + while (length >= sizeof(*blk)) { + blk = data; + data += sizeof(*blk); + length -= sizeof(*blk); + + if (blk->length > length) + break; + + switch (blk->type) { + case INIT_HID_DESCRIPTOR: + dchid_handle_descriptor(iface, data, blk->length); + break; + + case INIT_GPIO_REQUEST: { + struct dchid_gpio_request *req = data; + + if (sizeof(*req) > length) + break; + + if (iface->gpio_id) { + dev_err(dchid->dev, + "Cannot request more than one GPIO per interface!\n"); + break; + } + + strscpy(iface->gpio_name, req->name, MAX_GPIO_NAME); + iface->gpio_id = req->id; + break; + } + + case INIT_TERMINATOR: + break; + + case INIT_PRODUCT_NAME: { + char *product = data; + + if (product[blk->length - 1] != 0) { + dev_warn(dchid->dev, "Unterminated product name for %s\n", + iface->name); + } else { + dev_info(dchid->dev, "Product name for %s: %s\n", + iface->name, product); + } + break; + } + + default: + dev_warn(dchid->dev, "Unknown init packet %d for %s\n", + blk->type, iface->name); + break; + } + + data += blk->length; + length -= blk->length; + + if (blk->type == INIT_TERMINATOR) + break; + } + + if (hdr->more_packets) + return; + + /* We need to enable STM first, since it'll give us the device IDs */ + if (iface->dchid->id_ready || !strcmp(iface->name, "stm")) { + dchid_create_interface(iface); + } else { + iface->deferred = true; + } +} + +static void dchid_handle_gpio(struct dockchannel_hid *dchid, void *data, size_t length) +{ + struct dchid_gpio_cmd *cmd = data; + struct dchid_iface *iface; + u32 retcode = 0xe000f00d; /* Give it a random Apple-style error code */ + struct dchid_gpio_ack *ack; + + if (length < sizeof(*cmd)) + return; + + if (cmd->iface >= MAX_INTERFACES || !(iface = dchid->ifaces[cmd->iface])) { + dev_err(dchid->dev, "Got GPIO command for bad inteface %d\n", cmd->iface); + goto err; + } + + if (dchid_request_gpio(iface) < 0) + goto err; + + if (!iface->gpio || cmd->gpio != iface->gpio_id) { + dev_err(dchid->dev, "Got GPIO command for bad GPIO %s#%d\n", + iface->name, cmd->gpio); + goto err; + } + + dev_info(dchid->dev, "GPIO command: %s#%d: %d\n", iface->name, cmd->gpio, cmd->cmd); + + switch (cmd->cmd) { + case 3: + /* Pulse. */ + gpiod_set_value_cansleep(iface->gpio, 1); + msleep(10); /* Random guess... */ + gpiod_set_value_cansleep(iface->gpio, 0); + retcode = 0; + break; + default: + dev_err(dchid->dev, "Unknown GPIO command %d\n", cmd->cmd ); + break; + } + +err: + /* Ack it */ + ack = kzalloc(sizeof(*ack) + length, GFP_KERNEL); + if (!ack) + return; + + ack->type = CMD_ACK_GPIO_CMD; + ack->retcode = retcode; + memcpy(ack->cmd, data, length); + + if (dchid_comm_cmd(dchid, ack, sizeof(*ack) + length) < 0) + dev_err(dchid->dev, "Failed to ACK GPIO command\n"); + + kfree(ack); +} + +static void dchid_handle_event(struct dockchannel_hid *dchid, void *data, size_t length) +{ + u8 *p = data; + switch (*p) { + case EVENT_INIT: + dchid_handle_init(dchid, data, length); + break; + case EVENT_READY: + dchid_handle_ready(dchid, data, length); + break; + case EVENT_GPIO_CMD: + dchid_handle_gpio(dchid, data, length); + break; + } +} + +static void dchid_handle_report(struct dchid_iface *iface, void *data, size_t length) +{ + struct dockchannel_hid *dchid = iface->dchid; + + if (!iface->hid) { + dev_warn(dchid->dev, "Report received but %s is not initialized!\n", iface->name); + return; + } + + if (!iface->open) + return; + + hid_input_report(iface->hid, HID_INPUT_REPORT, data, length, 1); +} + +static void dchid_packet_work(struct work_struct *ws) +{ + struct dchid_work *work = container_of(ws, struct dchid_work, work); + struct dchid_subhdr *shdr = (void *)work->data; + struct dockchannel_hid *dchid = work->iface->dchid; + int type = FIELD_GET(FLAGS_GROUP, shdr->flags); + u8 *payload = work->data + sizeof(*shdr); + + if (shdr->length + sizeof(*shdr) > work->hdr.length) { + dev_err(dchid->dev, "Bad sub header length (%d > %zu)\n", + shdr->length, work->hdr.length - sizeof(*shdr)); + return; + } + + switch (type) { + case HID_INPUT_REPORT: + if (work->hdr.iface == IFACE_COMM) + dchid_handle_event(dchid, payload, shdr->length); + else + dchid_handle_report(work->iface, payload, shdr->length); + break; + default: + dev_err(dchid->dev, "Received unknown packet type %d\n", type); + break; + } + + kfree(work); +} + +static void dchid_handle_ack(struct dchid_iface *iface, struct dchid_hdr *hdr, void *data) +{ + struct dchid_subhdr *shdr = (void *)data; + u8 *payload = data + sizeof(*shdr); + + if (shdr->length + sizeof(*shdr) > hdr->length) { + dev_err(iface->dchid->dev, "Bad sub header length (%d > %ld)\n", + shdr->length, hdr->length - sizeof(*shdr)); + return; + } + if (shdr->flags != iface->out_flags) { + dev_err(iface->dchid->dev, + "Received unexpected flags 0x%x on ACK channel (expFected 0x%x)\n", + shdr->flags, iface->out_flags); + return; + } + + if (shdr->length < 1) { + dev_err(iface->dchid->dev, "Received length 0 output report ack\n"); + return; + } + if (iface->tx_seq != hdr->seq) { + dev_err(iface->dchid->dev, "Received ACK with bad seq (expected %d, got %d)\n", + iface->tx_seq, hdr->seq); + return; + } + if (iface->out_report != payload[0]) { + dev_err(iface->dchid->dev, "Received ACK with bad report (expected %d, got %d\n", + iface->out_report, payload[0]); + return; + } + + if (iface->resp_buf && iface->resp_size) + memcpy(iface->resp_buf, payload + 1, min((size_t)shdr->length - 1, iface->resp_size)); + + iface->resp_size = shdr->length; + iface->out_report = -1; + iface->retcode = shdr->retcode; + complete(&iface->out_complete); +} + +static void dchid_handle_packet(void *cookie, size_t avail) +{ + struct dockchannel_hid *dchid = cookie; + struct dchid_hdr hdr; + struct dchid_work *work; + struct dchid_iface *iface; + u32 checksum; + + if (dockchannel_recv(dchid->dc, &hdr, sizeof(hdr)) != sizeof(hdr)) { + dev_err(dchid->dev, "Read failed (header)\n"); + return; + } + + if (hdr.hdr_len != sizeof(hdr)) { + dev_err(dchid->dev, "Bad header length %d\n", hdr.hdr_len); + goto done; + } + + if (dockchannel_recv(dchid->dc, dchid->pkt_buf, hdr.length + 4) != (hdr.length + 4)) { + dev_err(dchid->dev, "Read failed (body)\n"); + goto done; + } + + checksum = dchid_checksum(&hdr, sizeof(hdr)); + checksum += dchid_checksum(dchid->pkt_buf, hdr.length + 4); + + if (checksum != 0xffffffff) { + dev_err(dchid->dev, "Checksum mismatch (iface %d): 0x%08x != 0xffffffff\n", + hdr.iface, checksum); + goto done; + } + + + if (hdr.iface >= MAX_INTERFACES) { + dev_err(dchid->dev, "Bad iface %d\n", hdr.iface); + } + + iface = dchid->ifaces[hdr.iface]; + + if (!iface) { + dev_err(dchid->dev, "Received packet for uninitialized iface %d\n", hdr.iface); + goto done; + } + + switch (hdr.channel) { + case DCHID_CHANNEL_CMD: + dchid_handle_ack(iface, &hdr, dchid->pkt_buf); + goto done; + case DCHID_CHANNEL_REPORT: + break; + default: + dev_warn(dchid->dev, "Unknown channel 0x%x, treating as report...\n", + hdr.channel); + break; + } + + work = kzalloc(sizeof(*work) + hdr.length, GFP_KERNEL); + if (!work) + return; + + work->hdr = hdr; + work->iface = iface; + memcpy(work->data, dchid->pkt_buf, hdr.length); + INIT_WORK(&work->work, dchid_packet_work); + + queue_work(iface->wq, &work->work); + +done: + dockchannel_await(dchid->dc, dchid_handle_packet, dchid, sizeof(struct dchid_hdr)); +} + +static int dockchannel_hid_probe(struct platform_device *pdev) +{ + struct device *dev = &pdev->dev; + struct dockchannel_hid *dchid; + struct device_node *child, *helper; + struct platform_device *helper_pdev; + struct property *prop; + int ret; + + ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); + if (ret) + return ret; + + dchid = devm_kzalloc(dev, sizeof(*dchid), GFP_KERNEL); + if (!dchid) { + return -ENOMEM; + } + + dchid->dev = dev; + + /* + * First make sure all the GPIOs are available, in cased we need to defer. + * This is necessary because MTP will request them by name later, and by then + * it's too late to defer the probe. + */ + + for_each_child_of_node(dev->of_node, child) { + for_each_property_of_node(child, prop) { + size_t len = strlen(prop->name); + struct gpio_desc *gpio; + + if (len < 12 || strncmp("apple,", prop->name, 6) || + strcmp("-gpios", prop->name + len - 6)) + continue; + + gpio = fwnode_gpiod_get_index(&child->fwnode, prop->name, 0, GPIOD_ASIS, + prop->name); + if (IS_ERR_OR_NULL(gpio)) { + if (PTR_ERR(gpio) == -EPROBE_DEFER) { + of_node_put(child); + return -EPROBE_DEFER; + } + } else { + gpiod_put(gpio); + } + } + } + + /* + * Make sure we also have the MTP coprocessor available, and + * defer probe if the helper hasn't probed yet. + */ + helper = of_parse_phandle(dev->of_node, "apple,helper-cpu", 0); + if (!helper) { + dev_err(dev, "Missing apple,helper-cpu property"); + return -EINVAL; + } + + helper_pdev = of_find_device_by_node(helper); + of_node_put(helper); + if (!helper_pdev) { + dev_err(dev, "Failed to find helper device"); + return -EINVAL; + } + + dchid->helper_link = device_link_add(dev, &helper_pdev->dev, + DL_FLAG_AUTOREMOVE_CONSUMER); + put_device(&helper_pdev->dev); + if (!dchid->helper_link) { + dev_err(dev, "Failed to link to helper device"); + return -EINVAL; + } + + if (dchid->helper_link->supplier->links.status != DL_DEV_DRIVER_BOUND) + return -EPROBE_DEFER; + + /* Now it is safe to begin initializing */ + dchid->dc = dockchannel_init(pdev); + if (IS_ERR_OR_NULL(dchid->dc)) { + return PTR_ERR(dchid->dc); + } + dchid->new_iface_wq = alloc_workqueue("dchid-new", WQ_MEM_RECLAIM, 0); + if (!dchid->new_iface_wq) + return -ENOMEM; + + dchid->comm = dchid_get_interface(dchid, IFACE_COMM, "comm"); + if (!dchid->comm) { + dev_err(dchid->dev, "Failed to initialize comm interface"); + return -EIO; + } + + dev_info(dchid->dev, "Initialized, awaiting packets\n"); + dockchannel_await(dchid->dc, dchid_handle_packet, dchid, sizeof(struct dchid_hdr)); + + return 0; +} + +static void dockchannel_hid_remove(struct platform_device *pdev) +{ + BUG_ON(1); +} + +static const struct of_device_id dockchannel_hid_of_match[] = { + { .compatible = "apple,dockchannel-hid" }, + {}, +}; +MODULE_DEVICE_TABLE(of, dockchannel_hid_of_match); +MODULE_FIRMWARE("apple/tpmtfw-*.bin"); + +static struct platform_driver dockchannel_hid_driver = { + .driver = { + .name = "dockchannel-hid", + .of_match_table = dockchannel_hid_of_match, + }, + .probe = dockchannel_hid_probe, + .remove = dockchannel_hid_remove, +}; +module_platform_driver(dockchannel_hid_driver); + +MODULE_DESCRIPTION("Apple DockChannel HID transport driver"); +MODULE_AUTHOR("Hector Martin "); +MODULE_LICENSE("Dual MIT/GPL"); diff --git a/drivers/hid/hid-apple.c b/drivers/hid/hid-apple.c index 57da4f86a9fa..a95ef2ddeafe 100644 --- a/drivers/hid/hid-apple.c +++ b/drivers/hid/hid-apple.c @@ -30,7 +30,7 @@ #include "hid-ids.h" #define APPLE_RDESC_JIS BIT(0) -/* BIT(1) reserved, was: APPLE_IGNORE_MOUSE */ +#define APPLE_IGNORE_MOUSE BIT(1) #define APPLE_HAS_FN BIT(2) /* BIT(3) reserved, was: APPLE_HIDDEV */ #define APPLE_ISO_TILDE_QUIRK BIT(4) @@ -518,6 +518,16 @@ static int hidinput_apple_event(struct hid_device *hid, struct input_dev *input, table = macbookair_fn_keys; else if (hid->product < 0x21d || hid->product >= 0x300) table = powerbook_fn_keys; + else if (hid->bus == BUS_HOST || hid->bus == BUS_SPI) + switch (hid->product) { + case SPI_DEVICE_ID_APPLE_MACBOOK_PRO13_2020: + case HOST_DEVICE_ID_APPLE_MACBOOK_PRO13_2022: + table = macbookpro_dedicated_esc_fn_keys; + break; + default: + table = magic_keyboard_2021_and_2024_fn_keys; + break; + } else table = apple_fn_keys; } @@ -938,6 +948,13 @@ static int apple_probe(struct hid_device *hdev, struct apple_sc *asc; int ret; + if ((id->bus == BUS_SPI || id->bus == BUS_HOST) && id->vendor == SPI_VENDOR_ID_APPLE && + hdev->type != HID_TYPE_SPI_KEYBOARD) + return -ENODEV; + + if (quirks & APPLE_IGNORE_MOUSE && hdev->type == HID_TYPE_USBMOUSE) + return -ENODEV; + asc = devm_kzalloc(&hdev->dev, sizeof(*asc), GFP_KERNEL); if (asc == NULL) { hid_err(hdev, "can't alloc apple descriptor\n"); @@ -1160,27 +1177,31 @@ static const struct hid_device_id apple_devices[] = { { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRING9_JIS), .driver_data = APPLE_HAS_FN | APPLE_RDESC_JIS }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J140K), - .driver_data = APPLE_HAS_FN | APPLE_BACKLIGHT_CTL | APPLE_ISO_TILDE_QUIRK }, + .driver_data = APPLE_HAS_FN | APPLE_BACKLIGHT_CTL | APPLE_ISO_TILDE_QUIRK | + APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J132), .driver_data = APPLE_HAS_FN | APPLE_BACKLIGHT_CTL | APPLE_ISO_TILDE_QUIRK | - APPLE_DISABLE_FKEYS }, + APPLE_DISABLE_FKEYS | APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680), .driver_data = APPLE_HAS_FN | APPLE_BACKLIGHT_CTL | APPLE_ISO_TILDE_QUIRK | - APPLE_DISABLE_FKEYS }, + APPLE_DISABLE_FKEYS | APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680_ALT), .driver_data = APPLE_HAS_FN | APPLE_BACKLIGHT_CTL | APPLE_ISO_TILDE_QUIRK | - APPLE_DISABLE_FKEYS }, + APPLE_DISABLE_FKEYS | APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J213), .driver_data = APPLE_HAS_FN | APPLE_BACKLIGHT_CTL | APPLE_ISO_TILDE_QUIRK | - APPLE_DISABLE_FKEYS }, + APPLE_DISABLE_FKEYS | APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J214K), - .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_DISABLE_FKEYS }, + .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_DISABLE_FKEYS | + APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J223), - .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_DISABLE_FKEYS }, + .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_DISABLE_FKEYS | + APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J230K), - .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK }, + .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_IGNORE_MOUSE }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_WELLSPRINGT2_J152F), - .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_DISABLE_FKEYS }, + .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_DISABLE_FKEYS | + APPLE_IGNORE_MOUSE }, { HID_BLUETOOTH_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_ALU_WIRELESS_2009_ANSI), .driver_data = APPLE_NUMLOCK_EMULATION | APPLE_HAS_FN }, { HID_BLUETOOTH_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_ALU_WIRELESS_2009_ISO), @@ -1216,6 +1237,10 @@ static const struct hid_device_id apple_devices[] = { .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK | APPLE_RDESC_BATTERY }, { HID_BLUETOOTH_DEVICE(BT_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_MAGIC_KEYBOARD_NUMPAD_2024), .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK }, + { HID_SPI_DEVICE(SPI_VENDOR_ID_APPLE, HID_ANY_ID), + .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK }, + { HID_DEVICE(BUS_HOST, HID_GROUP_ANY, HOST_VENDOR_ID_APPLE, HID_ANY_ID), + .driver_data = APPLE_HAS_FN | APPLE_ISO_TILDE_QUIRK }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_TOUCHBAR_BACKLIGHT), .driver_data = APPLE_MAGIC_BACKLIGHT }, diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c index a5b3a8ca2fcb..b353fd819989 100644 --- a/drivers/hid/hid-core.c +++ b/drivers/hid/hid-core.c @@ -468,7 +468,10 @@ static int hid_parser_global(struct hid_parser *parser, struct hid_item *item) case HID_GLOBAL_ITEM_TAG_REPORT_SIZE: parser->global.report_size = item_udata(item); - if (parser->global.report_size > 256) { + /* Arbitrary maximum. Some Apple devices have 16384 here. + * This * HID_MAX_USAGES must fit in a signed integer. + */ + if (parser->global.report_size > 16384) { hid_err(parser->device, "invalid report_size %d\n", parser->global.report_size); return -1; @@ -2316,6 +2319,12 @@ int hid_connect(struct hid_device *hdev, unsigned int connect_mask) case BUS_I2C: bus = "I2C"; break; + case BUS_SPI: + bus = "SPI"; + break; + case BUS_HOST: + bus = "HOST"; + break; case BUS_SDW: bus = "SOUNDWIRE"; break; diff --git a/drivers/hid/hid-ids.h b/drivers/hid/hid-ids.h index 9c2bf584d9f6..adad71cc8bf0 100644 --- a/drivers/hid/hid-ids.h +++ b/drivers/hid/hid-ids.h @@ -93,6 +93,8 @@ #define USB_VENDOR_ID_APPLE 0x05ac #define BT_VENDOR_ID_APPLE 0x004c +#define SPI_VENDOR_ID_APPLE 0x05ac +#define HOST_VENDOR_ID_APPLE 0x05ac #define USB_DEVICE_ID_APPLE_MIGHTYMOUSE 0x0304 #define USB_DEVICE_ID_APPLE_MAGICMOUSE 0x030d #define USB_DEVICE_ID_APPLE_MAGICMOUSE2 0x0269 @@ -197,6 +199,14 @@ #define USB_DEVICE_ID_APPLE_IRCONTROL5 0x8243 #define USB_DEVICE_ID_APPLE_TOUCHBAR_BACKLIGHT 0x8102 #define USB_DEVICE_ID_APPLE_TOUCHBAR_DISPLAY 0x8302 +#define SPI_DEVICE_ID_APPLE_MACBOOK_AIR_2020 0x0281 +#define SPI_DEVICE_ID_APPLE_MACBOOK_PRO13_2020 0x0341 +#define SPI_DEVICE_ID_APPLE_MACBOOK_PRO14_2021 0x0342 +#define SPI_DEVICE_ID_APPLE_MACBOOK_PRO16_2021 0x0343 +#define HOST_DEVICE_ID_APPLE_MACBOOK_AIR13_2022 0x0351 +#define HOST_DEVICE_ID_APPLE_MACBOOK_PRO14_2023 0x0352 +#define HOST_DEVICE_ID_APPLE_MACBOOK_PRO16_2023 0x0353 +#define HOST_DEVICE_ID_APPLE_MACBOOK_PRO13_2022 0x0354 #define USB_VENDOR_ID_ASETEK 0x2433 #define USB_DEVICE_ID_ASETEK_INVICTA 0xf300 diff --git a/drivers/hid/hid-magicmouse.c b/drivers/hid/hid-magicmouse.c index 7d4a25c6de0e..4bf49d7cad0e 100644 --- a/drivers/hid/hid-magicmouse.c +++ b/drivers/hid/hid-magicmouse.c @@ -60,8 +60,14 @@ MODULE_PARM_DESC(report_undeciphered, "Report undeciphered multi-touch state fie #define MOUSE_REPORT_ID 0x29 #define MOUSE2_REPORT_ID 0x12 #define DOUBLE_REPORT_ID 0xf7 +#define SPI_REPORT_ID 0x02 +#define SPI_RESET_REPORT_ID 0x60 +#define MTP_REPORT_ID 0x75 +#define SENSOR_DIMENSIONS_REPORT_ID 0xd9 #define USB_BATTERY_TIMEOUT_SEC 60 +#define MAX_CONTACTS 16 + /* These definitions are not precise, but they're close enough. (Bits * 0x03 seem to indicate the aspect ratio of the touch, bits 0x70 seem * to be some kind of bit mask -- 0x20 may be a near-field reading, @@ -112,30 +118,156 @@ MODULE_PARM_DESC(report_undeciphered, "Report undeciphered multi-touch state fie #define TRACKPAD2_RES_Y \ ((TRACKPAD2_MAX_Y - TRACKPAD2_MIN_Y) / (TRACKPAD2_DIMENSION_Y / 100)) +#define J140K_TP_DIMENSION_X (float)12100 +#define J140K_TP_MIN_X -5318 +#define J140K_TP_MAX_X 5787 +#define J140K_TP_RES_X \ + ((J140K_TP_MAX_X - J140K_TP_MIN_X) / (J140K_TP_DIMENSION_X / 100)) +#define J140K_TP_DIMENSION_Y (float)8200 +#define J140K_TP_MIN_Y -157 +#define J140K_TP_MAX_Y 7102 +#define J140K_TP_RES_Y \ + ((J140K_TP_MAX_Y - J140K_TP_MIN_Y) / (J140K_TP_DIMENSION_Y / 100)) + +#define J132_TP_DIMENSION_X (float)13500 +#define J132_TP_MIN_X -6243 +#define J132_TP_MAX_X 6749 +#define J132_TP_RES_X \ + ((J132_TP_MAX_X - J132_TP_MIN_X) / (J132_TP_DIMENSION_X / 100)) +#define J132_TP_DIMENSION_Y (float)8400 +#define J132_TP_MIN_Y -170 +#define J132_TP_MAX_Y 7685 +#define J132_TP_RES_Y \ + ((J132_TP_MAX_Y - J132_TP_MIN_Y) / (J132_TP_DIMENSION_Y / 100)) + +#define J680_TP_DIMENSION_X (float)16000 +#define J680_TP_MIN_X -7456 +#define J680_TP_MAX_X 7976 +#define J680_TP_RES_X \ + ((J680_TP_MAX_X - J680_TP_MIN_X) / (J680_TP_DIMENSION_X / 100)) +#define J680_TP_DIMENSION_Y (float)10000 +#define J680_TP_MIN_Y -163 +#define J680_TP_MAX_Y 9283 +#define J680_TP_RES_Y \ + ((J680_TP_MAX_Y - J680_TP_MIN_Y) / (J680_TP_DIMENSION_Y / 100)) + +#define J680_ALT_TP_DIMENSION_X (float)16000 +#define J680_ALT_TP_MIN_X -7456 +#define J680_ALT_TP_MAX_X 7976 +#define J680_ALT_TP_RES_X \ + ((J680_ALT_TP_MAX_X - J680_ALT_TP_MIN_X) / (J680_ALT_TP_DIMENSION_X / 100)) +#define J680_ALT_TP_DIMENSION_Y (float)10000 +#define J680_ALT_TP_MIN_Y -163 +#define J680_ALT_TP_MAX_Y 9283 +#define J680_ALT_TP_RES_Y \ + ((J680_ALT_TP_MAX_Y - J680_ALT_TP_MIN_Y) / (J680_ALT_TP_DIMENSION_Y / 100)) + +#define J213_TP_DIMENSION_X (float)13500 +#define J213_TP_MIN_X -6243 +#define J213_TP_MAX_X 6749 +#define J213_TP_RES_X \ + ((J213_TP_MAX_X - J213_TP_MIN_X) / (J213_TP_DIMENSION_X / 100)) +#define J213_TP_DIMENSION_Y (float)8400 +#define J213_TP_MIN_Y -170 +#define J213_TP_MAX_Y 7685 +#define J213_TP_RES_Y \ + ((J213_TP_MAX_Y - J213_TP_MIN_Y) / (J213_TP_DIMENSION_Y / 100)) + +#define J214K_TP_DIMENSION_X (float)13200 +#define J214K_TP_MIN_X -6046 +#define J214K_TP_MAX_X 6536 +#define J214K_TP_RES_X \ + ((J214K_TP_MAX_X - J214K_TP_MIN_X) / (J214K_TP_DIMENSION_X / 100)) +#define J214K_TP_DIMENSION_Y (float)8200 +#define J214K_TP_MIN_Y -164 +#define J214K_TP_MAX_Y 7439 +#define J214K_TP_RES_Y \ + ((J214K_TP_MAX_Y - J214K_TP_MIN_Y) / (J214K_TP_DIMENSION_Y / 100)) + +#define J223_TP_DIMENSION_X (float)13200 +#define J223_TP_MIN_X -6046 +#define J223_TP_MAX_X 6536 +#define J223_TP_RES_X \ + ((J223_TP_MAX_X - J223_TP_MIN_X) / (J223_TP_DIMENSION_X / 100)) +#define J223_TP_DIMENSION_Y (float)8200 +#define J223_TP_MIN_Y -164 +#define J223_TP_MAX_Y 7439 +#define J223_TP_RES_Y \ + ((J223_TP_MAX_Y - J223_TP_MIN_Y) / (J223_TP_DIMENSION_Y / 100)) + +#define J230K_TP_DIMENSION_X (float)12100 +#define J230K_TP_MIN_X -5318 +#define J230K_TP_MAX_X 5787 +#define J230K_TP_RES_X \ + ((J230K_TP_MAX_X - J230K_TP_MIN_X) / (J230K_TP_DIMENSION_X / 100)) +#define J230K_TP_DIMENSION_Y (float)8200 +#define J230K_TP_MIN_Y -157 +#define J230K_TP_MAX_Y 7102 +#define J230K_TP_RES_Y \ + ((J230K_TP_MAX_Y - J230K_TP_MIN_Y) / (J230K_TP_DIMENSION_Y / 100)) + +#define J152F_TP_DIMENSION_X (float)16000 +#define J152F_TP_MIN_X -7456 +#define J152F_TP_MAX_X 7976 +#define J152F_TP_RES_X \ + ((J152F_TP_MAX_X - J152F_TP_MIN_X) / (J152F_TP_DIMENSION_X / 100)) +#define J152F_TP_DIMENSION_Y (float)10000 +#define J152F_TP_MIN_Y -163 +#define J152F_TP_MAX_Y 9283 +#define J152F_TP_RES_Y \ + ((J152F_TP_MAX_Y - J152F_TP_MIN_Y) / (J152F_TP_DIMENSION_Y / 100)) + +/* These are fallback values, since the real values will be queried from the device. */ +#define J314_TP_DIMENSION_X (float)13000 +#define J314_TP_MIN_X -5900 +#define J314_TP_MAX_X 6500 +#define J314_TP_RES_X \ + ((J314_TP_MAX_X - J314_TP_MIN_X) / (J314_TP_DIMENSION_X / 100)) +#define J314_TP_DIMENSION_Y (float)8100 +#define J314_TP_MIN_Y -200 +#define J314_TP_MAX_Y 7400 +#define J314_TP_RES_Y \ + ((J314_TP_MAX_Y - J314_TP_MIN_Y) / (J314_TP_DIMENSION_Y / 100)) + +#define T2_TOUCHPAD_ENTRY(model) \ + { USB_DEVICE_ID_APPLE_WELLSPRINGT2_##model, model##_TP_MIN_X, model##_TP_MIN_Y, \ +model##_TP_MAX_X, model##_TP_MAX_Y, model##_TP_RES_X, model##_TP_RES_Y } + +#define INTERNAL_TP_MAX_FINGER_ORIENTATION 16384 + +struct magicmouse_input_ops { + int (*raw_event)(struct hid_device *hdev, + struct hid_report *report, u8 *data, int size); + int (*setup_input)(struct input_dev *input, struct hid_device *hdev); +}; + /** * struct magicmouse_sc - Tracks Magic Mouse-specific data. * @input: Input device through which we report events. * @quirks: Currently unused. + * @query_dimensions: Whether to query and update dimensions on first open * @ntouches: Number of touches in most recent touch report. * @scroll_accel: Number of consecutive scroll motions. * @scroll_jiffies: Time of last scroll motion. + * @pos: multi touch position data of the last report. * @touches: Most recent data for a touch, indexed by tracking ID. * @tracking_ids: Mapping of current touch input data to @touches. * @hdev: Pointer to the underlying HID device. * @work: Workqueue to handle initialization retry for quirky devices. * @battery_timer: Timer for obtaining battery level information. + * @input_ops: Input ops based on device type. */ struct magicmouse_sc { struct input_dev *input; unsigned long quirks; + bool query_dimensions; int ntouches; int scroll_accel; unsigned long scroll_jiffies; + struct input_mt_pos pos[MAX_CONTACTS]; struct { - short x; - short y; short scroll_x; short scroll_y; short scroll_x_hr; @@ -143,14 +275,112 @@ struct magicmouse_sc { u8 size; bool scroll_x_active; bool scroll_y_active; - } touches[16]; - int tracking_ids[16]; + } touches[MAX_CONTACTS]; + int tracking_ids[MAX_CONTACTS]; struct hid_device *hdev; struct delayed_work work; struct timer_list battery_timer; + struct magicmouse_input_ops input_ops; }; +static int magicmouse_enable_multitouch(struct hid_device *hdev); + +static inline int le16_to_int(__le16 x) +{ + return (signed short)le16_to_cpu(x); +} + +static int magicmouse_open(struct input_dev *dev) +{ + struct hid_device *hdev = input_get_drvdata(dev); + struct magicmouse_sc *msc = hid_get_drvdata(hdev); + int ret; + + ret = hid_hw_open(hdev); + if (ret) + return ret; + + /* + * Some devices repond with 'invalid report id' when feature + * report switching it into multitouch mode is sent to it. + * + * This results in -EIO from the _raw low-level transport callback, + * but there seems to be no other way of switching the mode. + * Thus the super-ugly hacky success check below. + * + * MTP devices do not need this. + */ + if (hdev->bus != BUS_HOST) { + ret = magicmouse_enable_multitouch(hdev); + if (ret != -EIO && ret < 0) { + hid_err(hdev, "unable to request touch data (%d)\n", ret); + return ret; + } + if (ret == -EIO && (hdev->product == USB_DEVICE_ID_APPLE_MAGICMOUSE2 || + hdev->product == USB_DEVICE_ID_APPLE_MAGICMOUSE2_USBC)) { + schedule_delayed_work(&msc->work, msecs_to_jiffies(500)); + } + } + + /* + * For Apple Silicon trackpads, we want to query the dimensions on + * device open. This is because doing so requires the firmware, but + * we don't want to force a firmware load until the device is opened + * for the first time. So do that here and update the input properties + * just in time before userspace queries them. + */ + if (msc->query_dimensions) { + struct input_dev *input = msc->input; + u8 buf[32]; + struct { + __le32 width; + __le32 height; + __le16 min_x; + __le16 min_y; + __le16 max_x; + __le16 max_y; + } dim; + uint32_t x_span, y_span; + + ret = hid_hw_raw_request(hdev, SENSOR_DIMENSIONS_REPORT_ID, buf, sizeof(buf), HID_FEATURE_REPORT, HID_REQ_GET_REPORT); + if (ret < (int)(1 + sizeof(dim))) { + hid_err(hdev, "unable to request dimensions (%d)\n", ret); + return ret; + } + + memcpy(&dim, buf + 1, sizeof(dim)); + + /* finger position */ + input_set_abs_params(input, ABS_MT_POSITION_X, + le16_to_int(dim.min_x), le16_to_int(dim.max_x), 0, 0); + /* Y axis is inverted */ + input_set_abs_params(input, ABS_MT_POSITION_Y, + -le16_to_int(dim.max_y), -le16_to_int(dim.min_y), 0, 0); + x_span = le16_to_int(dim.max_x) - le16_to_int(dim.min_x); + y_span = le16_to_int(dim.max_y) - le16_to_int(dim.min_y); + + /* X/Y resolution */ + input_abs_set_res(input, ABS_MT_POSITION_X, 100 * x_span / le32_to_cpu(dim.width) ); + input_abs_set_res(input, ABS_MT_POSITION_Y, 100 * y_span / le32_to_cpu(dim.height) ); + + /* copy info, as input_mt_init_slots() does */ + dev->absinfo[ABS_X] = dev->absinfo[ABS_MT_POSITION_X]; + dev->absinfo[ABS_Y] = dev->absinfo[ABS_MT_POSITION_Y]; + + msc->query_dimensions = false; + } + + return 0; +} + +static void magicmouse_close(struct input_dev *dev) +{ + struct hid_device *hdev = input_get_drvdata(dev); + + hid_hw_close(hdev); +} + static int magicmouse_firm_touch(struct magicmouse_sc *msc) { int touch = -1; @@ -192,7 +422,7 @@ static void magicmouse_emit_buttons(struct magicmouse_sc *msc, int state) } else if (last_state != 0) { state = last_state; } else if ((id = magicmouse_firm_touch(msc)) >= 0) { - int x = msc->touches[id].x; + int x = msc->pos[id].x; if (x < middle_button_start) state = 1; else if (x > middle_button_stop) @@ -256,8 +486,8 @@ static void magicmouse_emit_touch(struct magicmouse_sc *msc, int raw_id, u8 *tda /* Store tracking ID and other fields. */ msc->tracking_ids[raw_id] = id; - msc->touches[id].x = x; - msc->touches[id].y = y; + msc->pos[id].x = x; + msc->pos[id].y = y; msc->touches[id].size = size; /* If requested, emulate a scroll wheel by detecting small @@ -387,6 +617,14 @@ static int magicmouse_raw_event(struct hid_device *hdev, struct hid_report *report, u8 *data, int size) { struct magicmouse_sc *msc = hid_get_drvdata(hdev); + + return msc->input_ops.raw_event(hdev, report, data, size); +} + +static int magicmouse_raw_event_usb(struct hid_device *hdev, + struct hid_report *report, u8 *data, int size) +{ + struct magicmouse_sc *msc = hid_get_drvdata(hdev); struct input_dev *input = msc->input; int x = 0, y = 0, ii, clicks = 0, npoints; @@ -518,6 +756,188 @@ static int magicmouse_raw_event(struct hid_device *hdev, return 1; } +/** + * struct tp_finger - single trackpad finger structure, le16-aligned + * + * @unknown1: unknown + * @unknown2: unknown + * @abs_x: absolute x coordinate + * @abs_y: absolute y coordinate + * @rel_x: relative x coordinate + * @rel_y: relative y coordinate + * @tool_major: tool area, major axis + * @tool_minor: tool area, minor axis + * @orientation: 16384 when point, else 15 bit angle + * @touch_major: touch area, major axis + * @touch_minor: touch area, minor axis + * @unused: zeros + * @pressure: pressure on forcetouch touchpad + * @multi: one finger: varies, more fingers: constant + */ +struct tp_finger { + __le16 unknown1; + __le16 unknown2; + __le16 abs_x; + __le16 abs_y; + __le16 rel_x; + __le16 rel_y; + __le16 tool_major; + __le16 tool_minor; + __le16 orientation; + __le16 touch_major; + __le16 touch_minor; + __le16 unused[2]; + __le16 pressure; + __le16 multi; +} __attribute__((packed, aligned(2))); + +/** + * vendor trackpad report + * + * @num_fingers: the number of fingers being reported in @fingers + * @buttons: same as HID buttons + */ +struct tp_header { + // HID vendor part, up to 1751 bytes + u8 unknown[22]; + u8 num_fingers; + u8 buttons; + u8 unknown3[14]; +}; + +/** + * standard HID mouse report + * + * @report_id: reportid + * @buttons: HID Usage Buttons 3 1-bit reports + */ +struct tp_mouse_report { + // HID mouse report + u8 report_id; + u8 buttons; + u8 rel_x; + u8 rel_y; + u8 padding[4]; +}; + +static void report_finger_data(struct input_dev *input, int slot, + const struct input_mt_pos *pos, + const struct tp_finger *f) +{ + input_mt_slot(input, slot); + input_mt_report_slot_state(input, MT_TOOL_FINGER, true); + + input_report_abs(input, ABS_MT_TOUCH_MAJOR, + le16_to_int(f->touch_major) << 1); + input_report_abs(input, ABS_MT_TOUCH_MINOR, + le16_to_int(f->touch_minor) << 1); + input_report_abs(input, ABS_MT_WIDTH_MAJOR, + le16_to_int(f->tool_major) << 1); + input_report_abs(input, ABS_MT_WIDTH_MINOR, + le16_to_int(f->tool_minor) << 1); + input_report_abs(input, ABS_MT_ORIENTATION, + INTERNAL_TP_MAX_FINGER_ORIENTATION - le16_to_int(f->orientation)); + input_report_abs(input, ABS_MT_PRESSURE, le16_to_int(f->pressure)); + input_report_abs(input, ABS_MT_POSITION_X, pos->x); + input_report_abs(input, ABS_MT_POSITION_Y, pos->y); +} + +static int magicmouse_raw_event_mtp(struct hid_device *hdev, + struct hid_report *report, u8 *data, int size) +{ + struct magicmouse_sc *msc = hid_get_drvdata(hdev); + struct input_dev *input = msc->input; + struct tp_header *tp_hdr; + struct tp_finger *f; + int i, n; + u32 npoints; + const size_t hdr_sz = sizeof(struct tp_header); + const size_t touch_sz = sizeof(struct tp_finger); + u8 map_contacs[MAX_CONTACTS]; + + // hid_warn(hdev, "%s\n", __func__); + // print_hex_dump_debug("appleft ev: ", DUMP_PREFIX_OFFSET, 16, 1, data, + // size, false); + + /* Expect 46 bytes of prefix, and N * 30 bytes of touch data. */ + if (size < hdr_sz || ((size - hdr_sz) % touch_sz) != 0) + return 0; + + tp_hdr = (struct tp_header *)data; + + npoints = (size - hdr_sz) / touch_sz; + if (npoints < tp_hdr->num_fingers || npoints > MAX_CONTACTS) { + hid_warn(hdev, + "unexpected number of touches (%u) for " + "report\n", + npoints); + return 0; + } + + n = 0; + for (i = 0; i < tp_hdr->num_fingers; i++) { + f = (struct tp_finger *)(data + hdr_sz + i * touch_sz); + if (le16_to_int(f->touch_major) == 0) + continue; + + hid_dbg(hdev, "ev x:%04x y:%04x\n", le16_to_int(f->abs_x), + le16_to_int(f->abs_y)); + msc->pos[n].x = le16_to_int(f->abs_x); + msc->pos[n].y = -le16_to_int(f->abs_y); + map_contacs[n] = i; + n++; + } + + input_mt_assign_slots(input, msc->tracking_ids, msc->pos, n, 0); + + for (i = 0; i < n; i++) { + int idx = map_contacs[i]; + f = (struct tp_finger *)(data + hdr_sz + idx * touch_sz); + report_finger_data(input, msc->tracking_ids[i], &msc->pos[i], f); + } + + input_mt_sync_frame(input); + input_report_key(input, BTN_MOUSE, tp_hdr->buttons & 1); + + input_sync(input); + return 1; +} + +static int magicmouse_raw_event_spi(struct hid_device *hdev, + struct hid_report *report, u8 *data, int size) +{ + struct magicmouse_sc *msc = hid_get_drvdata(hdev); + const size_t hdr_sz = sizeof(struct tp_mouse_report); + + if (!size) + return 0; + + if (data[0] == SPI_RESET_REPORT_ID) { + hid_info(hdev, "Touch controller was reset, re-enabling touch mode\n"); + schedule_delayed_work(&msc->work, msecs_to_jiffies(10)); + return 1; + } + + if (data[0] != SPI_REPORT_ID || size < hdr_sz) + return 0; + + return magicmouse_raw_event_mtp(hdev, report, data + hdr_sz, size - hdr_sz); +} + +static int magicmouse_raw_event_t2(struct hid_device *hdev, + struct hid_report *report, u8 *data, int size) +{ + const size_t hdr_sz = sizeof(struct tp_mouse_report); + + if (!size) + return 0; + + if (data[0] != TRACKPAD2_USB_REPORT_ID || size < hdr_sz) + return 0; + + return magicmouse_raw_event_mtp(hdev, report, data + hdr_sz, size - hdr_sz); +} + static int magicmouse_event(struct hid_device *hdev, struct hid_field *field, struct hid_usage *usage, __s32 value) { @@ -536,7 +956,17 @@ static int magicmouse_event(struct hid_device *hdev, struct hid_field *field, return 0; } -static int magicmouse_setup_input(struct input_dev *input, struct hid_device *hdev) + +static int magicmouse_setup_input(struct input_dev *input, + struct hid_device *hdev) +{ + struct magicmouse_sc *msc = hid_get_drvdata(hdev); + + return msc->input_ops.setup_input(input, hdev); +} + +static int magicmouse_setup_input_usb(struct input_dev *input, + struct hid_device *hdev) { int error; int mt_flags = 0; @@ -615,7 +1045,7 @@ static int magicmouse_setup_input(struct input_dev *input, struct hid_device *hd __set_bit(EV_ABS, input->evbit); - error = input_mt_init_slots(input, 16, mt_flags); + error = input_mt_init_slots(input, MAX_CONTACTS, mt_flags); if (error) return error; input_set_abs_params(input, ABS_MT_TOUCH_MAJOR, 0, 255 << 2, @@ -695,6 +1125,171 @@ static int magicmouse_setup_input(struct input_dev *input, struct hid_device *hd */ __clear_bit(EV_REP, input->evbit); + /* + * This isn't strictly speaking needed for USB, but enabling MT on + * device open is probably more robust than only doing it once on probe + * even if USB devices are not known to suffer from the SPI reset issue. + */ + input->open = magicmouse_open; + input->close = magicmouse_close; + return 0; +} + +struct magicmouse_t2_properties { + u32 id; + int min_x; + int min_y; + int max_x; + int max_y; + int res_x; + int res_y; +}; + +static const struct magicmouse_t2_properties magicmouse_t2_configs[] = { + T2_TOUCHPAD_ENTRY(J140K), + T2_TOUCHPAD_ENTRY(J132), + T2_TOUCHPAD_ENTRY(J680), + T2_TOUCHPAD_ENTRY(J680_ALT), + T2_TOUCHPAD_ENTRY(J213), + T2_TOUCHPAD_ENTRY(J214K), + T2_TOUCHPAD_ENTRY(J223), + T2_TOUCHPAD_ENTRY(J230K), + T2_TOUCHPAD_ENTRY(J152F), +}; + +static int magicmouse_setup_input_int_tpd(struct input_dev *input, + struct hid_device *hdev, int min_x, int min_y, + int max_x, int max_y, int res_x, int res_y, + bool query_dimensions) +{ + int error; + int mt_flags = 0; + struct magicmouse_sc *msc = hid_get_drvdata(hdev); + + __set_bit(INPUT_PROP_BUTTONPAD, input->propbit); + __clear_bit(BTN_0, input->keybit); + __clear_bit(BTN_RIGHT, input->keybit); + __clear_bit(BTN_MIDDLE, input->keybit); + __clear_bit(EV_REL, input->evbit); + __clear_bit(REL_X, input->relbit); + __clear_bit(REL_Y, input->relbit); + + mt_flags = INPUT_MT_POINTER | INPUT_MT_DROP_UNUSED | INPUT_MT_TRACK; + + /* finger touch area */ + input_set_abs_params(input, ABS_MT_TOUCH_MAJOR, 0, 5000, 0, 0); + input_set_abs_params(input, ABS_MT_TOUCH_MINOR, 0, 5000, 0, 0); + + /* finger approach area */ + input_set_abs_params(input, ABS_MT_WIDTH_MAJOR, 0, 5000, 0, 0); + input_set_abs_params(input, ABS_MT_WIDTH_MINOR, 0, 5000, 0, 0); + + /* Note: Touch Y position from the device is inverted relative + * to how pointer motion is reported (and relative to how USB + * HID recommends the coordinates work). This driver keeps + * the origin at the same position, and just uses the additive + * inverse of the reported Y. + */ + + input_set_abs_params(input, ABS_MT_PRESSURE, 0, 6000, 0, 0); + + /* + * This makes libinput recognize this as a PressurePad and + * stop trying to use pressure for touch size. Pressure unit + * seems to be ~grams on these touchpads. + */ + input_abs_set_res(input, ABS_MT_PRESSURE, 1); + + /* finger orientation */ + input_set_abs_params(input, ABS_MT_ORIENTATION, -INTERNAL_TP_MAX_FINGER_ORIENTATION, + INTERNAL_TP_MAX_FINGER_ORIENTATION, 0, 0); + + /* finger position */ + input_set_abs_params(input, ABS_MT_POSITION_X, min_x, max_x, 0, 0); + /* Y axis is inverted */ + input_set_abs_params(input, ABS_MT_POSITION_Y, -max_y, -min_y, 0, 0); + + /* X/Y resolution */ + input_abs_set_res(input, ABS_MT_POSITION_X, res_x); + input_abs_set_res(input, ABS_MT_POSITION_Y, res_y); + + input_set_events_per_packet(input, 60); + + /* touchpad button */ + input_set_capability(input, EV_KEY, BTN_MOUSE); + + /* + * hid-input may mark device as using autorepeat, but the trackpad does + * not actually want it. + */ + __clear_bit(EV_REP, input->evbit); + + error = input_mt_init_slots(input, MAX_CONTACTS, mt_flags); + if (error) + return error; + + /* + * Override the default input->open function to send the MT + * enable every time the device is opened. This ensures it works + * even if we missed a reset event due to the device being closed. + * input->close is overridden for symmetry. + * + * This also takes care of the dimensions query. + */ + input->open = magicmouse_open; + input->close = magicmouse_close; + msc->query_dimensions = query_dimensions; + + return 0; +} + +static int magicmouse_setup_input_mtp(struct input_dev *input, + struct hid_device *hdev) +{ + int ret = magicmouse_setup_input_int_tpd(input, hdev, J314_TP_MIN_X, + J314_TP_MIN_Y, J314_TP_MAX_X, + J314_TP_MAX_Y, J314_TP_RES_X, + J314_TP_RES_Y, true); + if (ret) + return ret; + + return 0; +} + +static int magicmouse_setup_input_spi(struct input_dev *input, + struct hid_device *hdev) +{ + int ret = magicmouse_setup_input_int_tpd(input, hdev, J314_TP_MIN_X, + J314_TP_MIN_Y, J314_TP_MAX_X, + J314_TP_MAX_Y, J314_TP_RES_X, + J314_TP_RES_Y, true); + if (ret) + return ret; + + return 0; +} + +static int magicmouse_setup_input_t2(struct input_dev *input, + struct hid_device *hdev) +{ + int min_x, min_y, max_x, max_y, res_x, res_y; + + for (size_t i = 0; i < ARRAY_SIZE(magicmouse_t2_configs); i++) { + if (magicmouse_t2_configs[i].id == hdev->product) { + min_x = magicmouse_t2_configs[i].min_x; + min_y = magicmouse_t2_configs[i].min_y; + max_x = magicmouse_t2_configs[i].max_x; + max_y = magicmouse_t2_configs[i].max_y; + res_x = magicmouse_t2_configs[i].res_x; + res_y = magicmouse_t2_configs[i].res_y; + } + } + + int ret = magicmouse_setup_input_int_tpd(input, hdev, min_x, min_y, + max_x, max_y, res_x, res_y, false); + if (ret) + return ret; + return 0; } @@ -748,6 +1343,10 @@ static int magicmouse_enable_multitouch(struct hid_device *hdev) int feature_size; switch (hdev->product) { + case SPI_DEVICE_ID_APPLE_MACBOOK_AIR_2020: + case SPI_DEVICE_ID_APPLE_MACBOOK_PRO13_2020: + case SPI_DEVICE_ID_APPLE_MACBOOK_PRO14_2021: + case SPI_DEVICE_ID_APPLE_MACBOOK_PRO16_2021: case USB_DEVICE_ID_APPLE_MAGICTRACKPAD2: case USB_DEVICE_ID_APPLE_MAGICTRACKPAD2_USBC: switch (hdev->vendor) { @@ -755,11 +1354,23 @@ static int magicmouse_enable_multitouch(struct hid_device *hdev) feature_size = sizeof(feature_mt_trackpad2_bt); feature = feature_mt_trackpad2_bt; break; - default: /* USB_VENDOR_ID_APPLE */ + default: /* USB_VENDOR_ID_APPLE || SPI_VENDOR_ID_APPLE */ feature_size = sizeof(feature_mt_trackpad2_usb); feature = feature_mt_trackpad2_usb; } break; + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J140K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J132: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680_ALT: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J213: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J214K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J223: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J230K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J152F: + feature_size = sizeof(feature_mt_trackpad2_usb); + feature = feature_mt_trackpad2_usb; + break; case USB_DEVICE_ID_APPLE_MAGICMOUSE2: case USB_DEVICE_ID_APPLE_MAGICMOUSE2_USBC: feature_size = sizeof(feature_mt_mouse2); @@ -852,12 +1463,62 @@ static int magicmouse_probe(struct hid_device *hdev, struct hid_report *report; int ret; + if ((id->bus == BUS_SPI || id->bus == BUS_HOST) && id->vendor == SPI_VENDOR_ID_APPLE && + hdev->type != HID_TYPE_SPI_MOUSE) + return -ENODEV; + + switch (id->product) { + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J140K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J132: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680_ALT: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J213: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J214K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J223: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J230K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J152F: + if (hdev->type != HID_TYPE_USBMOUSE) + return -ENODEV; + break; + } + msc = devm_kzalloc(&hdev->dev, sizeof(*msc), GFP_KERNEL); if (msc == NULL) { hid_err(hdev, "can't alloc magicmouse descriptor\n"); return -ENOMEM; } + // internal trackpad use a data format use input ops to avoid + // conflicts with the report ID. + switch (id->bus) { + case BUS_HOST: + msc->input_ops.raw_event = magicmouse_raw_event_mtp; + msc->input_ops.setup_input = magicmouse_setup_input_mtp; + break; + case BUS_SPI: + msc->input_ops.raw_event = magicmouse_raw_event_spi; + msc->input_ops.setup_input = magicmouse_setup_input_spi; + break; + default: + switch (id->product) { + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J140K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J132: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680_ALT: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J213: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J214K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J223: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J230K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J152F: + msc->input_ops.raw_event = magicmouse_raw_event_t2; + msc->input_ops.setup_input = magicmouse_setup_input_t2; + break; + default: + msc->input_ops.raw_event = magicmouse_raw_event_usb; + msc->input_ops.setup_input = magicmouse_setup_input_usb; + } + } + msc->scroll_accel = SCROLL_ACCEL_DEFAULT; msc->hdev = hdev; INIT_DEFERRABLE_WORK(&msc->work, magicmouse_enable_mt_work); @@ -916,11 +1577,32 @@ static int magicmouse_probe(struct hid_device *hdev, TRACKPAD2_USB_REPORT_ID, 0); } break; - default: /* USB_DEVICE_ID_APPLE_MAGICTRACKPAD */ - report = hid_register_report(hdev, HID_INPUT_REPORT, - TRACKPAD_REPORT_ID, 0); + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J140K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J132: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680_ALT: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J213: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J214K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J223: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J230K: + case USB_DEVICE_ID_APPLE_WELLSPRINGT2_J152F: report = hid_register_report(hdev, HID_INPUT_REPORT, - DOUBLE_REPORT_ID, 0); + TRACKPAD2_USB_REPORT_ID, 0); + break; + default: + switch (id->bus) { + case BUS_HOST: + report = hid_register_report(hdev, HID_INPUT_REPORT, MTP_REPORT_ID, 0); + break; + case BUS_SPI: + report = hid_register_report(hdev, HID_INPUT_REPORT, SPI_REPORT_ID, 0); + break; + default: /* USB_DEVICE_ID_APPLE_MAGICTRACKPAD */ + report = hid_register_report(hdev, HID_INPUT_REPORT, + TRACKPAD_REPORT_ID, 0); + report = hid_register_report(hdev, HID_INPUT_REPORT, + DOUBLE_REPORT_ID, 0); + } } if (!report) { @@ -930,22 +1612,14 @@ static int magicmouse_probe(struct hid_device *hdev, } report->size = 6; - /* - * Some devices repond with 'invalid report id' when feature - * report switching it into multitouch mode is sent to it. - * - * This results in -EIO from the _raw low-level transport callback, - * but there seems to be no other way of switching the mode. - * Thus the super-ugly hacky success check below. - */ - ret = magicmouse_enable_multitouch(hdev); - if (ret != -EIO && ret < 0) { - hid_err(hdev, "unable to request touch data (%d)\n", ret); - goto err_stop_hw; - } - if (ret == -EIO && (id->product == USB_DEVICE_ID_APPLE_MAGICMOUSE2 || - id->product == USB_DEVICE_ID_APPLE_MAGICMOUSE2_USBC)) { - schedule_delayed_work(&msc->work, msecs_to_jiffies(500)); + /* MTP devices do not need the MT enable, this is handled by the MTP driver */ + if (id->bus == BUS_HOST) + return 0; + + /* SPI devices need to watch for reset events to re-send the MT enable */ + if (id->bus == BUS_SPI) { + report = hid_register_report(hdev, HID_INPUT_REPORT, SPI_RESET_REPORT_ID, 0); + report->size = 2; } return 0; @@ -1023,10 +1697,42 @@ static const struct hid_device_id magic_mice[] = { USB_DEVICE_ID_APPLE_MAGICTRACKPAD2_USBC), .driver_data = 0 }, { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, USB_DEVICE_ID_APPLE_MAGICTRACKPAD2_USBC), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J140K), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J132), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J680_ALT), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J213), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J214K), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J223), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J230K), .driver_data = 0 }, + { HID_USB_DEVICE(USB_VENDOR_ID_APPLE, + USB_DEVICE_ID_APPLE_WELLSPRINGT2_J152F), .driver_data = 0 }, + { HID_SPI_DEVICE(SPI_VENDOR_ID_APPLE, HID_ANY_ID), + .driver_data = 0 }, + { HID_DEVICE(BUS_HOST, HID_GROUP_ANY, HOST_VENDOR_ID_APPLE, + HID_ANY_ID), .driver_data = 0 }, { } }; MODULE_DEVICE_TABLE(hid, magic_mice); +#ifdef CONFIG_PM +static int magicmouse_reset_resume(struct hid_device *hdev) +{ + if (hdev->bus == BUS_SPI) + return magicmouse_enable_multitouch(hdev); + + return 0; +} +#endif + static struct hid_driver magicmouse_driver = { .name = "magicmouse", .id_table = magic_mice, @@ -1037,6 +1743,10 @@ static struct hid_driver magicmouse_driver = { .event = magicmouse_event, .input_mapping = magicmouse_input_mapping, .input_configured = magicmouse_input_configured, +#ifdef CONFIG_PM + .reset_resume = magicmouse_reset_resume, +#endif + }; module_hid_driver(magicmouse_driver); diff --git a/drivers/hid/spi-hid/Kconfig b/drivers/hid/spi-hid/Kconfig new file mode 100644 index 000000000000..8e37f0fec28a --- /dev/null +++ b/drivers/hid/spi-hid/Kconfig @@ -0,0 +1,26 @@ +# SPDX-License-Identifier: GPL-2.0-only +menu "SPI HID support" + depends on SPI + +config SPI_HID_APPLE_OF + tristate "HID over SPI transport layer for Apple Silicon SoCs" + default ARCH_APPLE + depends on SPI && INPUT && OF + help + Say Y here if you use Apple Silicon based laptop. The keyboard and + touchpad are HID based devices connected via SPI. + + If unsure, say N. + + This support is also available as a module. If so, the module + will be called spi-hid-apple-of. It will also build/depend on the + module spi-hid-apple. + +endmenu + +config SPI_HID_APPLE_CORE + tristate + default y if SPI_HID_APPLE_OF=y + default m if SPI_HID_APPLE_OF=m + select HID + select CRC16 diff --git a/drivers/hid/spi-hid/Makefile b/drivers/hid/spi-hid/Makefile new file mode 100644 index 000000000000..f276ee12cb94 --- /dev/null +++ b/drivers/hid/spi-hid/Makefile @@ -0,0 +1,10 @@ +# SPDX-License-Identifier: GPL-2.0-only +# +# Makefile for SPI HID tarnsport drivers +# + +obj-$(CONFIG_SPI_HID_APPLE_CORE) += spi-hid-apple.o + +spi-hid-apple-objs = spi-hid-apple-core.o + +obj-$(CONFIG_SPI_HID_APPLE_OF) += spi-hid-apple-of.o diff --git a/drivers/hid/spi-hid/spi-hid-apple-core.c b/drivers/hid/spi-hid/spi-hid-apple-core.c new file mode 100644 index 000000000000..2ed909895391 --- /dev/null +++ b/drivers/hid/spi-hid/spi-hid-apple-core.c @@ -0,0 +1,1194 @@ +/* + * SPDX-License-Identifier: GPL-2.0 + * + * Apple SPI HID transport driver + * + * Copyright (C) The Asahi Linux Contributors + * + * Based on: drivers/input/applespi.c + * + * MacBook (Pro) SPI keyboard and touchpad driver + * + * Copyright (c) 2015-2018 Federico Lorenzi + * Copyright (c) 2017-2018 Ronald Tschalär + * + */ + +//#define DEBUG 2 + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "spi-hid-apple.h" + +#define SPIHID_DEF_WAIT msecs_to_jiffies(1000) + +#define SPIHID_MAX_INPUT_REPORT_SIZE 0x800 + +/* support only keyboard, trackpad and management dev for now */ +#define SPIHID_MAX_DEVICES 3 + +#define SPIHID_DEVICE_ID_MNGT 0x0 +#define SPIHID_DEVICE_ID_KBD 0x1 +#define SPIHID_DEVICE_ID_TP 0x2 +#define SPIHID_DEVICE_ID_INFO 0xd0 + +#define SPIHID_READ_PACKET 0x20 +#define SPIHID_WRITE_PACKET 0x40 + +#define SPIHID_DESC_MAX 512 + +#define SPIHID_SET_LEDS 0x0151 /* caps lock */ + +#define SPI_RW_CHG_DELAY_US 200 /* 'Inter Stage Us'? */ + +static const u8 spi_hid_apple_booted[4] = { 0xa0, 0x80, 0x00, 0x00 }; +static const u8 spi_hid_apple_status_ok[4] = { 0xac, 0x27, 0x68, 0xd5 }; + +struct spihid_interface { + struct hid_device *hid; + u8 *hid_desc; + u32 hid_desc_len; + u32 id; + unsigned country; + u32 max_control_report_len; + u32 max_input_report_len; + u32 max_output_report_len; + u8 name[32]; + u8 reply_buf[SPIHID_DESC_MAX]; + u32 reply_len; + bool ready; +}; + +struct spihid_input_report { + u8 *buf; + u32 length; + u32 offset; + u8 device; + u8 flags; +}; + +struct spihid_apple { + struct spi_device *spidev; + + struct spihid_apple_ops *ops; + + struct spihid_interface mngt; + struct spihid_interface kbd; + struct spihid_interface tp; + + wait_queue_head_t wait; + struct mutex tx_lock; //< protects against concurrent SPI writes + + struct spi_message rx_msg; + struct spi_message tx_msg; + struct spi_transfer rx_transfer; + struct spi_transfer tx_transfer; + struct spi_transfer status_transfer; + + u8 *rx_buf; + u8 *tx_buf; + u8 *status_buf; + + u8 vendor[32]; + u8 product[64]; + u8 serial[32]; + + u32 num_devices; + + u32 vendor_id; + u32 product_id; + u32 version_number; + + u8 msg_id; + + /* fragmented HID report */ + struct spihid_input_report report; + + /* state tracking flags */ + bool status_booted; + +#ifdef IRQ_WAKE_SUPPORT + bool irq_wake_enabled; +#endif +}; + +/** + * struct spihid_msg_hdr - common header of protocol messages. + * + * Each message begins with fixed header, followed by a message-type specific + * payload, and ends with a 16-bit crc. Because of the varying lengths of the + * payload, the crc is defined at the end of each payload struct, rather than + * in this struct. + * + * @unknown0: request type? output, input (0x10), feature, protocol + * @unknown1: maybe report id? + * @unknown2: mostly zero, in info request maybe device num + * @id: incremented on each message, rolls over after 255; there is a + * separate counter for each message type. + * @rsplen: response length (the exact nature of this field is quite + * speculative). On a request/write this is often the same as + * @length, though in some cases it has been seen to be much larger + * (e.g. 0x400); on a response/read this the same as on the + * request; for reads that are not responses it is 0. + * @length: length of the remainder of the data in the whole message + * structure (after re-assembly in case of being split over + * multiple spi-packets), minus the trailing crc. The total size + * of a message is therefore @length + 10. + */ + +struct spihid_msg_hdr { + u8 unknown0; + u8 unknown1; + u8 unknown2; + u8 id; + __le16 rsplen; + __le16 length; +}; + +/** + * struct spihid_transfer_packet - a complete spi packet; always 256 bytes. This carries + * the (parts of the) message in the data. But note that this does not + * necessarily contain a complete message, as in some cases (e.g. many + * fingers pressed) the message is split over multiple packets (see the + * @offset, @remain, and @length fields). In general the data parts in + * spihid_transfer_packet's are concatenated until @remaining is 0, and the + * result is an message. + * + * @flags: 0x40 = write (to device), 0x20 = read (from device); note that + * the response to a write still has 0x40. + * @device: 1 = keyboard, 2 = touchpad + * @offset: specifies the offset of this packet's data in the complete + * message; i.e. > 0 indicates this is a continuation packet (in + * the second packet for a message split over multiple packets + * this would then be the same as the @length in the first packet) + * @remain: number of message bytes remaining in subsequents packets (in + * the first packet of a message split over two packets this would + * then be the same as the @length in the second packet) + * @length: length of the valid data in the @data in this packet + * @data: all or part of a message + * @crc16: crc over this whole structure minus this @crc16 field. This + * covers just this packet, even on multi-packet messages (in + * contrast to the crc in the message). + */ +struct spihid_transfer_packet { + u8 flags; + u8 device; + __le16 offset; + __le16 remain; + __le16 length; + u8 data[246]; + __le16 crc16; +}; + +/* + * how HID is mapped onto the protocol is not fully clear. This are the known + * reports/request: + * + * pkt.flags pkt.dev? msg.u0 msg.u1 msg.u2 + * info 0x40 0xd0 0x20 0x01 0xd0 + * + * info mngt: 0x40 0xd0 0x20 0x10 0x00 + * info kbd: 0x40 0xd0 0x20 0x10 0x01 + * info tp: 0x40 0xd0 0x20 0x10 0x02 + * + * desc kbd: 0x40 0xd0 0x20 0x10 0x01 + * desc trackpad: 0x40 0xd0 0x20 0x10 0x02 + * + * mt mode: 0x40 0x02 0x52 0x02 0x00 set protocol? + * capslock led 0x40 0x01 0x51 0x01 0x00 output report + * + * report kbd: 0x20 0x01 0x10 0x01 0x00 input report + * report tp: 0x20 0x02 0x10 0x02 0x00 input report + * + */ + + +static int spihid_apple_request(struct spihid_apple *spihid, u8 target, u8 unk0, + u8 unk1, u8 unk2, u16 resp_len, u8 *buf, + size_t len) +{ + struct spihid_transfer_packet *pkt; + struct spihid_msg_hdr *hdr; + u16 crc; + int err; + + /* know reports are small enoug to fit in a single packet */ + if (len > sizeof(pkt->data) - sizeof(*hdr) - sizeof(__le16)) + return -EINVAL; + + err = mutex_lock_interruptible(&spihid->tx_lock); + if (err < 0) + return err; + + pkt = (struct spihid_transfer_packet *)spihid->tx_buf; + + memset(pkt, 0, sizeof(*pkt)); + pkt->flags = SPIHID_WRITE_PACKET; + pkt->device = target; + pkt->length = cpu_to_le16(sizeof(*hdr) + len + sizeof(__le16)); + + hdr = (struct spihid_msg_hdr *)&pkt->data[0]; + hdr->unknown0 = unk0; + hdr->unknown1 = unk1; + hdr->unknown2 = unk2; + hdr->id = spihid->msg_id++; + hdr->rsplen = cpu_to_le16(resp_len); + hdr->length = cpu_to_le16(len); + + if (len) + memcpy(pkt->data + sizeof(*hdr), buf, len); + crc = crc16(0, &pkt->data[0], sizeof(*hdr) + len); + put_unaligned_le16(crc, pkt->data + sizeof(*hdr) + len); + + pkt->crc16 = cpu_to_le16(crc16(0, spihid->tx_buf, + offsetof(struct spihid_transfer_packet, crc16))); + + memset(spihid->status_buf, 0, sizeof(spi_hid_apple_status_ok)); + + err = spi_sync(spihid->spidev, &spihid->tx_msg); + + if (memcmp(spihid->status_buf, spi_hid_apple_status_ok, + sizeof(spi_hid_apple_status_ok))) { + u8 *b = spihid->status_buf; + dev_warn_ratelimited(&spihid->spidev->dev, "status message " + "mismatch: %02x %02x %02x %02x\n", + b[0], b[1], b[2], b[3]); + } + mutex_unlock(&spihid->tx_lock); + if (err < 0) + return err; + + return (int)len; +} + +static struct spihid_apple *spihid_get_data(struct spihid_interface *idev) +{ + switch (idev->id) { + case SPIHID_DEVICE_ID_KBD: + return container_of(idev, struct spihid_apple, kbd); + case SPIHID_DEVICE_ID_TP: + return container_of(idev, struct spihid_apple, tp); + default: + return NULL; + } +} + +static int apple_ll_start(struct hid_device *hdev) +{ + /* no-op SPI transport is already setup */ + return 0; +}; + +static void apple_ll_stop(struct hid_device *hdev) +{ + /* no-op, devices will be desstroyed on driver destruction */ +} + +static int apple_ll_open(struct hid_device *hdev) +{ + struct spihid_apple *spihid; + struct spihid_interface *idev = hdev->driver_data; + + if (idev->hid_desc_len == 0) { + spihid = spihid_get_data(idev); + dev_warn(&spihid->spidev->dev, + "HID descriptor missing for dev %u", idev->id); + } else + idev->ready = true; + + return 0; +} + +static void apple_ll_close(struct hid_device *hdev) +{ + struct spihid_interface *idev = hdev->driver_data; + idev->ready = false; +} + +static int apple_ll_parse(struct hid_device *hdev) +{ + struct spihid_interface *idev = hdev->driver_data; + + return hid_parse_report(hdev, idev->hid_desc, idev->hid_desc_len); +} + +static int apple_ll_raw_request(struct hid_device *hdev, + unsigned char reportnum, __u8 *buf, size_t len, + unsigned char rtype, int reqtype) +{ + struct spihid_interface *idev = hdev->driver_data; + struct spihid_apple *spihid = spihid_get_data(idev); + int ret; + + dev_dbg(&spihid->spidev->dev, + "apple_ll_raw_request: device:%u reportnum:%hhu rtype:%hhu", + idev->id, reportnum, rtype); + + switch (reqtype) { + case HID_REQ_GET_REPORT: + if (rtype != HID_FEATURE_REPORT) + return -EINVAL; + + idev->reply_len = 0; + ret = spihid_apple_request(spihid, idev->id, 0x32, reportnum, 0x00, len, NULL, 0); + if (ret < 0) + return ret; + + ret = wait_event_interruptible_timeout(spihid->wait, idev->reply_len, + SPIHID_DEF_WAIT); + if (ret == 0) + ret = -ETIMEDOUT; + if (ret < 0) { + dev_err(&spihid->spidev->dev, "waiting for get report failed: %d", ret); + return ret; + } + memcpy(buf, idev->reply_buf, max_t(size_t, len, idev->reply_len)); + return idev->reply_len; + + case HID_REQ_SET_REPORT: + if (buf[0] != reportnum) + return -EINVAL; + if (reportnum != idev->id) { + dev_warn(&spihid->spidev->dev, + "device:%u reportnum:" + "%hhu mismatch", + idev->id, reportnum); + return -EINVAL; + } + return spihid_apple_request(spihid, idev->id, 0x52, reportnum, 0x00, 2, buf, len); + default: + return -EIO; + } +} + +static int apple_ll_output_report(struct hid_device *hdev, __u8 *buf, + size_t len) +{ + struct spihid_interface *idev = hdev->driver_data; + struct spihid_apple *spihid = spihid_get_data(idev); + if (!spihid) + return -1; + + dev_dbg(&spihid->spidev->dev, + "apple_ll_output_report: device:%u len:%zu:", + idev->id, len); + // second idev->id should maybe be buf[0]? + return spihid_apple_request(spihid, idev->id, 0x51, idev->id, 0x00, 0, buf, len); +} + +static struct hid_ll_driver apple_hid_ll = { + .start = &apple_ll_start, + .stop = &apple_ll_stop, + .open = &apple_ll_open, + .close = &apple_ll_close, + .parse = &apple_ll_parse, + .raw_request = &apple_ll_raw_request, + .output_report = &apple_ll_output_report, + .max_buffer_size = SPIHID_MAX_INPUT_REPORT_SIZE, +}; + +static struct spihid_interface *spihid_get_iface(struct spihid_apple *spihid, + u32 iface) +{ + switch (iface) { + case SPIHID_DEVICE_ID_MNGT: + return &spihid->mngt; + case SPIHID_DEVICE_ID_KBD: + return &spihid->kbd; + case SPIHID_DEVICE_ID_TP: + return &spihid->tp; + default: + return NULL; + } +} + +static int spihid_verify_msg(struct spihid_apple *spihid, u8 *buf, size_t len) +{ + u16 msg_crc, crc; + struct device *dev = &spihid->spidev->dev; + + crc = crc16(0, buf, len - sizeof(__le16)); + msg_crc = get_unaligned_le16(buf + len - sizeof(__le16)); + if (crc != msg_crc) { + dev_warn_ratelimited(dev, "Read message crc mismatch\n"); + return 0; + } + return 1; +} + +static bool spihid_status_report(struct spihid_apple *spihid, u8 *pl, + size_t len) +{ + struct device *dev = &spihid->spidev->dev; + dev_dbg(dev, "%s: len: %zu", __func__, len); + if (len == 5 && pl[0] == 0xe0) + return true; + + return false; +} + +static bool spihid_process_input_report(struct spihid_apple *spihid, u32 device, + struct spihid_msg_hdr *hdr, u8 *payload, + size_t len) +{ + //dev_dbg(&spihid>spidev->dev, "input report: req:%hx iface:%u ", hdr->unknown0, device); + if (hdr->unknown0 != 0x10) + return false; + + /* HID device as well but Vendor usage only, handle it internally for now */ + if (device == 0) { + if (hdr->unknown1 == 0xe0) { + return spihid_status_report(spihid, payload, len); + } + } else if (device < SPIHID_MAX_DEVICES) { + struct spihid_interface *iface = + spihid_get_iface(spihid, device); + if (iface && iface->hid && iface->ready) { + hid_input_report(iface->hid, HID_INPUT_REPORT, payload, + len, 1); + return true; + } + } else + dev_dbg(&spihid->spidev->dev, + "unexpected iface:%u for input report", device); + + return false; +} + +struct spihid_device_info { + __le16 u0[2]; + __le16 num_devices; + __le16 vendor_id; + __le16 product_id; + __le16 version_number; + __le16 vendor_str[2]; //< offset and string length + __le16 product_str[2]; //< offset and string length + __le16 serial_str[2]; //< offset and string length +}; + +static bool spihid_process_device_info(struct spihid_apple *spihid, u32 iface, + u8 *payload, size_t len) +{ + struct device *dev = &spihid->spidev->dev; + + if (iface != SPIHID_DEVICE_ID_INFO) + return false; + + if (spihid->vendor_id == 0 && + len >= sizeof(struct spihid_device_info)) { + struct spihid_device_info *info = + (struct spihid_device_info *)payload; + u16 voff, vlen, poff, plen, soff, slen; + u32 num_devices; + + num_devices = __le16_to_cpu(info->num_devices); + + if (num_devices < SPIHID_MAX_DEVICES) { + dev_err(dev, + "Device info reports %u devices, expecting at least 3", + num_devices); + return false; + } + spihid->num_devices = num_devices; + + if (spihid->num_devices > SPIHID_MAX_DEVICES) { + dev_info( + dev, + "limiting the number of devices to mngt, kbd and mouse"); + spihid->num_devices = SPIHID_MAX_DEVICES; + } + + spihid->vendor_id = __le16_to_cpu(info->vendor_id); + spihid->product_id = __le16_to_cpu(info->product_id); + spihid->version_number = __le16_to_cpu(info->version_number); + + voff = __le16_to_cpu(info->vendor_str[0]); + vlen = __le16_to_cpu(info->vendor_str[1]); + + if (voff < len && vlen <= len - voff && + vlen < sizeof(spihid->vendor)) { + memcpy(spihid->vendor, payload + voff, vlen); + spihid->vendor[vlen] = '\0'; + } + + poff = __le16_to_cpu(info->product_str[0]); + plen = __le16_to_cpu(info->product_str[1]); + + if (poff < len && plen <= len - poff && + plen < sizeof(spihid->product)) { + memcpy(spihid->product, payload + poff, plen); + spihid->product[plen] = '\0'; + } + + soff = __le16_to_cpu(info->serial_str[0]); + slen = __le16_to_cpu(info->serial_str[1]); + + if (soff < len && slen <= len - soff && + slen < sizeof(spihid->serial)) { + memcpy(spihid->vendor, payload + soff, slen); + spihid->serial[slen] = '\0'; + } + + wake_up_interruptible(&spihid->wait); + } + return true; +} + +struct spihid_iface_info { + u8 u_0; + u8 interface_num; + u8 u_2; + u8 u_3; + u8 u_4; + u8 country_code; + __le16 max_input_report_len; + __le16 max_output_report_len; + __le16 max_control_report_len; + __le16 name_offset; + __le16 name_length; +}; + +static bool spihid_process_iface_info(struct spihid_apple *spihid, u32 num, + u8 *payload, size_t len) +{ + struct spihid_iface_info *info; + struct spihid_interface *iface = spihid_get_iface(spihid, num); + u32 name_off, name_len; + + if (!iface) + return false; + + if (!iface->max_input_report_len) { + if (len < sizeof(*info)) + return false; + + info = (struct spihid_iface_info *)payload; + + iface->max_input_report_len = + le16_to_cpu(info->max_input_report_len); + iface->max_output_report_len = + le16_to_cpu(info->max_output_report_len); + iface->max_control_report_len = + le16_to_cpu(info->max_control_report_len); + iface->country = info->country_code; + + name_off = le16_to_cpu(info->name_offset); + name_len = le16_to_cpu(info->name_length); + + if (name_off < len && name_len <= len - name_off && + name_len < sizeof(iface->name)) { + memcpy(iface->name, payload + name_off, name_len); + iface->name[name_len] = '\0'; + } + + dev_dbg(&spihid->spidev->dev, "Info for %s, country code: 0x%x", + iface->name, iface->country); + + wake_up_interruptible(&spihid->wait); + } + + return true; +} + +static int spihid_register_hid_device(struct spihid_apple *spihid, + struct spihid_interface *idev, u8 device); + +static bool spihid_process_iface_hid_report_desc(struct spihid_apple *spihid, + u32 num, u8 *payload, + size_t len) +{ + struct spihid_interface *iface = spihid_get_iface(spihid, num); + + if (!iface) + return false; + + if (iface->hid_desc_len == 0) { + if (len > SPIHID_DESC_MAX) + return false; + memcpy(iface->hid_desc, payload, len); + iface->hid_desc_len = len; + + /* do not register the mngt iface as HID device */ + if (num > 0) + spihid_register_hid_device(spihid, iface, num); + + wake_up_interruptible(&spihid->wait); + } + return true; +} + +static bool spihid_process_iface_get_report(struct spihid_apple *spihid, + u32 device, u8 report, + u8 *payload, size_t len) +{ + struct spihid_interface *iface = spihid_get_iface(spihid, device); + + if (!iface) + return false; + + if (len > sizeof(iface->reply_buf) || len < 1) + return false; + + memcpy(iface->reply_buf, payload, len); + iface->reply_len = len; + + wake_up_interruptible(&spihid->wait); + + return true; +} + +static bool spihid_process_response(struct spihid_apple *spihid, u32 device, + struct spihid_msg_hdr *hdr, u8 *payload, + size_t len) +{ + if (hdr->unknown0 == 0x20) { + switch (hdr->unknown1) { + case 0x01: + return spihid_process_device_info(spihid, hdr->unknown2, + payload, len); + case 0x02: + return spihid_process_iface_info(spihid, hdr->unknown2, + payload, len); + case 0x10: + return spihid_process_iface_hid_report_desc( + spihid, hdr->unknown2, payload, len); + default: + break; + } + } + + if (hdr->unknown0 == 0x32) { + return spihid_process_iface_get_report(spihid, device, hdr->unknown1, payload, len); + } + + return false; +} + +static void spihid_process_message(struct spihid_apple *spihid, u8 *data, + size_t length, u8 device, u8 flags) +{ + struct device *dev = &spihid->spidev->dev; + struct spihid_msg_hdr *hdr; + bool handled = false; + size_t payload_len; + u8 *payload; + + if (!spihid_verify_msg(spihid, data, length)) + return; + + hdr = (struct spihid_msg_hdr *)data; + payload_len = le16_to_cpu(hdr->length); + + if (payload_len == 0 || + (payload_len + sizeof(struct spihid_msg_hdr) + 2) > length) + return; + + payload = data + sizeof(struct spihid_msg_hdr); + + switch (flags) { + case SPIHID_READ_PACKET: + handled = spihid_process_input_report(spihid, device, hdr, + payload, payload_len); + break; + case SPIHID_WRITE_PACKET: + handled = spihid_process_response(spihid, device, hdr, payload, + payload_len); + break; + default: + break; + } + +#if defined(DEBUG) && DEBUG > 1 + { + dev_dbg(dev, + "R msg: req:%02hhx rep:%02hhx dev:%02hhx id:%hu len:%hu\n", + hdr->unknown0, hdr->unknown1, hdr->unknown2, hdr->id, + hdr->length); + print_hex_dump_debug("spihid msg: ", DUMP_PREFIX_OFFSET, 16, 1, + payload, le16_to_cpu(hdr->length), true); + } +#else + if (!handled) { + dev_dbg(dev, + "R unhandled msg: req:%02hhx rep:%02hhx dev:%02hhx id:%hu len:%hu\n", + hdr->unknown0, hdr->unknown1, hdr->unknown2, hdr->id, + hdr->length); + print_hex_dump_debug("spihid msg: ", DUMP_PREFIX_OFFSET, 16, 1, + payload, le16_to_cpu(hdr->length), true); + } +#endif +} + +static void spihid_assemble_message(struct spihid_apple *spihid, + struct spihid_transfer_packet *pkt) +{ + size_t length, offset, remain; + struct device *dev = &spihid->spidev->dev; + struct spihid_input_report *rep = &spihid->report; + + length = le16_to_cpu(pkt->length); + remain = le16_to_cpu(pkt->remain); + offset = le16_to_cpu(pkt->offset); + + if (offset + length + remain > U16_MAX) { + return; + } + + if (pkt->device != rep->device || pkt->flags != rep->flags || + offset != rep->offset) { + rep->device = 0; + rep->flags = 0; + rep->offset = 0; + rep->length = 0; + } + + if (offset == 0) { + if (rep->offset != 0) { + dev_warn(dev, "incomplete report off:%u len:%u", + rep->offset, rep->length); + } + memcpy(rep->buf, pkt->data, length); + rep->offset = length; + rep->length = length + remain; + rep->device = pkt->device; + rep->flags = pkt->flags; + } else if (offset == rep->offset) { + if (offset + length + remain != rep->length) { + dev_warn(dev, "incomplete report off:%u len:%u", + rep->offset, rep->length); + return; + } + memcpy(rep->buf + offset, pkt->data, length); + rep->offset += length; + + if (rep->offset == rep->length) { + spihid_process_message(spihid, rep->buf, rep->length, + rep->device, rep->flags); + rep->device = 0; + rep->flags = 0; + rep->offset = 0; + rep->length = 0; + } + } +} + +static void spihid_process_read(struct spihid_apple *spihid) +{ + u16 crc; + size_t length; + struct device *dev = &spihid->spidev->dev; + struct spihid_transfer_packet *pkt; + + pkt = (struct spihid_transfer_packet *)spihid->rx_buf; + + /* check transfer packet crc */ + crc = crc16(0, spihid->rx_buf, + offsetof(struct spihid_transfer_packet, crc16)); + if (crc != le16_to_cpu(pkt->crc16)) { + dev_warn_ratelimited(dev, "Read package crc mismatch\n"); + return; + } + + length = le16_to_cpu(pkt->length); + + if (length < sizeof(struct spihid_msg_hdr) + 2) { + if (length == sizeof(spi_hid_apple_booted) && + !memcmp(pkt->data, spi_hid_apple_booted, length)) { + if (!spihid->status_booted) { + spihid->status_booted = true; + wake_up_interruptible(&spihid->wait); + } + } else { + dev_info(dev, "R short packet: len:%zu\n", length); + print_hex_dump(KERN_INFO, "spihid pkt:", + DUMP_PREFIX_OFFSET, 16, 1, pkt->data, + length, false); + } + return; + } + +#if defined(DEBUG) && DEBUG > 1 + dev_dbg(dev, + "R pkt: flags:%02hhx dev:%02hhx off:%hu remain:%hu, len:%zu\n", + pkt->flags, pkt->device, pkt->offset, pkt->remain, length); +#if defined(DEBUG) && DEBUG > 2 + print_hex_dump_debug("spihid pkt: ", DUMP_PREFIX_OFFSET, 16, 1, + spihid->rx_buf, + sizeof(struct spihid_transfer_packet), true); +#endif +#endif + + if (length > sizeof(pkt->data)) { + dev_warn_ratelimited(dev, "Invalid pkt len:%zu", length); + return; + } + + /* short message */ + if (pkt->offset == 0 && pkt->remain == 0) { + spihid_process_message(spihid, pkt->data, length, pkt->device, + pkt->flags); + } else { + spihid_assemble_message(spihid, pkt); + } +} + +static void spihid_read_packet_sync(struct spihid_apple *spihid) +{ + int err; + + err = spi_sync(spihid->spidev, &spihid->rx_msg); + if (!err) { + spihid_process_read(spihid); + } else { + dev_warn(&spihid->spidev->dev, "RX failed: %d\n", err); + } +} + +irqreturn_t spihid_apple_core_irq(int irq, void *data) +{ + struct spi_device *spi = data; + struct spihid_apple *spihid = spi_get_drvdata(spi); + + spihid_read_packet_sync(spihid); + + return IRQ_HANDLED; +} +EXPORT_SYMBOL_GPL(spihid_apple_core_irq); + +static void spihid_apple_setup_spi_msgs(struct spihid_apple *spihid) +{ + memset(&spihid->rx_transfer, 0, sizeof(spihid->rx_transfer)); + + spihid->rx_transfer.rx_buf = spihid->rx_buf; + spihid->rx_transfer.len = sizeof(struct spihid_transfer_packet); + + spi_message_init(&spihid->rx_msg); + spi_message_add_tail(&spihid->rx_transfer, &spihid->rx_msg); + + memset(&spihid->tx_transfer, 0, sizeof(spihid->rx_transfer)); + memset(&spihid->status_transfer, 0, sizeof(spihid->status_transfer)); + + spihid->tx_transfer.tx_buf = spihid->tx_buf; + spihid->tx_transfer.len = sizeof(struct spihid_transfer_packet); + spihid->tx_transfer.delay.unit = SPI_DELAY_UNIT_USECS; + spihid->tx_transfer.delay.value = SPI_RW_CHG_DELAY_US; + + spihid->status_transfer.rx_buf = spihid->status_buf; + spihid->status_transfer.len = sizeof(spi_hid_apple_status_ok); + + spi_message_init(&spihid->tx_msg); + spi_message_add_tail(&spihid->tx_transfer, &spihid->tx_msg); + spi_message_add_tail(&spihid->status_transfer, &spihid->tx_msg); +} + +static int spihid_apple_setup_spi(struct spihid_apple *spihid) +{ + spihid_apple_setup_spi_msgs(spihid); + + return spihid->ops->power_on(spihid->ops); +} + +static int spihid_register_hid_device(struct spihid_apple *spihid, + struct spihid_interface *iface, u8 device) +{ + int ret; + char *suffix; + struct hid_device *hid; + + iface->id = device; + + hid = hid_allocate_device(); + if (IS_ERR(hid)) + return PTR_ERR(hid); + + /* + * Use 'Apple SPI Keyboard' and 'Apple SPI Trackpad' as input device + * names. The device names need to be distinct since at least Kwin uses + * the tripple Vendor ID, Product ID, Name to identify devices. + */ + snprintf(hid->name, sizeof(hid->name), "Apple SPI %s", iface->name); + // strip ' / Boot' suffix from the name + suffix = strstr(hid->name, " / Boot"); + if (suffix) + suffix[0] = '\0'; + snprintf(hid->phys, sizeof(hid->phys), "%s (%hhx)", + dev_name(&spihid->spidev->dev), device); + strscpy(hid->uniq, spihid->serial, sizeof(hid->uniq)); + + hid->ll_driver = &apple_hid_ll; + hid->bus = BUS_SPI; + hid->vendor = spihid->vendor_id; + hid->product = spihid->product_id; + hid->version = spihid->version_number; + + if (device == SPIHID_DEVICE_ID_KBD) + hid->type = HID_TYPE_SPI_KEYBOARD; + else if (device == SPIHID_DEVICE_ID_TP) + hid->type = HID_TYPE_SPI_MOUSE; + + hid->country = iface->country; + hid->dev.parent = &spihid->spidev->dev; + hid->driver_data = iface; + + ret = hid_add_device(hid); + if (ret < 0) { + hid_destroy_device(hid); + dev_warn(&spihid->spidev->dev, + "Failed to register hid device %hhu", device); + return ret; + } + + iface->hid = hid; + + return 0; +} + +static void spihid_destroy_hid_device(struct spihid_interface *iface) +{ + if (iface->hid) { + hid_destroy_device(iface->hid); + iface->hid = NULL; + } + iface->ready = false; +} + +int spihid_apple_core_probe(struct spi_device *spi, struct spihid_apple_ops *ops) +{ + struct device *dev = &spi->dev; + struct spihid_apple *spihid; + int err, i; + + if (!ops || !ops->power_on || !ops->power_off || !ops->enable_irq || !ops->disable_irq) + return -EINVAL; + + spihid = devm_kzalloc(dev, sizeof(*spihid), GFP_KERNEL); + if (!spihid) + return -ENOMEM; + + spihid->ops = ops; + spihid->spidev = spi; + + // init spi + spi_set_drvdata(spi, spihid); + + /* + * allocate SPI buffers + * Overallocate the receice buffer since it passed directly into + * hid_input_report / hid_report_raw_event. The later expects the buffer + * to be HID_MAX_BUFFER_SIZE (16k) or hid_ll_driver.max_buffer_size if + * set. + */ + spihid->rx_buf = devm_kmalloc( + &spi->dev, SPIHID_MAX_INPUT_REPORT_SIZE, GFP_KERNEL); + spihid->tx_buf = devm_kmalloc( + &spi->dev, sizeof(struct spihid_transfer_packet), GFP_KERNEL); + spihid->status_buf = devm_kmalloc( + &spi->dev, sizeof(spi_hid_apple_status_ok), GFP_KERNEL); + + if (!spihid->rx_buf || !spihid->tx_buf || !spihid->status_buf) + return -ENOMEM; + + spihid->report.buf = + devm_kmalloc(dev, SPIHID_MAX_INPUT_REPORT_SIZE, GFP_KERNEL); + + spihid->kbd.hid_desc = devm_kmalloc(dev, SPIHID_DESC_MAX, GFP_KERNEL); + spihid->tp.hid_desc = devm_kmalloc(dev, SPIHID_DESC_MAX, GFP_KERNEL); + + if (!spihid->report.buf || !spihid->kbd.hid_desc || + !spihid->tp.hid_desc) + return -ENOMEM; + + init_waitqueue_head(&spihid->wait); + + mutex_init(&spihid->tx_lock); + + /* Init spi transfer buffers and power device on */ + err = spihid_apple_setup_spi(spihid); + if (err < 0) + goto error; + + /* enable HID irq */ + spihid->ops->enable_irq(spihid->ops); + + // wait for boot message + err = wait_event_interruptible_timeout(spihid->wait, + spihid->status_booted, + msecs_to_jiffies(1000)); + if (err == 0) + err = -ENODEV; + if (err < 0) { + dev_err(dev, "waiting for device boot failed: %d", err); + goto error; + } + + /* request device information */ + dev_dbg(dev, "request device info"); + spihid_apple_request(spihid, 0xd0, 0x20, 0x01, 0xd0, 0, NULL, 0); + err = wait_event_interruptible_timeout(spihid->wait, spihid->vendor_id, + SPIHID_DEF_WAIT); + if (err == 0) + err = -ENODEV; + if (err < 0) { + dev_err(dev, "waiting for device info failed: %d", err); + goto error; + } + + /* request interface information */ + for (i = 0; i < spihid->num_devices; i++) { + struct spihid_interface *iface = spihid_get_iface(spihid, i); + if (!iface) + continue; + dev_dbg(dev, "request interface info 0x%02x", i); + spihid_apple_request(spihid, 0xd0, 0x20, 0x02, i, + SPIHID_DESC_MAX, NULL, 0); + err = wait_event_interruptible_timeout( + spihid->wait, iface->max_input_report_len, + SPIHID_DEF_WAIT); + } + + /* request HID report descriptors */ + for (i = 1; i < spihid->num_devices; i++) { + struct spihid_interface *iface = spihid_get_iface(spihid, i); + if (!iface) + continue; + dev_dbg(dev, "request hid report desc 0x%02x", i); + spihid_apple_request(spihid, 0xd0, 0x20, 0x10, i, + SPIHID_DESC_MAX, NULL, 0); + wait_event_interruptible_timeout( + spihid->wait, iface->hid_desc_len, SPIHID_DEF_WAIT); + } + + return 0; +error: + return err; +} +EXPORT_SYMBOL_GPL(spihid_apple_core_probe); + +void spihid_apple_core_remove(struct spi_device *spi) +{ + struct spihid_apple *spihid = spi_get_drvdata(spi); + + /* destroy input devices */ + + spihid_destroy_hid_device(&spihid->tp); + spihid_destroy_hid_device(&spihid->kbd); + + /* disable irq */ + spihid->ops->disable_irq(spihid->ops); + + /* power SPI device down */ + spihid->ops->power_off(spihid->ops); +} +EXPORT_SYMBOL_GPL(spihid_apple_core_remove); + +void spihid_apple_core_shutdown(struct spi_device *spi) +{ + struct spihid_apple *spihid = spi_get_drvdata(spi); + + /* disable irq */ + spihid->ops->disable_irq(spihid->ops); + + /* power SPI device down */ + spihid->ops->power_off(spihid->ops); +} +EXPORT_SYMBOL_GPL(spihid_apple_core_shutdown); + +#ifdef CONFIG_PM_SLEEP +static int spihid_apple_core_suspend(struct device *dev) +{ + int ret; +#ifdef IRQ_WAKE_SUPPORT + int wake_status; +#endif + struct spihid_apple *spihid = spi_get_drvdata(to_spi_device(dev)); + + if (spihid->tp.hid) { + ret = hid_driver_suspend(spihid->tp.hid, PMSG_SUSPEND); + if (ret < 0) + return ret; + } + + if (spihid->kbd.hid) { + ret = hid_driver_suspend(spihid->kbd.hid, PMSG_SUSPEND); + if (ret < 0) { + if (spihid->tp.hid) + hid_driver_resume(spihid->tp.hid); + return ret; + } + } + + /* Save some power */ + spihid->ops->disable_irq(spihid->ops); + +#ifdef IRQ_WAKE_SUPPORT + if (device_may_wakeup(dev)) { + wake_status = spihid->ops->enable_irq_wake(spihid->ops); + if (!wake_status) + spihid->irq_wake_enabled = true; + else + dev_warn(dev, "Failed to enable irq wake: %d\n", + wake_status); + } else { + spihid->ops->power_off(spihid->ops); + } +#else + spihid->ops->power_off(spihid->ops); +#endif + + return 0; +} + +static int spihid_apple_core_resume(struct device *dev) +{ + int ret_tp = 0, ret_kbd = 0; + struct spihid_apple *spihid = spi_get_drvdata(to_spi_device(dev)); +#ifdef IRQ_WAKE_SUPPORT + int wake_status; + + if (!device_may_wakeup(dev)) { + spihid->ops->power_on(spihid->ops); + } else if (spihid->irq_wake_enabled) { + wake_status = spihid->ops->disable_irq_wake(spihid->ops); + if (!wake_status) + spihid->irq_wake_enabled = false; + else + dev_warn(dev, "Failed to disable irq wake: %d\n", + wake_status); + } +#endif + + spihid->ops->enable_irq(spihid->ops); + spihid->ops->power_on(spihid->ops); + + if (spihid->tp.hid) + ret_tp = hid_driver_reset_resume(spihid->tp.hid); + if (spihid->kbd.hid) + ret_kbd = hid_driver_reset_resume(spihid->kbd.hid); + + if (ret_tp < 0) + return ret_tp; + + return ret_kbd; +} +#endif + +const struct dev_pm_ops spihid_apple_core_pm = { + SET_SYSTEM_SLEEP_PM_OPS(spihid_apple_core_suspend, + spihid_apple_core_resume) +}; +EXPORT_SYMBOL_GPL(spihid_apple_core_pm); + +MODULE_DESCRIPTION("Apple SPI HID transport driver"); +MODULE_AUTHOR("Janne Grunau "); +MODULE_LICENSE("GPL"); diff --git a/drivers/hid/spi-hid/spi-hid-apple-of.c b/drivers/hid/spi-hid/spi-hid-apple-of.c new file mode 100644 index 000000000000..b631212b836d --- /dev/null +++ b/drivers/hid/spi-hid/spi-hid-apple-of.c @@ -0,0 +1,153 @@ +/* + * SPDX-License-Identifier: GPL-2.0 + * + * Apple SPI HID transport driver - Open Firmware + * + * Copyright (C) The Asahi Linux Contributors + */ + +#include +#include +#include +#include + +#include "spi-hid-apple.h" + + +struct spihid_apple_of { + struct spihid_apple_ops ops; + + struct gpio_desc *enable_gpio; + int irq; +}; + +static int spihid_apple_of_power_on(struct spihid_apple_ops *ops) +{ + struct spihid_apple_of *sh_of = container_of(ops, struct spihid_apple_of, ops); + + /* reset the controller on boot */ + gpiod_direction_output(sh_of->enable_gpio, 1); + msleep(5); + gpiod_direction_output(sh_of->enable_gpio, 0); + msleep(5); + /* turn SPI device on */ + gpiod_direction_output(sh_of->enable_gpio, 1); + msleep(50); + + return 0; +} + +static int spihid_apple_of_power_off(struct spihid_apple_ops *ops) +{ + struct spihid_apple_of *sh_of = container_of(ops, struct spihid_apple_of, ops); + + /* turn SPI device off */ + gpiod_direction_output(sh_of->enable_gpio, 0); + + return 0; +} + +static int spihid_apple_of_enable_irq(struct spihid_apple_ops *ops) +{ + struct spihid_apple_of *sh_of = container_of(ops, struct spihid_apple_of, ops); + + enable_irq(sh_of->irq); + + return 0; +} + +static int spihid_apple_of_disable_irq(struct spihid_apple_ops *ops) +{ + struct spihid_apple_of *sh_of = container_of(ops, struct spihid_apple_of, ops); + + disable_irq(sh_of->irq); + + return 0; +} + +static int spihid_apple_of_enable_irq_wake(struct spihid_apple_ops *ops) +{ + struct spihid_apple_of *sh_of = container_of(ops, struct spihid_apple_of, ops); + + return enable_irq_wake(sh_of->irq); +} + +static int spihid_apple_of_disable_irq_wake(struct spihid_apple_ops *ops) +{ + struct spihid_apple_of *sh_of = container_of(ops, struct spihid_apple_of, ops); + + return disable_irq_wake(sh_of->irq); +} + +static int spihid_apple_of_probe(struct spi_device *spi) +{ + struct device *dev = &spi->dev; + struct spihid_apple_of *spihid_of; + int err; + + spihid_of = devm_kzalloc(dev, sizeof(*spihid_of), GFP_KERNEL); + if (!spihid_of) + return -ENOMEM; + + spihid_of->ops.power_on = spihid_apple_of_power_on; + spihid_of->ops.power_off = spihid_apple_of_power_off; + spihid_of->ops.enable_irq = spihid_apple_of_enable_irq; + spihid_of->ops.disable_irq = spihid_apple_of_disable_irq; + spihid_of->ops.enable_irq_wake = spihid_apple_of_enable_irq_wake; + spihid_of->ops.disable_irq_wake = spihid_apple_of_disable_irq_wake; + + spihid_of->enable_gpio = devm_gpiod_get_index(dev, "spien", 0, 0); + if (IS_ERR(spihid_of->enable_gpio)) { + err = PTR_ERR(spihid_of->enable_gpio); + dev_err(dev, "failed to get 'spien' gpio pin: %d", err); + return err; + } + + spihid_of->irq = of_irq_get(dev->of_node, 0); + if (spihid_of->irq < 0) { + err = spihid_of->irq; + dev_err(dev, "failed to get 'extended-irq': %d", err); + return err; + } + err = devm_request_threaded_irq(dev, spihid_of->irq, NULL, + spihid_apple_core_irq, IRQF_ONESHOT | IRQF_NO_AUTOEN, + "spi-hid-apple-irq", spi); + if (err < 0) { + dev_err(dev, "failed to request extended-irq %d: %d", + spihid_of->irq, err); + return err; + } + + return spihid_apple_core_probe(spi, &spihid_of->ops); +} + +static const struct of_device_id spihid_apple_of_match[] = { + { .compatible = "apple,spi-hid-transport" }, + {}, +}; +MODULE_DEVICE_TABLE(of, spihid_apple_of_match); + +static struct spi_device_id spihid_apple_of_id[] = { + { "spi-hid-transport", 0 }, + {} +}; +MODULE_DEVICE_TABLE(spi, spihid_apple_of_id); + +static struct spi_driver spihid_apple_of_driver = { + .driver = { + .name = "spi-hid-apple-of", + .pm = &spihid_apple_core_pm, + .of_match_table = of_match_ptr(spihid_apple_of_match), + }, + + .id_table = spihid_apple_of_id, + .probe = spihid_apple_of_probe, + .remove = spihid_apple_core_remove, + .shutdown = spihid_apple_core_shutdown, +}; + +module_spi_driver(spihid_apple_of_driver); + +MODULE_DESCRIPTION("Apple SPI HID transport driver for OpenFirmware systems"); +MODULE_AUTHOR("Janne Grunau "); +MODULE_LICENSE("GPL"); diff --git a/drivers/hid/spi-hid/spi-hid-apple.h b/drivers/hid/spi-hid/spi-hid-apple.h new file mode 100644 index 000000000000..9abecd1ba780 --- /dev/null +++ b/drivers/hid/spi-hid/spi-hid-apple.h @@ -0,0 +1,35 @@ +/* SPDX-License-Identifier: GPL-2.0-only OR MIT */ + +#ifndef SPI_HID_APPLE_H +#define SPI_HID_APPLE_H + +#include +#include + +/** + * struct spihid_apple_ops - Ops to control the device from the core driver. + * + * @power_on: reset and power the device on. + * @power_off: power the device off. + * @enable_irq: enable irq or ACPI gpe. + * @disable_irq: disable irq or ACPI gpe. + */ + +struct spihid_apple_ops { + int (*power_on)(struct spihid_apple_ops *ops); + int (*power_off)(struct spihid_apple_ops *ops); + int (*enable_irq)(struct spihid_apple_ops *ops); + int (*disable_irq)(struct spihid_apple_ops *ops); + int (*enable_irq_wake)(struct spihid_apple_ops *ops); + int (*disable_irq_wake)(struct spihid_apple_ops *ops); +}; + +irqreturn_t spihid_apple_core_irq(int irq, void *data); + +int spihid_apple_core_probe(struct spi_device *spi, struct spihid_apple_ops *ops); +void spihid_apple_core_remove(struct spi_device *spi); +void spihid_apple_core_shutdown(struct spi_device *spi); + +extern const struct dev_pm_ops spihid_apple_core_pm; + +#endif /* SPI_HID_APPLE_H */ diff --git a/drivers/hwmon/applesmc.c b/drivers/hwmon/applesmc.c index fc6d6a9053ce..698f44794453 100644 --- a/drivers/hwmon/applesmc.c +++ b/drivers/hwmon/applesmc.c @@ -6,6 +6,7 @@ * * Copyright (C) 2007 Nicolas Boichat * Copyright (C) 2010 Henrik Rydberg + * Copyright (C) 2019 Paul Pawlowski * * Based on hdaps.c driver: * Copyright (C) 2005 Robert Love @@ -18,7 +19,7 @@ #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt #include -#include +#include #include #include #include @@ -35,12 +36,24 @@ #include /* data port used by Apple SMC */ -#define APPLESMC_DATA_PORT 0x300 +#define APPLESMC_DATA_PORT 0 /* command/status port used by Apple SMC */ -#define APPLESMC_CMD_PORT 0x304 +#define APPLESMC_CMD_PORT 4 #define APPLESMC_NR_PORTS 32 /* 0x300-0x31f */ +#define APPLESMC_IOMEM_KEY_DATA 0 +#define APPLESMC_IOMEM_KEY_STATUS 0x4005 +#define APPLESMC_IOMEM_KEY_NAME 0x78 +#define APPLESMC_IOMEM_KEY_DATA_LEN 0x7D +#define APPLESMC_IOMEM_KEY_SMC_ID 0x7E +#define APPLESMC_IOMEM_KEY_CMD 0x7F +#define APPLESMC_IOMEM_MIN_SIZE 0x4006 + +#define APPLESMC_IOMEM_KEY_TYPE_CODE 0 +#define APPLESMC_IOMEM_KEY_TYPE_DATA_LEN 5 +#define APPLESMC_IOMEM_KEY_TYPE_FLAGS 6 + #define APPLESMC_MAX_DATA_LENGTH 32 /* Apple SMC status bits */ @@ -74,6 +87,7 @@ #define FAN_ID_FMT "F%dID" /* r-o char[16] */ #define TEMP_SENSOR_TYPE "sp78" +#define FLOAT_TYPE "flt " /* List of keys used to read/write fan speeds */ static const char *const fan_speed_fmt[] = { @@ -83,6 +97,7 @@ static const char *const fan_speed_fmt[] = { "F%dSf", /* safe speed - not all models */ "F%dTg", /* target speed (manual: rw) */ }; +#define FAN_MANUAL_FMT "F%dMd" #define INIT_TIMEOUT_MSECS 5000 /* wait up to 5s for device init ... */ #define INIT_WAIT_MSECS 50 /* ... in 50ms increments */ @@ -119,7 +134,7 @@ struct applesmc_entry { }; /* Register lookup and registers common to all SMCs */ -static struct applesmc_registers { +struct applesmc_registers { struct mutex mutex; /* register read/write mutex */ unsigned int key_count; /* number of SMC registers */ unsigned int fan_count; /* number of fans */ @@ -133,26 +148,38 @@ static struct applesmc_registers { bool init_complete; /* true when fully initialized */ struct applesmc_entry *cache; /* cached key entries */ const char **index; /* temperature key index */ -} smcreg = { - .mutex = __MUTEX_INITIALIZER(smcreg.mutex), }; -static const int debug; -static struct platform_device *pdev; -static s16 rest_x; -static s16 rest_y; -static u8 backlight_state[2]; +struct applesmc_device { + struct acpi_device *dev; + struct device *ldev; + struct applesmc_registers reg; -static struct device *hwmon_dev; -static struct input_dev *applesmc_idev; + bool port_base_set, iomem_base_set; + u16 port_base; + u8 *__iomem iomem_base; + u32 iomem_base_addr, iomem_base_size; -/* - * Last index written to key_at_index sysfs file, and value to use for all other - * key_at_index_* sysfs files. - */ -static unsigned int key_at_index; + s16 rest_x; + s16 rest_y; + + u8 backlight_state[2]; + + struct device *hwmon_dev; + struct input_dev *idev; + + /* + * Last index written to key_at_index sysfs file, and value to use for all other + * key_at_index_* sysfs files. + */ + unsigned int key_at_index; -static struct workqueue_struct *applesmc_led_wq; + struct workqueue_struct *backlight_wq; + struct work_struct backlight_work; + struct led_classdev backlight_dev; +}; + +static const int debug; /* * Wait for specific status bits with a mask on the SMC. @@ -162,7 +189,7 @@ static struct workqueue_struct *applesmc_led_wq; * run out past 500ms. */ -static int wait_status(u8 val, u8 mask) +static int port_wait_status(struct applesmc_device *smc, u8 val, u8 mask) { u8 status; int us; @@ -170,7 +197,7 @@ static int wait_status(u8 val, u8 mask) us = APPLESMC_MIN_WAIT; for (i = 0; i < 24 ; i++) { - status = inb(APPLESMC_CMD_PORT); + status = inb(smc->port_base + APPLESMC_CMD_PORT); if ((status & mask) == val) return 0; usleep_range(us, us * 2); @@ -180,13 +207,13 @@ static int wait_status(u8 val, u8 mask) return -EIO; } -/* send_byte - Write to SMC data port. Callers must hold applesmc_lock. */ +/* port_send_byte - Write to SMC data port. Callers must hold applesmc_lock. */ -static int send_byte(u8 cmd, u16 port) +static int port_send_byte(struct applesmc_device *smc, u8 cmd, u16 port) { int status; - status = wait_status(0, SMC_STATUS_IB_CLOSED); + status = port_wait_status(smc, 0, SMC_STATUS_IB_CLOSED); if (status) return status; /* @@ -195,24 +222,25 @@ static int send_byte(u8 cmd, u16 port) * this extra read may not happen if status returns both * simultaneously and this would appear to be required. */ - status = wait_status(SMC_STATUS_BUSY, SMC_STATUS_BUSY); + status = port_wait_status(smc, SMC_STATUS_BUSY, SMC_STATUS_BUSY); if (status) return status; - outb(cmd, port); + outb(cmd, smc->port_base + port); return 0; } -/* send_command - Write a command to the SMC. Callers must hold applesmc_lock. */ +/* port_send_command - Write a command to the SMC. Callers must hold applesmc_lock. */ -static int send_command(u8 cmd) +static int port_send_command(struct applesmc_device *smc, u8 cmd) { int ret; - ret = wait_status(0, SMC_STATUS_IB_CLOSED); + ret = port_wait_status(smc, 0, SMC_STATUS_IB_CLOSED); if (ret) return ret; - outb(cmd, APPLESMC_CMD_PORT); + + outb(cmd, smc->port_base + APPLESMC_CMD_PORT); return 0; } @@ -222,110 +250,304 @@ static int send_command(u8 cmd) * If busy is stuck high after the command then the SMC is jammed. */ -static int smc_sane(void) +static int port_smc_sane(struct applesmc_device *smc) { int ret; - ret = wait_status(0, SMC_STATUS_BUSY); + ret = port_wait_status(smc, 0, SMC_STATUS_BUSY); if (!ret) return ret; - ret = send_command(APPLESMC_READ_CMD); + ret = port_send_command(smc, APPLESMC_READ_CMD); if (ret) return ret; - return wait_status(0, SMC_STATUS_BUSY); + return port_wait_status(smc, 0, SMC_STATUS_BUSY); } -static int send_argument(const char *key) +static int port_send_argument(struct applesmc_device *smc, const char *key) { int i; for (i = 0; i < 4; i++) - if (send_byte(key[i], APPLESMC_DATA_PORT)) + if (port_send_byte(smc, key[i], APPLESMC_DATA_PORT)) return -EIO; return 0; } -static int read_smc(u8 cmd, const char *key, u8 *buffer, u8 len) +static int port_read_smc(struct applesmc_device *smc, u8 cmd, const char *key, + u8 *buffer, u8 len) { u8 status, data = 0; int i; int ret; - ret = smc_sane(); + ret = port_smc_sane(smc); if (ret) return ret; - if (send_command(cmd) || send_argument(key)) { + if (port_send_command(smc, cmd) || port_send_argument(smc, key)) { pr_warn("%.4s: read arg fail\n", key); return -EIO; } /* This has no effect on newer (2012) SMCs */ - if (send_byte(len, APPLESMC_DATA_PORT)) { + if (port_send_byte(smc, len, APPLESMC_DATA_PORT)) { pr_warn("%.4s: read len fail\n", key); return -EIO; } for (i = 0; i < len; i++) { - if (wait_status(SMC_STATUS_AWAITING_DATA | SMC_STATUS_BUSY, + if (port_wait_status(smc, + SMC_STATUS_AWAITING_DATA | SMC_STATUS_BUSY, SMC_STATUS_AWAITING_DATA | SMC_STATUS_BUSY)) { pr_warn("%.4s: read data[%d] fail\n", key, i); return -EIO; } - buffer[i] = inb(APPLESMC_DATA_PORT); + buffer[i] = inb(smc->port_base + APPLESMC_DATA_PORT); } /* Read the data port until bit0 is cleared */ for (i = 0; i < 16; i++) { udelay(APPLESMC_MIN_WAIT); - status = inb(APPLESMC_CMD_PORT); + status = inb(smc->port_base + APPLESMC_CMD_PORT); if (!(status & SMC_STATUS_AWAITING_DATA)) break; - data = inb(APPLESMC_DATA_PORT); + data = inb(smc->port_base + APPLESMC_DATA_PORT); } if (i) pr_warn("flushed %d bytes, last value is: %d\n", i, data); - return wait_status(0, SMC_STATUS_BUSY); + return port_wait_status(smc, 0, SMC_STATUS_BUSY); } -static int write_smc(u8 cmd, const char *key, const u8 *buffer, u8 len) +static int port_write_smc(struct applesmc_device *smc, u8 cmd, const char *key, + const u8 *buffer, u8 len) { int i; int ret; - ret = smc_sane(); + ret = port_smc_sane(smc); if (ret) return ret; - if (send_command(cmd) || send_argument(key)) { + if (port_send_command(smc, cmd) || port_send_argument(smc, key)) { pr_warn("%s: write arg fail\n", key); return -EIO; } - if (send_byte(len, APPLESMC_DATA_PORT)) { + if (port_send_byte(smc, len, APPLESMC_DATA_PORT)) { pr_warn("%.4s: write len fail\n", key); return -EIO; } for (i = 0; i < len; i++) { - if (send_byte(buffer[i], APPLESMC_DATA_PORT)) { + if (port_send_byte(smc, buffer[i], APPLESMC_DATA_PORT)) { pr_warn("%s: write data fail\n", key); return -EIO; } } - return wait_status(0, SMC_STATUS_BUSY); + return port_wait_status(smc, 0, SMC_STATUS_BUSY); } -static int read_register_count(unsigned int *count) +static int port_get_smc_key_info(struct applesmc_device *smc, + const char *key, struct applesmc_entry *info) { - __be32 be; int ret; + u8 raw[6]; - ret = read_smc(APPLESMC_READ_CMD, KEY_COUNT_KEY, (u8 *)&be, 4); + ret = port_read_smc(smc, APPLESMC_GET_KEY_TYPE_CMD, key, raw, 6); if (ret) return ret; + info->len = raw[0]; + memcpy(info->type, &raw[1], 4); + info->flags = raw[5]; + return 0; +} + + +/* + * MMIO based communication. + * TODO: Use updated mechanism for cmd timeout/retry + */ + +static void iomem_clear_status(struct applesmc_device *smc) +{ + if (ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_STATUS)) + iowrite8(0, smc->iomem_base + APPLESMC_IOMEM_KEY_STATUS); +} + +static int iomem_wait_read(struct applesmc_device *smc) +{ + u8 status; + int us; + int i; + + us = APPLESMC_MIN_WAIT; + for (i = 0; i < 24 ; i++) { + status = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_STATUS); + if (status & 0x20) + return 0; + usleep_range(us, us * 2); + if (i > 9) + us <<= 1; + } + + dev_warn(smc->ldev, "%s... timeout\n", __func__); + return -EIO; +} + +static int iomem_read_smc(struct applesmc_device *smc, u8 cmd, const char *key, + u8 *buffer, u8 len) +{ + u8 err, remote_len; + u32 key_int = *((u32 *) key); + + iomem_clear_status(smc); + iowrite32(key_int, smc->iomem_base + APPLESMC_IOMEM_KEY_NAME); + iowrite32(0, smc->iomem_base + APPLESMC_IOMEM_KEY_SMC_ID); + iowrite32(cmd, smc->iomem_base + APPLESMC_IOMEM_KEY_CMD); + + if (iomem_wait_read(smc)) + return -EIO; + + err = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_CMD); + if (err != 0) { + dev_warn(smc->ldev, "read_smc_mmio(%x %8x/%.4s) failed: %u\n", + cmd, key_int, key, err); + return -EIO; + } + + if (cmd == APPLESMC_READ_CMD) { + remote_len = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_DATA_LEN); + if (remote_len != len) { + dev_warn(smc->ldev, + "read_smc_mmio(%x %8x/%.4s) failed: buffer length mismatch (remote = %u, requested = %u)\n", + cmd, key_int, key, remote_len, len); + return -EINVAL; + } + } else { + remote_len = len; + } + + memcpy_fromio(buffer, smc->iomem_base + APPLESMC_IOMEM_KEY_DATA, + remote_len); + + dev_dbg(smc->ldev, "read_smc_mmio(%x %8x/%.4s): buflen=%u reslen=%u\n", + cmd, key_int, key, len, remote_len); + print_hex_dump_bytes("read_smc_mmio(): ", DUMP_PREFIX_NONE, buffer, remote_len); + return 0; +} + +static int iomem_get_smc_key_type(struct applesmc_device *smc, const char *key, + struct applesmc_entry *e) +{ + u8 err; + u8 cmd = APPLESMC_GET_KEY_TYPE_CMD; + u32 key_int = *((u32 *) key); + + iomem_clear_status(smc); + iowrite32(key_int, smc->iomem_base + APPLESMC_IOMEM_KEY_NAME); + iowrite32(0, smc->iomem_base + APPLESMC_IOMEM_KEY_SMC_ID); + iowrite32(cmd, smc->iomem_base + APPLESMC_IOMEM_KEY_CMD); + + if (iomem_wait_read(smc)) + return -EIO; + + err = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_CMD); + if (err != 0) { + dev_warn(smc->ldev, "get_smc_key_type_mmio(%.4s) failed: %u\n", key, err); + return -EIO; + } + + e->len = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_TYPE_DATA_LEN); + *((uint32_t *) e->type) = ioread32( + smc->iomem_base + APPLESMC_IOMEM_KEY_TYPE_CODE); + e->flags = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_TYPE_FLAGS); + + dev_dbg(smc->ldev, "get_smc_key_type_mmio(%.4s): len=%u type=%.4s flags=%x\n", + key, e->len, e->type, e->flags); + return 0; +} + +static int iomem_write_smc(struct applesmc_device *smc, u8 cmd, const char *key, + const u8 *buffer, u8 len) +{ + u8 err; + u32 key_int = *((u32 *) key); + + iomem_clear_status(smc); + iowrite32(key_int, smc->iomem_base + APPLESMC_IOMEM_KEY_NAME); + memcpy_toio(smc->iomem_base + APPLESMC_IOMEM_KEY_DATA, buffer, len); + iowrite32(len, smc->iomem_base + APPLESMC_IOMEM_KEY_DATA_LEN); + iowrite32(0, smc->iomem_base + APPLESMC_IOMEM_KEY_SMC_ID); + iowrite32(cmd, smc->iomem_base + APPLESMC_IOMEM_KEY_CMD); + + if (iomem_wait_read(smc)) + return -EIO; + + err = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_CMD); + if (err != 0) { + dev_warn(smc->ldev, "write_smc_mmio(%x %.4s) failed: %u\n", cmd, key, err); + print_hex_dump_bytes("write_smc_mmio(): ", DUMP_PREFIX_NONE, buffer, len); + return -EIO; + } + + dev_dbg(smc->ldev, "write_smc_mmio(%x %.4s): buflen=%u\n", cmd, key, len); + print_hex_dump_bytes("write_smc_mmio(): ", DUMP_PREFIX_NONE, buffer, len); + return 0; +} + + +static int read_smc(struct applesmc_device *smc, const char *key, + u8 *buffer, u8 len) +{ + if (smc->iomem_base_set) + return iomem_read_smc(smc, APPLESMC_READ_CMD, key, buffer, len); + else + return port_read_smc(smc, APPLESMC_READ_CMD, key, buffer, len); +} + +static int write_smc(struct applesmc_device *smc, const char *key, + const u8 *buffer, u8 len) +{ + if (smc->iomem_base_set) + return iomem_write_smc(smc, APPLESMC_WRITE_CMD, key, buffer, len); + else + return port_write_smc(smc, APPLESMC_WRITE_CMD, key, buffer, len); +} + +static int get_smc_key_by_index(struct applesmc_device *smc, + unsigned int index, char *key) +{ + __be32 be; + + be = cpu_to_be32(index); + if (smc->iomem_base_set) + return iomem_read_smc(smc, APPLESMC_GET_KEY_BY_INDEX_CMD, + (const char *) &be, (u8 *) key, 4); + else + return port_read_smc(smc, APPLESMC_GET_KEY_BY_INDEX_CMD, + (const char *) &be, (u8 *) key, 4); +} + +static int get_smc_key_info(struct applesmc_device *smc, const char *key, + struct applesmc_entry *info) +{ + if (smc->iomem_base_set) + return iomem_get_smc_key_type(smc, key, info); + else + return port_get_smc_key_info(smc, key, info); +} + +static int read_register_count(struct applesmc_device *smc, + unsigned int *count) +{ + __be32 be; + int ret; + + ret = read_smc(smc, KEY_COUNT_KEY, (u8 *)&be, 4); + if (ret < 0) + return ret; *count = be32_to_cpu(be); return 0; @@ -338,76 +560,73 @@ static int read_register_count(unsigned int *count) * All functions below are concurrency safe - callers should NOT hold lock. */ -static int applesmc_read_entry(const struct applesmc_entry *entry, - u8 *buf, u8 len) +static int applesmc_read_entry(struct applesmc_device *smc, + const struct applesmc_entry *entry, u8 *buf, u8 len) { int ret; if (entry->len != len) return -EINVAL; - mutex_lock(&smcreg.mutex); - ret = read_smc(APPLESMC_READ_CMD, entry->key, buf, len); - mutex_unlock(&smcreg.mutex); + mutex_lock(&smc->reg.mutex); + ret = read_smc(smc, entry->key, buf, len); + mutex_unlock(&smc->reg.mutex); return ret; } -static int applesmc_write_entry(const struct applesmc_entry *entry, - const u8 *buf, u8 len) +static int applesmc_write_entry(struct applesmc_device *smc, + const struct applesmc_entry *entry, const u8 *buf, u8 len) { int ret; if (entry->len != len) return -EINVAL; - mutex_lock(&smcreg.mutex); - ret = write_smc(APPLESMC_WRITE_CMD, entry->key, buf, len); - mutex_unlock(&smcreg.mutex); + mutex_lock(&smc->reg.mutex); + ret = write_smc(smc, entry->key, buf, len); + mutex_unlock(&smc->reg.mutex); return ret; } -static const struct applesmc_entry *applesmc_get_entry_by_index(int index) +static const struct applesmc_entry *applesmc_get_entry_by_index( + struct applesmc_device *smc, int index) { - struct applesmc_entry *cache = &smcreg.cache[index]; - u8 key[4], info[6]; - __be32 be; + struct applesmc_entry *cache = &smc->reg.cache[index]; + char key[4]; int ret = 0; if (cache->valid) return cache; - mutex_lock(&smcreg.mutex); + mutex_lock(&smc->reg.mutex); if (cache->valid) goto out; - be = cpu_to_be32(index); - ret = read_smc(APPLESMC_GET_KEY_BY_INDEX_CMD, (u8 *)&be, key, 4); + ret = get_smc_key_by_index(smc, index, key); if (ret) goto out; - ret = read_smc(APPLESMC_GET_KEY_TYPE_CMD, key, info, 6); + memcpy(cache->key, key, 4); + + ret = get_smc_key_info(smc, key, cache); if (ret) goto out; - - memcpy(cache->key, key, 4); - cache->len = info[0]; - memcpy(cache->type, &info[1], 4); - cache->flags = info[5]; cache->valid = true; out: - mutex_unlock(&smcreg.mutex); + mutex_unlock(&smc->reg.mutex); if (ret) return ERR_PTR(ret); return cache; } -static int applesmc_get_lower_bound(unsigned int *lo, const char *key) +static int applesmc_get_lower_bound(struct applesmc_device *smc, + unsigned int *lo, const char *key) { - int begin = 0, end = smcreg.key_count; + int begin = 0, end = smc->reg.key_count; const struct applesmc_entry *entry; while (begin != end) { int middle = begin + (end - begin) / 2; - entry = applesmc_get_entry_by_index(middle); + entry = applesmc_get_entry_by_index(smc, middle); if (IS_ERR(entry)) { *lo = 0; return PTR_ERR(entry); @@ -422,16 +641,17 @@ static int applesmc_get_lower_bound(unsigned int *lo, const char *key) return 0; } -static int applesmc_get_upper_bound(unsigned int *hi, const char *key) +static int applesmc_get_upper_bound(struct applesmc_device *smc, + unsigned int *hi, const char *key) { - int begin = 0, end = smcreg.key_count; + int begin = 0, end = smc->reg.key_count; const struct applesmc_entry *entry; while (begin != end) { int middle = begin + (end - begin) / 2; - entry = applesmc_get_entry_by_index(middle); + entry = applesmc_get_entry_by_index(smc, middle); if (IS_ERR(entry)) { - *hi = smcreg.key_count; + *hi = smc->reg.key_count; return PTR_ERR(entry); } if (strcmp(key, entry->key) < 0) @@ -444,50 +664,54 @@ static int applesmc_get_upper_bound(unsigned int *hi, const char *key) return 0; } -static const struct applesmc_entry *applesmc_get_entry_by_key(const char *key) +static const struct applesmc_entry *applesmc_get_entry_by_key( + struct applesmc_device *smc, const char *key) { int begin, end; int ret; - ret = applesmc_get_lower_bound(&begin, key); + ret = applesmc_get_lower_bound(smc, &begin, key); if (ret) return ERR_PTR(ret); - ret = applesmc_get_upper_bound(&end, key); + ret = applesmc_get_upper_bound(smc, &end, key); if (ret) return ERR_PTR(ret); if (end - begin != 1) return ERR_PTR(-EINVAL); - return applesmc_get_entry_by_index(begin); + return applesmc_get_entry_by_index(smc, begin); } -static int applesmc_read_key(const char *key, u8 *buffer, u8 len) +static int applesmc_read_key(struct applesmc_device *smc, + const char *key, u8 *buffer, u8 len) { const struct applesmc_entry *entry; - entry = applesmc_get_entry_by_key(key); + entry = applesmc_get_entry_by_key(smc, key); if (IS_ERR(entry)) return PTR_ERR(entry); - return applesmc_read_entry(entry, buffer, len); + return applesmc_read_entry(smc, entry, buffer, len); } -static int applesmc_write_key(const char *key, const u8 *buffer, u8 len) +static int applesmc_write_key(struct applesmc_device *smc, + const char *key, const u8 *buffer, u8 len) { const struct applesmc_entry *entry; - entry = applesmc_get_entry_by_key(key); + entry = applesmc_get_entry_by_key(smc, key); if (IS_ERR(entry)) return PTR_ERR(entry); - return applesmc_write_entry(entry, buffer, len); + return applesmc_write_entry(smc, entry, buffer, len); } -static int applesmc_has_key(const char *key, bool *value) +static int applesmc_has_key(struct applesmc_device *smc, + const char *key, bool *value) { const struct applesmc_entry *entry; - entry = applesmc_get_entry_by_key(key); + entry = applesmc_get_entry_by_key(smc, key); if (IS_ERR(entry) && PTR_ERR(entry) != -EINVAL) return PTR_ERR(entry); @@ -498,12 +722,13 @@ static int applesmc_has_key(const char *key, bool *value) /* * applesmc_read_s16 - Read 16-bit signed big endian register */ -static int applesmc_read_s16(const char *key, s16 *value) +static int applesmc_read_s16(struct applesmc_device *smc, + const char *key, s16 *value) { u8 buffer[2]; int ret; - ret = applesmc_read_key(key, buffer, 2); + ret = applesmc_read_key(smc, key, buffer, 2); if (ret) return ret; @@ -511,31 +736,68 @@ static int applesmc_read_s16(const char *key, s16 *value) return 0; } +/** + * applesmc_float_to_u32 - Retrieve the integral part of a float. + * This is needed because Apple made fans use float values in the T2. + * The fractional point is not significantly useful though, and the integral + * part can be easily extracted. + */ +static inline u32 applesmc_float_to_u32(u32 d) +{ + u8 sign = (u8) ((d >> 31) & 1); + s32 exp = (s32) ((d >> 23) & 0xff) - 0x7f; + u32 fr = d & ((1u << 23) - 1); + + if (sign || exp < 0) + return 0; + + return (u32) ((1u << exp) + (fr >> (23 - exp))); +} + +/** + * applesmc_u32_to_float - Convert an u32 into a float. + * See applesmc_float_to_u32 for a rationale. + */ +static inline u32 applesmc_u32_to_float(u32 d) +{ + u32 dc = d, bc = 0, exp; + + if (!d) + return 0; + + while (dc >>= 1) + ++bc; + exp = 0x7f + bc; + + return (u32) ((exp << 23) | + ((d << (23 - (exp - 0x7f))) & ((1u << 23) - 1))); +} /* * applesmc_device_init - initialize the accelerometer. Can sleep. */ -static void applesmc_device_init(void) +static void applesmc_device_init(struct applesmc_device *smc) { int total; u8 buffer[2]; - if (!smcreg.has_accelerometer) + if (!smc->reg.has_accelerometer) return; for (total = INIT_TIMEOUT_MSECS; total > 0; total -= INIT_WAIT_MSECS) { - if (!applesmc_read_key(MOTION_SENSOR_KEY, buffer, 2) && + if (!applesmc_read_key(smc, MOTION_SENSOR_KEY, buffer, 2) && (buffer[0] != 0x00 || buffer[1] != 0x00)) return; buffer[0] = 0xe0; buffer[1] = 0x00; - applesmc_write_key(MOTION_SENSOR_KEY, buffer, 2); + applesmc_write_key(smc, MOTION_SENSOR_KEY, buffer, 2); msleep(INIT_WAIT_MSECS); } pr_warn("failed to init the device\n"); } -static int applesmc_init_index(struct applesmc_registers *s) +static int applesmc_init_index(struct applesmc_device *smc, + struct applesmc_registers *s) { const struct applesmc_entry *entry; unsigned int i; @@ -548,7 +810,7 @@ static int applesmc_init_index(struct applesmc_registers *s) return -ENOMEM; for (i = s->temp_begin; i < s->temp_end; i++) { - entry = applesmc_get_entry_by_index(i); + entry = applesmc_get_entry_by_index(smc, i); if (IS_ERR(entry)) continue; if (strcmp(entry->type, TEMP_SENSOR_TYPE)) @@ -562,9 +824,9 @@ static int applesmc_init_index(struct applesmc_registers *s) /* * applesmc_init_smcreg_try - Try to initialize register cache. Idempotent. */ -static int applesmc_init_smcreg_try(void) +static int applesmc_init_smcreg_try(struct applesmc_device *smc) { - struct applesmc_registers *s = &smcreg; + struct applesmc_registers *s = &smc->reg; bool left_light_sensor = false, right_light_sensor = false; unsigned int count; u8 tmp[1]; @@ -573,7 +835,7 @@ static int applesmc_init_smcreg_try(void) if (s->init_complete) return 0; - ret = read_register_count(&count); + ret = read_register_count(smc, &count); if (ret) return ret; @@ -590,35 +852,35 @@ static int applesmc_init_smcreg_try(void) if (!s->cache) return -ENOMEM; - ret = applesmc_read_key(FANS_COUNT, tmp, 1); + ret = applesmc_read_key(smc, FANS_COUNT, tmp, 1); if (ret) return ret; s->fan_count = tmp[0]; if (s->fan_count > 10) s->fan_count = 10; - ret = applesmc_get_lower_bound(&s->temp_begin, "T"); + ret = applesmc_get_lower_bound(smc, &s->temp_begin, "T"); if (ret) return ret; - ret = applesmc_get_lower_bound(&s->temp_end, "U"); + ret = applesmc_get_lower_bound(smc, &s->temp_end, "U"); if (ret) return ret; s->temp_count = s->temp_end - s->temp_begin; - ret = applesmc_init_index(s); + ret = applesmc_init_index(smc, s); if (ret) return ret; - ret = applesmc_has_key(LIGHT_SENSOR_LEFT_KEY, &left_light_sensor); + ret = applesmc_has_key(smc, LIGHT_SENSOR_LEFT_KEY, &left_light_sensor); if (ret) return ret; - ret = applesmc_has_key(LIGHT_SENSOR_RIGHT_KEY, &right_light_sensor); + ret = applesmc_has_key(smc, LIGHT_SENSOR_RIGHT_KEY, &right_light_sensor); if (ret) return ret; - ret = applesmc_has_key(MOTION_SENSOR_KEY, &s->has_accelerometer); + ret = applesmc_has_key(smc, MOTION_SENSOR_KEY, &s->has_accelerometer); if (ret) return ret; - ret = applesmc_has_key(BACKLIGHT_KEY, &s->has_key_backlight); + ret = applesmc_has_key(smc, BACKLIGHT_KEY, &s->has_key_backlight); if (ret) return ret; @@ -634,13 +896,13 @@ static int applesmc_init_smcreg_try(void) return 0; } -static void applesmc_destroy_smcreg(void) +static void applesmc_destroy_smcreg(struct applesmc_device *smc) { - kfree(smcreg.index); - smcreg.index = NULL; - kfree(smcreg.cache); - smcreg.cache = NULL; - smcreg.init_complete = false; + kfree(smc->reg.index); + smc->reg.index = NULL; + kfree(smc->reg.cache); + smc->reg.cache = NULL; + smc->reg.init_complete = false; } /* @@ -649,12 +911,12 @@ static void applesmc_destroy_smcreg(void) * Retries until initialization is successful, or the operation times out. * */ -static int applesmc_init_smcreg(void) +static int applesmc_init_smcreg(struct applesmc_device *smc) { int ms, ret; for (ms = 0; ms < INIT_TIMEOUT_MSECS; ms += INIT_WAIT_MSECS) { - ret = applesmc_init_smcreg_try(); + ret = applesmc_init_smcreg_try(smc); if (!ret) { if (ms) pr_info("init_smcreg() took %d ms\n", ms); @@ -663,50 +925,223 @@ static int applesmc_init_smcreg(void) msleep(INIT_WAIT_MSECS); } - applesmc_destroy_smcreg(); + applesmc_destroy_smcreg(smc); return ret; } /* Device model stuff */ -static int applesmc_probe(struct platform_device *dev) + +static int applesmc_init_resources(struct applesmc_device *smc); +static void applesmc_free_resources(struct applesmc_device *smc); +static int applesmc_create_modules(struct applesmc_device *smc); +static void applesmc_destroy_modules(struct applesmc_device *smc); + +static int applesmc_add(struct acpi_device *dev) { + struct applesmc_device *smc; int ret; - ret = applesmc_init_smcreg(); + smc = kzalloc(sizeof(struct applesmc_device), GFP_KERNEL); + if (!smc) + return -ENOMEM; + smc->dev = dev; + smc->ldev = &dev->dev; + mutex_init(&smc->reg.mutex); + + dev_set_drvdata(&dev->dev, smc); + + ret = applesmc_init_resources(smc); if (ret) - return ret; + goto out_mem; + + ret = applesmc_init_smcreg(smc); + if (ret) + goto out_res; + + applesmc_device_init(smc); + + ret = applesmc_create_modules(smc); + if (ret) + goto out_reg; + + return 0; + +out_reg: + applesmc_destroy_smcreg(smc); +out_res: + applesmc_free_resources(smc); +out_mem: + dev_set_drvdata(&dev->dev, NULL); + mutex_destroy(&smc->reg.mutex); + kfree(smc); + + return ret; +} + +static void applesmc_remove(struct acpi_device *dev) +{ + struct applesmc_device *smc = dev_get_drvdata(&dev->dev); + + applesmc_destroy_modules(smc); + applesmc_destroy_smcreg(smc); + applesmc_free_resources(smc); - applesmc_device_init(); + mutex_destroy(&smc->reg.mutex); + kfree(smc); + + return; +} + +static acpi_status applesmc_walk_resources(struct acpi_resource *res, + void *data) +{ + struct applesmc_device *smc = data; + + switch (res->type) { + case ACPI_RESOURCE_TYPE_IO: + if (!smc->port_base_set) { + if (res->data.io.address_length < APPLESMC_NR_PORTS) + return AE_ERROR; + smc->port_base = res->data.io.minimum; + smc->port_base_set = true; + } + return AE_OK; + + case ACPI_RESOURCE_TYPE_FIXED_MEMORY32: + if (!smc->iomem_base_set) { + if (res->data.fixed_memory32.address_length < + APPLESMC_IOMEM_MIN_SIZE) { + dev_warn(smc->ldev, "found iomem but it's too small: %u\n", + res->data.fixed_memory32.address_length); + return AE_OK; + } + smc->iomem_base_addr = res->data.fixed_memory32.address; + smc->iomem_base_size = res->data.fixed_memory32.address_length; + smc->iomem_base_set = true; + } + return AE_OK; + + case ACPI_RESOURCE_TYPE_END_TAG: + if (smc->port_base_set) + return AE_OK; + else + return AE_NOT_FOUND; + + default: + return AE_OK; + } +} + +static int applesmc_try_enable_iomem(struct applesmc_device *smc); + +static int applesmc_init_resources(struct applesmc_device *smc) +{ + int ret; + + ret = acpi_walk_resources(smc->dev->handle, METHOD_NAME__CRS, + applesmc_walk_resources, smc); + if (ACPI_FAILURE(ret)) + return -ENXIO; + + if (!request_region(smc->port_base, APPLESMC_NR_PORTS, "applesmc")) + return -ENXIO; + + if (smc->iomem_base_set) { + if (applesmc_try_enable_iomem(smc)) + smc->iomem_base_set = false; + } + + return 0; +} + +static int applesmc_try_enable_iomem(struct applesmc_device *smc) +{ + u8 test_val, ldkn_version; + + dev_dbg(smc->ldev, "Trying to enable iomem based communication\n"); + smc->iomem_base = ioremap(smc->iomem_base_addr, smc->iomem_base_size); + if (!smc->iomem_base) + goto out; + + /* Apple's driver does this check for some reason */ + test_val = ioread8(smc->iomem_base + APPLESMC_IOMEM_KEY_STATUS); + if (test_val == 0xff) { + dev_warn(smc->ldev, + "iomem enable failed: initial status is 0xff (is %x)\n", + test_val); + goto out_iomem; + } + + if (read_smc(smc, "LDKN", &ldkn_version, 1)) { + dev_warn(smc->ldev, "iomem enable failed: ldkn read failed\n"); + goto out_iomem; + } + + if (ldkn_version < 2) { + dev_warn(smc->ldev, + "iomem enable failed: ldkn version %u is less than minimum (2)\n", + ldkn_version); + goto out_iomem; + } return 0; + +out_iomem: + iounmap(smc->iomem_base); + +out: + return -ENXIO; +} + +static void applesmc_free_resources(struct applesmc_device *smc) +{ + if (smc->iomem_base_set) + iounmap(smc->iomem_base); + release_region(smc->port_base, APPLESMC_NR_PORTS); } /* Synchronize device with memorized backlight state */ static int applesmc_pm_resume(struct device *dev) { - if (smcreg.has_key_backlight) - applesmc_write_key(BACKLIGHT_KEY, backlight_state, 2); + struct applesmc_device *smc = dev_get_drvdata(dev); + + if (smc->reg.has_key_backlight) + applesmc_write_key(smc, BACKLIGHT_KEY, smc->backlight_state, 2); + return 0; } /* Reinitialize device on resume from hibernation */ static int applesmc_pm_restore(struct device *dev) { - applesmc_device_init(); + struct applesmc_device *smc = dev_get_drvdata(dev); + + applesmc_device_init(smc); + return applesmc_pm_resume(dev); } +static const struct acpi_device_id applesmc_ids[] = { + {"APP0001", 0}, + {"", 0}, +}; + static const struct dev_pm_ops applesmc_pm_ops = { .resume = applesmc_pm_resume, .restore = applesmc_pm_restore, }; -static struct platform_driver applesmc_driver = { - .probe = applesmc_probe, - .driver = { - .name = "applesmc", - .pm = &applesmc_pm_ops, +static struct acpi_driver applesmc_driver = { + .name = "applesmc", + .class = "applesmc", + .ids = applesmc_ids, + .ops = { + .add = applesmc_add, + .remove = applesmc_remove + }, + .drv = { + .pm = &applesmc_pm_ops }, }; @@ -714,25 +1149,26 @@ static struct platform_driver applesmc_driver = { * applesmc_calibrate - Set our "resting" values. Callers must * hold applesmc_lock. */ -static void applesmc_calibrate(void) +static void applesmc_calibrate(struct applesmc_device *smc) { - applesmc_read_s16(MOTION_SENSOR_X_KEY, &rest_x); - applesmc_read_s16(MOTION_SENSOR_Y_KEY, &rest_y); - rest_x = -rest_x; + applesmc_read_s16(smc, MOTION_SENSOR_X_KEY, &smc->rest_x); + applesmc_read_s16(smc, MOTION_SENSOR_Y_KEY, &smc->rest_y); + smc->rest_x = -smc->rest_x; } static void applesmc_idev_poll(struct input_dev *idev) { + struct applesmc_device *smc = dev_get_drvdata(&idev->dev); s16 x, y; - if (applesmc_read_s16(MOTION_SENSOR_X_KEY, &x)) + if (applesmc_read_s16(smc, MOTION_SENSOR_X_KEY, &x)) return; - if (applesmc_read_s16(MOTION_SENSOR_Y_KEY, &y)) + if (applesmc_read_s16(smc, MOTION_SENSOR_Y_KEY, &y)) return; x = -x; - input_report_abs(idev, ABS_X, x - rest_x); - input_report_abs(idev, ABS_Y, y - rest_y); + input_report_abs(idev, ABS_X, x - smc->rest_x); + input_report_abs(idev, ABS_Y, y - smc->rest_y); input_sync(idev); } @@ -747,16 +1183,17 @@ static ssize_t applesmc_name_show(struct device *dev, static ssize_t applesmc_position_show(struct device *dev, struct device_attribute *attr, char *buf) { + struct applesmc_device *smc = dev_get_drvdata(dev); int ret; s16 x, y, z; - ret = applesmc_read_s16(MOTION_SENSOR_X_KEY, &x); + ret = applesmc_read_s16(smc, MOTION_SENSOR_X_KEY, &x); if (ret) goto out; - ret = applesmc_read_s16(MOTION_SENSOR_Y_KEY, &y); + ret = applesmc_read_s16(smc, MOTION_SENSOR_Y_KEY, &y); if (ret) goto out; - ret = applesmc_read_s16(MOTION_SENSOR_Z_KEY, &z); + ret = applesmc_read_s16(smc, MOTION_SENSOR_Z_KEY, &z); if (ret) goto out; @@ -770,6 +1207,7 @@ static ssize_t applesmc_position_show(struct device *dev, static ssize_t applesmc_light_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); const struct applesmc_entry *entry; static int data_length; int ret; @@ -777,7 +1215,7 @@ static ssize_t applesmc_light_show(struct device *dev, u8 buffer[10]; if (!data_length) { - entry = applesmc_get_entry_by_key(LIGHT_SENSOR_LEFT_KEY); + entry = applesmc_get_entry_by_key(smc, LIGHT_SENSOR_LEFT_KEY); if (IS_ERR(entry)) return PTR_ERR(entry); if (entry->len > 10) @@ -786,7 +1224,7 @@ static ssize_t applesmc_light_show(struct device *dev, pr_info("light sensor data length set to %d\n", data_length); } - ret = applesmc_read_key(LIGHT_SENSOR_LEFT_KEY, buffer, data_length); + ret = applesmc_read_key(smc, LIGHT_SENSOR_LEFT_KEY, buffer, data_length); if (ret) goto out; /* newer macbooks report a single 10-bit bigendian value */ @@ -796,7 +1234,7 @@ static ssize_t applesmc_light_show(struct device *dev, } left = buffer[2]; - ret = applesmc_read_key(LIGHT_SENSOR_RIGHT_KEY, buffer, data_length); + ret = applesmc_read_key(smc, LIGHT_SENSOR_RIGHT_KEY, buffer, data_length); if (ret) goto out; right = buffer[2]; @@ -812,7 +1250,8 @@ static ssize_t applesmc_light_show(struct device *dev, static ssize_t applesmc_show_sensor_label(struct device *dev, struct device_attribute *devattr, char *sysfsbuf) { - const char *key = smcreg.index[to_index(devattr)]; + struct applesmc_device *smc = dev_get_drvdata(dev); + const char *key = smc->reg.index[to_index(devattr)]; return sysfs_emit(sysfsbuf, "%s\n", key); } @@ -821,12 +1260,13 @@ static ssize_t applesmc_show_sensor_label(struct device *dev, static ssize_t applesmc_show_temperature(struct device *dev, struct device_attribute *devattr, char *sysfsbuf) { - const char *key = smcreg.index[to_index(devattr)]; + struct applesmc_device *smc = dev_get_drvdata(dev); + const char *key = smc->reg.index[to_index(devattr)]; int ret; s16 value; int temp; - ret = applesmc_read_s16(key, &value); + ret = applesmc_read_s16(smc, key, &value); if (ret) return ret; @@ -838,6 +1278,8 @@ static ssize_t applesmc_show_temperature(struct device *dev, static ssize_t applesmc_show_fan_speed(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); + const struct applesmc_entry *entry; int ret; unsigned int speed = 0; char newkey[5]; @@ -846,11 +1288,21 @@ static ssize_t applesmc_show_fan_speed(struct device *dev, scnprintf(newkey, sizeof(newkey), fan_speed_fmt[to_option(attr)], to_index(attr)); - ret = applesmc_read_key(newkey, buffer, 2); + entry = applesmc_get_entry_by_key(smc, newkey); + if (IS_ERR(entry)) + return PTR_ERR(entry); + + if (!strcmp(entry->type, FLOAT_TYPE)) { + ret = applesmc_read_entry(smc, entry, (u8 *) &speed, 4); + speed = applesmc_float_to_u32(speed); + } else { + ret = applesmc_read_entry(smc, entry, buffer, 2); + speed = ((buffer[0] << 8 | buffer[1]) >> 2); + } + if (ret) return ret; - speed = ((buffer[0] << 8 | buffer[1]) >> 2); return sysfs_emit(sysfsbuf, "%u\n", speed); } @@ -858,6 +1310,8 @@ static ssize_t applesmc_store_fan_speed(struct device *dev, struct device_attribute *attr, const char *sysfsbuf, size_t count) { + struct applesmc_device *smc = dev_get_drvdata(dev); + const struct applesmc_entry *entry; int ret; unsigned long speed; char newkey[5]; @@ -869,9 +1323,18 @@ static ssize_t applesmc_store_fan_speed(struct device *dev, scnprintf(newkey, sizeof(newkey), fan_speed_fmt[to_option(attr)], to_index(attr)); - buffer[0] = (speed >> 6) & 0xff; - buffer[1] = (speed << 2) & 0xff; - ret = applesmc_write_key(newkey, buffer, 2); + entry = applesmc_get_entry_by_key(smc, newkey); + if (IS_ERR(entry)) + return PTR_ERR(entry); + + if (!strcmp(entry->type, FLOAT_TYPE)) { + speed = applesmc_u32_to_float(speed); + ret = applesmc_write_entry(smc, entry, (u8 *) &speed, 4); + } else { + buffer[0] = (speed >> 6) & 0xff; + buffer[1] = (speed << 2) & 0xff; + ret = applesmc_write_key(smc, newkey, buffer, 2); + } if (ret) return ret; @@ -882,15 +1345,30 @@ static ssize_t applesmc_store_fan_speed(struct device *dev, static ssize_t applesmc_show_fan_manual(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); int ret; u16 manual = 0; u8 buffer[2]; + char newkey[5]; + bool has_newkey = false; + + scnprintf(newkey, sizeof(newkey), FAN_MANUAL_FMT, to_index(attr)); + + ret = applesmc_has_key(smc, newkey, &has_newkey); + if (ret) + return ret; + + if (has_newkey) { + ret = applesmc_read_key(smc, newkey, buffer, 1); + manual = buffer[0]; + } else { + ret = applesmc_read_key(smc, FANS_MANUAL, buffer, 2); + manual = ((buffer[0] << 8 | buffer[1]) >> to_index(attr)) & 0x01; + } - ret = applesmc_read_key(FANS_MANUAL, buffer, 2); if (ret) return ret; - manual = ((buffer[0] << 8 | buffer[1]) >> to_index(attr)) & 0x01; return sysfs_emit(sysfsbuf, "%d\n", manual); } @@ -898,29 +1376,42 @@ static ssize_t applesmc_store_fan_manual(struct device *dev, struct device_attribute *attr, const char *sysfsbuf, size_t count) { + struct applesmc_device *smc = dev_get_drvdata(dev); int ret; u8 buffer[2]; + char newkey[5]; + bool has_newkey = false; unsigned long input; u16 val; if (kstrtoul(sysfsbuf, 10, &input) < 0) return -EINVAL; - ret = applesmc_read_key(FANS_MANUAL, buffer, 2); + scnprintf(newkey, sizeof(newkey), FAN_MANUAL_FMT, to_index(attr)); + + ret = applesmc_has_key(smc, newkey, &has_newkey); if (ret) - goto out; + return ret; - val = (buffer[0] << 8 | buffer[1]); + if (has_newkey) { + buffer[0] = input & 1; + ret = applesmc_write_key(smc, newkey, buffer, 1); + } else { + ret = applesmc_read_key(smc, FANS_MANUAL, buffer, 2); + val = (buffer[0] << 8 | buffer[1]); + if (ret) + goto out; - if (input) - val = val | (0x01 << to_index(attr)); - else - val = val & ~(0x01 << to_index(attr)); + if (input) + val = val | (0x01 << to_index(attr)); + else + val = val & ~(0x01 << to_index(attr)); - buffer[0] = (val >> 8) & 0xFF; - buffer[1] = val & 0xFF; + buffer[0] = (val >> 8) & 0xFF; + buffer[1] = val & 0xFF; - ret = applesmc_write_key(FANS_MANUAL, buffer, 2); + ret = applesmc_write_key(smc, FANS_MANUAL, buffer, 2); + } out: if (ret) @@ -932,13 +1423,14 @@ static ssize_t applesmc_store_fan_manual(struct device *dev, static ssize_t applesmc_show_fan_position(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); int ret; char newkey[5]; u8 buffer[17]; scnprintf(newkey, sizeof(newkey), FAN_ID_FMT, to_index(attr)); - ret = applesmc_read_key(newkey, buffer, 16); + ret = applesmc_read_key(smc, newkey, buffer, 16); buffer[16] = 0; if (ret) @@ -950,43 +1442,79 @@ static ssize_t applesmc_show_fan_position(struct device *dev, static ssize_t applesmc_calibrate_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { - return sysfs_emit(sysfsbuf, "(%d,%d)\n", rest_x, rest_y); + struct applesmc_device *smc = dev_get_drvdata(dev); + + return sysfs_emit(sysfsbuf, "(%d,%d)\n", smc->rest_x, smc->rest_y); } static ssize_t applesmc_calibrate_store(struct device *dev, struct device_attribute *attr, const char *sysfsbuf, size_t count) { - applesmc_calibrate(); + struct applesmc_device *smc = dev_get_drvdata(dev); + + applesmc_calibrate(smc); return count; } static void applesmc_backlight_set(struct work_struct *work) { - applesmc_write_key(BACKLIGHT_KEY, backlight_state, 2); + struct applesmc_device *smc = container_of(work, struct applesmc_device, backlight_work); + + applesmc_write_key(smc, BACKLIGHT_KEY, smc->backlight_state, 2); } -static DECLARE_WORK(backlight_work, &applesmc_backlight_set); static void applesmc_brightness_set(struct led_classdev *led_cdev, enum led_brightness value) { + struct applesmc_device *smc = dev_get_drvdata(led_cdev->dev); int ret; - backlight_state[0] = value; - ret = queue_work(applesmc_led_wq, &backlight_work); + smc->backlight_state[0] = value; + ret = queue_work(smc->backlight_wq, &smc->backlight_work); if (debug && (!ret)) dev_dbg(led_cdev->dev, "work was already on the queue.\n"); } +static ssize_t applesmc_BCLM_store(struct device *dev, + struct device_attribute *attr, char *sysfsbuf, size_t count) +{ + struct applesmc_device *smc = dev_get_drvdata(dev); + u8 val; + + if (kstrtou8(sysfsbuf, 10, &val) < 0) + return -EINVAL; + + if (val < 0 || val > 100) + return -EINVAL; + + if (applesmc_write_key(smc, "BCLM", &val, 1)) + return -ENODEV; + return count; +} + +static ssize_t applesmc_BCLM_show(struct device *dev, + struct device_attribute *attr, char *sysfsbuf) +{ + struct applesmc_device *smc = dev_get_drvdata(dev); + u8 val; + + if (applesmc_read_key(smc, "BCLM", &val, 1)) + return -ENODEV; + + return sysfs_emit(sysfsbuf, "%d\n", val); +} + static ssize_t applesmc_key_count_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); int ret; u8 buffer[4]; u32 count; - ret = applesmc_read_key(KEY_COUNT_KEY, buffer, 4); + ret = applesmc_read_key(smc, KEY_COUNT_KEY, buffer, 4); if (ret) return ret; @@ -998,13 +1526,14 @@ static ssize_t applesmc_key_count_show(struct device *dev, static ssize_t applesmc_key_at_index_read_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); const struct applesmc_entry *entry; int ret; - entry = applesmc_get_entry_by_index(key_at_index); + entry = applesmc_get_entry_by_index(smc, smc->key_at_index); if (IS_ERR(entry)) return PTR_ERR(entry); - ret = applesmc_read_entry(entry, sysfsbuf, entry->len); + ret = applesmc_read_entry(smc, entry, sysfsbuf, entry->len); if (ret) return ret; @@ -1014,9 +1543,10 @@ static ssize_t applesmc_key_at_index_read_show(struct device *dev, static ssize_t applesmc_key_at_index_data_length_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); const struct applesmc_entry *entry; - entry = applesmc_get_entry_by_index(key_at_index); + entry = applesmc_get_entry_by_index(smc, smc->key_at_index); if (IS_ERR(entry)) return PTR_ERR(entry); @@ -1026,9 +1556,10 @@ static ssize_t applesmc_key_at_index_data_length_show(struct device *dev, static ssize_t applesmc_key_at_index_type_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); const struct applesmc_entry *entry; - entry = applesmc_get_entry_by_index(key_at_index); + entry = applesmc_get_entry_by_index(smc, smc->key_at_index); if (IS_ERR(entry)) return PTR_ERR(entry); @@ -1038,9 +1569,10 @@ static ssize_t applesmc_key_at_index_type_show(struct device *dev, static ssize_t applesmc_key_at_index_name_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { + struct applesmc_device *smc = dev_get_drvdata(dev); const struct applesmc_entry *entry; - entry = applesmc_get_entry_by_index(key_at_index); + entry = applesmc_get_entry_by_index(smc, smc->key_at_index); if (IS_ERR(entry)) return PTR_ERR(entry); @@ -1050,28 +1582,25 @@ static ssize_t applesmc_key_at_index_name_show(struct device *dev, static ssize_t applesmc_key_at_index_show(struct device *dev, struct device_attribute *attr, char *sysfsbuf) { - return sysfs_emit(sysfsbuf, "%d\n", key_at_index); + struct applesmc_device *smc = dev_get_drvdata(dev); + + return sysfs_emit(sysfsbuf, "%d\n", smc->key_at_index); } static ssize_t applesmc_key_at_index_store(struct device *dev, struct device_attribute *attr, const char *sysfsbuf, size_t count) { + struct applesmc_device *smc = dev_get_drvdata(dev); unsigned long newkey; if (kstrtoul(sysfsbuf, 10, &newkey) < 0 - || newkey >= smcreg.key_count) + || newkey >= smc->reg.key_count) return -EINVAL; - key_at_index = newkey; + smc->key_at_index = newkey; return count; } -static struct led_classdev applesmc_backlight = { - .name = "smc::kbd_backlight", - .default_trigger = "nand-disk", - .brightness_set = applesmc_brightness_set, -}; - static struct applesmc_node_group info_group[] = { { "name", applesmc_name_show }, { "key_count", applesmc_key_count_show }, @@ -1111,19 +1640,25 @@ static struct applesmc_node_group temp_group[] = { { } }; +static struct applesmc_node_group BCLM_group[] = { + { "battery_charge_limit", applesmc_BCLM_show, applesmc_BCLM_store }, + { } +}; + /* Module stuff */ /* * applesmc_destroy_nodes - remove files and free associated memory */ -static void applesmc_destroy_nodes(struct applesmc_node_group *groups) +static void applesmc_destroy_nodes(struct applesmc_device *smc, + struct applesmc_node_group *groups) { struct applesmc_node_group *grp; struct applesmc_dev_attr *node; for (grp = groups; grp->nodes; grp++) { for (node = grp->nodes; node->sda.dev_attr.attr.name; node++) - sysfs_remove_file(&pdev->dev.kobj, + sysfs_remove_file(&smc->dev->dev.kobj, &node->sda.dev_attr.attr); kfree(grp->nodes); grp->nodes = NULL; @@ -1133,7 +1668,8 @@ static void applesmc_destroy_nodes(struct applesmc_node_group *groups) /* * applesmc_create_nodes - create a two-dimensional group of sysfs files */ -static int applesmc_create_nodes(struct applesmc_node_group *groups, int num) +static int applesmc_create_nodes(struct applesmc_device *smc, + struct applesmc_node_group *groups, int num) { struct applesmc_node_group *grp; struct applesmc_dev_attr *node; @@ -1157,7 +1693,7 @@ static int applesmc_create_nodes(struct applesmc_node_group *groups, int num) sysfs_attr_init(attr); attr->name = node->name; attr->mode = 0444 | (grp->store ? 0200 : 0); - ret = sysfs_create_file(&pdev->dev.kobj, attr); + ret = sysfs_create_file(&smc->dev->dev.kobj, attr); if (ret) { attr->name = NULL; goto out; @@ -1167,57 +1703,56 @@ static int applesmc_create_nodes(struct applesmc_node_group *groups, int num) return 0; out: - applesmc_destroy_nodes(groups); + applesmc_destroy_nodes(smc, groups); return ret; } /* Create accelerometer resources */ -static int applesmc_create_accelerometer(void) +static int applesmc_create_accelerometer(struct applesmc_device *smc) { int ret; - - if (!smcreg.has_accelerometer) + if (!smc->reg.has_accelerometer) return 0; - ret = applesmc_create_nodes(accelerometer_group, 1); + ret = applesmc_create_nodes(smc, accelerometer_group, 1); if (ret) goto out; - applesmc_idev = input_allocate_device(); - if (!applesmc_idev) { + smc->idev = input_allocate_device(); + if (!smc->idev) { ret = -ENOMEM; goto out_sysfs; } /* initial calibrate for the input device */ - applesmc_calibrate(); + applesmc_calibrate(smc); /* initialize the input device */ - applesmc_idev->name = "applesmc"; - applesmc_idev->id.bustype = BUS_HOST; - applesmc_idev->dev.parent = &pdev->dev; - input_set_abs_params(applesmc_idev, ABS_X, + smc->idev->name = "applesmc"; + smc->idev->id.bustype = BUS_HOST; + smc->idev->dev.parent = &smc->dev->dev; + input_set_abs_params(smc->idev, ABS_X, -256, 256, APPLESMC_INPUT_FUZZ, APPLESMC_INPUT_FLAT); - input_set_abs_params(applesmc_idev, ABS_Y, + input_set_abs_params(smc->idev, ABS_Y, -256, 256, APPLESMC_INPUT_FUZZ, APPLESMC_INPUT_FLAT); - ret = input_setup_polling(applesmc_idev, applesmc_idev_poll); + ret = input_setup_polling(smc->idev, applesmc_idev_poll); if (ret) goto out_idev; - input_set_poll_interval(applesmc_idev, APPLESMC_POLL_INTERVAL); + input_set_poll_interval(smc->idev, APPLESMC_POLL_INTERVAL); - ret = input_register_device(applesmc_idev); + ret = input_register_device(smc->idev); if (ret) goto out_idev; return 0; out_idev: - input_free_device(applesmc_idev); + input_free_device(smc->idev); out_sysfs: - applesmc_destroy_nodes(accelerometer_group); + applesmc_destroy_nodes(smc, accelerometer_group); out: pr_warn("driver init failed (ret=%d)!\n", ret); @@ -1225,44 +1760,55 @@ static int applesmc_create_accelerometer(void) } /* Release all resources used by the accelerometer */ -static void applesmc_release_accelerometer(void) +static void applesmc_release_accelerometer(struct applesmc_device *smc) { - if (!smcreg.has_accelerometer) + if (!smc->reg.has_accelerometer) return; - input_unregister_device(applesmc_idev); - applesmc_destroy_nodes(accelerometer_group); + input_unregister_device(smc->idev); + applesmc_destroy_nodes(smc, accelerometer_group); } -static int applesmc_create_light_sensor(void) +static int applesmc_create_light_sensor(struct applesmc_device *smc) { - if (!smcreg.num_light_sensors) + if (!smc->reg.num_light_sensors) return 0; - return applesmc_create_nodes(light_sensor_group, 1); + return applesmc_create_nodes(smc, light_sensor_group, 1); } -static void applesmc_release_light_sensor(void) +static void applesmc_release_light_sensor(struct applesmc_device *smc) { - if (!smcreg.num_light_sensors) + if (!smc->reg.num_light_sensors) return; - applesmc_destroy_nodes(light_sensor_group); + applesmc_destroy_nodes(smc, light_sensor_group); } -static int applesmc_create_key_backlight(void) +static int applesmc_create_key_backlight(struct applesmc_device *smc) { - if (!smcreg.has_key_backlight) + int ret; + + if (!smc->reg.has_key_backlight) return 0; - applesmc_led_wq = create_singlethread_workqueue("applesmc-led"); - if (!applesmc_led_wq) + smc->backlight_wq = create_singlethread_workqueue("applesmc-led"); + if (!smc->backlight_wq) return -ENOMEM; - return led_classdev_register(&pdev->dev, &applesmc_backlight); + + INIT_WORK(&smc->backlight_work, applesmc_backlight_set); + smc->backlight_dev.name = "smc::kbd_backlight"; + smc->backlight_dev.default_trigger = "nand-disk"; + smc->backlight_dev.brightness_set = applesmc_brightness_set; + ret = led_classdev_register(&smc->dev->dev, &smc->backlight_dev); + if (ret) + destroy_workqueue(smc->backlight_wq); + + return ret; } -static void applesmc_release_key_backlight(void) +static void applesmc_release_key_backlight(struct applesmc_device *smc) { - if (!smcreg.has_key_backlight) + if (!smc->reg.has_key_backlight) return; - led_classdev_unregister(&applesmc_backlight); - destroy_workqueue(applesmc_led_wq); + led_classdev_unregister(&smc->backlight_dev); + destroy_workqueue(smc->backlight_wq); } static int applesmc_dmi_match(const struct dmi_system_id *id) @@ -1291,6 +1837,10 @@ static const struct dmi_system_id applesmc_whitelist[] __initconst = { DMI_MATCH(DMI_BOARD_VENDOR, "Apple"), DMI_MATCH(DMI_PRODUCT_NAME, "Macmini") }, }, + { applesmc_dmi_match, "Apple iMacPro", { + DMI_MATCH(DMI_BOARD_VENDOR, "Apple"), + DMI_MATCH(DMI_PRODUCT_NAME, "iMacPro") }, + }, { applesmc_dmi_match, "Apple MacPro", { DMI_MATCH(DMI_BOARD_VENDOR, "Apple"), DMI_MATCH(DMI_PRODUCT_NAME, "MacPro") }, @@ -1306,90 +1856,91 @@ static const struct dmi_system_id applesmc_whitelist[] __initconst = { { .ident = NULL } }; -static int __init applesmc_init(void) +static int applesmc_create_modules(struct applesmc_device *smc) { int ret; - if (!dmi_check_system(applesmc_whitelist)) { - pr_warn("supported laptop not found!\n"); - ret = -ENODEV; - goto out; - } - - if (!request_region(APPLESMC_DATA_PORT, APPLESMC_NR_PORTS, - "applesmc")) { - ret = -ENXIO; - goto out; - } - - ret = platform_driver_register(&applesmc_driver); - if (ret) - goto out_region; - - pdev = platform_device_register_simple("applesmc", APPLESMC_DATA_PORT, - NULL, 0); - if (IS_ERR(pdev)) { - ret = PTR_ERR(pdev); - goto out_driver; - } - - /* create register cache */ - ret = applesmc_init_smcreg(); + ret = applesmc_create_nodes(smc, info_group, 1); if (ret) - goto out_device; - - ret = applesmc_create_nodes(info_group, 1); + goto out; + ret = applesmc_create_nodes(smc, BCLM_group, 1); if (ret) - goto out_smcreg; + goto out_info; - ret = applesmc_create_nodes(fan_group, smcreg.fan_count); + ret = applesmc_create_nodes(smc, fan_group, smc->reg.fan_count); if (ret) - goto out_info; + goto out_bclm; - ret = applesmc_create_nodes(temp_group, smcreg.index_count); + ret = applesmc_create_nodes(smc, temp_group, smc->reg.index_count); if (ret) goto out_fans; - ret = applesmc_create_accelerometer(); + ret = applesmc_create_accelerometer(smc); if (ret) goto out_temperature; - ret = applesmc_create_light_sensor(); + ret = applesmc_create_light_sensor(smc); if (ret) goto out_accelerometer; - ret = applesmc_create_key_backlight(); + ret = applesmc_create_key_backlight(smc); if (ret) goto out_light_sysfs; - hwmon_dev = hwmon_device_register(&pdev->dev); - if (IS_ERR(hwmon_dev)) { - ret = PTR_ERR(hwmon_dev); + smc->hwmon_dev = hwmon_device_register(&smc->dev->dev); + if (IS_ERR(smc->hwmon_dev)) { + ret = PTR_ERR(smc->hwmon_dev); goto out_light_ledclass; } return 0; out_light_ledclass: - applesmc_release_key_backlight(); + applesmc_release_key_backlight(smc); out_light_sysfs: - applesmc_release_light_sensor(); + applesmc_release_light_sensor(smc); out_accelerometer: - applesmc_release_accelerometer(); + applesmc_release_accelerometer(smc); out_temperature: - applesmc_destroy_nodes(temp_group); + applesmc_destroy_nodes(smc, temp_group); out_fans: - applesmc_destroy_nodes(fan_group); + applesmc_destroy_nodes(smc, fan_group); +out_bclm: + applesmc_destroy_nodes(smc, BCLM_group); out_info: - applesmc_destroy_nodes(info_group); -out_smcreg: - applesmc_destroy_smcreg(); -out_device: - platform_device_unregister(pdev); -out_driver: - platform_driver_unregister(&applesmc_driver); -out_region: - release_region(APPLESMC_DATA_PORT, APPLESMC_NR_PORTS); + applesmc_destroy_nodes(smc, info_group); +out: + return ret; +} + +static void applesmc_destroy_modules(struct applesmc_device *smc) +{ + hwmon_device_unregister(smc->hwmon_dev); + applesmc_release_key_backlight(smc); + applesmc_release_light_sensor(smc); + applesmc_release_accelerometer(smc); + applesmc_destroy_nodes(smc, temp_group); + applesmc_destroy_nodes(smc, fan_group); + applesmc_destroy_nodes(smc, BCLM_group); + applesmc_destroy_nodes(smc, info_group); +} + +static int __init applesmc_init(void) +{ + int ret; + + if (!dmi_check_system(applesmc_whitelist)) { + pr_warn("supported laptop not found!\n"); + ret = -ENODEV; + goto out; + } + + ret = acpi_bus_register_driver(&applesmc_driver); + if (ret) + goto out; + + return 0; + out: pr_warn("driver init failed (ret=%d)!\n", ret); return ret; @@ -1397,23 +1948,14 @@ static int __init applesmc_init(void) static void __exit applesmc_exit(void) { - hwmon_device_unregister(hwmon_dev); - applesmc_release_key_backlight(); - applesmc_release_light_sensor(); - applesmc_release_accelerometer(); - applesmc_destroy_nodes(temp_group); - applesmc_destroy_nodes(fan_group); - applesmc_destroy_nodes(info_group); - applesmc_destroy_smcreg(); - platform_device_unregister(pdev); - platform_driver_unregister(&applesmc_driver); - release_region(APPLESMC_DATA_PORT, APPLESMC_NR_PORTS); + acpi_bus_unregister_driver(&applesmc_driver); } module_init(applesmc_init); module_exit(applesmc_exit); MODULE_AUTHOR("Nicolas Boichat"); +MODULE_AUTHOR("Paul Pawlowski"); MODULE_DESCRIPTION("Apple SMC"); MODULE_LICENSE("GPL v2"); MODULE_DEVICE_TABLE(dmi, applesmc_whitelist); diff --git a/drivers/pci/vgaarb.c b/drivers/pci/vgaarb.c index baa242b14099..d89a00080046 100644 --- a/drivers/pci/vgaarb.c +++ b/drivers/pci/vgaarb.c @@ -143,6 +143,7 @@ void vga_set_default_device(struct pci_dev *pdev) pci_dev_put(vga_default); vga_default = pci_dev_get(pdev); } +EXPORT_SYMBOL_GPL(vga_set_default_device); /** * vga_remove_vgacon - deactivate VGA console diff --git a/drivers/platform/x86/apple-gmux.c b/drivers/platform/x86/apple-gmux.c index 1417e230edbd..e69785af8e1d 100644 --- a/drivers/platform/x86/apple-gmux.c +++ b/drivers/platform/x86/apple-gmux.c @@ -21,6 +21,7 @@ #include #include #include +#include #include #include #include @@ -107,6 +108,10 @@ struct apple_gmux_config { # define MMIO_GMUX_MAX_BRIGHTNESS 0xffff +static bool force_igd; +module_param(force_igd, bool, 0); +MODULE_PARM_DESC(force_idg, "Switch gpu to igd on module load. Make sure that you have apple-set-os set up and the iGPU is in `lspci -s 00:02.0`. (default: false) (bool)"); + static u8 gmux_pio_read8(struct apple_gmux_data *gmux_data, int port) { return inb(gmux_data->iostart + port); @@ -945,6 +950,19 @@ static int gmux_probe(struct pnp_dev *pnp, const struct pnp_device_id *id) gmux_enable_interrupts(gmux_data); gmux_read_switch_state(gmux_data); + if (force_igd) { + struct pci_dev *pdev; + + pdev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(2, 0)); + if (pdev) { + pr_info("Switching to IGD"); + gmux_switchto(VGA_SWITCHEROO_IGD); + vga_set_default_device(pdev); + } else { + pr_err("force_idg is true, but couldn't find iGPU at 00:02.0! Is apple-set-os working?"); + } + } + /* * Retina MacBook Pros cannot switch the panel's AUX separately * and need eDP pre-calibration. They are distinguishable from diff --git a/drivers/soc/apple/Kconfig b/drivers/soc/apple/Kconfig index ad6736889231..90d1a448f1f4 100644 --- a/drivers/soc/apple/Kconfig +++ b/drivers/soc/apple/Kconfig @@ -4,6 +4,16 @@ if ARCH_APPLE || COMPILE_TEST menu "Apple SoC drivers" +config APPLE_DOCKCHANNEL + tristate "Apple DockChannel FIFO" + depends on ARCH_APPLE || COMPILE_TEST + default ARCH_APPLE + help + DockChannel is a simple FIFO used on Apple SoCs for debug and inter-processor + communications. + + Say 'y' here if you have an Apple SoC. + config APPLE_MAILBOX tristate "Apple SoC mailboxes" depends on PM @@ -28,6 +38,20 @@ config APPLE_RTKIT Say 'y' here if you have an Apple SoC. +config APPLE_RTKIT_HELPER + tristate "Apple Generic RTKit helper co-processor" + depends on APPLE_RTKIT + depends on ARCH_APPLE || COMPILE_TEST + default ARCH_APPLE + help + Apple SoCs such as the M1 come with various co-processors running + their proprietary RTKit operating system. This option enables support + for a generic co-processor that does not implement any additional + in-band communications. It can be used for testing purposes, or for + coprocessors such as MTP that communicate over a different interface. + + Say 'y' here if you have an Apple SoC. + config APPLE_SART tristate "Apple SART DMA address filter" depends on ARCH_APPLE || COMPILE_TEST diff --git a/drivers/soc/apple/Makefile b/drivers/soc/apple/Makefile index 4d9ab8f3037b..5e526a9edcf2 100644 --- a/drivers/soc/apple/Makefile +++ b/drivers/soc/apple/Makefile @@ -1,10 +1,16 @@ # SPDX-License-Identifier: GPL-2.0-only +obj-$(CONFIG_APPLE_DOCKCHANNEL) += apple-dockchannel.o +apple-dockchannel-y = dockchannel.o + obj-$(CONFIG_APPLE_MAILBOX) += apple-mailbox.o apple-mailbox-y = mailbox.o obj-$(CONFIG_APPLE_RTKIT) += apple-rtkit.o apple-rtkit-y = rtkit.o rtkit-crashlog.o +obj-$(CONFIG_APPLE_RTKIT_HELPER) += apple-rtkit-helper.o +apple-rtkit-helper-y = rtkit-helper.o + obj-$(CONFIG_APPLE_SART) += apple-sart.o apple-sart-y = sart.o diff --git a/drivers/soc/apple/dockchannel.c b/drivers/soc/apple/dockchannel.c new file mode 100644 index 000000000000..3a0d7964007c --- /dev/null +++ b/drivers/soc/apple/dockchannel.c @@ -0,0 +1,406 @@ +// SPDX-License-Identifier: GPL-2.0-only OR MIT +/* + * Apple DockChannel FIFO driver + * Copyright The Asahi Linux Contributors + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define DOCKCHANNEL_MAX_IRQ 32 + +#define DOCKCHANNEL_TX_TIMEOUT_MS 1000 +#define DOCKCHANNEL_RX_TIMEOUT_MS 1000 + +#define IRQ_MASK 0x0 +#define IRQ_FLAG 0x4 + +#define IRQ_TX BIT(0) +#define IRQ_RX BIT(1) + +#define CONFIG_TX_THRESH 0x0 +#define CONFIG_RX_THRESH 0x4 + +#define DATA_TX8 0x4 +#define DATA_TX16 0x8 +#define DATA_TX24 0xc +#define DATA_TX32 0x10 +#define DATA_TX_FREE 0x14 +#define DATA_RX8 0x1c +#define DATA_RX16 0x20 +#define DATA_RX24 0x24 +#define DATA_RX32 0x28 +#define DATA_RX_COUNT 0x2c + +struct dockchannel { + struct device *dev; + int tx_irq; + int rx_irq; + + void __iomem *config_base; + void __iomem *data_base; + + u32 fifo_size; + bool awaiting; + struct completion tx_comp; + struct completion rx_comp; + + void *cookie; + void (*data_available)(void *cookie, size_t avail); +}; + +struct dockchannel_common { + struct device *dev; + struct irq_domain *domain; + int irq; + + void __iomem *irq_base; +}; + +/* Dockchannel FIFO functions */ + +static irqreturn_t dockchannel_tx_irq(int irq, void *data) +{ + struct dockchannel *dockchannel = data; + + disable_irq_nosync(irq); + complete(&dockchannel->tx_comp); + + return IRQ_HANDLED; +} + +static irqreturn_t dockchannel_rx_irq(int irq, void *data) +{ + struct dockchannel *dockchannel = data; + + disable_irq_nosync(irq); + + if (dockchannel->awaiting) { + return IRQ_WAKE_THREAD; + } else { + complete(&dockchannel->rx_comp); + return IRQ_HANDLED; + } +} + +static irqreturn_t dockchannel_rx_irq_thread(int irq, void *data) +{ + struct dockchannel *dockchannel = data; + size_t avail = readl_relaxed(dockchannel->data_base + DATA_RX_COUNT); + + dockchannel->awaiting = false; + dockchannel->data_available(dockchannel->cookie, avail); + + return IRQ_HANDLED; +} + +int dockchannel_send(struct dockchannel *dockchannel, const void *buf, size_t count) +{ + size_t left = count; + const u8 *p = buf; + + while (left > 0) { + size_t avail = readl_relaxed(dockchannel->data_base + DATA_TX_FREE); + size_t block = min(left, avail); + + if (avail == 0) { + size_t threshold = min((size_t)(dockchannel->fifo_size / 2), left); + + writel_relaxed(threshold, dockchannel->config_base + CONFIG_TX_THRESH); + reinit_completion(&dockchannel->tx_comp); + enable_irq(dockchannel->tx_irq); + + if (!wait_for_completion_timeout(&dockchannel->tx_comp, + msecs_to_jiffies(DOCKCHANNEL_TX_TIMEOUT_MS))) { + disable_irq(dockchannel->tx_irq); + return -ETIMEDOUT; + } + + continue; + } + + while (block >= 4) { + writel_relaxed(get_unaligned_le32(p), dockchannel->data_base + DATA_TX32); + p += 4; + left -= 4; + block -= 4; + } + while (block > 0) { + writeb_relaxed(*p++, dockchannel->data_base + DATA_TX8); + left--; + block--; + } + } + + return count; +} +EXPORT_SYMBOL(dockchannel_send); + +int dockchannel_recv(struct dockchannel *dockchannel, void *buf, size_t count) +{ + size_t left = count; + u8 *p = buf; + + while (left > 0) { + size_t avail = readl_relaxed(dockchannel->data_base + DATA_RX_COUNT); + size_t block = min(left, avail); + + if (avail == 0) { + size_t threshold = min((size_t)(dockchannel->fifo_size / 2), left); + + writel_relaxed(threshold, dockchannel->config_base + CONFIG_RX_THRESH); + reinit_completion(&dockchannel->rx_comp); + enable_irq(dockchannel->rx_irq); + + if (!wait_for_completion_timeout(&dockchannel->rx_comp, + msecs_to_jiffies(DOCKCHANNEL_RX_TIMEOUT_MS))) { + disable_irq(dockchannel->rx_irq); + return -ETIMEDOUT; + } + + continue; + } + + while (block >= 4) { + put_unaligned_le32(readl_relaxed(dockchannel->data_base + DATA_RX32), p); + p += 4; + left -= 4; + block -= 4; + } + while (block > 0) { + *p++ = readl_relaxed(dockchannel->data_base + DATA_RX8) >> 8; + left--; + block--; + } + } + + return count; +} +EXPORT_SYMBOL(dockchannel_recv); + +int dockchannel_await(struct dockchannel *dockchannel, + void (*callback)(void *cookie, size_t avail), + void *cookie, size_t count) +{ + size_t threshold = min((size_t)dockchannel->fifo_size, count); + + if (!count) { + dockchannel->awaiting = false; + disable_irq(dockchannel->rx_irq); + return 0; + } + + dockchannel->data_available = callback; + dockchannel->cookie = cookie; + dockchannel->awaiting = true; + writel_relaxed(threshold, dockchannel->config_base + CONFIG_RX_THRESH); + enable_irq(dockchannel->rx_irq); + + return threshold; +} +EXPORT_SYMBOL(dockchannel_await); + +struct dockchannel *dockchannel_init(struct platform_device *pdev) +{ + struct device *dev = &pdev->dev; + struct dockchannel *dockchannel; + int ret; + + dockchannel = devm_kzalloc(dev, sizeof(*dockchannel), GFP_KERNEL); + if (!dockchannel) + return ERR_PTR(-ENOMEM); + + dockchannel->dev = dev; + dockchannel->config_base = devm_platform_ioremap_resource_byname(pdev, "config"); + if (IS_ERR(dockchannel->config_base)) + return (__force void *)dockchannel->config_base; + + dockchannel->data_base = devm_platform_ioremap_resource_byname(pdev, "data"); + if (IS_ERR(dockchannel->data_base)) + return (__force void *)dockchannel->data_base; + + ret = of_property_read_u32(dev->of_node, "apple,fifo-size", &dockchannel->fifo_size); + if (ret) + return ERR_PTR(dev_err_probe(dev, ret, "Missing apple,fifo-size property")); + + init_completion(&dockchannel->tx_comp); + init_completion(&dockchannel->rx_comp); + + dockchannel->tx_irq = platform_get_irq_byname(pdev, "tx"); + if (dockchannel->tx_irq <= 0) { + return ERR_PTR(dev_err_probe(dev, dockchannel->tx_irq, + "Failed to get TX IRQ")); + } + + dockchannel->rx_irq = platform_get_irq_byname(pdev, "rx"); + if (dockchannel->rx_irq <= 0) { + return ERR_PTR(dev_err_probe(dev, dockchannel->rx_irq, + "Failed to get RX IRQ")); + } + + ret = devm_request_irq(dev, dockchannel->tx_irq, dockchannel_tx_irq, IRQF_NO_AUTOEN, + "apple-dockchannel-tx", dockchannel); + if (ret) + return ERR_PTR(dev_err_probe(dev, ret, "Failed to request TX IRQ")); + + ret = devm_request_threaded_irq(dev, dockchannel->rx_irq, dockchannel_rx_irq, + dockchannel_rx_irq_thread, IRQF_NO_AUTOEN, + "apple-dockchannel-rx", dockchannel); + if (ret) + return ERR_PTR(dev_err_probe(dev, ret, "Failed to request RX IRQ")); + + return dockchannel; +} +EXPORT_SYMBOL(dockchannel_init); + + +/* Dockchannel IRQchip */ + +static void dockchannel_irq(struct irq_desc *desc) +{ + unsigned int irq = irq_desc_get_irq(desc); + struct irq_chip *chip = irq_desc_get_chip(desc); + struct dockchannel_common *dcc = irq_get_handler_data(irq); + unsigned long flags = readl_relaxed(dcc->irq_base + IRQ_FLAG); + int bit; + + chained_irq_enter(chip, desc); + + for_each_set_bit(bit, &flags, DOCKCHANNEL_MAX_IRQ) + generic_handle_domain_irq(dcc->domain, bit); + + chained_irq_exit(chip, desc); +} + +static void dockchannel_irq_ack(struct irq_data *data) +{ + struct dockchannel_common *dcc = irq_data_get_irq_chip_data(data); + unsigned int hwirq = data->hwirq; + + writel_relaxed(BIT(hwirq), dcc->irq_base + IRQ_FLAG); +} + +static void dockchannel_irq_mask(struct irq_data *data) +{ + struct dockchannel_common *dcc = irq_data_get_irq_chip_data(data); + unsigned int hwirq = data->hwirq; + u32 val = readl_relaxed(dcc->irq_base + IRQ_MASK); + + writel_relaxed(val & ~BIT(hwirq), dcc->irq_base + IRQ_MASK); +} + +static void dockchannel_irq_unmask(struct irq_data *data) +{ + struct dockchannel_common *dcc = irq_data_get_irq_chip_data(data); + unsigned int hwirq = data->hwirq; + u32 val = readl_relaxed(dcc->irq_base + IRQ_MASK); + + writel_relaxed(val | BIT(hwirq), dcc->irq_base + IRQ_MASK); +} + +static const struct irq_chip dockchannel_irqchip = { + .name = "dockchannel-irqc", + .irq_ack = dockchannel_irq_ack, + .irq_mask = dockchannel_irq_mask, + .irq_unmask = dockchannel_irq_unmask, +}; + +static int dockchannel_irq_domain_map(struct irq_domain *d, unsigned int virq, + irq_hw_number_t hw) +{ + irq_set_chip_data(virq, d->host_data); + irq_set_chip_and_handler(virq, &dockchannel_irqchip, handle_level_irq); + + return 0; +} + +static const struct irq_domain_ops dockchannel_irq_domain_ops = { + .xlate = irq_domain_xlate_twocell, + .map = dockchannel_irq_domain_map, +}; + +static int dockchannel_probe(struct platform_device *pdev) +{ + struct device *dev = &pdev->dev; + struct dockchannel_common *dcc; + struct device_node *child; + + dcc = devm_kzalloc(dev, sizeof(*dcc), GFP_KERNEL); + if (!dcc) + return -ENOMEM; + + dcc->dev = dev; + platform_set_drvdata(pdev, dcc); + + dcc->irq_base = devm_platform_ioremap_resource_byname(pdev, "irq"); + if (IS_ERR(dcc->irq_base)) + return PTR_ERR(dcc->irq_base); + + writel_relaxed(0, dcc->irq_base + IRQ_MASK); + writel_relaxed(~0, dcc->irq_base + IRQ_FLAG); + + dcc->domain = irq_domain_add_linear(dev->of_node, DOCKCHANNEL_MAX_IRQ, + &dockchannel_irq_domain_ops, dcc); + if (!dcc->domain) + return -ENOMEM; + + dcc->irq = platform_get_irq(pdev, 0); + if (dcc->irq <= 0) + return dev_err_probe(dev, dcc->irq, "Failed to get IRQ"); + + irq_set_handler_data(dcc->irq, dcc); + irq_set_chained_handler(dcc->irq, dockchannel_irq); + + for_each_child_of_node(dev->of_node, child) + of_platform_device_create(child, NULL, dev); + + return 0; +} + +static void dockchannel_remove(struct platform_device *pdev) +{ + struct dockchannel_common *dcc = platform_get_drvdata(pdev); + int hwirq; + + device_for_each_child(&pdev->dev, NULL, of_platform_device_destroy); + + irq_set_chained_handler_and_data(dcc->irq, NULL, NULL); + + for (hwirq = 0; hwirq < DOCKCHANNEL_MAX_IRQ; hwirq++) + irq_dispose_mapping(irq_find_mapping(dcc->domain, hwirq)); + + irq_domain_remove(dcc->domain); + + writel_relaxed(0, dcc->irq_base + IRQ_MASK); + writel_relaxed(~0, dcc->irq_base + IRQ_FLAG); +} + +static const struct of_device_id dockchannel_of_match[] = { + { .compatible = "apple,dockchannel" }, + {}, +}; +MODULE_DEVICE_TABLE(of, dockchannel_of_match); + +static struct platform_driver dockchannel_driver = { + .driver = { + .name = "dockchannel", + .of_match_table = dockchannel_of_match, + }, + .probe = dockchannel_probe, + .remove = dockchannel_remove, +}; +module_platform_driver(dockchannel_driver); + +MODULE_AUTHOR("Hector Martin "); +MODULE_LICENSE("Dual MIT/GPL"); +MODULE_DESCRIPTION("Apple DockChannel driver"); diff --git a/drivers/soc/apple/rtkit-helper.c b/drivers/soc/apple/rtkit-helper.c new file mode 100644 index 000000000000..080d083ed9bd --- /dev/null +++ b/drivers/soc/apple/rtkit-helper.c @@ -0,0 +1,151 @@ +// SPDX-License-Identifier: GPL-2.0-only OR MIT +/* + * Apple Generic RTKit helper coprocessor + * Copyright The Asahi Linux Contributors + */ + +#include +#include +#include +#include +#include +#include +#include +#include + +#define APPLE_ASC_CPU_CONTROL 0x44 +#define APPLE_ASC_CPU_CONTROL_RUN BIT(4) + +struct apple_rtkit_helper { + struct device *dev; + struct apple_rtkit *rtk; + + void __iomem *asc_base; + + struct resource *sram; + void __iomem *sram_base; +}; + +static int apple_rtkit_helper_shmem_setup(void *cookie, struct apple_rtkit_shmem *bfr) +{ + struct apple_rtkit_helper *helper = cookie; + struct resource res = { + .start = bfr->iova, + .end = bfr->iova + bfr->size - 1, + .name = "rtkit_map", + }; + + if (!bfr->iova) { + bfr->buffer = dma_alloc_coherent(helper->dev, bfr->size, + &bfr->iova, GFP_KERNEL); + if (!bfr->buffer) + return -ENOMEM; + return 0; + } + + if (!helper->sram) { + dev_err(helper->dev, + "RTKit buffer request with no SRAM region: %pR", &res); + return -EFAULT; + } + + res.flags = helper->sram->flags; + + if (res.end < res.start || !resource_contains(helper->sram, &res)) { + dev_err(helper->dev, + "RTKit buffer request outside SRAM region: %pR", &res); + return -EFAULT; + } + + bfr->iomem = helper->sram_base + (res.start - helper->sram->start); + bfr->is_mapped = true; + + return 0; +} + +static void apple_rtkit_helper_shmem_destroy(void *cookie, struct apple_rtkit_shmem *bfr) +{ + // no-op +} + +static const struct apple_rtkit_ops apple_rtkit_helper_ops = { + .shmem_setup = apple_rtkit_helper_shmem_setup, + .shmem_destroy = apple_rtkit_helper_shmem_destroy, +}; + +static int apple_rtkit_helper_probe(struct platform_device *pdev) +{ + struct device *dev = &pdev->dev; + struct apple_rtkit_helper *helper; + int ret; + + /* 44 bits for addresses in standard RTKit requests */ + ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(44)); + if (ret) + return ret; + + helper = devm_kzalloc(dev, sizeof(*helper), GFP_KERNEL); + if (!helper) + return -ENOMEM; + + helper->dev = dev; + platform_set_drvdata(pdev, helper); + + helper->asc_base = devm_platform_ioremap_resource_byname(pdev, "asc"); + if (IS_ERR(helper->asc_base)) + return PTR_ERR(helper->asc_base); + + helper->sram = platform_get_resource_byname(pdev, IORESOURCE_MEM, "sram"); + if (helper->sram) { + helper->sram_base = devm_ioremap_resource(dev, helper->sram); + if (IS_ERR(helper->sram_base)) + return dev_err_probe(dev, PTR_ERR(helper->sram_base), + "Failed to map SRAM region"); + } + + helper->rtk = + devm_apple_rtkit_init(dev, helper, NULL, 0, &apple_rtkit_helper_ops); + if (IS_ERR(helper->rtk)) + return dev_err_probe(dev, PTR_ERR(helper->rtk), + "Failed to intialize RTKit"); + + writel_relaxed(APPLE_ASC_CPU_CONTROL_RUN, + helper->asc_base + APPLE_ASC_CPU_CONTROL); + + /* Works for both wake and boot */ + ret = apple_rtkit_wake(helper->rtk); + if (ret != 0) + return dev_err_probe(dev, ret, "Failed to wake up coprocessor"); + + return 0; +} + +static void apple_rtkit_helper_remove(struct platform_device *pdev) +{ + struct apple_rtkit_helper *helper = platform_get_drvdata(pdev); + + if (apple_rtkit_is_running(helper->rtk)) + apple_rtkit_quiesce(helper->rtk); + + writel_relaxed(0, helper->asc_base + APPLE_ASC_CPU_CONTROL); +} + +static const struct of_device_id apple_rtkit_helper_of_match[] = { + { .compatible = "apple,rtk-helper-asc4" }, + {}, +}; +MODULE_DEVICE_TABLE(of, apple_rtkit_helper_of_match); + +static struct platform_driver apple_rtkit_helper_driver = { + .driver = { + .name = "rtkit-helper", + .of_match_table = apple_rtkit_helper_of_match, + }, + .probe = apple_rtkit_helper_probe, + .remove = apple_rtkit_helper_remove, +}; +module_platform_driver(apple_rtkit_helper_driver); + +MODULE_AUTHOR("Hector Martin "); +MODULE_LICENSE("Dual MIT/GPL"); +MODULE_DESCRIPTION("Apple RTKit helper driver"); diff --git a/drivers/staging/Kconfig b/drivers/staging/Kconfig index 2f92cd698bef..cefe34038c57 100644 --- a/drivers/staging/Kconfig +++ b/drivers/staging/Kconfig @@ -48,4 +48,6 @@ source "drivers/staging/axis-fifo/Kconfig" source "drivers/staging/vme_user/Kconfig" +source "drivers/staging/apple-bce/Kconfig" + endif # STAGING diff --git a/drivers/staging/Makefile b/drivers/staging/Makefile index f5b8876aa536..cf24b596cde9 100644 --- a/drivers/staging/Makefile +++ b/drivers/staging/Makefile @@ -13,3 +13,4 @@ obj-$(CONFIG_MOST) += most/ obj-$(CONFIG_GREYBUS) += greybus/ obj-$(CONFIG_BCM2835_VCHIQ) += vc04_services/ obj-$(CONFIG_XIL_AXIS_FIFO) += axis-fifo/ +obj-$(CONFIG_APPLE_BCE) += apple-bce/ diff --git a/drivers/staging/apple-bce/Kconfig b/drivers/staging/apple-bce/Kconfig new file mode 100644 index 000000000000..fe92bc441e89 --- /dev/null +++ b/drivers/staging/apple-bce/Kconfig @@ -0,0 +1,18 @@ +config APPLE_BCE + tristate "Apple BCE driver (VHCI and Audio support)" + default m + depends on X86 + select SOUND + select SND + select SND_PCM + select SND_JACK + help + VHCI and audio support on Apple MacBooks with the T2 Chip. + This driver is divided in three components: + - BCE (Buffer Copy Engine): which establishes a basic communication + channel with the T2 chip. This component is required by the other two: + - VHCI (Virtual Host Controller Interface): Access to keyboard, mouse + and other system devices depend on this virtual USB host controller + - Audio: a driver for the T2 audio interface. + + If "M" is selected, the module will be called apple-bce.' diff --git a/drivers/staging/apple-bce/Makefile b/drivers/staging/apple-bce/Makefile new file mode 100644 index 000000000000..8cfbd3f64af6 --- /dev/null +++ b/drivers/staging/apple-bce/Makefile @@ -0,0 +1,28 @@ +modname := apple-bce +obj-$(CONFIG_APPLE_BCE) += $(modname).o + +apple-bce-objs := apple_bce.o mailbox.o queue.o queue_dma.o vhci/vhci.o vhci/queue.o vhci/transfer.o audio/audio.o audio/protocol.o audio/protocol_bce.o audio/pcm.o + +MY_CFLAGS += -DWITHOUT_NVME_PATCH +#MY_CFLAGS += -g -DDEBUG +ccflags-y += ${MY_CFLAGS} +CC += ${MY_CFLAGS} + +KVERSION := $(KERNELRELEASE) +ifeq ($(origin KERNELRELEASE), undefined) +KVERSION := $(shell uname -r) +endif + +KDIR := /lib/modules/$(KVERSION)/build +PWD := $(shell pwd) + +.PHONY: all + +all: + $(MAKE) -C $(KDIR) M=$(PWD) modules + +clean: + $(MAKE) -C $(KDIR) M=$(PWD) clean + +install: + $(MAKE) -C $(KDIR) M=$(PWD) modules_install diff --git a/drivers/staging/apple-bce/apple_bce.c b/drivers/staging/apple-bce/apple_bce.c new file mode 100644 index 000000000000..4fd2415d7028 --- /dev/null +++ b/drivers/staging/apple-bce/apple_bce.c @@ -0,0 +1,445 @@ +#include "apple_bce.h" +#include +#include +#include "audio/audio.h" +#include + +static dev_t bce_chrdev; +static struct class *bce_class; + +struct apple_bce_device *global_bce; + +static int bce_create_command_queues(struct apple_bce_device *bce); +static void bce_free_command_queues(struct apple_bce_device *bce); +static irqreturn_t bce_handle_mb_irq(int irq, void *dev); +static irqreturn_t bce_handle_dma_irq(int irq, void *dev); +static int bce_fw_version_handshake(struct apple_bce_device *bce); +static int bce_register_command_queue(struct apple_bce_device *bce, struct bce_queue_memcfg *cfg, int is_sq); + +static int apple_bce_probe(struct pci_dev *dev, const struct pci_device_id *id) +{ + struct apple_bce_device *bce = NULL; + int status = 0; + int nvec; + + pr_info("apple-bce: capturing our device\n"); + + if (pci_enable_device(dev)) + return -ENODEV; + if (pci_request_regions(dev, "apple-bce")) { + status = -ENODEV; + goto fail; + } + pci_set_master(dev); + nvec = pci_alloc_irq_vectors(dev, 1, 8, PCI_IRQ_MSI); + if (nvec < 5) { + status = -EINVAL; + goto fail; + } + + bce = kzalloc(sizeof(struct apple_bce_device), GFP_KERNEL); + if (!bce) { + status = -ENOMEM; + goto fail; + } + + bce->pci = dev; + pci_set_drvdata(dev, bce); + + bce->devt = bce_chrdev; + bce->dev = device_create(bce_class, &dev->dev, bce->devt, NULL, "apple-bce"); + if (IS_ERR_OR_NULL(bce->dev)) { + status = PTR_ERR(bce_class); + goto fail; + } + + bce->reg_mem_mb = pci_iomap(dev, 4, 0); + bce->reg_mem_dma = pci_iomap(dev, 2, 0); + + if (IS_ERR_OR_NULL(bce->reg_mem_mb) || IS_ERR_OR_NULL(bce->reg_mem_dma)) { + dev_warn(&dev->dev, "apple-bce: Failed to pci_iomap required regions\n"); + goto fail; + } + + bce_mailbox_init(&bce->mbox, bce->reg_mem_mb); + bce_timestamp_init(&bce->timestamp, bce->reg_mem_mb); + + spin_lock_init(&bce->queues_lock); + ida_init(&bce->queue_ida); + + if ((status = pci_request_irq(dev, 0, bce_handle_mb_irq, NULL, dev, "bce_mbox"))) + goto fail; + if ((status = pci_request_irq(dev, 4, NULL, bce_handle_dma_irq, dev, "bce_dma"))) + goto fail_interrupt_0; + + if ((status = dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(37)))) { + dev_warn(&dev->dev, "dma: Setting mask failed\n"); + goto fail_interrupt; + } + + /* Gets the function 0's interface. This is needed because Apple only accepts DMA on our function if function 0 + is a bus master, so we need to work around this. */ + bce->pci0 = pci_get_slot(dev->bus, PCI_DEVFN(PCI_SLOT(dev->devfn), 0)); +#ifndef WITHOUT_NVME_PATCH + if ((status = pci_enable_device_mem(bce->pci0))) { + dev_warn(&dev->dev, "apple-bce: failed to enable function 0\n"); + goto fail_dev0; + } +#endif + pci_set_master(bce->pci0); + + bce_timestamp_start(&bce->timestamp, true); + + if ((status = bce_fw_version_handshake(bce))) + goto fail_ts; + pr_info("apple-bce: handshake done\n"); + + if ((status = bce_create_command_queues(bce))) { + pr_info("apple-bce: Creating command queues failed\n"); + goto fail_ts; + } + + global_bce = bce; + + bce_vhci_create(bce, &bce->vhci); + + return 0; + +fail_ts: + bce_timestamp_stop(&bce->timestamp); +#ifndef WITHOUT_NVME_PATCH + pci_disable_device(bce->pci0); +fail_dev0: +#endif + pci_dev_put(bce->pci0); +fail_interrupt: + pci_free_irq(dev, 4, dev); +fail_interrupt_0: + pci_free_irq(dev, 0, dev); +fail: + if (bce && bce->dev) { + device_destroy(bce_class, bce->devt); + + if (!IS_ERR_OR_NULL(bce->reg_mem_mb)) + pci_iounmap(dev, bce->reg_mem_mb); + if (!IS_ERR_OR_NULL(bce->reg_mem_dma)) + pci_iounmap(dev, bce->reg_mem_dma); + + kfree(bce); + } + + pci_free_irq_vectors(dev); + pci_release_regions(dev); + pci_disable_device(dev); + + if (!status) + status = -EINVAL; + return status; +} + +static int bce_create_command_queues(struct apple_bce_device *bce) +{ + int status; + struct bce_queue_memcfg *cfg; + + bce->cmd_cq = bce_alloc_cq(bce, 0, 0x20); + bce->cmd_cmdq = bce_alloc_cmdq(bce, 1, 0x20); + if (bce->cmd_cq == NULL || bce->cmd_cmdq == NULL) { + status = -ENOMEM; + goto err; + } + bce->queues[0] = (struct bce_queue *) bce->cmd_cq; + bce->queues[1] = (struct bce_queue *) bce->cmd_cmdq->sq; + + cfg = kzalloc(sizeof(struct bce_queue_memcfg), GFP_KERNEL); + if (!cfg) { + status = -ENOMEM; + goto err; + } + bce_get_cq_memcfg(bce->cmd_cq, cfg); + if ((status = bce_register_command_queue(bce, cfg, false))) + goto err; + bce_get_sq_memcfg(bce->cmd_cmdq->sq, bce->cmd_cq, cfg); + if ((status = bce_register_command_queue(bce, cfg, true))) + goto err; + kfree(cfg); + + return 0; + +err: + if (bce->cmd_cq) + bce_free_cq(bce, bce->cmd_cq); + if (bce->cmd_cmdq) + bce_free_cmdq(bce, bce->cmd_cmdq); + return status; +} + +static void bce_free_command_queues(struct apple_bce_device *bce) +{ + bce_free_cq(bce, bce->cmd_cq); + bce_free_cmdq(bce, bce->cmd_cmdq); + bce->cmd_cq = NULL; + bce->queues[0] = NULL; +} + +static irqreturn_t bce_handle_mb_irq(int irq, void *dev) +{ + struct apple_bce_device *bce = pci_get_drvdata(dev); + bce_mailbox_handle_interrupt(&bce->mbox); + return IRQ_HANDLED; +} + +static irqreturn_t bce_handle_dma_irq(int irq, void *dev) +{ + int i; + struct apple_bce_device *bce = pci_get_drvdata(dev); + spin_lock(&bce->queues_lock); + for (i = 0; i < BCE_MAX_QUEUE_COUNT; i++) + if (bce->queues[i] && bce->queues[i]->type == BCE_QUEUE_CQ) + bce_handle_cq_completions(bce, (struct bce_queue_cq *) bce->queues[i]); + spin_unlock(&bce->queues_lock); + return IRQ_HANDLED; +} + +static int bce_fw_version_handshake(struct apple_bce_device *bce) +{ + u64 result; + int status; + + if ((status = bce_mailbox_send(&bce->mbox, BCE_MB_MSG(BCE_MB_SET_FW_PROTOCOL_VERSION, BC_PROTOCOL_VERSION), + &result))) + return status; + if (BCE_MB_TYPE(result) != BCE_MB_SET_FW_PROTOCOL_VERSION || + BCE_MB_VALUE(result) != BC_PROTOCOL_VERSION) { + pr_err("apple-bce: FW version handshake failed %x:%llx\n", BCE_MB_TYPE(result), BCE_MB_VALUE(result)); + return -EINVAL; + } + return 0; +} + +static int bce_register_command_queue(struct apple_bce_device *bce, struct bce_queue_memcfg *cfg, int is_sq) +{ + int status; + int cmd_type; + u64 result; + // OS X uses an bidirectional direction, but that's not really needed + dma_addr_t a = dma_map_single(&bce->pci->dev, cfg, sizeof(struct bce_queue_memcfg), DMA_TO_DEVICE); + if (dma_mapping_error(&bce->pci->dev, a)) + return -ENOMEM; + cmd_type = is_sq ? BCE_MB_REGISTER_COMMAND_SQ : BCE_MB_REGISTER_COMMAND_CQ; + status = bce_mailbox_send(&bce->mbox, BCE_MB_MSG(cmd_type, a), &result); + dma_unmap_single(&bce->pci->dev, a, sizeof(struct bce_queue_memcfg), DMA_TO_DEVICE); + if (status) + return status; + if (BCE_MB_TYPE(result) != BCE_MB_REGISTER_COMMAND_QUEUE_REPLY) + return -EINVAL; + return 0; +} + +static void apple_bce_remove(struct pci_dev *dev) +{ + struct apple_bce_device *bce = pci_get_drvdata(dev); + bce->is_being_removed = true; + + bce_vhci_destroy(&bce->vhci); + + bce_timestamp_stop(&bce->timestamp); +#ifndef WITHOUT_NVME_PATCH + pci_disable_device(bce->pci0); +#endif + pci_dev_put(bce->pci0); + pci_free_irq(dev, 0, dev); + pci_free_irq(dev, 4, dev); + bce_free_command_queues(bce); + pci_iounmap(dev, bce->reg_mem_mb); + pci_iounmap(dev, bce->reg_mem_dma); + device_destroy(bce_class, bce->devt); + pci_free_irq_vectors(dev); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(bce); +} + +static int bce_save_state_and_sleep(struct apple_bce_device *bce) +{ + int attempt, status = 0; + u64 resp; + dma_addr_t dma_addr; + void *dma_ptr = NULL; + size_t size = max(PAGE_SIZE, 4096UL); + + for (attempt = 0; attempt < 5; ++attempt) { + pr_debug("apple-bce: suspend: attempt %i, buffer size %li\n", attempt, size); + dma_ptr = dma_alloc_coherent(&bce->pci->dev, size, &dma_addr, GFP_KERNEL); + if (!dma_ptr) { + pr_err("apple-bce: suspend failed (data alloc failed)\n"); + break; + } + BUG_ON((dma_addr % 4096) != 0); + status = bce_mailbox_send(&bce->mbox, + BCE_MB_MSG(BCE_MB_SAVE_STATE_AND_SLEEP, (dma_addr & ~(4096LLU - 1)) | (size / 4096)), &resp); + if (status) { + pr_err("apple-bce: suspend failed (mailbox send)\n"); + break; + } + if (BCE_MB_TYPE(resp) == BCE_MB_SAVE_RESTORE_STATE_COMPLETE) { + bce->saved_data_dma_addr = dma_addr; + bce->saved_data_dma_ptr = dma_ptr; + bce->saved_data_dma_size = size; + return 0; + } else if (BCE_MB_TYPE(resp) == BCE_MB_SAVE_STATE_AND_SLEEP_FAILURE) { + dma_free_coherent(&bce->pci->dev, size, dma_ptr, dma_addr); + /* The 0x10ff magic value was extracted from Apple's driver */ + size = (BCE_MB_VALUE(resp) + 0x10ff) & ~(4096LLU - 1); + pr_debug("apple-bce: suspend: device requested a larger buffer (%li)\n", size); + continue; + } else { + pr_err("apple-bce: suspend failed (invalid device response)\n"); + status = -EINVAL; + break; + } + } + if (dma_ptr) + dma_free_coherent(&bce->pci->dev, size, dma_ptr, dma_addr); + if (!status) + return bce_mailbox_send(&bce->mbox, BCE_MB_MSG(BCE_MB_SLEEP_NO_STATE, 0), &resp); + return status; +} + +static int bce_restore_state_and_wake(struct apple_bce_device *bce) +{ + int status; + u64 resp; + if (!bce->saved_data_dma_ptr) { + if ((status = bce_mailbox_send(&bce->mbox, BCE_MB_MSG(BCE_MB_RESTORE_NO_STATE, 0), &resp))) { + pr_err("apple-bce: resume with no state failed (mailbox send)\n"); + return status; + } + if (BCE_MB_TYPE(resp) != BCE_MB_RESTORE_NO_STATE) { + pr_err("apple-bce: resume with no state failed (invalid device response)\n"); + return -EINVAL; + } + return 0; + } + + if ((status = bce_mailbox_send(&bce->mbox, BCE_MB_MSG(BCE_MB_RESTORE_STATE_AND_WAKE, + (bce->saved_data_dma_addr & ~(4096LLU - 1)) | (bce->saved_data_dma_size / 4096)), &resp))) { + pr_err("apple-bce: resume with state failed (mailbox send)\n"); + goto finish_with_state; + } + if (BCE_MB_TYPE(resp) != BCE_MB_SAVE_RESTORE_STATE_COMPLETE) { + pr_err("apple-bce: resume with state failed (invalid device response)\n"); + status = -EINVAL; + goto finish_with_state; + } + +finish_with_state: + dma_free_coherent(&bce->pci->dev, bce->saved_data_dma_size, bce->saved_data_dma_ptr, bce->saved_data_dma_addr); + bce->saved_data_dma_ptr = NULL; + return status; +} + +static int apple_bce_suspend(struct device *dev) +{ + struct apple_bce_device *bce = pci_get_drvdata(to_pci_dev(dev)); + int status; + + bce_timestamp_stop(&bce->timestamp); + + if ((status = bce_save_state_and_sleep(bce))) + return status; + + return 0; +} + +static int apple_bce_resume(struct device *dev) +{ + struct apple_bce_device *bce = pci_get_drvdata(to_pci_dev(dev)); + int status; + + pci_set_master(bce->pci); + pci_set_master(bce->pci0); + + if ((status = bce_restore_state_and_wake(bce))) + return status; + + bce_timestamp_start(&bce->timestamp, false); + + return 0; +} + +static struct pci_device_id apple_bce_ids[ ] = { + { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x1801) }, + { 0, }, +}; + +MODULE_DEVICE_TABLE(pci, apple_bce_ids); + +struct dev_pm_ops apple_bce_pci_driver_pm = { + .suspend = apple_bce_suspend, + .resume = apple_bce_resume +}; +struct pci_driver apple_bce_pci_driver = { + .name = "apple-bce", + .id_table = apple_bce_ids, + .probe = apple_bce_probe, + .remove = apple_bce_remove, + .driver = { + .pm = &apple_bce_pci_driver_pm + } +}; + + +static int __init apple_bce_module_init(void) +{ + int result; + if ((result = alloc_chrdev_region(&bce_chrdev, 0, 1, "apple-bce"))) + goto fail_chrdev; +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,4,0) + bce_class = class_create(THIS_MODULE, "apple-bce"); +#else + bce_class = class_create("apple-bce"); +#endif + if (IS_ERR(bce_class)) { + result = PTR_ERR(bce_class); + goto fail_class; + } + if ((result = bce_vhci_module_init())) { + pr_err("apple-bce: bce-vhci init failed"); + goto fail_class; + } + + result = pci_register_driver(&apple_bce_pci_driver); + if (result) + goto fail_drv; + + aaudio_module_init(); + + return 0; + +fail_drv: + pci_unregister_driver(&apple_bce_pci_driver); +fail_class: + class_destroy(bce_class); +fail_chrdev: + unregister_chrdev_region(bce_chrdev, 1); + if (!result) + result = -EINVAL; + return result; +} +static void __exit apple_bce_module_exit(void) +{ + pci_unregister_driver(&apple_bce_pci_driver); + + aaudio_module_exit(); + bce_vhci_module_exit(); + class_destroy(bce_class); + unregister_chrdev_region(bce_chrdev, 1); +} + +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("MrARM"); +MODULE_DESCRIPTION("Apple BCE Driver"); +MODULE_VERSION("0.01"); +module_init(apple_bce_module_init); +module_exit(apple_bce_module_exit); diff --git a/drivers/staging/apple-bce/apple_bce.h b/drivers/staging/apple-bce/apple_bce.h new file mode 100644 index 000000000000..58dbeff79e43 --- /dev/null +++ b/drivers/staging/apple-bce/apple_bce.h @@ -0,0 +1,41 @@ +#ifndef APPLE_BCE_H +#define APPLE_BCE_H + +#include +#include +#include "mailbox.h" +#include "queue.h" +#include "vhci/vhci.h" + +#define BC_PROTOCOL_VERSION 0x20001 +#define BCE_MAX_QUEUE_COUNT 0x100 + +#define BCE_QUEUE_USER_MIN 2 +#define BCE_QUEUE_USER_MAX (BCE_MAX_QUEUE_COUNT - 1) + +struct apple_bce_device { + struct pci_dev *pci, *pci0; + dev_t devt; + struct device *dev; + void __iomem *reg_mem_mb; + void __iomem *reg_mem_dma; + struct bce_mailbox mbox; + struct bce_timestamp timestamp; + struct bce_queue *queues[BCE_MAX_QUEUE_COUNT]; + struct spinlock queues_lock; + struct ida queue_ida; + struct bce_queue_cq *cmd_cq; + struct bce_queue_cmdq *cmd_cmdq; + struct bce_queue_sq *int_sq_list[BCE_MAX_QUEUE_COUNT]; + bool is_being_removed; + + dma_addr_t saved_data_dma_addr; + void *saved_data_dma_ptr; + size_t saved_data_dma_size; + + struct bce_vhci vhci; +}; + +extern struct apple_bce_device *global_bce; + +#endif //APPLE_BCE_H diff --git a/drivers/staging/apple-bce/audio/audio.c b/drivers/staging/apple-bce/audio/audio.c new file mode 100644 index 000000000000..bd16ddd16c1d --- /dev/null +++ b/drivers/staging/apple-bce/audio/audio.c @@ -0,0 +1,711 @@ +#include +#include +#include +#include +#include +#include +#include +#include +#include "audio.h" +#include "pcm.h" +#include + +static int aaudio_alsa_index = SNDRV_DEFAULT_IDX1; +static char *aaudio_alsa_id = SNDRV_DEFAULT_STR1; + +static dev_t aaudio_chrdev; +static struct class *aaudio_class; + +static int aaudio_init_cmd(struct aaudio_device *a); +static int aaudio_init_bs(struct aaudio_device *a); +static void aaudio_init_dev(struct aaudio_device *a, aaudio_device_id_t dev_id); +static void aaudio_free_dev(struct aaudio_subdevice *sdev); + +static int aaudio_probe(struct pci_dev *dev, const struct pci_device_id *id) +{ + struct aaudio_device *aaudio = NULL; + struct aaudio_subdevice *sdev = NULL; + int status = 0; + u32 cfg; + + pr_info("aaudio: capturing our device\n"); + + if (pci_enable_device(dev)) + return -ENODEV; + if (pci_request_regions(dev, "aaudio")) { + status = -ENODEV; + goto fail; + } + pci_set_master(dev); + + aaudio = kzalloc(sizeof(struct aaudio_device), GFP_KERNEL); + if (!aaudio) { + status = -ENOMEM; + goto fail; + } + + aaudio->bce = global_bce; + if (!aaudio->bce) { + dev_warn(&dev->dev, "aaudio: No BCE available\n"); + status = -EINVAL; + goto fail; + } + + aaudio->pci = dev; + pci_set_drvdata(dev, aaudio); + + aaudio->devt = aaudio_chrdev; + aaudio->dev = device_create(aaudio_class, &dev->dev, aaudio->devt, NULL, "aaudio"); + if (IS_ERR_OR_NULL(aaudio->dev)) { + status = PTR_ERR(aaudio_class); + goto fail; + } + device_link_add(aaudio->dev, aaudio->bce->dev, DL_FLAG_PM_RUNTIME | DL_FLAG_AUTOREMOVE_CONSUMER); + + init_completion(&aaudio->remote_alive); + INIT_LIST_HEAD(&aaudio->subdevice_list); + + /* Init: set an unknown flag in the bitset */ + if (pci_read_config_dword(dev, 4, &cfg)) + dev_warn(&dev->dev, "aaudio: pci_read_config_dword fail\n"); + if (pci_write_config_dword(dev, 4, cfg | 6u)) + dev_warn(&dev->dev, "aaudio: pci_write_config_dword fail\n"); + + dev_info(aaudio->dev, "aaudio: bs len = %llx\n", pci_resource_len(dev, 0)); + aaudio->reg_mem_bs_dma = pci_resource_start(dev, 0); + aaudio->reg_mem_bs = pci_iomap(dev, 0, 0); + aaudio->reg_mem_cfg = pci_iomap(dev, 4, 0); + + aaudio->reg_mem_gpr = (u32 __iomem *) ((u8 __iomem *) aaudio->reg_mem_cfg + 0xC000); + + if (IS_ERR_OR_NULL(aaudio->reg_mem_bs) || IS_ERR_OR_NULL(aaudio->reg_mem_cfg)) { + dev_warn(&dev->dev, "aaudio: Failed to pci_iomap required regions\n"); + goto fail; + } + + if (aaudio_bce_init(aaudio)) { + dev_warn(&dev->dev, "aaudio: Failed to init BCE command transport\n"); + goto fail; + } + + if (snd_card_new(aaudio->dev, aaudio_alsa_index, aaudio_alsa_id, THIS_MODULE, 0, &aaudio->card)) { + dev_err(&dev->dev, "aaudio: Failed to create ALSA card\n"); + goto fail; + } + + strcpy(aaudio->card->shortname, "Apple T2 Audio"); + strcpy(aaudio->card->longname, "Apple T2 Audio"); + strcpy(aaudio->card->mixername, "Apple T2 Audio"); + /* Dynamic alsa ids start at 100 */ + aaudio->next_alsa_id = 100; + + if (aaudio_init_cmd(aaudio)) { + dev_err(&dev->dev, "aaudio: Failed to initialize over BCE\n"); + goto fail_snd; + } + + if (aaudio_init_bs(aaudio)) { + dev_err(&dev->dev, "aaudio: Failed to initialize BufferStruct\n"); + goto fail_snd; + } + + if ((status = aaudio_cmd_set_remote_access(aaudio, AAUDIO_REMOTE_ACCESS_ON))) { + dev_err(&dev->dev, "Failed to set remote access\n"); + return status; + } + + if (snd_card_register(aaudio->card)) { + dev_err(&dev->dev, "aaudio: Failed to register ALSA sound device\n"); + goto fail_snd; + } + + list_for_each_entry(sdev, &aaudio->subdevice_list, list) { + struct aaudio_buffer_struct_device *dev = &aaudio->bs->devices[sdev->buf_id]; + + if (sdev->out_stream_cnt == 1 && !strcmp(dev->name, "Speaker")) { + struct snd_pcm_hardware *hw = sdev->out_streams[0].alsa_hw_desc; + + snprintf(aaudio->card->driver, sizeof(aaudio->card->driver) / sizeof(char), "AppleT2x%d", hw->channels_min); + } + } + + return 0; + +fail_snd: + snd_card_free(aaudio->card); +fail: + if (aaudio && aaudio->dev) + device_destroy(aaudio_class, aaudio->devt); + kfree(aaudio); + + if (!IS_ERR_OR_NULL(aaudio->reg_mem_bs)) + pci_iounmap(dev, aaudio->reg_mem_bs); + if (!IS_ERR_OR_NULL(aaudio->reg_mem_cfg)) + pci_iounmap(dev, aaudio->reg_mem_cfg); + + pci_release_regions(dev); + pci_disable_device(dev); + + if (!status) + status = -EINVAL; + return status; +} + + + +static void aaudio_remove(struct pci_dev *dev) +{ + struct aaudio_subdevice *sdev; + struct aaudio_device *aaudio = pci_get_drvdata(dev); + + snd_card_free(aaudio->card); + while (!list_empty(&aaudio->subdevice_list)) { + sdev = list_first_entry(&aaudio->subdevice_list, struct aaudio_subdevice, list); + list_del(&sdev->list); + aaudio_free_dev(sdev); + } + pci_iounmap(dev, aaudio->reg_mem_bs); + pci_iounmap(dev, aaudio->reg_mem_cfg); + device_destroy(aaudio_class, aaudio->devt); + pci_free_irq_vectors(dev); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(aaudio); +} + +static int aaudio_suspend(struct device *dev) +{ + struct aaudio_device *aaudio = pci_get_drvdata(to_pci_dev(dev)); + + if (aaudio_cmd_set_remote_access(aaudio, AAUDIO_REMOTE_ACCESS_OFF)) + dev_warn(aaudio->dev, "Failed to reset remote access\n"); + + pci_disable_device(aaudio->pci); + return 0; +} + +static int aaudio_resume(struct device *dev) +{ + int status; + struct aaudio_device *aaudio = pci_get_drvdata(to_pci_dev(dev)); + + if ((status = pci_enable_device(aaudio->pci))) + return status; + pci_set_master(aaudio->pci); + + if ((status = aaudio_cmd_set_remote_access(aaudio, AAUDIO_REMOTE_ACCESS_ON))) { + dev_err(aaudio->dev, "Failed to set remote access\n"); + return status; + } + + return 0; +} + +static int aaudio_init_cmd(struct aaudio_device *a) +{ + int status; + struct aaudio_send_ctx sctx; + struct aaudio_msg buf; + u64 dev_cnt, dev_i; + aaudio_device_id_t *dev_l; + + if ((status = aaudio_send(a, &sctx, 500, + aaudio_msg_write_alive_notification, 1, 3))) { + dev_err(a->dev, "Sending alive notification failed\n"); + return status; + } + + if (wait_for_completion_timeout(&a->remote_alive, msecs_to_jiffies(500)) == 0) { + dev_err(a->dev, "Timed out waiting for remote\n"); + return -ETIMEDOUT; + } + dev_info(a->dev, "Continuing init\n"); + + buf = aaudio_reply_alloc(); + if ((status = aaudio_cmd_get_device_list(a, &buf, &dev_l, &dev_cnt))) { + dev_err(a->dev, "Failed to get device list\n"); + aaudio_reply_free(&buf); + return status; + } + for (dev_i = 0; dev_i < dev_cnt; ++dev_i) + aaudio_init_dev(a, dev_l[dev_i]); + aaudio_reply_free(&buf); + + return 0; +} + +static void aaudio_init_stream_info(struct aaudio_subdevice *sdev, struct aaudio_stream *strm); +static void aaudio_handle_jack_connection_change(struct aaudio_subdevice *sdev); + +static void aaudio_init_dev(struct aaudio_device *a, aaudio_device_id_t dev_id) +{ + struct aaudio_subdevice *sdev; + struct aaudio_msg buf = aaudio_reply_alloc(); + u64 uid_len, stream_cnt, i; + aaudio_object_id_t *stream_list; + char *uid; + + sdev = kzalloc(sizeof(struct aaudio_subdevice), GFP_KERNEL); + + if (aaudio_cmd_get_property(a, &buf, dev_id, dev_id, AAUDIO_PROP(AAUDIO_PROP_SCOPE_GLOBAL, AAUDIO_PROP_UID, 0), + NULL, 0, (void **) &uid, &uid_len) || uid_len > AAUDIO_DEVICE_MAX_UID_LEN) { + dev_err(a->dev, "Failed to get device uid for device %llx\n", dev_id); + goto fail; + } + dev_info(a->dev, "Remote device %llx %.*s\n", dev_id, (int) uid_len, uid); + + sdev->a = a; + INIT_LIST_HEAD(&sdev->list); + sdev->dev_id = dev_id; + sdev->buf_id = AAUDIO_BUFFER_ID_NONE; + strncpy(sdev->uid, uid, uid_len); + sdev->uid[uid_len + 1] = '\0'; + + if (aaudio_cmd_get_primitive_property(a, dev_id, dev_id, + AAUDIO_PROP(AAUDIO_PROP_SCOPE_INPUT, AAUDIO_PROP_LATENCY, 0), NULL, 0, &sdev->in_latency, sizeof(u32))) + dev_warn(a->dev, "Failed to query device input latency\n"); + if (aaudio_cmd_get_primitive_property(a, dev_id, dev_id, + AAUDIO_PROP(AAUDIO_PROP_SCOPE_OUTPUT, AAUDIO_PROP_LATENCY, 0), NULL, 0, &sdev->out_latency, sizeof(u32))) + dev_warn(a->dev, "Failed to query device output latency\n"); + + if (aaudio_cmd_get_input_stream_list(a, &buf, dev_id, &stream_list, &stream_cnt)) { + dev_err(a->dev, "Failed to get input stream list for device %llx\n", dev_id); + goto fail; + } + if (stream_cnt > AAUDIO_DEIVCE_MAX_INPUT_STREAMS) { + dev_warn(a->dev, "Device %s input stream count %llu is larger than the supported count of %u\n", + sdev->uid, stream_cnt, AAUDIO_DEIVCE_MAX_INPUT_STREAMS); + stream_cnt = AAUDIO_DEIVCE_MAX_INPUT_STREAMS; + } + sdev->in_stream_cnt = stream_cnt; + for (i = 0; i < stream_cnt; i++) { + sdev->in_streams[i].id = stream_list[i]; + sdev->in_streams[i].buffer_cnt = 0; + aaudio_init_stream_info(sdev, &sdev->in_streams[i]); + sdev->in_streams[i].latency += sdev->in_latency; + } + + if (aaudio_cmd_get_output_stream_list(a, &buf, dev_id, &stream_list, &stream_cnt)) { + dev_err(a->dev, "Failed to get output stream list for device %llx\n", dev_id); + goto fail; + } + if (stream_cnt > AAUDIO_DEIVCE_MAX_OUTPUT_STREAMS) { + dev_warn(a->dev, "Device %s input stream count %llu is larger than the supported count of %u\n", + sdev->uid, stream_cnt, AAUDIO_DEIVCE_MAX_OUTPUT_STREAMS); + stream_cnt = AAUDIO_DEIVCE_MAX_OUTPUT_STREAMS; + } + sdev->out_stream_cnt = stream_cnt; + for (i = 0; i < stream_cnt; i++) { + sdev->out_streams[i].id = stream_list[i]; + sdev->out_streams[i].buffer_cnt = 0; + aaudio_init_stream_info(sdev, &sdev->out_streams[i]); + sdev->out_streams[i].latency += sdev->in_latency; + } + + if (sdev->is_pcm) + aaudio_create_pcm(sdev); + /* Headphone Jack status */ + if (!strcmp(sdev->uid, "Codec Output")) { + if (snd_jack_new(a->card, sdev->uid, SND_JACK_HEADPHONE, &sdev->jack, true, false)) + dev_warn(a->dev, "Failed to create an attached jack for %s\n", sdev->uid); + aaudio_cmd_property_listener(a, sdev->dev_id, sdev->dev_id, + AAUDIO_PROP(AAUDIO_PROP_SCOPE_OUTPUT, AAUDIO_PROP_JACK_PLUGGED, 0)); + aaudio_handle_jack_connection_change(sdev); + } + + aaudio_reply_free(&buf); + + list_add_tail(&sdev->list, &a->subdevice_list); + return; + +fail: + aaudio_reply_free(&buf); + kfree(sdev); +} + +static void aaudio_init_stream_info(struct aaudio_subdevice *sdev, struct aaudio_stream *strm) +{ + if (aaudio_cmd_get_primitive_property(sdev->a, sdev->dev_id, strm->id, + AAUDIO_PROP(AAUDIO_PROP_SCOPE_GLOBAL, AAUDIO_PROP_PHYS_FORMAT, 0), NULL, 0, + &strm->desc, sizeof(strm->desc))) + dev_warn(sdev->a->dev, "Failed to query stream descriptor\n"); + if (aaudio_cmd_get_primitive_property(sdev->a, sdev->dev_id, strm->id, + AAUDIO_PROP(AAUDIO_PROP_SCOPE_GLOBAL, AAUDIO_PROP_LATENCY, 0), NULL, 0, &strm->latency, sizeof(u32))) + dev_warn(sdev->a->dev, "Failed to query stream latency\n"); + if (strm->desc.format_id == AAUDIO_FORMAT_LPCM) + sdev->is_pcm = true; +} + +static void aaudio_free_dev(struct aaudio_subdevice *sdev) +{ + size_t i; + for (i = 0; i < sdev->in_stream_cnt; i++) { + if (sdev->in_streams[i].alsa_hw_desc) + kfree(sdev->in_streams[i].alsa_hw_desc); + if (sdev->in_streams[i].buffers) + kfree(sdev->in_streams[i].buffers); + } + for (i = 0; i < sdev->out_stream_cnt; i++) { + if (sdev->out_streams[i].alsa_hw_desc) + kfree(sdev->out_streams[i].alsa_hw_desc); + if (sdev->out_streams[i].buffers) + kfree(sdev->out_streams[i].buffers); + } + kfree(sdev); +} + +static struct aaudio_subdevice *aaudio_find_dev_by_dev_id(struct aaudio_device *a, aaudio_device_id_t dev_id) +{ + struct aaudio_subdevice *sdev; + list_for_each_entry(sdev, &a->subdevice_list, list) { + if (dev_id == sdev->dev_id) + return sdev; + } + return NULL; +} + +static struct aaudio_subdevice *aaudio_find_dev_by_uid(struct aaudio_device *a, const char *uid) +{ + struct aaudio_subdevice *sdev; + list_for_each_entry(sdev, &a->subdevice_list, list) { + if (!strcmp(uid, sdev->uid)) + return sdev; + } + return NULL; +} + +static void aaudio_init_bs_stream(struct aaudio_device *a, struct aaudio_stream *strm, + struct aaudio_buffer_struct_stream *bs_strm); +static void aaudio_init_bs_stream_host(struct aaudio_device *a, struct aaudio_stream *strm, + struct aaudio_buffer_struct_stream *bs_strm); + +static int aaudio_init_bs(struct aaudio_device *a) +{ + int i, j; + struct aaudio_buffer_struct_device *dev; + struct aaudio_subdevice *sdev; + u32 ver, sig, bs_base; + + ver = ioread32(&a->reg_mem_gpr[0]); + if (ver < 3) { + dev_err(a->dev, "aaudio: Bad GPR version (%u)", ver); + return -EINVAL; + } + sig = ioread32(&a->reg_mem_gpr[1]); + if (sig != AAUDIO_SIG) { + dev_err(a->dev, "aaudio: Bad GPR sig (%x)", sig); + return -EINVAL; + } + bs_base = ioread32(&a->reg_mem_gpr[2]); + a->bs = (struct aaudio_buffer_struct *) ((u8 *) a->reg_mem_bs + bs_base); + if (a->bs->signature != AAUDIO_SIG) { + dev_err(a->dev, "aaudio: Bad BufferStruct sig (%x)", a->bs->signature); + return -EINVAL; + } + dev_info(a->dev, "aaudio: BufferStruct ver = %i\n", a->bs->version); + dev_info(a->dev, "aaudio: Num devices = %i\n", a->bs->num_devices); + for (i = 0; i < a->bs->num_devices; i++) { + dev = &a->bs->devices[i]; + dev_info(a->dev, "aaudio: Device %i %s\n", i, dev->name); + + sdev = aaudio_find_dev_by_uid(a, dev->name); + if (!sdev) { + dev_err(a->dev, "aaudio: Subdevice not found for BufferStruct device %s\n", dev->name); + continue; + } + sdev->buf_id = (u8) i; + dev->num_input_streams = 0; + for (j = 0; j < dev->num_output_streams; j++) { + dev_info(a->dev, "aaudio: Device %i Stream %i: Output; Buffer Count = %i\n", i, j, + dev->output_streams[j].num_buffers); + if (j < sdev->out_stream_cnt) + aaudio_init_bs_stream(a, &sdev->out_streams[j], &dev->output_streams[j]); + } + } + + list_for_each_entry(sdev, &a->subdevice_list, list) { + if (sdev->buf_id != AAUDIO_BUFFER_ID_NONE) + continue; + sdev->buf_id = i; + dev_info(a->dev, "aaudio: Created device %i %s\n", i, sdev->uid); + strcpy(a->bs->devices[i].name, sdev->uid); + a->bs->devices[i].num_input_streams = 0; + a->bs->devices[i].num_output_streams = 0; + a->bs->num_devices = ++i; + } + list_for_each_entry(sdev, &a->subdevice_list, list) { + if (sdev->in_stream_cnt == 1) { + dev_info(a->dev, "aaudio: Device %i Host Stream; Input\n", sdev->buf_id); + aaudio_init_bs_stream_host(a, &sdev->in_streams[0], &a->bs->devices[sdev->buf_id].input_streams[0]); + a->bs->devices[sdev->buf_id].num_input_streams = 1; + wmb(); + + if (aaudio_cmd_set_input_stream_address_ranges(a, sdev->dev_id)) { + dev_err(a->dev, "aaudio: Failed to set input stream address ranges\n"); + } + } + } + + return 0; +} + +static void aaudio_init_bs_stream(struct aaudio_device *a, struct aaudio_stream *strm, + struct aaudio_buffer_struct_stream *bs_strm) +{ + size_t i; + strm->buffer_cnt = bs_strm->num_buffers; + if (bs_strm->num_buffers > AAUDIO_DEIVCE_MAX_BUFFER_COUNT) { + dev_warn(a->dev, "BufferStruct buffer count %u exceeds driver limit of %u\n", bs_strm->num_buffers, + AAUDIO_DEIVCE_MAX_BUFFER_COUNT); + strm->buffer_cnt = AAUDIO_DEIVCE_MAX_BUFFER_COUNT; + } + if (!strm->buffer_cnt) + return; + strm->buffers = kmalloc_array(strm->buffer_cnt, sizeof(struct aaudio_dma_buf), GFP_KERNEL); + if (!strm->buffers) { + dev_err(a->dev, "Buffer list allocation failed\n"); + return; + } + for (i = 0; i < strm->buffer_cnt; i++) { + strm->buffers[i].dma_addr = a->reg_mem_bs_dma + (dma_addr_t) bs_strm->buffers[i].address; + strm->buffers[i].ptr = a->reg_mem_bs + bs_strm->buffers[i].address; + strm->buffers[i].size = bs_strm->buffers[i].size; + } + + if (strm->buffer_cnt == 1) { + strm->alsa_hw_desc = kmalloc(sizeof(struct snd_pcm_hardware), GFP_KERNEL); + if (aaudio_create_hw_info(&strm->desc, strm->alsa_hw_desc, strm->buffers[0].size)) { + kfree(strm->alsa_hw_desc); + strm->alsa_hw_desc = NULL; + } + } +} + +static void aaudio_init_bs_stream_host(struct aaudio_device *a, struct aaudio_stream *strm, + struct aaudio_buffer_struct_stream *bs_strm) +{ + size_t size; + dma_addr_t dma_addr; + void *dma_ptr; + size = strm->desc.bytes_per_packet * 16640; + dma_ptr = dma_alloc_coherent(&a->pci->dev, size, &dma_addr, GFP_KERNEL); + if (!dma_ptr) { + dev_err(a->dev, "dma_alloc_coherent failed\n"); + return; + } + bs_strm->buffers[0].address = dma_addr; + bs_strm->buffers[0].size = size; + bs_strm->num_buffers = 1; + + memset(dma_ptr, 0, size); + + strm->buffer_cnt = 1; + strm->buffers = kmalloc_array(strm->buffer_cnt, sizeof(struct aaudio_dma_buf), GFP_KERNEL); + if (!strm->buffers) { + dev_err(a->dev, "Buffer list allocation failed\n"); + return; + } + strm->buffers[0].dma_addr = dma_addr; + strm->buffers[0].ptr = dma_ptr; + strm->buffers[0].size = size; + + strm->alsa_hw_desc = kmalloc(sizeof(struct snd_pcm_hardware), GFP_KERNEL); + if (aaudio_create_hw_info(&strm->desc, strm->alsa_hw_desc, strm->buffers[0].size)) { + kfree(strm->alsa_hw_desc); + strm->alsa_hw_desc = NULL; + } +} + +static void aaudio_handle_prop_change(struct aaudio_device *a, struct aaudio_msg *msg); + +void aaudio_handle_notification(struct aaudio_device *a, struct aaudio_msg *msg) +{ + struct aaudio_send_ctx sctx; + struct aaudio_msg_base base; + if (aaudio_msg_read_base(msg, &base)) + return; + switch (base.msg) { + case AAUDIO_MSG_NOTIFICATION_BOOT: + dev_info(a->dev, "Received boot notification from remote\n"); + + /* Resend the alive notify */ + if (aaudio_send(a, &sctx, 500, + aaudio_msg_write_alive_notification, 1, 3)) { + pr_err("Sending alive notification failed\n"); + } + break; + case AAUDIO_MSG_NOTIFICATION_ALIVE: + dev_info(a->dev, "Received alive notification from remote\n"); + complete_all(&a->remote_alive); + break; + case AAUDIO_MSG_PROPERTY_CHANGED: + aaudio_handle_prop_change(a, msg); + break; + default: + dev_info(a->dev, "Unhandled notification %i", base.msg); + break; + } +} + +struct aaudio_prop_change_work_struct { + struct work_struct ws; + struct aaudio_device *a; + aaudio_device_id_t dev; + aaudio_object_id_t obj; + struct aaudio_prop_addr prop; +}; + +static void aaudio_handle_jack_connection_change(struct aaudio_subdevice *sdev) +{ + u32 plugged; + if (!sdev->jack) + return; + /* NOTE: Apple made the plug status scoped to the input and output streams. This makes no sense for us, so I just + * always pick the OUTPUT status. */ + if (aaudio_cmd_get_primitive_property(sdev->a, sdev->dev_id, sdev->dev_id, + AAUDIO_PROP(AAUDIO_PROP_SCOPE_OUTPUT, AAUDIO_PROP_JACK_PLUGGED, 0), NULL, 0, &plugged, sizeof(plugged))) { + dev_err(sdev->a->dev, "Failed to get jack enable status\n"); + return; + } + dev_dbg(sdev->a->dev, "Jack is now %s\n", plugged ? "plugged" : "unplugged"); + snd_jack_report(sdev->jack, plugged ? sdev->jack->type : 0); +} + +void aaudio_handle_prop_change_work(struct work_struct *ws) +{ + struct aaudio_prop_change_work_struct *work = container_of(ws, struct aaudio_prop_change_work_struct, ws); + struct aaudio_subdevice *sdev; + + sdev = aaudio_find_dev_by_dev_id(work->a, work->dev); + if (!sdev) { + dev_err(work->a->dev, "Property notification change: device not found\n"); + goto done; + } + dev_dbg(work->a->dev, "Property changed for device: %s\n", sdev->uid); + + if (work->prop.scope == AAUDIO_PROP_SCOPE_OUTPUT && work->prop.selector == AAUDIO_PROP_JACK_PLUGGED) { + aaudio_handle_jack_connection_change(sdev); + } + +done: + kfree(work); +} + +void aaudio_handle_prop_change(struct aaudio_device *a, struct aaudio_msg *msg) +{ + /* NOTE: This is a scheduled work because this callback will generally need to query device information and this + * is not possible when we are in the reply parsing code's context. */ + struct aaudio_prop_change_work_struct *work; + work = kmalloc(sizeof(struct aaudio_prop_change_work_struct), GFP_KERNEL); + work->a = a; + INIT_WORK(&work->ws, aaudio_handle_prop_change_work); + aaudio_msg_read_property_changed(msg, &work->dev, &work->obj, &work->prop); + schedule_work(&work->ws); +} + +#define aaudio_send_cmd_response(a, sctx, msg, fn, ...) \ + if (aaudio_send_with_tag(a, sctx, ((struct aaudio_msg_header *) msg->data)->tag, 500, fn, ##__VA_ARGS__)) \ + pr_err("aaudio: Failed to reply to a command\n"); + +void aaudio_handle_cmd_timestamp(struct aaudio_device *a, struct aaudio_msg *msg) +{ + ktime_t time_os = ktime_get_boottime(); + struct aaudio_send_ctx sctx; + struct aaudio_subdevice *sdev; + u64 devid, timestamp, update_seed; + aaudio_msg_read_update_timestamp(msg, &devid, ×tamp, &update_seed); + dev_dbg(a->dev, "Received timestamp update for dev=%llx ts=%llx seed=%llx\n", devid, timestamp, update_seed); + + sdev = aaudio_find_dev_by_dev_id(a, devid); + aaudio_handle_timestamp(sdev, time_os, timestamp); + + aaudio_send_cmd_response(a, &sctx, msg, + aaudio_msg_write_update_timestamp_response); +} + +void aaudio_handle_command(struct aaudio_device *a, struct aaudio_msg *msg) +{ + struct aaudio_msg_base base; + if (aaudio_msg_read_base(msg, &base)) + return; + switch (base.msg) { + case AAUDIO_MSG_UPDATE_TIMESTAMP: + aaudio_handle_cmd_timestamp(a, msg); + break; + default: + dev_info(a->dev, "Unhandled device command %i", base.msg); + break; + } +} + +static struct pci_device_id aaudio_ids[ ] = { + { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x1803) }, + { 0, }, +}; + +struct dev_pm_ops aaudio_pci_driver_pm = { + .suspend = aaudio_suspend, + .resume = aaudio_resume +}; +struct pci_driver aaudio_pci_driver = { + .name = "aaudio", + .id_table = aaudio_ids, + .probe = aaudio_probe, + .remove = aaudio_remove, + .driver = { + .pm = &aaudio_pci_driver_pm + } +}; + + +int aaudio_module_init(void) +{ + int result; + if ((result = alloc_chrdev_region(&aaudio_chrdev, 0, 1, "aaudio"))) + goto fail_chrdev; +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,4,0) + aaudio_class = class_create(THIS_MODULE, "aaudio"); +#else + aaudio_class = class_create("aaudio"); +#endif + if (IS_ERR(aaudio_class)) { + result = PTR_ERR(aaudio_class); + goto fail_class; + } + + result = pci_register_driver(&aaudio_pci_driver); + if (result) + goto fail_drv; + return 0; + +fail_drv: + pci_unregister_driver(&aaudio_pci_driver); +fail_class: + class_destroy(aaudio_class); +fail_chrdev: + unregister_chrdev_region(aaudio_chrdev, 1); + if (!result) + result = -EINVAL; + return result; +} + +void aaudio_module_exit(void) +{ + pci_unregister_driver(&aaudio_pci_driver); + class_destroy(aaudio_class); + unregister_chrdev_region(aaudio_chrdev, 1); +} + +struct aaudio_alsa_pcm_id_mapping aaudio_alsa_id_mappings[] = { + {"Speaker", 0}, + {"Digital Mic", 1}, + {"Codec Output", 2}, + {"Codec Input", 3}, + {"Bridge Loopback", 4}, + {} +}; + +module_param_named(index, aaudio_alsa_index, int, 0444); +MODULE_PARM_DESC(index, "Index value for Apple Internal Audio soundcard."); +module_param_named(id, aaudio_alsa_id, charp, 0444); +MODULE_PARM_DESC(id, "ID string for Apple Internal Audio soundcard."); diff --git a/drivers/staging/apple-bce/audio/audio.h b/drivers/staging/apple-bce/audio/audio.h new file mode 100644 index 000000000000..004bc1e22ea4 --- /dev/null +++ b/drivers/staging/apple-bce/audio/audio.h @@ -0,0 +1,125 @@ +#ifndef AAUDIO_H +#define AAUDIO_H + +#include +#include +#include "../apple_bce.h" +#include "protocol_bce.h" +#include "description.h" + +#define AAUDIO_SIG 0x19870423 + +#define AAUDIO_DEVICE_MAX_UID_LEN 128 +#define AAUDIO_DEIVCE_MAX_INPUT_STREAMS 1 +#define AAUDIO_DEIVCE_MAX_OUTPUT_STREAMS 1 +#define AAUDIO_DEIVCE_MAX_BUFFER_COUNT 1 + +#define AAUDIO_BUFFER_ID_NONE 0xffu + +struct snd_card; +struct snd_pcm; +struct snd_pcm_hardware; +struct snd_jack; + +struct __attribute__((packed)) __attribute__((aligned(4))) aaudio_buffer_struct_buffer { + size_t address; + size_t size; + size_t pad[4]; +}; +struct aaudio_buffer_struct_stream { + u8 num_buffers; + struct aaudio_buffer_struct_buffer buffers[100]; + char filler[32]; +}; +struct aaudio_buffer_struct_device { + char name[128]; + u8 num_input_streams; + u8 num_output_streams; + struct aaudio_buffer_struct_stream input_streams[5]; + struct aaudio_buffer_struct_stream output_streams[5]; + char filler[128]; +}; +struct aaudio_buffer_struct { + u32 version; + u32 signature; + u32 flags; + u8 num_devices; + struct aaudio_buffer_struct_device devices[20]; +}; + +struct aaudio_device; +struct aaudio_dma_buf { + dma_addr_t dma_addr; + void *ptr; + size_t size; +}; +struct aaudio_stream { + aaudio_object_id_t id; + size_t buffer_cnt; + struct aaudio_dma_buf *buffers; + + struct aaudio_apple_description desc; + struct snd_pcm_hardware *alsa_hw_desc; + u32 latency; + + bool waiting_for_first_ts; + + ktime_t remote_timestamp; + snd_pcm_sframes_t frame_min; + int started; +}; +struct aaudio_subdevice { + struct aaudio_device *a; + struct list_head list; + aaudio_device_id_t dev_id; + u32 in_latency, out_latency; + u8 buf_id; + int alsa_id; + char uid[AAUDIO_DEVICE_MAX_UID_LEN + 1]; + size_t in_stream_cnt; + struct aaudio_stream in_streams[AAUDIO_DEIVCE_MAX_INPUT_STREAMS]; + size_t out_stream_cnt; + struct aaudio_stream out_streams[AAUDIO_DEIVCE_MAX_OUTPUT_STREAMS]; + bool is_pcm; + struct snd_pcm *pcm; + struct snd_jack *jack; +}; +struct aaudio_alsa_pcm_id_mapping { + const char *name; + int alsa_id; +}; + +struct aaudio_device { + struct pci_dev *pci; + dev_t devt; + struct device *dev; + void __iomem *reg_mem_bs; + dma_addr_t reg_mem_bs_dma; + void __iomem *reg_mem_cfg; + + u32 __iomem *reg_mem_gpr; + + struct aaudio_buffer_struct *bs; + + struct apple_bce_device *bce; + struct aaudio_bce bcem; + + struct snd_card *card; + + struct list_head subdevice_list; + int next_alsa_id; + + struct completion remote_alive; +}; + +void aaudio_handle_notification(struct aaudio_device *a, struct aaudio_msg *msg); +void aaudio_handle_prop_change_work(struct work_struct *ws); +void aaudio_handle_cmd_timestamp(struct aaudio_device *a, struct aaudio_msg *msg); +void aaudio_handle_command(struct aaudio_device *a, struct aaudio_msg *msg); + +int aaudio_module_init(void); +void aaudio_module_exit(void); + +extern struct aaudio_alsa_pcm_id_mapping aaudio_alsa_id_mappings[]; + +#endif //AAUDIO_H diff --git a/drivers/staging/apple-bce/audio/description.h b/drivers/staging/apple-bce/audio/description.h new file mode 100644 index 000000000000..dfef3ab68f27 --- /dev/null +++ b/drivers/staging/apple-bce/audio/description.h @@ -0,0 +1,42 @@ +#ifndef AAUDIO_DESCRIPTION_H +#define AAUDIO_DESCRIPTION_H + +#include + +struct aaudio_apple_description { + u64 sample_rate_double; + u32 format_id; + u32 format_flags; + u32 bytes_per_packet; + u32 frames_per_packet; + u32 bytes_per_frame; + u32 channels_per_frame; + u32 bits_per_channel; + u32 reserved; +}; + +enum { + AAUDIO_FORMAT_LPCM = 0x6c70636d // 'lpcm' +}; + +enum { + AAUDIO_FORMAT_FLAG_FLOAT = 1, + AAUDIO_FORMAT_FLAG_BIG_ENDIAN = 2, + AAUDIO_FORMAT_FLAG_SIGNED = 4, + AAUDIO_FORMAT_FLAG_PACKED = 8, + AAUDIO_FORMAT_FLAG_ALIGNED_HIGH = 16, + AAUDIO_FORMAT_FLAG_NON_INTERLEAVED = 32, + AAUDIO_FORMAT_FLAG_NON_MIXABLE = 64 +}; + +static inline u64 aaudio_double_to_u64(u64 d) +{ + u8 sign = (u8) ((d >> 63) & 1); + s32 exp = (s32) ((d >> 52) & 0x7ff) - 1023; + u64 fr = d & ((1LL << 52) - 1); + if (sign || exp < 0) + return 0; + return (u64) ((1LL << exp) + (fr >> (52 - exp))); +} + +#endif //AAUDIO_DESCRIPTION_H diff --git a/drivers/staging/apple-bce/audio/pcm.c b/drivers/staging/apple-bce/audio/pcm.c new file mode 100644 index 000000000000..1026e10a9ac5 --- /dev/null +++ b/drivers/staging/apple-bce/audio/pcm.c @@ -0,0 +1,308 @@ +#include "pcm.h" +#include "audio.h" + +static u64 aaudio_get_alsa_fmtbit(struct aaudio_apple_description *desc) +{ + if (desc->format_flags & AAUDIO_FORMAT_FLAG_FLOAT) { + if (desc->bits_per_channel == 32) { + if (desc->format_flags & AAUDIO_FORMAT_FLAG_BIG_ENDIAN) + return SNDRV_PCM_FMTBIT_FLOAT_BE; + else + return SNDRV_PCM_FMTBIT_FLOAT_LE; + } else if (desc->bits_per_channel == 64) { + if (desc->format_flags & AAUDIO_FORMAT_FLAG_BIG_ENDIAN) + return SNDRV_PCM_FMTBIT_FLOAT64_BE; + else + return SNDRV_PCM_FMTBIT_FLOAT64_LE; + } else { + pr_err("aaudio: unsupported bits per channel for float format: %u\n", desc->bits_per_channel); + return 0; + } + } +#define DEFINE_BPC_OPTION(val, b) \ + case val: \ + if (desc->format_flags & AAUDIO_FORMAT_FLAG_BIG_ENDIAN) { \ + if (desc->format_flags & AAUDIO_FORMAT_FLAG_SIGNED) \ + return SNDRV_PCM_FMTBIT_S ## b ## BE; \ + else \ + return SNDRV_PCM_FMTBIT_U ## b ## BE; \ + } else { \ + if (desc->format_flags & AAUDIO_FORMAT_FLAG_SIGNED) \ + return SNDRV_PCM_FMTBIT_S ## b ## LE; \ + else \ + return SNDRV_PCM_FMTBIT_U ## b ## LE; \ + } + if (desc->format_flags & AAUDIO_FORMAT_FLAG_PACKED) { + switch (desc->bits_per_channel) { + case 8: + case 16: + case 32: + break; + DEFINE_BPC_OPTION(24, 24_3) + default: + pr_err("aaudio: unsupported bits per channel for packed format: %u\n", desc->bits_per_channel); + return 0; + } + } + if (desc->format_flags & AAUDIO_FORMAT_FLAG_ALIGNED_HIGH) { + switch (desc->bits_per_channel) { + DEFINE_BPC_OPTION(24, 32_) + default: + pr_err("aaudio: unsupported bits per channel for high-aligned format: %u\n", desc->bits_per_channel); + return 0; + } + } + switch (desc->bits_per_channel) { + case 8: + if (desc->format_flags & AAUDIO_FORMAT_FLAG_SIGNED) + return SNDRV_PCM_FMTBIT_S8; + else + return SNDRV_PCM_FMTBIT_U8; + DEFINE_BPC_OPTION(16, 16_) + DEFINE_BPC_OPTION(24, 24_) + DEFINE_BPC_OPTION(32, 32_) + default: + pr_err("aaudio: unsupported bits per channel: %u\n", desc->bits_per_channel); + return 0; + } +} +int aaudio_create_hw_info(struct aaudio_apple_description *desc, struct snd_pcm_hardware *alsa_hw, + size_t buf_size) +{ + uint rate; + alsa_hw->info = (SNDRV_PCM_INFO_MMAP | + SNDRV_PCM_INFO_BLOCK_TRANSFER | + SNDRV_PCM_INFO_MMAP_VALID | + SNDRV_PCM_INFO_DOUBLE); + if (desc->format_flags & AAUDIO_FORMAT_FLAG_NON_MIXABLE) + pr_warn("aaudio: unsupported hw flag: NON_MIXABLE\n"); + if (!(desc->format_flags & AAUDIO_FORMAT_FLAG_NON_INTERLEAVED)) + alsa_hw->info |= SNDRV_PCM_INFO_INTERLEAVED; + alsa_hw->formats = aaudio_get_alsa_fmtbit(desc); + if (!alsa_hw->formats) + return -EINVAL; + rate = (uint) aaudio_double_to_u64(desc->sample_rate_double); + alsa_hw->rates = snd_pcm_rate_to_rate_bit(rate); + alsa_hw->rate_min = rate; + alsa_hw->rate_max = rate; + alsa_hw->channels_min = desc->channels_per_frame; + alsa_hw->channels_max = desc->channels_per_frame; + alsa_hw->buffer_bytes_max = buf_size; + alsa_hw->period_bytes_min = desc->bytes_per_packet; + alsa_hw->period_bytes_max = desc->bytes_per_packet; + alsa_hw->periods_min = (uint) (buf_size / desc->bytes_per_packet); + alsa_hw->periods_max = (uint) (buf_size / desc->bytes_per_packet); + pr_debug("aaudio_create_hw_info: format = %llu, rate = %u/%u. channels = %u, periods = %u, period size = %lu\n", + alsa_hw->formats, alsa_hw->rate_min, alsa_hw->rates, alsa_hw->channels_min, alsa_hw->periods_min, + alsa_hw->period_bytes_min); + return 0; +} + +static struct aaudio_stream *aaudio_pcm_stream(struct snd_pcm_substream *substream) +{ + struct aaudio_subdevice *sdev = snd_pcm_substream_chip(substream); + if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) + return &sdev->out_streams[substream->number]; + else + return &sdev->in_streams[substream->number]; +} + +static int aaudio_pcm_open(struct snd_pcm_substream *substream) +{ + pr_debug("aaudio_pcm_open\n"); + substream->runtime->hw = *aaudio_pcm_stream(substream)->alsa_hw_desc; + + return 0; +} + +static int aaudio_pcm_close(struct snd_pcm_substream *substream) +{ + pr_debug("aaudio_pcm_close\n"); + return 0; +} + +static int aaudio_pcm_prepare(struct snd_pcm_substream *substream) +{ + return 0; +} + +static int aaudio_pcm_hw_params(struct snd_pcm_substream *substream, struct snd_pcm_hw_params *hw_params) +{ + struct aaudio_stream *astream = aaudio_pcm_stream(substream); + pr_debug("aaudio_pcm_hw_params\n"); + + if (!astream->buffer_cnt || !astream->buffers) + return -EINVAL; + + substream->runtime->dma_area = astream->buffers[0].ptr; + substream->runtime->dma_addr = astream->buffers[0].dma_addr; + substream->runtime->dma_bytes = astream->buffers[0].size; + return 0; +} + +static int aaudio_pcm_hw_free(struct snd_pcm_substream *substream) +{ + pr_debug("aaudio_pcm_hw_free\n"); + return 0; +} + +static void aaudio_pcm_start(struct snd_pcm_substream *substream) +{ + struct aaudio_subdevice *sdev = snd_pcm_substream_chip(substream); + struct aaudio_stream *stream = aaudio_pcm_stream(substream); + void *buf; + size_t s; + ktime_t time_start, time_end; + bool back_buffer; + time_start = ktime_get(); + + back_buffer = (substream->stream == SNDRV_PCM_STREAM_PLAYBACK); + + if (back_buffer) { + s = frames_to_bytes(substream->runtime, substream->runtime->control->appl_ptr); + buf = kmalloc(s, GFP_KERNEL); + memcpy_fromio(buf, substream->runtime->dma_area, s); + time_end = ktime_get(); + pr_debug("aaudio: Backed up the buffer in %lluns [%li]\n", ktime_to_ns(time_end - time_start), + substream->runtime->control->appl_ptr); + } + + stream->waiting_for_first_ts = true; + stream->frame_min = stream->latency; + + aaudio_cmd_start_io(sdev->a, sdev->dev_id); + if (back_buffer) + memcpy_toio(substream->runtime->dma_area, buf, s); + + time_end = ktime_get(); + pr_debug("aaudio: Started the audio device in %lluns\n", ktime_to_ns(time_end - time_start)); +} + +static int aaudio_pcm_trigger(struct snd_pcm_substream *substream, int cmd) +{ + struct aaudio_subdevice *sdev = snd_pcm_substream_chip(substream); + struct aaudio_stream *stream = aaudio_pcm_stream(substream); + pr_debug("aaudio_pcm_trigger %x\n", cmd); + + /* We only supports triggers on the #0 buffer */ + if (substream->number != 0) + return 0; + switch (cmd) { + case SNDRV_PCM_TRIGGER_START: + aaudio_pcm_start(substream); + stream->started = 1; + break; + case SNDRV_PCM_TRIGGER_STOP: + aaudio_cmd_stop_io(sdev->a, sdev->dev_id); + stream->started = 0; + break; + default: + return -EINVAL; + } + return 0; +} + +static snd_pcm_uframes_t aaudio_pcm_pointer(struct snd_pcm_substream *substream) +{ + struct aaudio_stream *stream = aaudio_pcm_stream(substream); + ktime_t time_from_start; + snd_pcm_sframes_t frames; + snd_pcm_sframes_t buffer_time_length; + + if (!stream->started || stream->waiting_for_first_ts) { + pr_warn("aaudio_pcm_pointer while not started\n"); + return 0; + } + + /* Approximate the pointer based on the last received timestamp */ + time_from_start = ktime_get_boottime() - stream->remote_timestamp; + buffer_time_length = NSEC_PER_SEC * substream->runtime->buffer_size / substream->runtime->rate; + frames = (ktime_to_ns(time_from_start) % buffer_time_length) * substream->runtime->buffer_size / buffer_time_length; + if (ktime_to_ns(time_from_start) < buffer_time_length) { + if (frames < stream->frame_min) + frames = stream->frame_min; + else + stream->frame_min = 0; + } else { + if (ktime_to_ns(time_from_start) < 2 * buffer_time_length) + stream->frame_min = frames; + else + stream->frame_min = 0; /* Heavy desync */ + } + frames -= stream->latency; + if (frames < 0) + frames += ((-frames - 1) / substream->runtime->buffer_size + 1) * substream->runtime->buffer_size; + return (snd_pcm_uframes_t) frames; +} + +static struct snd_pcm_ops aaudio_pcm_ops = { + .open = aaudio_pcm_open, + .close = aaudio_pcm_close, + .ioctl = snd_pcm_lib_ioctl, + .hw_params = aaudio_pcm_hw_params, + .hw_free = aaudio_pcm_hw_free, + .prepare = aaudio_pcm_prepare, + .trigger = aaudio_pcm_trigger, + .pointer = aaudio_pcm_pointer, + .mmap = snd_pcm_lib_mmap_iomem +}; + +int aaudio_create_pcm(struct aaudio_subdevice *sdev) +{ + struct snd_pcm *pcm; + struct aaudio_alsa_pcm_id_mapping *id_mapping; + int err; + + if (!sdev->is_pcm || (sdev->in_stream_cnt == 0 && sdev->out_stream_cnt == 0)) { + return -EINVAL; + } + + for (id_mapping = aaudio_alsa_id_mappings; id_mapping->name; id_mapping++) { + if (!strcmp(sdev->uid, id_mapping->name)) { + sdev->alsa_id = id_mapping->alsa_id; + break; + } + } + if (!id_mapping->name) + sdev->alsa_id = sdev->a->next_alsa_id++; + err = snd_pcm_new(sdev->a->card, sdev->uid, sdev->alsa_id, + (int) sdev->out_stream_cnt, (int) sdev->in_stream_cnt, &pcm); + if (err < 0) + return err; + pcm->private_data = sdev; + pcm->nonatomic = 1; + sdev->pcm = pcm; + strcpy(pcm->name, sdev->uid); + snd_pcm_set_ops(pcm, SNDRV_PCM_STREAM_PLAYBACK, &aaudio_pcm_ops); + snd_pcm_set_ops(pcm, SNDRV_PCM_STREAM_CAPTURE, &aaudio_pcm_ops); + return 0; +} + +static void aaudio_handle_stream_timestamp(struct snd_pcm_substream *substream, ktime_t timestamp) +{ + unsigned long flags; + struct aaudio_stream *stream; + + stream = aaudio_pcm_stream(substream); + snd_pcm_stream_lock_irqsave(substream, flags); + stream->remote_timestamp = timestamp; + if (stream->waiting_for_first_ts) { + stream->waiting_for_first_ts = false; + snd_pcm_stream_unlock_irqrestore(substream, flags); + return; + } + snd_pcm_stream_unlock_irqrestore(substream, flags); + snd_pcm_period_elapsed(substream); +} + +void aaudio_handle_timestamp(struct aaudio_subdevice *sdev, ktime_t os_timestamp, u64 dev_timestamp) +{ + struct snd_pcm_substream *substream; + + substream = sdev->pcm->streams[SNDRV_PCM_STREAM_PLAYBACK].substream; + if (substream) + aaudio_handle_stream_timestamp(substream, dev_timestamp); + substream = sdev->pcm->streams[SNDRV_PCM_STREAM_CAPTURE].substream; + if (substream) + aaudio_handle_stream_timestamp(substream, os_timestamp); +} diff --git a/drivers/staging/apple-bce/audio/pcm.h b/drivers/staging/apple-bce/audio/pcm.h new file mode 100644 index 000000000000..ea5f35fbe408 --- /dev/null +++ b/drivers/staging/apple-bce/audio/pcm.h @@ -0,0 +1,16 @@ +#ifndef AAUDIO_PCM_H +#define AAUDIO_PCM_H + +#include +#include + +struct aaudio_subdevice; +struct aaudio_apple_description; +struct snd_pcm_hardware; + +int aaudio_create_hw_info(struct aaudio_apple_description *desc, struct snd_pcm_hardware *alsa_hw, size_t buf_size); +int aaudio_create_pcm(struct aaudio_subdevice *sdev); + +void aaudio_handle_timestamp(struct aaudio_subdevice *sdev, ktime_t os_timestamp, u64 dev_timestamp); + +#endif //AAUDIO_PCM_H diff --git a/drivers/staging/apple-bce/audio/protocol.c b/drivers/staging/apple-bce/audio/protocol.c new file mode 100644 index 000000000000..2314813aeead --- /dev/null +++ b/drivers/staging/apple-bce/audio/protocol.c @@ -0,0 +1,347 @@ +#include "protocol.h" +#include "protocol_bce.h" +#include "audio.h" + +int aaudio_msg_read_base(struct aaudio_msg *msg, struct aaudio_msg_base *base) +{ + if (msg->size < sizeof(struct aaudio_msg_header) + sizeof(struct aaudio_msg_base) * 2) + return -EINVAL; + *base = *((struct aaudio_msg_base *) ((struct aaudio_msg_header *) msg->data + 1)); + return 0; +} + +#define READ_START(type) \ + size_t offset = sizeof(struct aaudio_msg_header) + sizeof(struct aaudio_msg_base); (void)offset; \ + if (((struct aaudio_msg_base *) ((struct aaudio_msg_header *) msg->data + 1))->msg != type) \ + return -EINVAL; +#define READ_DEVID_VAR(devid) *devid = ((struct aaudio_msg_header *) msg->data)->device_id +#define READ_VAL(type) ({ offset += sizeof(type); *((type *) ((u8 *) msg->data + offset - sizeof(type))); }) +#define READ_VAR(type, var) *var = READ_VAL(type) + +int aaudio_msg_read_start_io_response(struct aaudio_msg *msg) +{ + READ_START(AAUDIO_MSG_START_IO_RESPONSE); + return 0; +} + +int aaudio_msg_read_stop_io_response(struct aaudio_msg *msg) +{ + READ_START(AAUDIO_MSG_STOP_IO_RESPONSE); + return 0; +} + +int aaudio_msg_read_update_timestamp(struct aaudio_msg *msg, aaudio_device_id_t *devid, + u64 *timestamp, u64 *update_seed) +{ + READ_START(AAUDIO_MSG_UPDATE_TIMESTAMP); + READ_DEVID_VAR(devid); + READ_VAR(u64, timestamp); + READ_VAR(u64, update_seed); + return 0; +} + +int aaudio_msg_read_get_property_response(struct aaudio_msg *msg, aaudio_object_id_t *obj, + struct aaudio_prop_addr *prop, void **data, u64 *data_size) +{ + READ_START(AAUDIO_MSG_GET_PROPERTY_RESPONSE); + READ_VAR(aaudio_object_id_t, obj); + READ_VAR(u32, &prop->element); + READ_VAR(u32, &prop->scope); + READ_VAR(u32, &prop->selector); + READ_VAR(u64, data_size); + *data = ((u8 *) msg->data + offset); + /* offset += data_size; */ + return 0; +} + +int aaudio_msg_read_set_property_response(struct aaudio_msg *msg, aaudio_object_id_t *obj) +{ + READ_START(AAUDIO_MSG_SET_PROPERTY_RESPONSE); + READ_VAR(aaudio_object_id_t, obj); + return 0; +} + +int aaudio_msg_read_property_listener_response(struct aaudio_msg *msg, aaudio_object_id_t *obj, + struct aaudio_prop_addr *prop) +{ + READ_START(AAUDIO_MSG_PROPERTY_LISTENER_RESPONSE); + READ_VAR(aaudio_object_id_t, obj); + READ_VAR(u32, &prop->element); + READ_VAR(u32, &prop->scope); + READ_VAR(u32, &prop->selector); + return 0; +} + +int aaudio_msg_read_property_changed(struct aaudio_msg *msg, aaudio_device_id_t *devid, aaudio_object_id_t *obj, + struct aaudio_prop_addr *prop) +{ + READ_START(AAUDIO_MSG_PROPERTY_CHANGED); + READ_DEVID_VAR(devid); + READ_VAR(aaudio_object_id_t, obj); + READ_VAR(u32, &prop->element); + READ_VAR(u32, &prop->scope); + READ_VAR(u32, &prop->selector); + return 0; +} + +int aaudio_msg_read_set_input_stream_address_ranges_response(struct aaudio_msg *msg) +{ + READ_START(AAUDIO_MSG_SET_INPUT_STREAM_ADDRESS_RANGES_RESPONSE); + return 0; +} + +int aaudio_msg_read_get_input_stream_list_response(struct aaudio_msg *msg, aaudio_object_id_t **str_l, u64 *str_cnt) +{ + READ_START(AAUDIO_MSG_GET_INPUT_STREAM_LIST_RESPONSE); + READ_VAR(u64, str_cnt); + *str_l = (aaudio_device_id_t *) ((u8 *) msg->data + offset); + /* offset += str_cnt * sizeof(aaudio_object_id_t); */ + return 0; +} + +int aaudio_msg_read_get_output_stream_list_response(struct aaudio_msg *msg, aaudio_object_id_t **str_l, u64 *str_cnt) +{ + READ_START(AAUDIO_MSG_GET_OUTPUT_STREAM_LIST_RESPONSE); + READ_VAR(u64, str_cnt); + *str_l = (aaudio_device_id_t *) ((u8 *) msg->data + offset); + /* offset += str_cnt * sizeof(aaudio_object_id_t); */ + return 0; +} + +int aaudio_msg_read_set_remote_access_response(struct aaudio_msg *msg) +{ + READ_START(AAUDIO_MSG_SET_REMOTE_ACCESS_RESPONSE); + return 0; +} + +int aaudio_msg_read_get_device_list_response(struct aaudio_msg *msg, aaudio_device_id_t **dev_l, u64 *dev_cnt) +{ + READ_START(AAUDIO_MSG_GET_DEVICE_LIST_RESPONSE); + READ_VAR(u64, dev_cnt); + *dev_l = (aaudio_device_id_t *) ((u8 *) msg->data + offset); + /* offset += dev_cnt * sizeof(aaudio_device_id_t); */ + return 0; +} + +#define WRITE_START_OF_TYPE(typev, devid) \ + size_t offset = sizeof(struct aaudio_msg_header); (void) offset; \ + ((struct aaudio_msg_header *) msg->data)->type = (typev); \ + ((struct aaudio_msg_header *) msg->data)->device_id = (devid); +#define WRITE_START_COMMAND(devid) WRITE_START_OF_TYPE(AAUDIO_MSG_TYPE_COMMAND, devid) +#define WRITE_START_RESPONSE() WRITE_START_OF_TYPE(AAUDIO_MSG_TYPE_RESPONSE, 0) +#define WRITE_START_NOTIFICATION() WRITE_START_OF_TYPE(AAUDIO_MSG_TYPE_NOTIFICATION, 0) +#define WRITE_VAL(type, value) { *((type *) ((u8 *) msg->data + offset)) = value; offset += sizeof(value); } +#define WRITE_BIN(value, size) { memcpy((u8 *) msg->data + offset, value, size); offset += size; } +#define WRITE_BASE(type) WRITE_VAL(u32, type) WRITE_VAL(u32, 0) +#define WRITE_END() { msg->size = offset; } + +void aaudio_msg_write_start_io(struct aaudio_msg *msg, aaudio_device_id_t dev) +{ + WRITE_START_COMMAND(dev); + WRITE_BASE(AAUDIO_MSG_START_IO); + WRITE_END(); +} + +void aaudio_msg_write_stop_io(struct aaudio_msg *msg, aaudio_device_id_t dev) +{ + WRITE_START_COMMAND(dev); + WRITE_BASE(AAUDIO_MSG_STOP_IO); + WRITE_END(); +} + +void aaudio_msg_write_get_property(struct aaudio_msg *msg, aaudio_device_id_t dev, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size) +{ + WRITE_START_COMMAND(dev); + WRITE_BASE(AAUDIO_MSG_GET_PROPERTY); + WRITE_VAL(aaudio_object_id_t, obj); + WRITE_VAL(u32, prop.element); + WRITE_VAL(u32, prop.scope); + WRITE_VAL(u32, prop.selector); + WRITE_VAL(u64, qualifier_size); + WRITE_BIN(qualifier, qualifier_size); + WRITE_END(); +} + +void aaudio_msg_write_set_property(struct aaudio_msg *msg, aaudio_device_id_t dev, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *data, u64 data_size, void *qualifier, u64 qualifier_size) +{ + WRITE_START_COMMAND(dev); + WRITE_BASE(AAUDIO_MSG_SET_PROPERTY); + WRITE_VAL(aaudio_object_id_t, obj); + WRITE_VAL(u32, prop.element); + WRITE_VAL(u32, prop.scope); + WRITE_VAL(u32, prop.selector); + WRITE_VAL(u64, data_size); + WRITE_BIN(data, data_size); + WRITE_VAL(u64, qualifier_size); + WRITE_BIN(qualifier, qualifier_size); + WRITE_END(); +} + +void aaudio_msg_write_property_listener(struct aaudio_msg *msg, aaudio_device_id_t dev, aaudio_object_id_t obj, + struct aaudio_prop_addr prop) +{ + WRITE_START_COMMAND(dev); + WRITE_BASE(AAUDIO_MSG_PROPERTY_LISTENER); + WRITE_VAL(aaudio_object_id_t, obj); + WRITE_VAL(u32, prop.element); + WRITE_VAL(u32, prop.scope); + WRITE_VAL(u32, prop.selector); + WRITE_END(); +} + +void aaudio_msg_write_set_input_stream_address_ranges(struct aaudio_msg *msg, aaudio_device_id_t devid) +{ + WRITE_START_COMMAND(devid); + WRITE_BASE(AAUDIO_MSG_SET_INPUT_STREAM_ADDRESS_RANGES); + WRITE_END(); +} + +void aaudio_msg_write_get_input_stream_list(struct aaudio_msg *msg, aaudio_device_id_t devid) +{ + WRITE_START_COMMAND(devid); + WRITE_BASE(AAUDIO_MSG_GET_INPUT_STREAM_LIST); + WRITE_END(); +} + +void aaudio_msg_write_get_output_stream_list(struct aaudio_msg *msg, aaudio_device_id_t devid) +{ + WRITE_START_COMMAND(devid); + WRITE_BASE(AAUDIO_MSG_GET_OUTPUT_STREAM_LIST); + WRITE_END(); +} + +void aaudio_msg_write_set_remote_access(struct aaudio_msg *msg, u64 mode) +{ + WRITE_START_COMMAND(0); + WRITE_BASE(AAUDIO_MSG_SET_REMOTE_ACCESS); + WRITE_VAL(u64, mode); + WRITE_END(); +} + +void aaudio_msg_write_alive_notification(struct aaudio_msg *msg, u32 proto_ver, u32 msg_ver) +{ + WRITE_START_NOTIFICATION(); + WRITE_BASE(AAUDIO_MSG_NOTIFICATION_ALIVE); + WRITE_VAL(u32, proto_ver); + WRITE_VAL(u32, msg_ver); + WRITE_END(); +} + +void aaudio_msg_write_update_timestamp_response(struct aaudio_msg *msg) +{ + WRITE_START_RESPONSE(); + WRITE_BASE(AAUDIO_MSG_UPDATE_TIMESTAMP_RESPONSE); + WRITE_END(); +} + +void aaudio_msg_write_get_device_list(struct aaudio_msg *msg) +{ + WRITE_START_COMMAND(0); + WRITE_BASE(AAUDIO_MSG_GET_DEVICE_LIST); + WRITE_END(); +} + +#define CMD_SHARED_VARS_NO_REPLY \ + int status = 0; \ + struct aaudio_send_ctx sctx; +#define CMD_SHARED_VARS \ + CMD_SHARED_VARS_NO_REPLY \ + struct aaudio_msg reply = aaudio_reply_alloc(); \ + struct aaudio_msg *buf = &reply; +#define CMD_SEND_REQUEST(fn, ...) \ + if ((status = aaudio_send_cmd_sync(a, &sctx, buf, 500, fn, ##__VA_ARGS__))) \ + return status; +#define CMD_DEF_SHARED_AND_SEND(fn, ...) \ + CMD_SHARED_VARS \ + CMD_SEND_REQUEST(fn, ##__VA_ARGS__); +#define CMD_DEF_SHARED_NO_REPLY_AND_SEND(fn, ...) \ + CMD_SHARED_VARS_NO_REPLY \ + CMD_SEND_REQUEST(fn, ##__VA_ARGS__); +#define CMD_HNDL_REPLY_NO_FREE(fn, ...) \ + status = fn(buf, ##__VA_ARGS__); \ + return status; +#define CMD_HNDL_REPLY_AND_FREE(fn, ...) \ + status = fn(buf, ##__VA_ARGS__); \ + aaudio_reply_free(&reply); \ + return status; + +int aaudio_cmd_start_io(struct aaudio_device *a, aaudio_device_id_t devid) +{ + CMD_DEF_SHARED_AND_SEND(aaudio_msg_write_start_io, devid); + CMD_HNDL_REPLY_AND_FREE(aaudio_msg_read_start_io_response); +} +int aaudio_cmd_stop_io(struct aaudio_device *a, aaudio_device_id_t devid) +{ + CMD_DEF_SHARED_AND_SEND(aaudio_msg_write_stop_io, devid); + CMD_HNDL_REPLY_AND_FREE(aaudio_msg_read_stop_io_response); +} +int aaudio_cmd_get_property(struct aaudio_device *a, struct aaudio_msg *buf, + aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size, void **data, u64 *data_size) +{ + CMD_DEF_SHARED_NO_REPLY_AND_SEND(aaudio_msg_write_get_property, devid, obj, prop, qualifier, qualifier_size); + CMD_HNDL_REPLY_NO_FREE(aaudio_msg_read_get_property_response, &obj, &prop, data, data_size); +} +int aaudio_cmd_get_primitive_property(struct aaudio_device *a, + aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size, void *data, u64 data_size) +{ + int status; + struct aaudio_msg reply = aaudio_reply_alloc(); + void *r_data; + u64 r_data_size; + if ((status = aaudio_cmd_get_property(a, &reply, devid, obj, prop, qualifier, qualifier_size, + &r_data, &r_data_size))) + goto finish; + if (r_data_size != data_size) { + status = -EINVAL; + goto finish; + } + memcpy(data, r_data, data_size); +finish: + aaudio_reply_free(&reply); + return status; +} +int aaudio_cmd_set_property(struct aaudio_device *a, aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size, void *data, u64 data_size) +{ + CMD_DEF_SHARED_AND_SEND(aaudio_msg_write_set_property, devid, obj, prop, data, data_size, + qualifier, qualifier_size); + CMD_HNDL_REPLY_AND_FREE(aaudio_msg_read_set_property_response, &obj); +} +int aaudio_cmd_property_listener(struct aaudio_device *a, aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop) +{ + CMD_DEF_SHARED_AND_SEND(aaudio_msg_write_property_listener, devid, obj, prop); + CMD_HNDL_REPLY_AND_FREE(aaudio_msg_read_property_listener_response, &obj, &prop); +} +int aaudio_cmd_set_input_stream_address_ranges(struct aaudio_device *a, aaudio_device_id_t devid) +{ + CMD_DEF_SHARED_AND_SEND(aaudio_msg_write_set_input_stream_address_ranges, devid); + CMD_HNDL_REPLY_AND_FREE(aaudio_msg_read_set_input_stream_address_ranges_response); +} +int aaudio_cmd_get_input_stream_list(struct aaudio_device *a, struct aaudio_msg *buf, aaudio_device_id_t devid, + aaudio_object_id_t **str_l, u64 *str_cnt) +{ + CMD_DEF_SHARED_NO_REPLY_AND_SEND(aaudio_msg_write_get_input_stream_list, devid); + CMD_HNDL_REPLY_NO_FREE(aaudio_msg_read_get_input_stream_list_response, str_l, str_cnt); +} +int aaudio_cmd_get_output_stream_list(struct aaudio_device *a, struct aaudio_msg *buf, aaudio_device_id_t devid, + aaudio_object_id_t **str_l, u64 *str_cnt) +{ + CMD_DEF_SHARED_NO_REPLY_AND_SEND(aaudio_msg_write_get_output_stream_list, devid); + CMD_HNDL_REPLY_NO_FREE(aaudio_msg_read_get_output_stream_list_response, str_l, str_cnt); +} +int aaudio_cmd_set_remote_access(struct aaudio_device *a, u64 mode) +{ + CMD_DEF_SHARED_AND_SEND(aaudio_msg_write_set_remote_access, mode); + CMD_HNDL_REPLY_AND_FREE(aaudio_msg_read_set_remote_access_response); +} +int aaudio_cmd_get_device_list(struct aaudio_device *a, struct aaudio_msg *buf, + aaudio_device_id_t **dev_l, u64 *dev_cnt) +{ + CMD_DEF_SHARED_NO_REPLY_AND_SEND(aaudio_msg_write_get_device_list); + CMD_HNDL_REPLY_NO_FREE(aaudio_msg_read_get_device_list_response, dev_l, dev_cnt); +} \ No newline at end of file diff --git a/drivers/staging/apple-bce/audio/protocol.h b/drivers/staging/apple-bce/audio/protocol.h new file mode 100644 index 000000000000..3427486f3f57 --- /dev/null +++ b/drivers/staging/apple-bce/audio/protocol.h @@ -0,0 +1,147 @@ +#ifndef AAUDIO_PROTOCOL_H +#define AAUDIO_PROTOCOL_H + +#include + +struct aaudio_device; + +typedef u64 aaudio_device_id_t; +typedef u64 aaudio_object_id_t; + +struct aaudio_msg { + void *data; + size_t size; +}; + +struct __attribute__((packed)) aaudio_msg_header { + char tag[4]; + u8 type; + aaudio_device_id_t device_id; // Idk, use zero for commands? +}; +struct __attribute__((packed)) aaudio_msg_base { + u32 msg; + u32 status; +}; + +struct aaudio_prop_addr { + u32 scope; + u32 selector; + u32 element; +}; +#define AAUDIO_PROP(scope, sel, el) (struct aaudio_prop_addr) { scope, sel, el } + +enum { + AAUDIO_MSG_TYPE_COMMAND = 1, + AAUDIO_MSG_TYPE_RESPONSE = 2, + AAUDIO_MSG_TYPE_NOTIFICATION = 3 +}; + +enum { + AAUDIO_MSG_START_IO = 0, + AAUDIO_MSG_START_IO_RESPONSE = 1, + AAUDIO_MSG_STOP_IO = 2, + AAUDIO_MSG_STOP_IO_RESPONSE = 3, + AAUDIO_MSG_UPDATE_TIMESTAMP = 4, + AAUDIO_MSG_GET_PROPERTY = 7, + AAUDIO_MSG_GET_PROPERTY_RESPONSE = 8, + AAUDIO_MSG_SET_PROPERTY = 9, + AAUDIO_MSG_SET_PROPERTY_RESPONSE = 10, + AAUDIO_MSG_PROPERTY_LISTENER = 11, + AAUDIO_MSG_PROPERTY_LISTENER_RESPONSE = 12, + AAUDIO_MSG_PROPERTY_CHANGED = 13, + AAUDIO_MSG_SET_INPUT_STREAM_ADDRESS_RANGES = 18, + AAUDIO_MSG_SET_INPUT_STREAM_ADDRESS_RANGES_RESPONSE = 19, + AAUDIO_MSG_GET_INPUT_STREAM_LIST = 24, + AAUDIO_MSG_GET_INPUT_STREAM_LIST_RESPONSE = 25, + AAUDIO_MSG_GET_OUTPUT_STREAM_LIST = 26, + AAUDIO_MSG_GET_OUTPUT_STREAM_LIST_RESPONSE = 27, + AAUDIO_MSG_SET_REMOTE_ACCESS = 32, + AAUDIO_MSG_SET_REMOTE_ACCESS_RESPONSE = 33, + AAUDIO_MSG_UPDATE_TIMESTAMP_RESPONSE = 34, + + AAUDIO_MSG_NOTIFICATION_ALIVE = 100, + AAUDIO_MSG_GET_DEVICE_LIST = 101, + AAUDIO_MSG_GET_DEVICE_LIST_RESPONSE = 102, + AAUDIO_MSG_NOTIFICATION_BOOT = 104 +}; + +enum { + AAUDIO_REMOTE_ACCESS_OFF = 0, + AAUDIO_REMOTE_ACCESS_ON = 2 +}; + +enum { + AAUDIO_PROP_SCOPE_GLOBAL = 0x676c6f62, // 'glob' + AAUDIO_PROP_SCOPE_INPUT = 0x696e7074, // 'inpt' + AAUDIO_PROP_SCOPE_OUTPUT = 0x6f757470 // 'outp' +}; + +enum { + AAUDIO_PROP_UID = 0x75696420, // 'uid ' + AAUDIO_PROP_BOOL_VALUE = 0x6263766c, // 'bcvl' + AAUDIO_PROP_JACK_PLUGGED = 0x6a61636b, // 'jack' + AAUDIO_PROP_SEL_VOLUME = 0x64656176, // 'deav' + AAUDIO_PROP_LATENCY = 0x6c746e63, // 'ltnc' + AAUDIO_PROP_PHYS_FORMAT = 0x70667420 // 'pft ' +}; + +int aaudio_msg_read_base(struct aaudio_msg *msg, struct aaudio_msg_base *base); + +int aaudio_msg_read_start_io_response(struct aaudio_msg *msg); +int aaudio_msg_read_stop_io_response(struct aaudio_msg *msg); +int aaudio_msg_read_update_timestamp(struct aaudio_msg *msg, aaudio_device_id_t *devid, + u64 *timestamp, u64 *update_seed); +int aaudio_msg_read_get_property_response(struct aaudio_msg *msg, aaudio_object_id_t *obj, + struct aaudio_prop_addr *prop, void **data, u64 *data_size); +int aaudio_msg_read_set_property_response(struct aaudio_msg *msg, aaudio_object_id_t *obj); +int aaudio_msg_read_property_listener_response(struct aaudio_msg *msg,aaudio_object_id_t *obj, + struct aaudio_prop_addr *prop); +int aaudio_msg_read_property_changed(struct aaudio_msg *msg, aaudio_device_id_t *devid, aaudio_object_id_t *obj, + struct aaudio_prop_addr *prop); +int aaudio_msg_read_set_input_stream_address_ranges_response(struct aaudio_msg *msg); +int aaudio_msg_read_get_input_stream_list_response(struct aaudio_msg *msg, aaudio_object_id_t **str_l, u64 *str_cnt); +int aaudio_msg_read_get_output_stream_list_response(struct aaudio_msg *msg, aaudio_object_id_t **str_l, u64 *str_cnt); +int aaudio_msg_read_set_remote_access_response(struct aaudio_msg *msg); +int aaudio_msg_read_get_device_list_response(struct aaudio_msg *msg, aaudio_device_id_t **dev_l, u64 *dev_cnt); + +void aaudio_msg_write_start_io(struct aaudio_msg *msg, aaudio_device_id_t dev); +void aaudio_msg_write_stop_io(struct aaudio_msg *msg, aaudio_device_id_t dev); +void aaudio_msg_write_get_property(struct aaudio_msg *msg, aaudio_device_id_t dev, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size); +void aaudio_msg_write_set_property(struct aaudio_msg *msg, aaudio_device_id_t dev, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *data, u64 data_size, void *qualifier, u64 qualifier_size); +void aaudio_msg_write_property_listener(struct aaudio_msg *msg, aaudio_device_id_t dev, aaudio_object_id_t obj, + struct aaudio_prop_addr prop); +void aaudio_msg_write_set_input_stream_address_ranges(struct aaudio_msg *msg, aaudio_device_id_t devid); +void aaudio_msg_write_get_input_stream_list(struct aaudio_msg *msg, aaudio_device_id_t devid); +void aaudio_msg_write_get_output_stream_list(struct aaudio_msg *msg, aaudio_device_id_t devid); +void aaudio_msg_write_set_remote_access(struct aaudio_msg *msg, u64 mode); +void aaudio_msg_write_alive_notification(struct aaudio_msg *msg, u32 proto_ver, u32 msg_ver); +void aaudio_msg_write_update_timestamp_response(struct aaudio_msg *msg); +void aaudio_msg_write_get_device_list(struct aaudio_msg *msg); + + +int aaudio_cmd_start_io(struct aaudio_device *a, aaudio_device_id_t devid); +int aaudio_cmd_stop_io(struct aaudio_device *a, aaudio_device_id_t devid); +int aaudio_cmd_get_property(struct aaudio_device *a, struct aaudio_msg *buf, + aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size, void **data, u64 *data_size); +int aaudio_cmd_get_primitive_property(struct aaudio_device *a, + aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size, void *data, u64 data_size); +int aaudio_cmd_set_property(struct aaudio_device *a, aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop, void *qualifier, u64 qualifier_size, void *data, u64 data_size); +int aaudio_cmd_property_listener(struct aaudio_device *a, aaudio_device_id_t devid, aaudio_object_id_t obj, + struct aaudio_prop_addr prop); +int aaudio_cmd_set_input_stream_address_ranges(struct aaudio_device *a, aaudio_device_id_t devid); +int aaudio_cmd_get_input_stream_list(struct aaudio_device *a, struct aaudio_msg *buf, aaudio_device_id_t devid, + aaudio_object_id_t **str_l, u64 *str_cnt); +int aaudio_cmd_get_output_stream_list(struct aaudio_device *a, struct aaudio_msg *buf, aaudio_device_id_t devid, + aaudio_object_id_t **str_l, u64 *str_cnt); +int aaudio_cmd_set_remote_access(struct aaudio_device *a, u64 mode); +int aaudio_cmd_get_device_list(struct aaudio_device *a, struct aaudio_msg *buf, + aaudio_device_id_t **dev_l, u64 *dev_cnt); + + + +#endif //AAUDIO_PROTOCOL_H diff --git a/drivers/staging/apple-bce/audio/protocol_bce.c b/drivers/staging/apple-bce/audio/protocol_bce.c new file mode 100644 index 000000000000..28f2dfd44d67 --- /dev/null +++ b/drivers/staging/apple-bce/audio/protocol_bce.c @@ -0,0 +1,226 @@ +#include "protocol_bce.h" + +#include "audio.h" + +static void aaudio_bce_out_queue_completion(struct bce_queue_sq *sq); +static void aaudio_bce_in_queue_completion(struct bce_queue_sq *sq); +static int aaudio_bce_queue_init(struct aaudio_device *dev, struct aaudio_bce_queue *q, const char *name, int direction, + bce_sq_completion cfn); +void aaudio_bce_in_queue_submit_pending(struct aaudio_bce_queue *q, size_t count); + +int aaudio_bce_init(struct aaudio_device *dev) +{ + int status; + struct aaudio_bce *bce = &dev->bcem; + bce->cq = bce_create_cq(dev->bce, 0x80); + spin_lock_init(&bce->spinlock); + if (!bce->cq) + return -EINVAL; + if ((status = aaudio_bce_queue_init(dev, &bce->qout, "com.apple.BridgeAudio.IntelToARM", DMA_TO_DEVICE, + aaudio_bce_out_queue_completion))) { + return status; + } + if ((status = aaudio_bce_queue_init(dev, &bce->qin, "com.apple.BridgeAudio.ARMToIntel", DMA_FROM_DEVICE, + aaudio_bce_in_queue_completion))) { + return status; + } + aaudio_bce_in_queue_submit_pending(&bce->qin, bce->qin.el_count); + return 0; +} + +int aaudio_bce_queue_init(struct aaudio_device *dev, struct aaudio_bce_queue *q, const char *name, int direction, + bce_sq_completion cfn) +{ + q->cq = dev->bcem.cq; + q->el_size = AAUDIO_BCE_QUEUE_ELEMENT_SIZE; + q->el_count = AAUDIO_BCE_QUEUE_ELEMENT_COUNT; + /* NOTE: The Apple impl uses 0x80 as the queue size, however we use 21 (in fact 20) to simplify the impl */ + q->sq = bce_create_sq(dev->bce, q->cq, name, (u32) (q->el_count + 1), direction, cfn, dev); + if (!q->sq) + return -EINVAL; + + q->data = dma_alloc_coherent(&dev->bce->pci->dev, q->el_size * q->el_count, &q->dma_addr, GFP_KERNEL); + if (!q->data) { + bce_destroy_sq(dev->bce, q->sq); + return -EINVAL; + } + return 0; +} + +static void aaudio_send_create_tag(struct aaudio_bce *b, int *tagn, char tag[4]) +{ + char tag_zero[5]; + b->tag_num = (b->tag_num + 1) % AAUDIO_BCE_QUEUE_TAG_COUNT; + *tagn = b->tag_num; + snprintf(tag_zero, 5, "S%03d", b->tag_num); + *((u32 *) tag) = *((u32 *) tag_zero); +} + +int __aaudio_send_prepare(struct aaudio_bce *b, struct aaudio_send_ctx *ctx, char *tag) +{ + int status; + size_t index; + void *dptr; + struct aaudio_msg_header *header; + if ((status = bce_reserve_submission(b->qout.sq, &ctx->timeout))) + return status; + spin_lock_irqsave(&b->spinlock, ctx->irq_flags); + index = b->qout.data_tail; + dptr = (u8 *) b->qout.data + index * b->qout.el_size; + ctx->msg.data = dptr; + header = dptr; + if (tag) + *((u32 *) header->tag) = *((u32 *) tag); + else + aaudio_send_create_tag(b, &ctx->tag_n, header->tag); + return 0; +} + +void __aaudio_send(struct aaudio_bce *b, struct aaudio_send_ctx *ctx) +{ + struct bce_qe_submission *s = bce_next_submission(b->qout.sq); +#ifdef DEBUG + pr_debug("aaudio: Sending command data\n"); + print_hex_dump(KERN_DEBUG, "aaudio:OUT ", DUMP_PREFIX_NONE, 32, 1, ctx->msg.data, ctx->msg.size, true); +#endif + bce_set_submission_single(s, b->qout.dma_addr + (dma_addr_t) (ctx->msg.data - b->qout.data), ctx->msg.size); + bce_submit_to_device(b->qout.sq); + b->qout.data_tail = (b->qout.data_tail + 1) % b->qout.el_count; + spin_unlock_irqrestore(&b->spinlock, ctx->irq_flags); +} + +int __aaudio_send_cmd_sync(struct aaudio_bce *b, struct aaudio_send_ctx *ctx, struct aaudio_msg *reply) +{ + struct aaudio_bce_queue_entry ent; + DECLARE_COMPLETION_ONSTACK(cmpl); + ent.msg = reply; + ent.cmpl = &cmpl; + b->pending_entries[ctx->tag_n] = &ent; + __aaudio_send(b, ctx); /* unlocks the spinlock */ + ctx->timeout = wait_for_completion_timeout(&cmpl, ctx->timeout); + if (ctx->timeout == 0) { + /* Remove the pending queue entry; this will be normally handled by the completion route but + * during a timeout it won't */ + spin_lock_irqsave(&b->spinlock, ctx->irq_flags); + if (b->pending_entries[ctx->tag_n] == &ent) + b->pending_entries[ctx->tag_n] = NULL; + spin_unlock_irqrestore(&b->spinlock, ctx->irq_flags); + return -ETIMEDOUT; + } + return 0; +} + +static void aaudio_handle_reply(struct aaudio_bce *b, struct aaudio_msg *reply) +{ + const char *tag; + int tagn; + unsigned long irq_flags; + char tag_zero[5]; + struct aaudio_bce_queue_entry *entry; + + tag = ((struct aaudio_msg_header *) reply->data)->tag; + if (tag[0] != 'S') { + pr_err("aaudio_handle_reply: Unexpected tag: %.4s\n", tag); + return; + } + *((u32 *) tag_zero) = *((u32 *) tag); + tag_zero[4] = 0; + if (kstrtoint(&tag_zero[1], 10, &tagn)) { + pr_err("aaudio_handle_reply: Tag parse failed: %.4s\n", tag); + return; + } + + spin_lock_irqsave(&b->spinlock, irq_flags); + entry = b->pending_entries[tagn]; + if (entry) { + if (reply->size < entry->msg->size) + entry->msg->size = reply->size; + memcpy(entry->msg->data, reply->data, entry->msg->size); + complete(entry->cmpl); + + b->pending_entries[tagn] = NULL; + } else { + pr_err("aaudio_handle_reply: No queued item found for tag: %.4s\n", tag); + } + spin_unlock_irqrestore(&b->spinlock, irq_flags); +} + +static void aaudio_bce_out_queue_completion(struct bce_queue_sq *sq) +{ + while (bce_next_completion(sq)) { + //pr_info("aaudio: Send confirmed\n"); + bce_notify_submission_complete(sq); + } +} + +static void aaudio_bce_in_queue_handle_msg(struct aaudio_device *a, struct aaudio_msg *msg); + +static void aaudio_bce_in_queue_completion(struct bce_queue_sq *sq) +{ + struct aaudio_msg msg; + struct aaudio_device *dev = sq->userdata; + struct aaudio_bce_queue *q = &dev->bcem.qin; + struct bce_sq_completion_data *c; + size_t cnt = 0; + + mb(); + while ((c = bce_next_completion(sq))) { + msg.data = (u8 *) q->data + q->data_head * q->el_size; + msg.size = c->data_size; +#ifdef DEBUG + pr_debug("aaudio: Received command data %llx\n", c->data_size); + print_hex_dump(KERN_DEBUG, "aaudio:IN ", DUMP_PREFIX_NONE, 32, 1, msg.data, min(msg.size, 128UL), true); +#endif + aaudio_bce_in_queue_handle_msg(dev, &msg); + + q->data_head = (q->data_head + 1) % q->el_count; + + bce_notify_submission_complete(sq); + ++cnt; + } + aaudio_bce_in_queue_submit_pending(q, cnt); +} + +static void aaudio_bce_in_queue_handle_msg(struct aaudio_device *a, struct aaudio_msg *msg) +{ + struct aaudio_msg_header *header = (struct aaudio_msg_header *) msg->data; + if (msg->size < sizeof(struct aaudio_msg_header)) { + pr_err("aaudio: Msg size smaller than header (%lx)", msg->size); + return; + } + if (header->type == AAUDIO_MSG_TYPE_RESPONSE) { + aaudio_handle_reply(&a->bcem, msg); + } else if (header->type == AAUDIO_MSG_TYPE_COMMAND) { + aaudio_handle_command(a, msg); + } else if (header->type == AAUDIO_MSG_TYPE_NOTIFICATION) { + aaudio_handle_notification(a, msg); + } +} + +void aaudio_bce_in_queue_submit_pending(struct aaudio_bce_queue *q, size_t count) +{ + struct bce_qe_submission *s; + while (count--) { + if (bce_reserve_submission(q->sq, NULL)) { + pr_err("aaudio: Failed to reserve an event queue submission\n"); + break; + } + s = bce_next_submission(q->sq); + bce_set_submission_single(s, q->dma_addr + (dma_addr_t) (q->data_tail * q->el_size), q->el_size); + q->data_tail = (q->data_tail + 1) % q->el_count; + } + bce_submit_to_device(q->sq); +} + +struct aaudio_msg aaudio_reply_alloc(void) +{ + struct aaudio_msg ret; + ret.size = AAUDIO_BCE_QUEUE_ELEMENT_SIZE; + ret.data = kmalloc(ret.size, GFP_KERNEL); + return ret; +} + +void aaudio_reply_free(struct aaudio_msg *reply) +{ + kfree(reply->data); +} diff --git a/drivers/staging/apple-bce/audio/protocol_bce.h b/drivers/staging/apple-bce/audio/protocol_bce.h new file mode 100644 index 000000000000..14d26c05ddf9 --- /dev/null +++ b/drivers/staging/apple-bce/audio/protocol_bce.h @@ -0,0 +1,72 @@ +#ifndef AAUDIO_PROTOCOL_BCE_H +#define AAUDIO_PROTOCOL_BCE_H + +#include "protocol.h" +#include "../queue.h" + +#define AAUDIO_BCE_QUEUE_ELEMENT_SIZE 0x1000 +#define AAUDIO_BCE_QUEUE_ELEMENT_COUNT 20 + +#define AAUDIO_BCE_QUEUE_TAG_COUNT 1000 + +struct aaudio_device; + +struct aaudio_bce_queue_entry { + struct aaudio_msg *msg; + struct completion *cmpl; +}; +struct aaudio_bce_queue { + struct bce_queue_cq *cq; + struct bce_queue_sq *sq; + void *data; + dma_addr_t dma_addr; + size_t data_head, data_tail; + size_t el_size, el_count; +}; +struct aaudio_bce { + struct bce_queue_cq *cq; + struct aaudio_bce_queue qin; + struct aaudio_bce_queue qout; + int tag_num; + struct aaudio_bce_queue_entry *pending_entries[AAUDIO_BCE_QUEUE_TAG_COUNT]; + struct spinlock spinlock; +}; + +struct aaudio_send_ctx { + int status; + int tag_n; + unsigned long irq_flags; + struct aaudio_msg msg; + unsigned long timeout; +}; + +int aaudio_bce_init(struct aaudio_device *dev); +int __aaudio_send_prepare(struct aaudio_bce *b, struct aaudio_send_ctx *ctx, char *tag); +void __aaudio_send(struct aaudio_bce *b, struct aaudio_send_ctx *ctx); +int __aaudio_send_cmd_sync(struct aaudio_bce *b, struct aaudio_send_ctx *ctx, struct aaudio_msg *reply); + +#define aaudio_send_with_tag(a, ctx, tag, tout, fn, ...) ({ \ + (ctx)->timeout = msecs_to_jiffies(tout); \ + (ctx)->status = __aaudio_send_prepare(&(a)->bcem, (ctx), (tag)); \ + if (!(ctx)->status) { \ + fn(&(ctx)->msg, ##__VA_ARGS__); \ + __aaudio_send(&(a)->bcem, (ctx)); \ + } \ + (ctx)->status; \ +}) +#define aaudio_send(a, ctx, tout, fn, ...) aaudio_send_with_tag(a, ctx, NULL, tout, fn, ##__VA_ARGS__) + +#define aaudio_send_cmd_sync(a, ctx, reply, tout, fn, ...) ({ \ + (ctx)->timeout = msecs_to_jiffies(tout); \ + (ctx)->status = __aaudio_send_prepare(&(a)->bcem, (ctx), NULL); \ + if (!(ctx)->status) { \ + fn(&(ctx)->msg, ##__VA_ARGS__); \ + (ctx)->status = __aaudio_send_cmd_sync(&(a)->bcem, (ctx), (reply)); \ + } \ + (ctx)->status; \ +}) + +struct aaudio_msg aaudio_reply_alloc(void); +void aaudio_reply_free(struct aaudio_msg *reply); + +#endif //AAUDIO_PROTOCOL_BCE_H diff --git a/drivers/staging/apple-bce/mailbox.c b/drivers/staging/apple-bce/mailbox.c new file mode 100644 index 000000000000..07ce11928988 --- /dev/null +++ b/drivers/staging/apple-bce/mailbox.c @@ -0,0 +1,155 @@ +#include "mailbox.h" +#include +#include "apple_bce.h" +#include + +#define REG_MBOX_OUT_BASE 0x820 +#define REG_MBOX_REPLY_COUNTER 0x108 +#define REG_MBOX_REPLY_BASE 0x810 +#define REG_TIMESTAMP_BASE 0xC000 + +#define BCE_MBOX_TIMEOUT_MS 200 + +void bce_mailbox_init(struct bce_mailbox *mb, void __iomem *reg_mb) +{ + mb->reg_mb = reg_mb; + init_completion(&mb->mb_completion); +} + +int bce_mailbox_send(struct bce_mailbox *mb, u64 msg, u64* recv) +{ + u32 __iomem *regb; + + if (atomic_cmpxchg(&mb->mb_status, 0, 1) != 0) { + return -EEXIST; // We don't support two messages at once + } + reinit_completion(&mb->mb_completion); + + pr_debug("bce_mailbox_send: %llx\n", msg); + regb = (u32*) ((u8*) mb->reg_mb + REG_MBOX_OUT_BASE); + iowrite32((u32) msg, regb); + iowrite32((u32) (msg >> 32), regb + 1); + iowrite32(0, regb + 2); + iowrite32(0, regb + 3); + + wait_for_completion_timeout(&mb->mb_completion, msecs_to_jiffies(BCE_MBOX_TIMEOUT_MS)); + if (atomic_read(&mb->mb_status) != 2) { // Didn't get the reply + atomic_set(&mb->mb_status, 0); + return -ETIMEDOUT; + } + + *recv = mb->mb_result; + pr_debug("bce_mailbox_send: reply %llx\n", *recv); + + atomic_set(&mb->mb_status, 0); + return 0; +} + +static int bce_mailbox_retrive_response(struct bce_mailbox *mb) +{ + u32 __iomem *regb; + u32 lo, hi; + int count, counter; + u32 res = ioread32((u8*) mb->reg_mb + REG_MBOX_REPLY_COUNTER); + count = (res >> 20) & 0xf; + counter = count; + pr_debug("bce_mailbox_retrive_response count=%i\n", count); + while (counter--) { + regb = (u32*) ((u8*) mb->reg_mb + REG_MBOX_REPLY_BASE); + lo = ioread32(regb); + hi = ioread32(regb + 1); + ioread32(regb + 2); + ioread32(regb + 3); + pr_debug("bce_mailbox_retrive_response %llx\n", ((u64) hi << 32) | lo); + mb->mb_result = ((u64) hi << 32) | lo; + } + return count > 0 ? 0 : -ENODATA; +} + +int bce_mailbox_handle_interrupt(struct bce_mailbox *mb) +{ + int status = bce_mailbox_retrive_response(mb); + if (!status) { + atomic_set(&mb->mb_status, 2); + complete(&mb->mb_completion); + } + return status; +} + +static void bc_send_timestamp(struct timer_list *tl); + +void bce_timestamp_init(struct bce_timestamp *ts, void __iomem *reg) +{ + u32 __iomem *regb; + + spin_lock_init(&ts->stop_sl); + ts->stopped = false; + + ts->reg = reg; + + regb = (u32*) ((u8*) ts->reg + REG_TIMESTAMP_BASE); + + ioread32(regb); + mb(); + + timer_setup(&ts->timer, bc_send_timestamp, 0); +} + +void bce_timestamp_start(struct bce_timestamp *ts, bool is_initial) +{ + unsigned long flags; + u32 __iomem *regb = (u32*) ((u8*) ts->reg + REG_TIMESTAMP_BASE); + + if (is_initial) { + iowrite32((u32) -4, regb + 2); + iowrite32((u32) -1, regb); + } else { + iowrite32((u32) -3, regb + 2); + iowrite32((u32) -1, regb); + } + + spin_lock_irqsave(&ts->stop_sl, flags); + ts->stopped = false; + spin_unlock_irqrestore(&ts->stop_sl, flags); + mod_timer(&ts->timer, jiffies + msecs_to_jiffies(150)); +} + +void bce_timestamp_stop(struct bce_timestamp *ts) +{ + unsigned long flags; + u32 __iomem *regb = (u32*) ((u8*) ts->reg + REG_TIMESTAMP_BASE); + + spin_lock_irqsave(&ts->stop_sl, flags); + ts->stopped = true; + spin_unlock_irqrestore(&ts->stop_sl, flags); +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,15,0) + del_timer_sync(&ts->timer); +#else + timer_delete_sync(&ts->timer); +#endif + iowrite32((u32) -2, regb + 2); + iowrite32((u32) -1, regb); +} + +static void bc_send_timestamp(struct timer_list *tl) +{ + struct bce_timestamp *ts; + unsigned long flags; + u32 __iomem *regb; + ktime_t bt; + + ts = container_of(tl, struct bce_timestamp, timer); + regb = (u32*) ((u8*) ts->reg + REG_TIMESTAMP_BASE); + local_irq_save(flags); + ioread32(regb + 2); + mb(); + bt = ktime_get_boottime(); + iowrite32((u32) bt, regb + 2); + iowrite32((u32) (bt >> 32), regb); + + spin_lock(&ts->stop_sl); + if (!ts->stopped) + mod_timer(&ts->timer, jiffies + msecs_to_jiffies(150)); + spin_unlock(&ts->stop_sl); + local_irq_restore(flags); +} diff --git a/drivers/staging/apple-bce/mailbox.h b/drivers/staging/apple-bce/mailbox.h new file mode 100644 index 000000000000..f3323f95ba51 --- /dev/null +++ b/drivers/staging/apple-bce/mailbox.h @@ -0,0 +1,53 @@ +#ifndef BCE_MAILBOX_H +#define BCE_MAILBOX_H + +#include +#include +#include + +struct bce_mailbox { + void __iomem *reg_mb; + + atomic_t mb_status; // possible statuses: 0 (no msg), 1 (has active msg), 2 (got reply) + struct completion mb_completion; + uint64_t mb_result; +}; + +enum bce_message_type { + BCE_MB_REGISTER_COMMAND_SQ = 0x7, // to-device + BCE_MB_REGISTER_COMMAND_CQ = 0x8, // to-device + BCE_MB_REGISTER_COMMAND_QUEUE_REPLY = 0xB, // to-host + BCE_MB_SET_FW_PROTOCOL_VERSION = 0xC, // both + BCE_MB_SLEEP_NO_STATE = 0x14, // to-device + BCE_MB_RESTORE_NO_STATE = 0x15, // to-device + BCE_MB_SAVE_STATE_AND_SLEEP = 0x17, // to-device + BCE_MB_RESTORE_STATE_AND_WAKE = 0x18, // to-device + BCE_MB_SAVE_STATE_AND_SLEEP_FAILURE = 0x19, // from-device + BCE_MB_SAVE_RESTORE_STATE_COMPLETE = 0x1A, // from-device +}; + +#define BCE_MB_MSG(type, value) (((u64) (type) << 58) | ((value) & 0x3FFFFFFFFFFFFFFLL)) +#define BCE_MB_TYPE(v) ((u32) (v >> 58)) +#define BCE_MB_VALUE(v) (v & 0x3FFFFFFFFFFFFFFLL) + +void bce_mailbox_init(struct bce_mailbox *mb, void __iomem *reg_mb); + +int bce_mailbox_send(struct bce_mailbox *mb, u64 msg, u64* recv); + +int bce_mailbox_handle_interrupt(struct bce_mailbox *mb); + + +struct bce_timestamp { + void __iomem *reg; + struct timer_list timer; + struct spinlock stop_sl; + bool stopped; +}; + +void bce_timestamp_init(struct bce_timestamp *ts, void __iomem *reg); + +void bce_timestamp_start(struct bce_timestamp *ts, bool is_initial); + +void bce_timestamp_stop(struct bce_timestamp *ts); + +#endif //BCEDRIVER_MAILBOX_H diff --git a/drivers/staging/apple-bce/queue.c b/drivers/staging/apple-bce/queue.c new file mode 100644 index 000000000000..948286157e3a --- /dev/null +++ b/drivers/staging/apple-bce/queue.c @@ -0,0 +1,415 @@ +#include "queue.h" +#include "apple_bce.h" +#include + +#define REG_DOORBELL_BASE 0x44000 + +struct bce_queue_cq *bce_alloc_cq(struct apple_bce_device *dev, int qid, u32 el_count) +{ + struct bce_queue_cq *q; + q = kzalloc(sizeof(struct bce_queue_cq), GFP_KERNEL); + q->qid = qid; + q->type = BCE_QUEUE_CQ; + q->el_count = el_count; + q->data = dma_alloc_coherent(&dev->pci->dev, el_count * sizeof(struct bce_qe_completion), + &q->dma_handle, GFP_KERNEL); + if (!q->data) { + pr_err("DMA queue memory alloc failed\n"); + kfree(q); + return NULL; + } + return q; +} + +void bce_get_cq_memcfg(struct bce_queue_cq *cq, struct bce_queue_memcfg *cfg) +{ + cfg->qid = (u16) cq->qid; + cfg->el_count = (u16) cq->el_count; + cfg->vector_or_cq = 0; + cfg->_pad = 0; + cfg->addr = cq->dma_handle; + cfg->length = cq->el_count * sizeof(struct bce_qe_completion); +} + +void bce_free_cq(struct apple_bce_device *dev, struct bce_queue_cq *cq) +{ + dma_free_coherent(&dev->pci->dev, cq->el_count * sizeof(struct bce_qe_completion), cq->data, cq->dma_handle); + kfree(cq); +} + +static void bce_handle_cq_completion(struct apple_bce_device *dev, struct bce_qe_completion *e, size_t *ce) +{ + struct bce_queue *target; + struct bce_queue_sq *target_sq; + struct bce_sq_completion_data *cmpl; + if (e->qid >= BCE_MAX_QUEUE_COUNT) { + pr_err("Device sent a response for qid (%u) >= BCE_MAX_QUEUE_COUNT\n", e->qid); + return; + } + target = dev->queues[e->qid]; + if (!target || target->type != BCE_QUEUE_SQ) { + pr_err("Device sent a response for qid (%u), which does not exist\n", e->qid); + return; + } + target_sq = (struct bce_queue_sq *) target; + if (target_sq->completion_tail != e->completion_index) { + pr_err("Completion index mismatch; this is likely going to make this driver unusable\n"); + return; + } + if (!target_sq->has_pending_completions) { + target_sq->has_pending_completions = true; + dev->int_sq_list[(*ce)++] = target_sq; + } + cmpl = &target_sq->completion_data[e->completion_index]; + cmpl->status = e->status; + cmpl->data_size = e->data_size; + cmpl->result = e->result; + wmb(); + target_sq->completion_tail = (target_sq->completion_tail + 1) % target_sq->el_count; +} + +void bce_handle_cq_completions(struct apple_bce_device *dev, struct bce_queue_cq *cq) +{ + size_t ce = 0; + struct bce_qe_completion *e; + struct bce_queue_sq *sq; + e = bce_cq_element(cq, cq->index); + if (!(e->flags & BCE_COMPLETION_FLAG_PENDING)) + return; + mb(); + while (true) { + e = bce_cq_element(cq, cq->index); + if (!(e->flags & BCE_COMPLETION_FLAG_PENDING)) + break; + // pr_info("apple-bce: compl: %i: %i %llx %llx", e->qid, e->status, e->data_size, e->result); + bce_handle_cq_completion(dev, e, &ce); + e->flags = 0; + cq->index = (cq->index + 1) % cq->el_count; + } + mb(); + iowrite32(cq->index, (u32 *) ((u8 *) dev->reg_mem_dma + REG_DOORBELL_BASE) + cq->qid); + while (ce) { + --ce; + sq = dev->int_sq_list[ce]; + sq->completion(sq); + sq->has_pending_completions = false; + } +} + + +struct bce_queue_sq *bce_alloc_sq(struct apple_bce_device *dev, int qid, u32 el_size, u32 el_count, + bce_sq_completion compl, void *userdata) +{ + struct bce_queue_sq *q; + q = kzalloc(sizeof(struct bce_queue_sq), GFP_KERNEL); + q->qid = qid; + q->type = BCE_QUEUE_SQ; + q->el_size = el_size; + q->el_count = el_count; + q->data = dma_alloc_coherent(&dev->pci->dev, el_count * el_size, + &q->dma_handle, GFP_KERNEL); + q->completion = compl; + q->userdata = userdata; + q->completion_data = kzalloc(sizeof(struct bce_sq_completion_data) * el_count, GFP_KERNEL); + q->reg_mem_dma = dev->reg_mem_dma; + atomic_set(&q->available_commands, el_count - 1); + init_completion(&q->available_command_completion); + atomic_set(&q->available_command_completion_waiting_count, 0); + if (!q->data) { + pr_err("DMA queue memory alloc failed\n"); + kfree(q); + return NULL; + } + return q; +} + +void bce_get_sq_memcfg(struct bce_queue_sq *sq, struct bce_queue_cq *cq, struct bce_queue_memcfg *cfg) +{ + cfg->qid = (u16) sq->qid; + cfg->el_count = (u16) sq->el_count; + cfg->vector_or_cq = (u16) cq->qid; + cfg->_pad = 0; + cfg->addr = sq->dma_handle; + cfg->length = sq->el_count * sq->el_size; +} + +void bce_free_sq(struct apple_bce_device *dev, struct bce_queue_sq *sq) +{ + dma_free_coherent(&dev->pci->dev, sq->el_count * sq->el_size, sq->data, sq->dma_handle); + kfree(sq); +} + +int bce_reserve_submission(struct bce_queue_sq *sq, unsigned long *timeout) +{ + while (atomic_dec_if_positive(&sq->available_commands) < 0) { + if (!timeout || !*timeout) + return -EAGAIN; + atomic_inc(&sq->available_command_completion_waiting_count); + *timeout = wait_for_completion_timeout(&sq->available_command_completion, *timeout); + if (!*timeout) { + if (atomic_dec_if_positive(&sq->available_command_completion_waiting_count) < 0) + try_wait_for_completion(&sq->available_command_completion); /* consume the pending completion */ + } + } + return 0; +} + +void bce_cancel_submission_reservation(struct bce_queue_sq *sq) +{ + atomic_inc(&sq->available_commands); +} + +void *bce_next_submission(struct bce_queue_sq *sq) +{ + void *ret = bce_sq_element(sq, sq->tail); + sq->tail = (sq->tail + 1) % sq->el_count; + return ret; +} + +void bce_submit_to_device(struct bce_queue_sq *sq) +{ + mb(); + iowrite32(sq->tail, (u32 *) ((u8 *) sq->reg_mem_dma + REG_DOORBELL_BASE) + sq->qid); +} + +void bce_notify_submission_complete(struct bce_queue_sq *sq) +{ + sq->head = (sq->head + 1) % sq->el_count; + atomic_inc(&sq->available_commands); + if (atomic_dec_if_positive(&sq->available_command_completion_waiting_count) >= 0) { + complete(&sq->available_command_completion); + } +} + +void bce_set_submission_single(struct bce_qe_submission *element, dma_addr_t addr, size_t size) +{ + element->addr = addr; + element->length = size; + element->segl_addr = element->segl_length = 0; +} + +static void bce_cmdq_completion(struct bce_queue_sq *q); + +struct bce_queue_cmdq *bce_alloc_cmdq(struct apple_bce_device *dev, int qid, u32 el_count) +{ + struct bce_queue_cmdq *q; + q = kzalloc(sizeof(struct bce_queue_cmdq), GFP_KERNEL); + q->sq = bce_alloc_sq(dev, qid, BCE_CMD_SIZE, el_count, bce_cmdq_completion, q); + if (!q->sq) { + kfree(q); + return NULL; + } + spin_lock_init(&q->lck); + q->tres = kzalloc(sizeof(struct bce_queue_cmdq_result_el*) * el_count, GFP_KERNEL); + if (!q->tres) { + kfree(q); + return NULL; + } + return q; +} + +void bce_free_cmdq(struct apple_bce_device *dev, struct bce_queue_cmdq *cmdq) +{ + bce_free_sq(dev, cmdq->sq); + kfree(cmdq->tres); + kfree(cmdq); +} + +void bce_cmdq_completion(struct bce_queue_sq *q) +{ + struct bce_queue_cmdq_result_el *el; + struct bce_queue_cmdq *cmdq = q->userdata; + struct bce_sq_completion_data *result; + + spin_lock(&cmdq->lck); + while ((result = bce_next_completion(q))) { + el = cmdq->tres[cmdq->sq->head]; + if (el) { + el->result = result->result; + el->status = result->status; + mb(); + complete(&el->cmpl); + } else { + pr_err("apple-bce: Unexpected command queue completion\n"); + } + cmdq->tres[cmdq->sq->head] = NULL; + bce_notify_submission_complete(q); + } + spin_unlock(&cmdq->lck); +} + +static __always_inline void *bce_cmd_start(struct bce_queue_cmdq *cmdq, struct bce_queue_cmdq_result_el *res) +{ + void *ret; + unsigned long timeout; + init_completion(&res->cmpl); + mb(); + + timeout = msecs_to_jiffies(1000L * 60 * 5); /* wait for up to ~5 minutes */ + if (bce_reserve_submission(cmdq->sq, &timeout)) + return NULL; + + spin_lock(&cmdq->lck); + cmdq->tres[cmdq->sq->tail] = res; + ret = bce_next_submission(cmdq->sq); + return ret; +} + +static __always_inline void bce_cmd_finish(struct bce_queue_cmdq *cmdq, struct bce_queue_cmdq_result_el *res) +{ + bce_submit_to_device(cmdq->sq); + spin_unlock(&cmdq->lck); + + wait_for_completion(&res->cmpl); + mb(); +} + +u32 bce_cmd_register_queue(struct bce_queue_cmdq *cmdq, struct bce_queue_memcfg *cfg, const char *name, bool isdirout) +{ + struct bce_queue_cmdq_result_el res; + struct bce_cmdq_register_memory_queue_cmd *cmd = bce_cmd_start(cmdq, &res); + if (!cmd) + return (u32) -1; + cmd->cmd = BCE_CMD_REGISTER_MEMORY_QUEUE; + cmd->flags = (u16) ((name ? 2 : 0) | (isdirout ? 1 : 0)); + cmd->qid = cfg->qid; + cmd->el_count = cfg->el_count; + cmd->vector_or_cq = cfg->vector_or_cq; + memset(cmd->name, 0, sizeof(cmd->name)); + if (name) { + cmd->name_len = (u16) min(strlen(name), (size_t) sizeof(cmd->name)); + memcpy(cmd->name, name, cmd->name_len); + } else { + cmd->name_len = 0; + } + cmd->addr = cfg->addr; + cmd->length = cfg->length; + + bce_cmd_finish(cmdq, &res); + return res.status; +} + +u32 bce_cmd_unregister_memory_queue(struct bce_queue_cmdq *cmdq, u16 qid) +{ + struct bce_queue_cmdq_result_el res; + struct bce_cmdq_simple_memory_queue_cmd *cmd = bce_cmd_start(cmdq, &res); + if (!cmd) + return (u32) -1; + cmd->cmd = BCE_CMD_UNREGISTER_MEMORY_QUEUE; + cmd->flags = 0; + cmd->qid = qid; + bce_cmd_finish(cmdq, &res); + return res.status; +} + +u32 bce_cmd_flush_memory_queue(struct bce_queue_cmdq *cmdq, u16 qid) +{ + struct bce_queue_cmdq_result_el res; + struct bce_cmdq_simple_memory_queue_cmd *cmd = bce_cmd_start(cmdq, &res); + if (!cmd) + return (u32) -1; + cmd->cmd = BCE_CMD_FLUSH_MEMORY_QUEUE; + cmd->flags = 0; + cmd->qid = qid; + bce_cmd_finish(cmdq, &res); + return res.status; +} + + +struct bce_queue_cq *bce_create_cq(struct apple_bce_device *dev, u32 el_count) +{ + struct bce_queue_cq *cq; + struct bce_queue_memcfg cfg; +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,18,0) + int qid = ida_simple_get(&dev->queue_ida, BCE_QUEUE_USER_MIN, BCE_QUEUE_USER_MAX, GFP_KERNEL); +#else + int qid = ida_alloc_range(&dev->queue_ida, BCE_QUEUE_USER_MIN, BCE_QUEUE_USER_MAX - 1, GFP_KERNEL); +#endif + if (qid < 0) + return NULL; + cq = bce_alloc_cq(dev, qid, el_count); + if (!cq) + return NULL; + bce_get_cq_memcfg(cq, &cfg); + if (bce_cmd_register_queue(dev->cmd_cmdq, &cfg, NULL, false) != 0) { + pr_err("apple-bce: CQ registration failed (%i)", qid); + bce_free_cq(dev, cq); +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,18,0) + ida_simple_remove(&dev->queue_ida, (uint) qid); +#else + ida_free(&dev->queue_ida, (uint) qid); +#endif + return NULL; + } + dev->queues[qid] = (struct bce_queue *) cq; + return cq; +} + +struct bce_queue_sq *bce_create_sq(struct apple_bce_device *dev, struct bce_queue_cq *cq, const char *name, u32 el_count, + int direction, bce_sq_completion compl, void *userdata) +{ + struct bce_queue_sq *sq; + struct bce_queue_memcfg cfg; + int qid; + if (cq == NULL) + return NULL; /* cq can not be null */ + if (name == NULL) + return NULL; /* name can not be null */ + if (direction != DMA_TO_DEVICE && direction != DMA_FROM_DEVICE) + return NULL; /* unsupported direction */ +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,18,0) + qid = ida_simple_get(&dev->queue_ida, BCE_QUEUE_USER_MIN, BCE_QUEUE_USER_MAX, GFP_KERNEL); +#else + qid = ida_alloc_range(&dev->queue_ida, BCE_QUEUE_USER_MIN, BCE_QUEUE_USER_MAX - 1, GFP_KERNEL); +#endif + if (qid < 0) + return NULL; + sq = bce_alloc_sq(dev, qid, sizeof(struct bce_qe_submission), el_count, compl, userdata); + if (!sq) + return NULL; + bce_get_sq_memcfg(sq, cq, &cfg); + if (bce_cmd_register_queue(dev->cmd_cmdq, &cfg, name, direction != DMA_FROM_DEVICE) != 0) { + pr_err("apple-bce: SQ registration failed (%i)", qid); + bce_free_sq(dev, sq); +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,18,0) + ida_simple_remove(&dev->queue_ida, (uint) qid); +#else + ida_free(&dev->queue_ida, (uint) qid); +#endif + return NULL; + } + spin_lock(&dev->queues_lock); + dev->queues[qid] = (struct bce_queue *) sq; + spin_unlock(&dev->queues_lock); + return sq; +} + +void bce_destroy_cq(struct apple_bce_device *dev, struct bce_queue_cq *cq) +{ + if (!dev->is_being_removed && bce_cmd_unregister_memory_queue(dev->cmd_cmdq, (u16) cq->qid)) + pr_err("apple-bce: CQ unregister failed"); + spin_lock(&dev->queues_lock); + dev->queues[cq->qid] = NULL; + spin_unlock(&dev->queues_lock); +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,18,0) + ida_simple_remove(&dev->queue_ida, (uint) cq->qid); +#else + ida_free(&dev->queue_ida, (uint) cq->qid); +#endif + bce_free_cq(dev, cq); +} + +void bce_destroy_sq(struct apple_bce_device *dev, struct bce_queue_sq *sq) +{ + if (!dev->is_being_removed && bce_cmd_unregister_memory_queue(dev->cmd_cmdq, (u16) sq->qid)) + pr_err("apple-bce: CQ unregister failed"); + spin_lock(&dev->queues_lock); + dev->queues[sq->qid] = NULL; + spin_unlock(&dev->queues_lock); +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,18,0) + ida_simple_remove(&dev->queue_ida, (uint) sq->qid); +#else + ida_free(&dev->queue_ida, (uint) sq->qid); +#endif + bce_free_sq(dev, sq); +} \ No newline at end of file diff --git a/drivers/staging/apple-bce/queue.h b/drivers/staging/apple-bce/queue.h new file mode 100644 index 000000000000..8368ac5dfca8 --- /dev/null +++ b/drivers/staging/apple-bce/queue.h @@ -0,0 +1,177 @@ +#ifndef BCE_QUEUE_H +#define BCE_QUEUE_H + +#include +#include + +#define BCE_CMD_SIZE 0x40 + +struct apple_bce_device; + +enum bce_queue_type { + BCE_QUEUE_CQ, BCE_QUEUE_SQ +}; +struct bce_queue { + int qid; + int type; +}; +struct bce_queue_cq { + int qid; + int type; + u32 el_count; + dma_addr_t dma_handle; + void *data; + + u32 index; +}; +struct bce_queue_sq; +typedef void (*bce_sq_completion)(struct bce_queue_sq *q); +struct bce_sq_completion_data { + u32 status; + u64 data_size; + u64 result; +}; +struct bce_queue_sq { + int qid; + int type; + u32 el_size; + u32 el_count; + dma_addr_t dma_handle; + void *data; + void *userdata; + void __iomem *reg_mem_dma; + + atomic_t available_commands; + struct completion available_command_completion; + atomic_t available_command_completion_waiting_count; + u32 head, tail; + + u32 completion_cidx, completion_tail; + struct bce_sq_completion_data *completion_data; + bool has_pending_completions; + bce_sq_completion completion; +}; + +struct bce_queue_cmdq_result_el { + struct completion cmpl; + u32 status; + u64 result; +}; +struct bce_queue_cmdq { + struct bce_queue_sq *sq; + struct spinlock lck; + struct bce_queue_cmdq_result_el **tres; +}; + +struct bce_queue_memcfg { + u16 qid; + u16 el_count; + u16 vector_or_cq; + u16 _pad; + u64 addr; + u64 length; +}; + +enum bce_qe_completion_status { + BCE_COMPLETION_SUCCESS = 0, + BCE_COMPLETION_ERROR = 1, + BCE_COMPLETION_ABORTED = 2, + BCE_COMPLETION_NO_SPACE = 3, + BCE_COMPLETION_OVERRUN = 4 +}; +enum bce_qe_completion_flags { + BCE_COMPLETION_FLAG_PENDING = 0x8000 +}; +struct bce_qe_completion { + u64 result; + u64 data_size; + u16 qid; + u16 completion_index; + u16 status; // bce_qe_completion_status + u16 flags; // bce_qe_completion_flags +}; + +struct bce_qe_submission { + u64 length; + u64 addr; + + u64 segl_addr; + u64 segl_length; +}; + +enum bce_cmdq_command { + BCE_CMD_REGISTER_MEMORY_QUEUE = 0x20, + BCE_CMD_UNREGISTER_MEMORY_QUEUE = 0x30, + BCE_CMD_FLUSH_MEMORY_QUEUE = 0x40, + BCE_CMD_SET_MEMORY_QUEUE_PROPERTY = 0x50 +}; +struct bce_cmdq_simple_memory_queue_cmd { + u16 cmd; // bce_cmdq_command + u16 flags; + u16 qid; +}; +struct bce_cmdq_register_memory_queue_cmd { + u16 cmd; // bce_cmdq_command + u16 flags; + u16 qid; + u16 _pad; + u16 el_count; + u16 vector_or_cq; + u16 _pad2; + u16 name_len; + char name[0x20]; + u64 addr; + u64 length; +}; + +static __always_inline void *bce_sq_element(struct bce_queue_sq *q, int i) { + return (void *) ((u8 *) q->data + q->el_size * i); +} +static __always_inline void *bce_cq_element(struct bce_queue_cq *q, int i) { + return (void *) ((struct bce_qe_completion *) q->data + i); +} + +static __always_inline struct bce_sq_completion_data *bce_next_completion(struct bce_queue_sq *sq) { + struct bce_sq_completion_data *res; + rmb(); + if (sq->completion_cidx == sq->completion_tail) + return NULL; + res = &sq->completion_data[sq->completion_cidx]; + sq->completion_cidx = (sq->completion_cidx + 1) % sq->el_count; + return res; +} + +struct bce_queue_cq *bce_alloc_cq(struct apple_bce_device *dev, int qid, u32 el_count); +void bce_get_cq_memcfg(struct bce_queue_cq *cq, struct bce_queue_memcfg *cfg); +void bce_free_cq(struct apple_bce_device *dev, struct bce_queue_cq *cq); +void bce_handle_cq_completions(struct apple_bce_device *dev, struct bce_queue_cq *cq); + +struct bce_queue_sq *bce_alloc_sq(struct apple_bce_device *dev, int qid, u32 el_size, u32 el_count, + bce_sq_completion compl, void *userdata); +void bce_get_sq_memcfg(struct bce_queue_sq *sq, struct bce_queue_cq *cq, struct bce_queue_memcfg *cfg); +void bce_free_sq(struct apple_bce_device *dev, struct bce_queue_sq *sq); +int bce_reserve_submission(struct bce_queue_sq *sq, unsigned long *timeout); +void bce_cancel_submission_reservation(struct bce_queue_sq *sq); +void *bce_next_submission(struct bce_queue_sq *sq); +void bce_submit_to_device(struct bce_queue_sq *sq); +void bce_notify_submission_complete(struct bce_queue_sq *sq); + +void bce_set_submission_single(struct bce_qe_submission *element, dma_addr_t addr, size_t size); + +struct bce_queue_cmdq *bce_alloc_cmdq(struct apple_bce_device *dev, int qid, u32 el_count); +void bce_free_cmdq(struct apple_bce_device *dev, struct bce_queue_cmdq *cmdq); + +u32 bce_cmd_register_queue(struct bce_queue_cmdq *cmdq, struct bce_queue_memcfg *cfg, const char *name, bool isdirout); +u32 bce_cmd_unregister_memory_queue(struct bce_queue_cmdq *cmdq, u16 qid); +u32 bce_cmd_flush_memory_queue(struct bce_queue_cmdq *cmdq, u16 qid); + + +/* User API - Creates and registers the queue */ + +struct bce_queue_cq *bce_create_cq(struct apple_bce_device *dev, u32 el_count); +struct bce_queue_sq *bce_create_sq(struct apple_bce_device *dev, struct bce_queue_cq *cq, const char *name, u32 el_count, + int direction, bce_sq_completion compl, void *userdata); +void bce_destroy_cq(struct apple_bce_device *dev, struct bce_queue_cq *cq); +void bce_destroy_sq(struct apple_bce_device *dev, struct bce_queue_sq *sq); + +#endif //BCEDRIVER_MAILBOX_H diff --git a/drivers/staging/apple-bce/queue_dma.c b/drivers/staging/apple-bce/queue_dma.c new file mode 100644 index 000000000000..b236613285c0 --- /dev/null +++ b/drivers/staging/apple-bce/queue_dma.c @@ -0,0 +1,220 @@ +#include "queue_dma.h" +#include +#include +#include "queue.h" + +static int bce_alloc_scatterlist_from_vm(struct sg_table *tbl, void *data, size_t len); +static struct bce_segment_list_element_hostinfo *bce_map_segment_list( + struct device *dev, struct scatterlist *pages, int pagen); +static void bce_unmap_segement_list(struct device *dev, struct bce_segment_list_element_hostinfo *list); + +int bce_map_dma_buffer(struct device *dev, struct bce_dma_buffer *buf, struct sg_table scatterlist, + enum dma_data_direction dir) +{ + int cnt; + + buf->direction = dir; + buf->scatterlist = scatterlist; + buf->seglist_hostinfo = NULL; + + cnt = dma_map_sg(dev, buf->scatterlist.sgl, buf->scatterlist.nents, dir); + if (cnt != buf->scatterlist.nents) { + pr_err("apple-bce: DMA scatter list mapping returned an unexpected count: %i\n", cnt); + dma_unmap_sg(dev, buf->scatterlist.sgl, buf->scatterlist.nents, dir); + return -EIO; + } + if (cnt == 1) + return 0; + + buf->seglist_hostinfo = bce_map_segment_list(dev, buf->scatterlist.sgl, buf->scatterlist.nents); + if (!buf->seglist_hostinfo) { + pr_err("apple-bce: Creating segment list failed\n"); + dma_unmap_sg(dev, buf->scatterlist.sgl, buf->scatterlist.nents, dir); + return -EIO; + } + return 0; +} + +int bce_map_dma_buffer_vm(struct device *dev, struct bce_dma_buffer *buf, void *data, size_t len, + enum dma_data_direction dir) +{ + int status; + struct sg_table scatterlist; + if ((status = bce_alloc_scatterlist_from_vm(&scatterlist, data, len))) + return status; + if ((status = bce_map_dma_buffer(dev, buf, scatterlist, dir))) { + sg_free_table(&scatterlist); + return status; + } + return 0; +} + +int bce_map_dma_buffer_km(struct device *dev, struct bce_dma_buffer *buf, void *data, size_t len, + enum dma_data_direction dir) +{ + /* Kernel memory is continuous which is great for us. */ + int status; + struct sg_table scatterlist; + if ((status = sg_alloc_table(&scatterlist, 1, GFP_KERNEL))) { + sg_free_table(&scatterlist); + return status; + } + sg_set_buf(scatterlist.sgl, data, (uint) len); + if ((status = bce_map_dma_buffer(dev, buf, scatterlist, dir))) { + sg_free_table(&scatterlist); + return status; + } + return 0; +} + +void bce_unmap_dma_buffer(struct device *dev, struct bce_dma_buffer *buf) +{ + dma_unmap_sg(dev, buf->scatterlist.sgl, buf->scatterlist.nents, buf->direction); + bce_unmap_segement_list(dev, buf->seglist_hostinfo); +} + + +static int bce_alloc_scatterlist_from_vm(struct sg_table *tbl, void *data, size_t len) +{ + int status, i; + struct page **pages; + size_t off, start_page, end_page, page_count; + off = (size_t) data % PAGE_SIZE; + start_page = (size_t) data / PAGE_SIZE; + end_page = ((size_t) data + len - 1) / PAGE_SIZE; + page_count = end_page - start_page + 1; + + if (page_count > PAGE_SIZE / sizeof(struct page *)) + pages = vmalloc(page_count * sizeof(struct page *)); + else + pages = kmalloc(page_count * sizeof(struct page *), GFP_KERNEL); + + for (i = 0; i < page_count; i++) + pages[i] = vmalloc_to_page((void *) ((start_page + i) * PAGE_SIZE)); + + if ((status = sg_alloc_table_from_pages(tbl, pages, page_count, (unsigned int) off, len, GFP_KERNEL))) { + sg_free_table(tbl); + } + + if (page_count > PAGE_SIZE / sizeof(struct page *)) + vfree(pages); + else + kfree(pages); + return status; +} + +#define BCE_ELEMENTS_PER_PAGE ((PAGE_SIZE - sizeof(struct bce_segment_list_header)) \ + / sizeof(struct bce_segment_list_element)) +#define BCE_ELEMENTS_PER_ADDITIONAL_PAGE (PAGE_SIZE / sizeof(struct bce_segment_list_element)) + +static struct bce_segment_list_element_hostinfo *bce_map_segment_list( + struct device *dev, struct scatterlist *pages, int pagen) +{ + size_t ptr, pptr = 0; + struct bce_segment_list_header theader; /* a temp header, to store the initial seg */ + struct bce_segment_list_header *header; + struct bce_segment_list_element *el, *el_end; + struct bce_segment_list_element_hostinfo *out, *pout, *out_root; + struct scatterlist *sg; + int i; + header = &theader; + out = out_root = NULL; + el = el_end = NULL; + for_each_sg(pages, sg, pagen, i) { + if (el >= el_end) { + /* allocate a new page, this will be also done for the first element */ + ptr = __get_free_page(GFP_KERNEL); + if (pptr && ptr == pptr + PAGE_SIZE) { + out->page_count++; + header->element_count += BCE_ELEMENTS_PER_ADDITIONAL_PAGE; + el_end += BCE_ELEMENTS_PER_ADDITIONAL_PAGE; + } else { + header = (void *) ptr; + header->element_count = BCE_ELEMENTS_PER_PAGE; + header->data_size = 0; + header->next_segl_addr = 0; + header->next_segl_length = 0; + el = (void *) (header + 1); + el_end = el + BCE_ELEMENTS_PER_PAGE; + + if (out) { + out->next = kmalloc(sizeof(struct bce_segment_list_element_hostinfo), GFP_KERNEL); + out = out->next; + } else { + out_root = out = kmalloc(sizeof(struct bce_segment_list_element_hostinfo), GFP_KERNEL); + } + out->page_start = (void *) ptr; + out->page_count = 1; + out->dma_start = DMA_MAPPING_ERROR; + out->next = NULL; + } + pptr = ptr; + } + el->addr = sg->dma_address; + el->length = sg->length; + header->data_size += el->length; + } + + /* DMA map */ + out = out_root; + pout = NULL; + while (out) { + out->dma_start = dma_map_single(dev, out->page_start, out->page_count * PAGE_SIZE, DMA_TO_DEVICE); + if (dma_mapping_error(dev, out->dma_start)) + goto error; + if (pout) { + header = pout->page_start; + header->next_segl_addr = out->dma_start; + header->next_segl_length = out->page_count * PAGE_SIZE; + } + pout = out; + out = out->next; + } + return out_root; + + error: + bce_unmap_segement_list(dev, out_root); + return NULL; +} + +static void bce_unmap_segement_list(struct device *dev, struct bce_segment_list_element_hostinfo *list) +{ + struct bce_segment_list_element_hostinfo *next; + while (list) { + if (list->dma_start != DMA_MAPPING_ERROR) + dma_unmap_single(dev, list->dma_start, list->page_count * PAGE_SIZE, DMA_TO_DEVICE); + next = list->next; + kfree(list); + list = next; + } +} + +int bce_set_submission_buf(struct bce_qe_submission *element, struct bce_dma_buffer *buf, size_t offset, size_t length) +{ + struct bce_segment_list_element_hostinfo *seg; + struct bce_segment_list_header *seg_header; + + seg = buf->seglist_hostinfo; + if (!seg) { + element->addr = buf->scatterlist.sgl->dma_address + offset; + element->length = length; + element->segl_addr = 0; + element->segl_length = 0; + return 0; + } + + while (seg) { + seg_header = seg->page_start; + if (offset <= seg_header->data_size) + break; + offset -= seg_header->data_size; + seg = seg->next; + } + if (!seg) + return -EINVAL; + element->addr = offset; + element->length = buf->scatterlist.sgl->dma_length; + element->segl_addr = seg->dma_start; + element->segl_length = seg->page_count * PAGE_SIZE; + return 0; +} \ No newline at end of file diff --git a/drivers/staging/apple-bce/queue_dma.h b/drivers/staging/apple-bce/queue_dma.h new file mode 100644 index 000000000000..f8a57e50e7a3 --- /dev/null +++ b/drivers/staging/apple-bce/queue_dma.h @@ -0,0 +1,50 @@ +#ifndef BCE_QUEUE_DMA_H +#define BCE_QUEUE_DMA_H + +#include + +struct bce_qe_submission; + +struct bce_segment_list_header { + u64 element_count; + u64 data_size; + + u64 next_segl_addr; + u64 next_segl_length; +}; +struct bce_segment_list_element { + u64 addr; + u64 length; +}; + +struct bce_segment_list_element_hostinfo { + struct bce_segment_list_element_hostinfo *next; + void *page_start; + size_t page_count; + dma_addr_t dma_start; +}; + + +struct bce_dma_buffer { + enum dma_data_direction direction; + struct sg_table scatterlist; + struct bce_segment_list_element_hostinfo *seglist_hostinfo; +}; + +/* NOTE: Takes ownership of the sg_table if it succeeds. Ownership is not transferred on failure. */ +int bce_map_dma_buffer(struct device *dev, struct bce_dma_buffer *buf, struct sg_table scatterlist, + enum dma_data_direction dir); + +/* Creates a buffer from virtual memory (vmalloc) */ +int bce_map_dma_buffer_vm(struct device *dev, struct bce_dma_buffer *buf, void *data, size_t len, + enum dma_data_direction dir); + +/* Creates a buffer from kernel memory (kmalloc) */ +int bce_map_dma_buffer_km(struct device *dev, struct bce_dma_buffer *buf, void *data, size_t len, + enum dma_data_direction dir); + +void bce_unmap_dma_buffer(struct device *dev, struct bce_dma_buffer *buf); + +int bce_set_submission_buf(struct bce_qe_submission *element, struct bce_dma_buffer *buf, size_t offset, size_t length); + +#endif //BCE_QUEUE_DMA_H diff --git a/drivers/staging/apple-bce/vhci/command.h b/drivers/staging/apple-bce/vhci/command.h new file mode 100644 index 000000000000..26619e0bccfa --- /dev/null +++ b/drivers/staging/apple-bce/vhci/command.h @@ -0,0 +1,204 @@ +#ifndef BCE_VHCI_COMMAND_H +#define BCE_VHCI_COMMAND_H + +#include "queue.h" +#include +#include + +#define BCE_VHCI_CMD_TIMEOUT_SHORT msecs_to_jiffies(2000) +#define BCE_VHCI_CMD_TIMEOUT_LONG msecs_to_jiffies(30000) + +#define BCE_VHCI_BULK_MAX_ACTIVE_URBS_POW2 2 +#define BCE_VHCI_BULK_MAX_ACTIVE_URBS (1 << BCE_VHCI_BULK_MAX_ACTIVE_URBS_POW2) + +typedef u8 bce_vhci_port_t; +typedef u8 bce_vhci_device_t; + +enum bce_vhci_command { + BCE_VHCI_CMD_CONTROLLER_ENABLE = 1, + BCE_VHCI_CMD_CONTROLLER_DISABLE = 2, + BCE_VHCI_CMD_CONTROLLER_START = 3, + BCE_VHCI_CMD_CONTROLLER_PAUSE = 4, + + BCE_VHCI_CMD_PORT_POWER_ON = 0x10, + BCE_VHCI_CMD_PORT_POWER_OFF = 0x11, + BCE_VHCI_CMD_PORT_RESUME = 0x12, + BCE_VHCI_CMD_PORT_SUSPEND = 0x13, + BCE_VHCI_CMD_PORT_RESET = 0x14, + BCE_VHCI_CMD_PORT_DISABLE = 0x15, + BCE_VHCI_CMD_PORT_STATUS = 0x16, + + BCE_VHCI_CMD_DEVICE_CREATE = 0x30, + BCE_VHCI_CMD_DEVICE_DESTROY = 0x31, + + BCE_VHCI_CMD_ENDPOINT_CREATE = 0x40, + BCE_VHCI_CMD_ENDPOINT_DESTROY = 0x41, + BCE_VHCI_CMD_ENDPOINT_SET_STATE = 0x42, + BCE_VHCI_CMD_ENDPOINT_RESET = 0x44, + + /* Device to host only */ + BCE_VHCI_CMD_ENDPOINT_REQUEST_STATE = 0x43, + BCE_VHCI_CMD_TRANSFER_REQUEST = 0x1000, + BCE_VHCI_CMD_CONTROL_TRANSFER_STATUS = 0x1005 +}; + +enum bce_vhci_endpoint_state { + BCE_VHCI_ENDPOINT_ACTIVE = 0, + BCE_VHCI_ENDPOINT_PAUSED = 1, + BCE_VHCI_ENDPOINT_STALLED = 2 +}; + +static inline int bce_vhci_cmd_controller_enable(struct bce_vhci_command_queue *q, u8 busNum, u16 *portMask) +{ + int status; + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_CONTROLLER_ENABLE; + cmd.param1 = 0x7100u | busNum; + status = bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_LONG); + if (!status) + *portMask = (u16) res.param2; + return status; +} +static inline int bce_vhci_cmd_controller_disable(struct bce_vhci_command_queue *q) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_CONTROLLER_DISABLE; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_LONG); +} +static inline int bce_vhci_cmd_controller_start(struct bce_vhci_command_queue *q) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_CONTROLLER_START; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_LONG); +} +static inline int bce_vhci_cmd_controller_pause(struct bce_vhci_command_queue *q) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_CONTROLLER_PAUSE; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_LONG); +} + +static inline int bce_vhci_cmd_port_power_on(struct bce_vhci_command_queue *q, bce_vhci_port_t port) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_PORT_POWER_ON; + cmd.param1 = port; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); +} +static inline int bce_vhci_cmd_port_power_off(struct bce_vhci_command_queue *q, bce_vhci_port_t port) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_PORT_POWER_OFF; + cmd.param1 = port; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); +} +static inline int bce_vhci_cmd_port_resume(struct bce_vhci_command_queue *q, bce_vhci_port_t port) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_PORT_RESUME; + cmd.param1 = port; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_LONG); +} +static inline int bce_vhci_cmd_port_suspend(struct bce_vhci_command_queue *q, bce_vhci_port_t port) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_PORT_SUSPEND; + cmd.param1 = port; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_LONG); +} +static inline int bce_vhci_cmd_port_reset(struct bce_vhci_command_queue *q, bce_vhci_port_t port, u32 timeout) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_PORT_RESET; + cmd.param1 = port; + cmd.param2 = timeout; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); +} +static inline int bce_vhci_cmd_port_disable(struct bce_vhci_command_queue *q, bce_vhci_port_t port) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_PORT_DISABLE; + cmd.param1 = port; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); +} +static inline int bce_vhci_cmd_port_status(struct bce_vhci_command_queue *q, bce_vhci_port_t port, + u32 clearFlags, u32 *resStatus) +{ + int status; + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_PORT_STATUS; + cmd.param1 = port; + cmd.param2 = clearFlags & 0x560000; + status = bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); + if (status >= 0) + *resStatus = (u32) res.param2; + return status; +} + +static inline int bce_vhci_cmd_device_create(struct bce_vhci_command_queue *q, bce_vhci_port_t port, + bce_vhci_device_t *dev) +{ + int status; + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_DEVICE_CREATE; + cmd.param1 = port; + status = bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); + if (!status) + *dev = (bce_vhci_device_t) res.param2; + return status; +} +static inline int bce_vhci_cmd_device_destroy(struct bce_vhci_command_queue *q, bce_vhci_device_t dev) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_DEVICE_DESTROY; + cmd.param1 = dev; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_LONG); +} + +static inline int bce_vhci_cmd_endpoint_create(struct bce_vhci_command_queue *q, bce_vhci_device_t dev, + struct usb_endpoint_descriptor *desc) +{ + struct bce_vhci_message cmd, res; + int endpoint_type = usb_endpoint_type(desc); + int maxp = usb_endpoint_maxp(desc); + int maxp_burst = usb_endpoint_maxp_mult(desc) * maxp; + u8 max_active_requests_pow2 = 0; + cmd.cmd = BCE_VHCI_CMD_ENDPOINT_CREATE; + cmd.param1 = dev | ((desc->bEndpointAddress & 0x8Fu) << 8); + if (endpoint_type == USB_ENDPOINT_XFER_BULK) + max_active_requests_pow2 = BCE_VHCI_BULK_MAX_ACTIVE_URBS_POW2; + cmd.param2 = endpoint_type | ((max_active_requests_pow2 & 0xf) << 4) | (maxp << 16) | ((u64) maxp_burst << 32); + if (endpoint_type == USB_ENDPOINT_XFER_INT) + cmd.param2 |= (desc->bInterval - 1) << 8; + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); +} +static inline int bce_vhci_cmd_endpoint_destroy(struct bce_vhci_command_queue *q, bce_vhci_device_t dev, u8 endpoint) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_ENDPOINT_DESTROY; + cmd.param1 = dev | (endpoint << 8); + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); +} +static inline int bce_vhci_cmd_endpoint_set_state(struct bce_vhci_command_queue *q, bce_vhci_device_t dev, u8 endpoint, + enum bce_vhci_endpoint_state newState, enum bce_vhci_endpoint_state *retState) +{ + int status; + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_ENDPOINT_SET_STATE; + cmd.param1 = dev | (endpoint << 8); + cmd.param2 = (u64) newState; + status = bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); + if (status != BCE_VHCI_INTERNAL_ERROR && status != BCE_VHCI_NO_POWER) + *retState = (enum bce_vhci_endpoint_state) res.param2; + return status; +} +static inline int bce_vhci_cmd_endpoint_reset(struct bce_vhci_command_queue *q, bce_vhci_device_t dev, u8 endpoint) +{ + struct bce_vhci_message cmd, res; + cmd.cmd = BCE_VHCI_CMD_ENDPOINT_RESET; + cmd.param1 = dev | (endpoint << 8); + return bce_vhci_command_queue_execute(q, &cmd, &res, BCE_VHCI_CMD_TIMEOUT_SHORT); +} + + +#endif //BCE_VHCI_COMMAND_H diff --git a/drivers/staging/apple-bce/vhci/queue.c b/drivers/staging/apple-bce/vhci/queue.c new file mode 100644 index 000000000000..7b0b5027157b --- /dev/null +++ b/drivers/staging/apple-bce/vhci/queue.c @@ -0,0 +1,268 @@ +#include "queue.h" +#include "vhci.h" +#include "../apple_bce.h" + + +static void bce_vhci_message_queue_completion(struct bce_queue_sq *sq); + +int bce_vhci_message_queue_create(struct bce_vhci *vhci, struct bce_vhci_message_queue *ret, const char *name) +{ + int status; + ret->cq = bce_create_cq(vhci->dev, VHCI_EVENT_QUEUE_EL_COUNT); + if (!ret->cq) + return -EINVAL; + ret->sq = bce_create_sq(vhci->dev, ret->cq, name, VHCI_EVENT_QUEUE_EL_COUNT, DMA_TO_DEVICE, + bce_vhci_message_queue_completion, ret); + if (!ret->sq) { + status = -EINVAL; + goto fail_cq; + } + ret->data = dma_alloc_coherent(&vhci->dev->pci->dev, sizeof(struct bce_vhci_message) * VHCI_EVENT_QUEUE_EL_COUNT, + &ret->dma_addr, GFP_KERNEL); + if (!ret->data) { + status = -EINVAL; + goto fail_sq; + } + return 0; + +fail_sq: + bce_destroy_sq(vhci->dev, ret->sq); + ret->sq = NULL; +fail_cq: + bce_destroy_cq(vhci->dev, ret->cq); + ret->cq = NULL; + return status; +} + +void bce_vhci_message_queue_destroy(struct bce_vhci *vhci, struct bce_vhci_message_queue *q) +{ + if (!q->cq) + return; + dma_free_coherent(&vhci->dev->pci->dev, sizeof(struct bce_vhci_message) * VHCI_EVENT_QUEUE_EL_COUNT, + q->data, q->dma_addr); + bce_destroy_sq(vhci->dev, q->sq); + bce_destroy_cq(vhci->dev, q->cq); +} + +void bce_vhci_message_queue_write(struct bce_vhci_message_queue *q, struct bce_vhci_message *req) +{ + int sidx; + struct bce_qe_submission *s; + sidx = q->sq->tail; + s = bce_next_submission(q->sq); + pr_debug("bce-vhci: Send message: %x s=%x p1=%x p2=%llx\n", req->cmd, req->status, req->param1, req->param2); + q->data[sidx] = *req; + bce_set_submission_single(s, q->dma_addr + sizeof(struct bce_vhci_message) * sidx, + sizeof(struct bce_vhci_message)); + bce_submit_to_device(q->sq); +} + +static void bce_vhci_message_queue_completion(struct bce_queue_sq *sq) +{ + while (bce_next_completion(sq)) + bce_notify_submission_complete(sq); +} + + + +static void bce_vhci_event_queue_completion(struct bce_queue_sq *sq); + +int __bce_vhci_event_queue_create(struct bce_vhci *vhci, struct bce_vhci_event_queue *ret, const char *name, + bce_sq_completion compl) +{ + ret->vhci = vhci; + + ret->sq = bce_create_sq(vhci->dev, vhci->ev_cq, name, VHCI_EVENT_QUEUE_EL_COUNT, DMA_FROM_DEVICE, compl, ret); + if (!ret->sq) + return -EINVAL; + ret->data = dma_alloc_coherent(&vhci->dev->pci->dev, sizeof(struct bce_vhci_message) * VHCI_EVENT_QUEUE_EL_COUNT, + &ret->dma_addr, GFP_KERNEL); + if (!ret->data) { + bce_destroy_sq(vhci->dev, ret->sq); + ret->sq = NULL; + return -EINVAL; + } + + init_completion(&ret->queue_empty_completion); + bce_vhci_event_queue_submit_pending(ret, VHCI_EVENT_PENDING_COUNT); + return 0; +} + +int bce_vhci_event_queue_create(struct bce_vhci *vhci, struct bce_vhci_event_queue *ret, const char *name, + bce_vhci_event_queue_callback cb) +{ + ret->cb = cb; + return __bce_vhci_event_queue_create(vhci, ret, name, bce_vhci_event_queue_completion); +} + +void bce_vhci_event_queue_destroy(struct bce_vhci *vhci, struct bce_vhci_event_queue *q) +{ + if (!q->sq) + return; + dma_free_coherent(&vhci->dev->pci->dev, sizeof(struct bce_vhci_message) * VHCI_EVENT_QUEUE_EL_COUNT, + q->data, q->dma_addr); + bce_destroy_sq(vhci->dev, q->sq); +} + +static void bce_vhci_event_queue_completion(struct bce_queue_sq *sq) +{ + struct bce_sq_completion_data *cd; + struct bce_vhci_event_queue *ev = sq->userdata; + struct bce_vhci_message *msg; + size_t cnt = 0; + + while ((cd = bce_next_completion(sq))) { + if (cd->status == BCE_COMPLETION_ABORTED) { /* We flushed the queue */ + bce_notify_submission_complete(sq); + continue; + } + msg = &ev->data[sq->head]; + pr_debug("bce-vhci: Got event: %x s=%x p1=%x p2=%llx\n", msg->cmd, msg->status, msg->param1, msg->param2); + ev->cb(ev, msg); + + bce_notify_submission_complete(sq); + ++cnt; + } + bce_vhci_event_queue_submit_pending(ev, cnt); + if (atomic_read(&sq->available_commands) == sq->el_count - 1) + complete(&ev->queue_empty_completion); +} + +void bce_vhci_event_queue_submit_pending(struct bce_vhci_event_queue *q, size_t count) +{ + int idx; + struct bce_qe_submission *s; + while (count--) { + if (bce_reserve_submission(q->sq, NULL)) { + pr_err("bce-vhci: Failed to reserve an event queue submission\n"); + break; + } + idx = q->sq->tail; + s = bce_next_submission(q->sq); + bce_set_submission_single(s, + q->dma_addr + idx * sizeof(struct bce_vhci_message), sizeof(struct bce_vhci_message)); + } + bce_submit_to_device(q->sq); +} + +void bce_vhci_event_queue_pause(struct bce_vhci_event_queue *q) +{ + unsigned long timeout; + reinit_completion(&q->queue_empty_completion); + if (bce_cmd_flush_memory_queue(q->vhci->dev->cmd_cmdq, q->sq->qid)) + pr_warn("bce-vhci: failed to flush event queue\n"); + timeout = msecs_to_jiffies(5000); + while (atomic_read(&q->sq->available_commands) != q->sq->el_count - 1) { + timeout = wait_for_completion_timeout(&q->queue_empty_completion, timeout); + if (timeout == 0) { + pr_err("bce-vhci: waiting for queue to be flushed timed out\n"); + break; + } + } +} + +void bce_vhci_event_queue_resume(struct bce_vhci_event_queue *q) +{ + if (atomic_read(&q->sq->available_commands) != q->sq->el_count - 1) { + pr_err("bce-vhci: resume of a queue with pending submissions\n"); + return; + } + bce_vhci_event_queue_submit_pending(q, VHCI_EVENT_PENDING_COUNT); +} + +void bce_vhci_command_queue_create(struct bce_vhci_command_queue *ret, struct bce_vhci_message_queue *mq) +{ + ret->mq = mq; + ret->completion.result = NULL; + init_completion(&ret->completion.completion); + spin_lock_init(&ret->completion_lock); + mutex_init(&ret->mutex); +} + +void bce_vhci_command_queue_destroy(struct bce_vhci_command_queue *cq) +{ + spin_lock(&cq->completion_lock); + if (cq->completion.result) { + memset(cq->completion.result, 0, sizeof(struct bce_vhci_message)); + cq->completion.result->status = BCE_VHCI_ABORT; + complete(&cq->completion.completion); + cq->completion.result = NULL; + } + spin_unlock(&cq->completion_lock); + mutex_lock(&cq->mutex); + mutex_unlock(&cq->mutex); + mutex_destroy(&cq->mutex); +} + +void bce_vhci_command_queue_deliver_completion(struct bce_vhci_command_queue *cq, struct bce_vhci_message *msg) +{ + struct bce_vhci_command_queue_completion *c = &cq->completion; + + spin_lock(&cq->completion_lock); + if (c->result) { + *c->result = *msg; + complete(&c->completion); + c->result = NULL; + } + spin_unlock(&cq->completion_lock); +} + +static int __bce_vhci_command_queue_execute(struct bce_vhci_command_queue *cq, struct bce_vhci_message *req, + struct bce_vhci_message *res, unsigned long timeout) +{ + int status; + struct bce_vhci_command_queue_completion *c; + struct bce_vhci_message creq; + c = &cq->completion; + + if ((status = bce_reserve_submission(cq->mq->sq, &timeout))) + return status; + + spin_lock(&cq->completion_lock); + c->result = res; + reinit_completion(&c->completion); + spin_unlock(&cq->completion_lock); + + bce_vhci_message_queue_write(cq->mq, req); + + if (!wait_for_completion_timeout(&c->completion, timeout)) { + /* we ran out of time, send cancellation */ + pr_debug("bce-vhci: command timed out req=%x\n", req->cmd); + if ((status = bce_reserve_submission(cq->mq->sq, &timeout))) + return status; + + creq = *req; + creq.cmd |= 0x4000; + bce_vhci_message_queue_write(cq->mq, &creq); + + if (!wait_for_completion_timeout(&c->completion, 1000)) { + pr_err("bce-vhci: Possible desync, cmd cancel timed out\n"); + + spin_lock(&cq->completion_lock); + c->result = NULL; + spin_unlock(&cq->completion_lock); + return -ETIMEDOUT; + } + if ((res->cmd & ~0x8000) == creq.cmd) + return -ETIMEDOUT; + /* reply for the previous command most likely arrived */ + } + + if ((res->cmd & ~0x8000) != req->cmd) { + pr_err("bce-vhci: Possible desync, cmd reply mismatch req=%x, res=%x\n", req->cmd, res->cmd); + return -EIO; + } + if (res->status == BCE_VHCI_SUCCESS) + return 0; + return res->status; +} + +int bce_vhci_command_queue_execute(struct bce_vhci_command_queue *cq, struct bce_vhci_message *req, + struct bce_vhci_message *res, unsigned long timeout) +{ + int status; + mutex_lock(&cq->mutex); + status = __bce_vhci_command_queue_execute(cq, req, res, timeout); + mutex_unlock(&cq->mutex); + return status; +} diff --git a/drivers/staging/apple-bce/vhci/queue.h b/drivers/staging/apple-bce/vhci/queue.h new file mode 100644 index 000000000000..adb705b6ba1d --- /dev/null +++ b/drivers/staging/apple-bce/vhci/queue.h @@ -0,0 +1,76 @@ +#ifndef BCE_VHCI_QUEUE_H +#define BCE_VHCI_QUEUE_H + +#include +#include "../queue.h" + +#define VHCI_EVENT_QUEUE_EL_COUNT 256 +#define VHCI_EVENT_PENDING_COUNT 32 + +struct bce_vhci; +struct bce_vhci_event_queue; + +enum bce_vhci_message_status { + BCE_VHCI_SUCCESS = 1, + BCE_VHCI_ERROR = 2, + BCE_VHCI_USB_PIPE_STALL = 3, + BCE_VHCI_ABORT = 4, + BCE_VHCI_BAD_ARGUMENT = 5, + BCE_VHCI_OVERRUN = 6, + BCE_VHCI_INTERNAL_ERROR = 7, + BCE_VHCI_NO_POWER = 8, + BCE_VHCI_UNSUPPORTED = 9 +}; +struct bce_vhci_message { + u16 cmd; + u16 status; // bce_vhci_message_status + u32 param1; + u64 param2; +}; + +struct bce_vhci_message_queue { + struct bce_queue_cq *cq; + struct bce_queue_sq *sq; + struct bce_vhci_message *data; + dma_addr_t dma_addr; +}; +typedef void (*bce_vhci_event_queue_callback)(struct bce_vhci_event_queue *q, struct bce_vhci_message *msg); +struct bce_vhci_event_queue { + struct bce_vhci *vhci; + struct bce_queue_sq *sq; + struct bce_vhci_message *data; + dma_addr_t dma_addr; + bce_vhci_event_queue_callback cb; + struct completion queue_empty_completion; +}; +struct bce_vhci_command_queue_completion { + struct bce_vhci_message *result; + struct completion completion; +}; +struct bce_vhci_command_queue { + struct bce_vhci_message_queue *mq; + struct bce_vhci_command_queue_completion completion; + struct spinlock completion_lock; + struct mutex mutex; +}; + +int bce_vhci_message_queue_create(struct bce_vhci *vhci, struct bce_vhci_message_queue *ret, const char *name); +void bce_vhci_message_queue_destroy(struct bce_vhci *vhci, struct bce_vhci_message_queue *q); +void bce_vhci_message_queue_write(struct bce_vhci_message_queue *q, struct bce_vhci_message *req); + +int __bce_vhci_event_queue_create(struct bce_vhci *vhci, struct bce_vhci_event_queue *ret, const char *name, + bce_sq_completion compl); +int bce_vhci_event_queue_create(struct bce_vhci *vhci, struct bce_vhci_event_queue *ret, const char *name, + bce_vhci_event_queue_callback cb); +void bce_vhci_event_queue_destroy(struct bce_vhci *vhci, struct bce_vhci_event_queue *q); +void bce_vhci_event_queue_submit_pending(struct bce_vhci_event_queue *q, size_t count); +void bce_vhci_event_queue_pause(struct bce_vhci_event_queue *q); +void bce_vhci_event_queue_resume(struct bce_vhci_event_queue *q); + +void bce_vhci_command_queue_create(struct bce_vhci_command_queue *ret, struct bce_vhci_message_queue *mq); +void bce_vhci_command_queue_destroy(struct bce_vhci_command_queue *cq); +int bce_vhci_command_queue_execute(struct bce_vhci_command_queue *cq, struct bce_vhci_message *req, + struct bce_vhci_message *res, unsigned long timeout); +void bce_vhci_command_queue_deliver_completion(struct bce_vhci_command_queue *cq, struct bce_vhci_message *msg); + +#endif //BCE_VHCI_QUEUE_H diff --git a/drivers/staging/apple-bce/vhci/transfer.c b/drivers/staging/apple-bce/vhci/transfer.c new file mode 100644 index 000000000000..d77220707492 --- /dev/null +++ b/drivers/staging/apple-bce/vhci/transfer.c @@ -0,0 +1,673 @@ +#include "transfer.h" +#include "../queue.h" +#include "vhci.h" +#include "../apple_bce.h" +#include + +static void bce_vhci_transfer_queue_completion(struct bce_queue_sq *sq); +static void bce_vhci_transfer_queue_giveback(struct bce_vhci_transfer_queue *q); +static void bce_vhci_transfer_queue_remove_pending(struct bce_vhci_transfer_queue *q); + +static int bce_vhci_urb_init(struct bce_vhci_urb *vurb); +static int bce_vhci_urb_update(struct bce_vhci_urb *urb, struct bce_vhci_message *msg); +static int bce_vhci_urb_transfer_completion(struct bce_vhci_urb *urb, struct bce_sq_completion_data *c); + +static void bce_vhci_transfer_queue_reset_w(struct work_struct *work); + +void bce_vhci_create_transfer_queue(struct bce_vhci *vhci, struct bce_vhci_transfer_queue *q, + struct usb_host_endpoint *endp, bce_vhci_device_t dev_addr, enum dma_data_direction dir) +{ + char name[0x21]; + INIT_LIST_HEAD(&q->evq); + INIT_LIST_HEAD(&q->giveback_urb_list); + spin_lock_init(&q->urb_lock); + mutex_init(&q->pause_lock); + q->vhci = vhci; + q->endp = endp; + q->dev_addr = dev_addr; + q->endp_addr = (u8) (endp->desc.bEndpointAddress & 0x8F); + q->state = BCE_VHCI_ENDPOINT_ACTIVE; + q->active = true; + q->stalled = false; + q->max_active_requests = 1; + if (usb_endpoint_type(&endp->desc) == USB_ENDPOINT_XFER_BULK) + q->max_active_requests = BCE_VHCI_BULK_MAX_ACTIVE_URBS; + q->remaining_active_requests = q->max_active_requests; + q->cq = bce_create_cq(vhci->dev, 0x100); + INIT_WORK(&q->w_reset, bce_vhci_transfer_queue_reset_w); + q->sq_in = NULL; + if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) { + snprintf(name, sizeof(name), "VHC1-%i-%02x", dev_addr, 0x80 | usb_endpoint_num(&endp->desc)); + q->sq_in = bce_create_sq(vhci->dev, q->cq, name, 0x100, DMA_FROM_DEVICE, + bce_vhci_transfer_queue_completion, q); + } + q->sq_out = NULL; + if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) { + snprintf(name, sizeof(name), "VHC1-%i-%02x", dev_addr, usb_endpoint_num(&endp->desc)); + q->sq_out = bce_create_sq(vhci->dev, q->cq, name, 0x100, DMA_TO_DEVICE, + bce_vhci_transfer_queue_completion, q); + } +} + +void bce_vhci_destroy_transfer_queue(struct bce_vhci *vhci, struct bce_vhci_transfer_queue *q) +{ + bce_vhci_transfer_queue_giveback(q); + bce_vhci_transfer_queue_remove_pending(q); + if (q->sq_in) + bce_destroy_sq(vhci->dev, q->sq_in); + if (q->sq_out) + bce_destroy_sq(vhci->dev, q->sq_out); + bce_destroy_cq(vhci->dev, q->cq); +} + +static inline bool bce_vhci_transfer_queue_can_init_urb(struct bce_vhci_transfer_queue *q) +{ + return q->remaining_active_requests > 0; +} + +static void bce_vhci_transfer_queue_defer_event(struct bce_vhci_transfer_queue *q, struct bce_vhci_message *msg) +{ + struct bce_vhci_list_message *lm; + lm = kmalloc(sizeof(struct bce_vhci_list_message), GFP_KERNEL); + INIT_LIST_HEAD(&lm->list); + lm->msg = *msg; + list_add_tail(&lm->list, &q->evq); +} + +static void bce_vhci_transfer_queue_giveback(struct bce_vhci_transfer_queue *q) +{ + unsigned long flags; + struct urb *urb; + spin_lock_irqsave(&q->urb_lock, flags); + while (!list_empty(&q->giveback_urb_list)) { + urb = list_first_entry(&q->giveback_urb_list, struct urb, urb_list); + list_del(&urb->urb_list); + + spin_unlock_irqrestore(&q->urb_lock, flags); + usb_hcd_giveback_urb(q->vhci->hcd, urb, urb->status); + spin_lock_irqsave(&q->urb_lock, flags); + } + spin_unlock_irqrestore(&q->urb_lock, flags); +} + +static void bce_vhci_transfer_queue_init_pending_urbs(struct bce_vhci_transfer_queue *q); + +static void bce_vhci_transfer_queue_deliver_pending(struct bce_vhci_transfer_queue *q) +{ + struct urb *urb; + struct bce_vhci_list_message *lm; + + while (!list_empty(&q->endp->urb_list) && !list_empty(&q->evq)) { + urb = list_first_entry(&q->endp->urb_list, struct urb, urb_list); + + lm = list_first_entry(&q->evq, struct bce_vhci_list_message, list); + if (bce_vhci_urb_update(urb->hcpriv, &lm->msg) == -EAGAIN) + break; + list_del(&lm->list); + kfree(lm); + } + + /* some of the URBs could have been completed, so initialize more URBs if possible */ + bce_vhci_transfer_queue_init_pending_urbs(q); +} + +static void bce_vhci_transfer_queue_remove_pending(struct bce_vhci_transfer_queue *q) +{ + unsigned long flags; + struct bce_vhci_list_message *lm; + spin_lock_irqsave(&q->urb_lock, flags); + while (!list_empty(&q->evq)) { + lm = list_first_entry(&q->evq, struct bce_vhci_list_message, list); + list_del(&lm->list); + kfree(lm); + } + spin_unlock_irqrestore(&q->urb_lock, flags); +} + +void bce_vhci_transfer_queue_event(struct bce_vhci_transfer_queue *q, struct bce_vhci_message *msg) +{ + unsigned long flags; + struct bce_vhci_urb *turb; + struct urb *urb; + spin_lock_irqsave(&q->urb_lock, flags); + bce_vhci_transfer_queue_deliver_pending(q); + + if (msg->cmd == BCE_VHCI_CMD_TRANSFER_REQUEST && + (!list_empty(&q->evq) || list_empty(&q->endp->urb_list))) { + bce_vhci_transfer_queue_defer_event(q, msg); + goto complete; + } + if (list_empty(&q->endp->urb_list)) { + pr_err("bce-vhci: [%02x] Unexpected transfer queue event\n", q->endp_addr); + goto complete; + } + urb = list_first_entry(&q->endp->urb_list, struct urb, urb_list); + turb = urb->hcpriv; + if (bce_vhci_urb_update(turb, msg) == -EAGAIN) { + bce_vhci_transfer_queue_defer_event(q, msg); + } else { + bce_vhci_transfer_queue_init_pending_urbs(q); + } + +complete: + spin_unlock_irqrestore(&q->urb_lock, flags); + bce_vhci_transfer_queue_giveback(q); +} + +static void bce_vhci_transfer_queue_completion(struct bce_queue_sq *sq) +{ + unsigned long flags; + struct bce_sq_completion_data *c; + struct urb *urb; + struct bce_vhci_transfer_queue *q = sq->userdata; + spin_lock_irqsave(&q->urb_lock, flags); + while ((c = bce_next_completion(sq))) { + if (c->status == BCE_COMPLETION_ABORTED) { /* We flushed the queue */ + pr_debug("bce-vhci: [%02x] Got an abort completion\n", q->endp_addr); + bce_notify_submission_complete(sq); + continue; + } + if (list_empty(&q->endp->urb_list)) { + pr_err("bce-vhci: [%02x] Got a completion while no requests are pending\n", q->endp_addr); + continue; + } + pr_debug("bce-vhci: [%02x] Got a transfer queue completion\n", q->endp_addr); + urb = list_first_entry(&q->endp->urb_list, struct urb, urb_list); + bce_vhci_urb_transfer_completion(urb->hcpriv, c); + bce_notify_submission_complete(sq); + } + bce_vhci_transfer_queue_deliver_pending(q); + spin_unlock_irqrestore(&q->urb_lock, flags); + bce_vhci_transfer_queue_giveback(q); +} + +int bce_vhci_transfer_queue_do_pause(struct bce_vhci_transfer_queue *q) +{ + unsigned long flags; + int status; + u8 endp_addr = (u8) (q->endp->desc.bEndpointAddress & 0x8F); + spin_lock_irqsave(&q->urb_lock, flags); + q->active = false; + spin_unlock_irqrestore(&q->urb_lock, flags); + if (q->sq_out) { + pr_err("bce-vhci: Not implemented: wait for pending output requests\n"); + } + bce_vhci_transfer_queue_remove_pending(q); + if ((status = bce_vhci_cmd_endpoint_set_state( + &q->vhci->cq, q->dev_addr, endp_addr, BCE_VHCI_ENDPOINT_PAUSED, &q->state))) + return status; + if (q->state != BCE_VHCI_ENDPOINT_PAUSED) + return -EINVAL; + if (q->sq_in) + bce_cmd_flush_memory_queue(q->vhci->dev->cmd_cmdq, (u16) q->sq_in->qid); + if (q->sq_out) + bce_cmd_flush_memory_queue(q->vhci->dev->cmd_cmdq, (u16) q->sq_out->qid); + return 0; +} + +static void bce_vhci_urb_resume(struct bce_vhci_urb *urb); + +int bce_vhci_transfer_queue_do_resume(struct bce_vhci_transfer_queue *q) +{ + unsigned long flags; + int status; + struct urb *urb, *urbt; + struct bce_vhci_urb *vurb; + u8 endp_addr = (u8) (q->endp->desc.bEndpointAddress & 0x8F); + if ((status = bce_vhci_cmd_endpoint_set_state( + &q->vhci->cq, q->dev_addr, endp_addr, BCE_VHCI_ENDPOINT_ACTIVE, &q->state))) + return status; + if (q->state != BCE_VHCI_ENDPOINT_ACTIVE) + return -EINVAL; + spin_lock_irqsave(&q->urb_lock, flags); + q->active = true; + list_for_each_entry_safe(urb, urbt, &q->endp->urb_list, urb_list) { + vurb = urb->hcpriv; + if (vurb->state == BCE_VHCI_URB_INIT_PENDING) { + if (!bce_vhci_transfer_queue_can_init_urb(q)) + break; + bce_vhci_urb_init(vurb); + } else { + bce_vhci_urb_resume(vurb); + } + } + bce_vhci_transfer_queue_deliver_pending(q); + spin_unlock_irqrestore(&q->urb_lock, flags); + return 0; +} + +int bce_vhci_transfer_queue_pause(struct bce_vhci_transfer_queue *q, enum bce_vhci_pause_source src) +{ + int ret = 0; + mutex_lock(&q->pause_lock); + if ((q->paused_by & src) != src) { + if (!q->paused_by) + ret = bce_vhci_transfer_queue_do_pause(q); + if (!ret) + q->paused_by |= src; + } + mutex_unlock(&q->pause_lock); + return ret; +} + +int bce_vhci_transfer_queue_resume(struct bce_vhci_transfer_queue *q, enum bce_vhci_pause_source src) +{ + int ret = 0; + mutex_lock(&q->pause_lock); + if (q->paused_by & src) { + if (!(q->paused_by & ~src)) + ret = bce_vhci_transfer_queue_do_resume(q); + if (!ret) + q->paused_by &= ~src; + } + mutex_unlock(&q->pause_lock); + return ret; +} + +static void bce_vhci_transfer_queue_reset_w(struct work_struct *work) +{ + unsigned long flags; + struct bce_vhci_transfer_queue *q = container_of(work, struct bce_vhci_transfer_queue, w_reset); + + mutex_lock(&q->pause_lock); + spin_lock_irqsave(&q->urb_lock, flags); + if (!q->stalled) { + spin_unlock_irqrestore(&q->urb_lock, flags); + mutex_unlock(&q->pause_lock); + return; + } + q->active = false; + spin_unlock_irqrestore(&q->urb_lock, flags); + q->paused_by |= BCE_VHCI_PAUSE_INTERNAL_WQ; + bce_vhci_transfer_queue_remove_pending(q); + if (q->sq_in) + bce_cmd_flush_memory_queue(q->vhci->dev->cmd_cmdq, (u16) q->sq_in->qid); + if (q->sq_out) + bce_cmd_flush_memory_queue(q->vhci->dev->cmd_cmdq, (u16) q->sq_out->qid); + bce_vhci_cmd_endpoint_reset(&q->vhci->cq, q->dev_addr, (u8) (q->endp->desc.bEndpointAddress & 0x8F)); + spin_lock_irqsave(&q->urb_lock, flags); + q->stalled = false; + spin_unlock_irqrestore(&q->urb_lock, flags); + mutex_unlock(&q->pause_lock); + bce_vhci_transfer_queue_resume(q, BCE_VHCI_PAUSE_INTERNAL_WQ); +} + +void bce_vhci_transfer_queue_request_reset(struct bce_vhci_transfer_queue *q) +{ + queue_work(q->vhci->tq_state_wq, &q->w_reset); +} + +static void bce_vhci_transfer_queue_init_pending_urbs(struct bce_vhci_transfer_queue *q) +{ + struct urb *urb, *urbt; + struct bce_vhci_urb *vurb; + list_for_each_entry_safe(urb, urbt, &q->endp->urb_list, urb_list) { + vurb = urb->hcpriv; + if (!bce_vhci_transfer_queue_can_init_urb(q)) + break; + if (vurb->state == BCE_VHCI_URB_INIT_PENDING) + bce_vhci_urb_init(vurb); + } +} + + + +static int bce_vhci_urb_data_start(struct bce_vhci_urb *urb, unsigned long *timeout); + +int bce_vhci_urb_create(struct bce_vhci_transfer_queue *q, struct urb *urb) +{ + unsigned long flags; + int status = 0; + struct bce_vhci_urb *vurb; + vurb = kzalloc(sizeof(struct bce_vhci_urb), GFP_KERNEL); + urb->hcpriv = vurb; + + vurb->q = q; + vurb->urb = urb; + vurb->dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE; + vurb->is_control = (usb_endpoint_num(&urb->ep->desc) == 0); + + spin_lock_irqsave(&q->urb_lock, flags); + status = usb_hcd_link_urb_to_ep(q->vhci->hcd, urb); + if (status) { + spin_unlock_irqrestore(&q->urb_lock, flags); + urb->hcpriv = NULL; + kfree(vurb); + return status; + } + + if (q->active) { + if (bce_vhci_transfer_queue_can_init_urb(vurb->q)) + status = bce_vhci_urb_init(vurb); + else + vurb->state = BCE_VHCI_URB_INIT_PENDING; + } else { + if (q->stalled) + bce_vhci_transfer_queue_request_reset(q); + vurb->state = BCE_VHCI_URB_INIT_PENDING; + } + if (status) { + usb_hcd_unlink_urb_from_ep(q->vhci->hcd, urb); + urb->hcpriv = NULL; + kfree(vurb); + } else { + bce_vhci_transfer_queue_deliver_pending(q); + } + spin_unlock_irqrestore(&q->urb_lock, flags); + pr_debug("bce-vhci: [%02x] URB enqueued (dir = %s, size = %i)\n", q->endp_addr, + usb_urb_dir_in(urb) ? "IN" : "OUT", urb->transfer_buffer_length); + return status; +} + +static int bce_vhci_urb_init(struct bce_vhci_urb *vurb) +{ + int status = 0; + + if (vurb->q->remaining_active_requests == 0) { + pr_err("bce-vhci: cannot init request (remaining_active_requests = 0)\n"); + return -EINVAL; + } + + if (vurb->is_control) { + vurb->state = BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_REQUEST; + } else { + status = bce_vhci_urb_data_start(vurb, NULL); + } + + if (!status) { + --vurb->q->remaining_active_requests; + } + return status; +} + +static void bce_vhci_urb_complete(struct bce_vhci_urb *urb, int status) +{ + struct bce_vhci_transfer_queue *q = urb->q; + struct bce_vhci *vhci = q->vhci; + struct urb *real_urb = urb->urb; + pr_debug("bce-vhci: [%02x] URB complete %i\n", q->endp_addr, status); + usb_hcd_unlink_urb_from_ep(vhci->hcd, real_urb); + real_urb->hcpriv = NULL; + real_urb->status = status; + if (urb->state != BCE_VHCI_URB_INIT_PENDING) + ++urb->q->remaining_active_requests; + kfree(urb); + list_add_tail(&real_urb->urb_list, &q->giveback_urb_list); +} + +int bce_vhci_urb_request_cancel(struct bce_vhci_transfer_queue *q, struct urb *urb, int status) +{ + struct bce_vhci_urb *vurb; + unsigned long flags; + int ret; + enum bce_vhci_urb_state old_state; + + spin_lock_irqsave(&q->urb_lock, flags); + if ((ret = usb_hcd_check_unlink_urb(q->vhci->hcd, urb, status))) { + spin_unlock_irqrestore(&q->urb_lock, flags); + return ret; + } + + vurb = urb->hcpriv; + + old_state = vurb->state; /* save old state to use later because we'll set state as cancelled */ + + if (old_state == BCE_VHCI_URB_CANCELLED) { + spin_unlock_irqrestore(&q->urb_lock, flags); + pr_debug("bce-vhci: URB %p is already cancelled, skipping\n", urb); + return 0; + } + + vurb->state = BCE_VHCI_URB_CANCELLED; + + /* If the URB wasn't posted to the device yet, we can still remove it on the host without pausing the queue. */ + if (old_state != BCE_VHCI_URB_INIT_PENDING) { + pr_debug("bce-vhci: [%02x] Cancelling URB\n", q->endp_addr); + + spin_unlock_irqrestore(&q->urb_lock, flags); + bce_vhci_transfer_queue_pause(q, BCE_VHCI_PAUSE_INTERNAL_WQ); + spin_lock_irqsave(&q->urb_lock, flags); + + ++q->remaining_active_requests; + } + + usb_hcd_unlink_urb_from_ep(q->vhci->hcd, urb); + + spin_unlock_irqrestore(&q->urb_lock, flags); + + usb_hcd_giveback_urb(q->vhci->hcd, urb, status); + + if (old_state != BCE_VHCI_URB_INIT_PENDING) + bce_vhci_transfer_queue_resume(q, BCE_VHCI_PAUSE_INTERNAL_WQ); + + kfree(vurb); + + return 0; +} + +static int bce_vhci_urb_data_transfer_in(struct bce_vhci_urb *urb, unsigned long *timeout) +{ + struct bce_vhci_message msg; + struct bce_qe_submission *s; + u32 tr_len; + int reservation1, reservation2 = -EFAULT; + + pr_debug("bce-vhci: [%02x] DMA from device %llx %x\n", urb->q->endp_addr, + (u64) urb->urb->transfer_dma, urb->urb->transfer_buffer_length); + + /* Reserve both a message and a submission, so we don't run into issues later. */ + reservation1 = bce_reserve_submission(urb->q->vhci->msg_asynchronous.sq, timeout); + if (!reservation1) + reservation2 = bce_reserve_submission(urb->q->sq_in, timeout); + if (reservation1 || reservation2) { + pr_err("bce-vhci: Failed to reserve a submission for URB data transfer\n"); + if (!reservation1) + bce_cancel_submission_reservation(urb->q->vhci->msg_asynchronous.sq); + return -ENOMEM; + } + + urb->send_offset = urb->receive_offset; + + tr_len = urb->urb->transfer_buffer_length - urb->send_offset; + + spin_lock(&urb->q->vhci->msg_asynchronous_lock); + msg.cmd = BCE_VHCI_CMD_TRANSFER_REQUEST; + msg.status = 0; + msg.param1 = ((urb->urb->ep->desc.bEndpointAddress & 0x8Fu) << 8) | urb->q->dev_addr; + msg.param2 = tr_len; + bce_vhci_message_queue_write(&urb->q->vhci->msg_asynchronous, &msg); + spin_unlock(&urb->q->vhci->msg_asynchronous_lock); + + s = bce_next_submission(urb->q->sq_in); + bce_set_submission_single(s, urb->urb->transfer_dma + urb->send_offset, tr_len); + bce_submit_to_device(urb->q->sq_in); + + urb->state = BCE_VHCI_URB_WAITING_FOR_COMPLETION; + return 0; +} + +static int bce_vhci_urb_data_start(struct bce_vhci_urb *urb, unsigned long *timeout) +{ + if (urb->dir == DMA_TO_DEVICE) { + if (urb->urb->transfer_buffer_length > 0) + urb->state = BCE_VHCI_URB_WAITING_FOR_TRANSFER_REQUEST; + else + urb->state = BCE_VHCI_URB_DATA_TRANSFER_COMPLETE; + return 0; + } else { + return bce_vhci_urb_data_transfer_in(urb, timeout); + } +} + +static int bce_vhci_urb_send_out_data(struct bce_vhci_urb *urb, dma_addr_t addr, size_t size) +{ + struct bce_qe_submission *s; + unsigned long timeout = 0; + if (bce_reserve_submission(urb->q->sq_out, &timeout)) { + pr_err("bce-vhci: Failed to reserve a submission for URB data transfer\n"); + return -EPIPE; + } + + pr_debug("bce-vhci: [%02x] DMA to device %llx %lx\n", urb->q->endp_addr, (u64) addr, size); + + s = bce_next_submission(urb->q->sq_out); + bce_set_submission_single(s, addr, size); + bce_submit_to_device(urb->q->sq_out); + return 0; +} + +static int bce_vhci_urb_data_update(struct bce_vhci_urb *urb, struct bce_vhci_message *msg) +{ + u32 tr_len; + int status; + if (urb->state == BCE_VHCI_URB_WAITING_FOR_TRANSFER_REQUEST) { + if (msg->cmd == BCE_VHCI_CMD_TRANSFER_REQUEST) { + tr_len = min(urb->urb->transfer_buffer_length - urb->send_offset, (u32) msg->param2); + if ((status = bce_vhci_urb_send_out_data(urb, urb->urb->transfer_dma + urb->send_offset, tr_len))) + return status; + urb->send_offset += tr_len; + urb->state = BCE_VHCI_URB_WAITING_FOR_COMPLETION; + return 0; + } + } + + /* 0x1000 in out queues aren't really unexpected */ + if (msg->cmd == BCE_VHCI_CMD_TRANSFER_REQUEST && urb->q->sq_out != NULL) + return -EAGAIN; + pr_err("bce-vhci: [%02x] %s URB unexpected message (state = %x, msg: %x %x %x %llx)\n", + urb->q->endp_addr, (urb->is_control ? "Control (data update)" : "Data"), urb->state, + msg->cmd, msg->status, msg->param1, msg->param2); + return -EAGAIN; +} + +static int bce_vhci_urb_data_transfer_completion(struct bce_vhci_urb *urb, struct bce_sq_completion_data *c) +{ + if (urb->state == BCE_VHCI_URB_WAITING_FOR_COMPLETION) { + urb->receive_offset += c->data_size; + if (urb->dir == DMA_FROM_DEVICE || urb->receive_offset >= urb->urb->transfer_buffer_length) { + urb->urb->actual_length = (u32) urb->receive_offset; + urb->state = BCE_VHCI_URB_DATA_TRANSFER_COMPLETE; + if (!urb->is_control) { + bce_vhci_urb_complete(urb, 0); + return -ENOENT; + } + } + } else { + pr_err("bce-vhci: [%02x] Data URB unexpected completion\n", urb->q->endp_addr); + } + return 0; +} + + +static int bce_vhci_urb_control_check_status(struct bce_vhci_urb *urb) +{ + struct bce_vhci_transfer_queue *q = urb->q; + if (urb->received_status == 0) + return 0; + if (urb->state == BCE_VHCI_URB_DATA_TRANSFER_COMPLETE || + (urb->received_status != BCE_VHCI_SUCCESS && urb->state != BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_REQUEST && + urb->state != BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_COMPLETION)) { + urb->state = BCE_VHCI_URB_CONTROL_COMPLETE; + if (urb->received_status != BCE_VHCI_SUCCESS) { + pr_err("bce-vhci: [%02x] URB failed: %x\n", urb->q->endp_addr, urb->received_status); + urb->q->active = false; + urb->q->stalled = true; + bce_vhci_urb_complete(urb, -EPIPE); + if (!list_empty(&q->endp->urb_list)) + bce_vhci_transfer_queue_request_reset(q); + return -ENOENT; + } + bce_vhci_urb_complete(urb, 0); + return -ENOENT; + } + return 0; +} + +static int bce_vhci_urb_control_update(struct bce_vhci_urb *urb, struct bce_vhci_message *msg) +{ + int status; + if (msg->cmd == BCE_VHCI_CMD_CONTROL_TRANSFER_STATUS) { + urb->received_status = msg->status; + return bce_vhci_urb_control_check_status(urb); + } + + if (urb->state == BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_REQUEST) { + if (msg->cmd == BCE_VHCI_CMD_TRANSFER_REQUEST) { + if (bce_vhci_urb_send_out_data(urb, urb->urb->setup_dma, sizeof(struct usb_ctrlrequest))) { + pr_err("bce-vhci: [%02x] Failed to start URB setup transfer\n", urb->q->endp_addr); + return 0; /* TODO: fail the URB? */ + } + urb->state = BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_COMPLETION; + pr_debug("bce-vhci: [%02x] Sent setup %llx\n", urb->q->endp_addr, urb->urb->setup_dma); + return 0; + } + } else if (urb->state == BCE_VHCI_URB_WAITING_FOR_TRANSFER_REQUEST || + urb->state == BCE_VHCI_URB_WAITING_FOR_COMPLETION) { + if ((status = bce_vhci_urb_data_update(urb, msg))) + return status; + return bce_vhci_urb_control_check_status(urb); + } + + /* 0x1000 in out queues aren't really unexpected */ + if (msg->cmd == BCE_VHCI_CMD_TRANSFER_REQUEST && urb->q->sq_out != NULL) + return -EAGAIN; + pr_err("bce-vhci: [%02x] Control URB unexpected message (state = %x, msg: %x %x %x %llx)\n", urb->q->endp_addr, + urb->state, msg->cmd, msg->status, msg->param1, msg->param2); + return -EAGAIN; +} + +static int bce_vhci_urb_control_transfer_completion(struct bce_vhci_urb *urb, struct bce_sq_completion_data *c) +{ + int status; + unsigned long timeout; + + if (urb->state == BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_COMPLETION) { + if (c->data_size != sizeof(struct usb_ctrlrequest)) + pr_err("bce-vhci: [%02x] transfer complete data size mistmatch for usb_ctrlrequest (%llx instead of %lx)\n", + urb->q->endp_addr, c->data_size, sizeof(struct usb_ctrlrequest)); + + timeout = 1000; + status = bce_vhci_urb_data_start(urb, &timeout); + if (status) { + bce_vhci_urb_complete(urb, status); + return -ENOENT; + } + return 0; + } else if (urb->state == BCE_VHCI_URB_WAITING_FOR_TRANSFER_REQUEST || + urb->state == BCE_VHCI_URB_WAITING_FOR_COMPLETION) { + if ((status = bce_vhci_urb_data_transfer_completion(urb, c))) + return status; + return bce_vhci_urb_control_check_status(urb); + } else { + pr_err("bce-vhci: [%02x] Control URB unexpected completion (state = %x)\n", urb->q->endp_addr, urb->state); + } + return 0; +} + +static int bce_vhci_urb_update(struct bce_vhci_urb *urb, struct bce_vhci_message *msg) +{ + if (urb->state == BCE_VHCI_URB_INIT_PENDING) + return -EAGAIN; + if (urb->is_control) + return bce_vhci_urb_control_update(urb, msg); + else + return bce_vhci_urb_data_update(urb, msg); +} + +static int bce_vhci_urb_transfer_completion(struct bce_vhci_urb *urb, struct bce_sq_completion_data *c) +{ + if (urb->is_control) + return bce_vhci_urb_control_transfer_completion(urb, c); + else + return bce_vhci_urb_data_transfer_completion(urb, c); +} + +static void bce_vhci_urb_resume(struct bce_vhci_urb *urb) +{ + int status = 0; + if (urb->state == BCE_VHCI_URB_WAITING_FOR_COMPLETION) { + status = bce_vhci_urb_data_transfer_in(urb, NULL); + } + if (status) + bce_vhci_urb_complete(urb, status); +} diff --git a/drivers/staging/apple-bce/vhci/transfer.h b/drivers/staging/apple-bce/vhci/transfer.h new file mode 100644 index 000000000000..b5403d5703ed --- /dev/null +++ b/drivers/staging/apple-bce/vhci/transfer.h @@ -0,0 +1,75 @@ +#ifndef BCEDRIVER_TRANSFER_H +#define BCEDRIVER_TRANSFER_H + +#include +#include "queue.h" +#include "command.h" +#include "../queue.h" + +struct bce_vhci_list_message { + struct list_head list; + struct bce_vhci_message msg; +}; +enum bce_vhci_pause_source { + BCE_VHCI_PAUSE_INTERNAL_WQ = 1, + BCE_VHCI_PAUSE_FIRMWARE = 2, + BCE_VHCI_PAUSE_SUSPEND = 4, + BCE_VHCI_PAUSE_SHUTDOWN = 8 +}; +struct bce_vhci_transfer_queue { + struct bce_vhci *vhci; + struct usb_host_endpoint *endp; + enum bce_vhci_endpoint_state state; + u32 max_active_requests, remaining_active_requests; + bool active, stalled; + u32 paused_by; + bce_vhci_device_t dev_addr; + u8 endp_addr; + struct bce_queue_cq *cq; + struct bce_queue_sq *sq_in; + struct bce_queue_sq *sq_out; + struct list_head evq; + struct spinlock urb_lock; + struct mutex pause_lock; + struct list_head giveback_urb_list; + + struct work_struct w_reset; +}; +enum bce_vhci_urb_state { + BCE_VHCI_URB_INIT_PENDING, + + BCE_VHCI_URB_WAITING_FOR_TRANSFER_REQUEST, + BCE_VHCI_URB_WAITING_FOR_COMPLETION, + BCE_VHCI_URB_DATA_TRANSFER_COMPLETE, + + BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_REQUEST, + BCE_VHCI_URB_CONTROL_WAITING_FOR_SETUP_COMPLETION, + BCE_VHCI_URB_CONTROL_COMPLETE, + + BCE_VHCI_URB_CANCELLED +}; +struct bce_vhci_urb { + struct urb *urb; + struct bce_vhci_transfer_queue *q; + enum dma_data_direction dir; + bool is_control; + enum bce_vhci_urb_state state; + int received_status; + u32 send_offset; + u32 receive_offset; +}; + +void bce_vhci_create_transfer_queue(struct bce_vhci *vhci, struct bce_vhci_transfer_queue *q, + struct usb_host_endpoint *endp, bce_vhci_device_t dev_addr, enum dma_data_direction dir); +void bce_vhci_destroy_transfer_queue(struct bce_vhci *vhci, struct bce_vhci_transfer_queue *q); +void bce_vhci_transfer_queue_event(struct bce_vhci_transfer_queue *q, struct bce_vhci_message *msg); +int bce_vhci_transfer_queue_do_pause(struct bce_vhci_transfer_queue *q); +int bce_vhci_transfer_queue_do_resume(struct bce_vhci_transfer_queue *q); +int bce_vhci_transfer_queue_pause(struct bce_vhci_transfer_queue *q, enum bce_vhci_pause_source src); +int bce_vhci_transfer_queue_resume(struct bce_vhci_transfer_queue *q, enum bce_vhci_pause_source src); +void bce_vhci_transfer_queue_request_reset(struct bce_vhci_transfer_queue *q); + +int bce_vhci_urb_create(struct bce_vhci_transfer_queue *q, struct urb *urb); +int bce_vhci_urb_request_cancel(struct bce_vhci_transfer_queue *q, struct urb *urb, int status); + +#endif //BCEDRIVER_TRANSFER_H diff --git a/drivers/staging/apple-bce/vhci/vhci.c b/drivers/staging/apple-bce/vhci/vhci.c new file mode 100644 index 000000000000..3394e0aa2eef --- /dev/null +++ b/drivers/staging/apple-bce/vhci/vhci.c @@ -0,0 +1,763 @@ +#include "vhci.h" +#include "../apple_bce.h" +#include "command.h" +#include +#include +#include +#include + +static dev_t bce_vhci_chrdev; +static struct class *bce_vhci_class; +static const struct hc_driver bce_vhci_driver; +static u16 bce_vhci_port_mask = U16_MAX; + +static int bce_vhci_create_event_queues(struct bce_vhci *vhci); +static void bce_vhci_destroy_event_queues(struct bce_vhci *vhci); +static int bce_vhci_create_message_queues(struct bce_vhci *vhci); +static void bce_vhci_destroy_message_queues(struct bce_vhci *vhci); +static void bce_vhci_handle_firmware_events_w(struct work_struct *ws); +static void bce_vhci_firmware_event_completion(struct bce_queue_sq *sq); + +int bce_vhci_create(struct apple_bce_device *dev, struct bce_vhci *vhci) +{ + int status; + + spin_lock_init(&vhci->hcd_spinlock); + + vhci->dev = dev; + + vhci->vdevt = bce_vhci_chrdev; + vhci->vdev = device_create(bce_vhci_class, dev->dev, vhci->vdevt, NULL, "bce-vhci"); + if (IS_ERR_OR_NULL(vhci->vdev)) { + status = PTR_ERR(vhci->vdev); + goto fail_dev; + } + + if ((status = bce_vhci_create_message_queues(vhci))) + goto fail_mq; + if ((status = bce_vhci_create_event_queues(vhci))) + goto fail_eq; + + vhci->tq_state_wq = alloc_ordered_workqueue("bce-vhci-tq-state", 0); + INIT_WORK(&vhci->w_fw_events, bce_vhci_handle_firmware_events_w); + + vhci->hcd = usb_create_hcd(&bce_vhci_driver, vhci->vdev, "bce-vhci"); + if (!vhci->hcd) { + status = -ENOMEM; + goto fail_hcd; + } + vhci->hcd->self.sysdev = &dev->pci->dev; +#if LINUX_VERSION_CODE < KERNEL_VERSION(5,4,0) + vhci->hcd->self.uses_dma = 1; +#endif + *((struct bce_vhci **) vhci->hcd->hcd_priv) = vhci; + vhci->hcd->speed = HCD_USB2; + + if ((status = usb_add_hcd(vhci->hcd, 0, 0))) + goto fail_hcd; + + return 0; + +fail_hcd: + bce_vhci_destroy_event_queues(vhci); +fail_eq: + bce_vhci_destroy_message_queues(vhci); +fail_mq: + device_destroy(bce_vhci_class, vhci->vdevt); +fail_dev: + if (!status) + status = -EINVAL; + return status; +} + +void bce_vhci_destroy(struct bce_vhci *vhci) +{ + usb_remove_hcd(vhci->hcd); + bce_vhci_destroy_event_queues(vhci); + bce_vhci_destroy_message_queues(vhci); + device_destroy(bce_vhci_class, vhci->vdevt); +} + +struct bce_vhci *bce_vhci_from_hcd(struct usb_hcd *hcd) +{ + return *((struct bce_vhci **) hcd->hcd_priv); +} + +int bce_vhci_start(struct usb_hcd *hcd) +{ + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + int status; + u16 port_mask = 0; + bce_vhci_port_t port_no = 0; + if ((status = bce_vhci_cmd_controller_enable(&vhci->cq, 1, &port_mask))) + return status; + vhci->port_mask = port_mask; + vhci->port_power_mask = 0; + if ((status = bce_vhci_cmd_controller_start(&vhci->cq))) + return status; + port_mask = vhci->port_mask; + while (port_mask) { + port_no += 1; + port_mask >>= 1; + } + vhci->port_count = port_no; + return 0; +} + +void bce_vhci_stop(struct usb_hcd *hcd) +{ + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + bce_vhci_cmd_controller_disable(&vhci->cq); +} + +static int bce_vhci_hub_status_data(struct usb_hcd *hcd, char *buf) +{ + return 0; +} + +static int bce_vhci_reset_device(struct bce_vhci *vhci, int index, u16 timeout); + +static int bce_vhci_hub_control(struct usb_hcd *hcd, u16 typeReq, u16 wValue, u16 wIndex, char *buf, u16 wLength) +{ + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + int status; + struct usb_hub_descriptor *hd; + struct usb_hub_status *hs; + struct usb_port_status *ps; + u32 port_status; + // pr_info("bce-vhci: bce_vhci_hub_control %x %i %i [bufl=%i]\n", typeReq, wValue, wIndex, wLength); + if (typeReq == GetHubDescriptor && wLength >= sizeof(struct usb_hub_descriptor)) { + hd = (struct usb_hub_descriptor *) buf; + memset(hd, 0, sizeof(*hd)); + hd->bDescLength = sizeof(struct usb_hub_descriptor); + hd->bDescriptorType = USB_DT_HUB; + hd->bNbrPorts = (u8) vhci->port_count; + hd->wHubCharacteristics = HUB_CHAR_INDV_PORT_LPSM | HUB_CHAR_INDV_PORT_OCPM; + hd->bPwrOn2PwrGood = 0; + hd->bHubContrCurrent = 0; + return 0; + } else if (typeReq == GetHubStatus && wLength >= sizeof(struct usb_hub_status)) { + hs = (struct usb_hub_status *) buf; + memset(hs, 0, sizeof(*hs)); + hs->wHubStatus = 0; + hs->wHubChange = 0; + return 0; + } else if (typeReq == GetPortStatus && wLength >= 4 /* usb 2.0 */) { + ps = (struct usb_port_status *) buf; + ps->wPortStatus = 0; + ps->wPortChange = 0; + + if (vhci->port_power_mask & BIT(wIndex)) + ps->wPortStatus |= USB_PORT_STAT_POWER; + + if (!(bce_vhci_port_mask & BIT(wIndex))) + return 0; + + if ((status = bce_vhci_cmd_port_status(&vhci->cq, (u8) wIndex, 0, &port_status))) + return status; + + if (port_status & 16) + ps->wPortStatus |= USB_PORT_STAT_ENABLE | USB_PORT_STAT_HIGH_SPEED; + if (port_status & 4) + ps->wPortStatus |= USB_PORT_STAT_CONNECTION; + if (port_status & 2) + ps->wPortStatus |= USB_PORT_STAT_OVERCURRENT; + if (port_status & 8) + ps->wPortStatus |= USB_PORT_STAT_RESET; + if (port_status & 0x60) + ps->wPortStatus |= USB_PORT_STAT_SUSPEND; + + if (port_status & 0x40000) + ps->wPortChange |= USB_PORT_STAT_C_CONNECTION; + + pr_debug("bce-vhci: Translated status %x to %x:%x\n", port_status, ps->wPortStatus, ps->wPortChange); + return 0; + } else if (typeReq == SetPortFeature) { + if (wValue == USB_PORT_FEAT_POWER) { + status = bce_vhci_cmd_port_power_on(&vhci->cq, (u8) wIndex); + /* As far as I am aware, power status is not part of the port status so store it separately */ + if (!status) + vhci->port_power_mask |= BIT(wIndex); + return status; + } + if (wValue == USB_PORT_FEAT_RESET) { + return bce_vhci_reset_device(vhci, wIndex, wValue); + } + if (wValue == USB_PORT_FEAT_SUSPEND) { + /* TODO: Am I supposed to also suspend the endpoints? */ + pr_debug("bce-vhci: Suspending port %i\n", wIndex); + return bce_vhci_cmd_port_suspend(&vhci->cq, (u8) wIndex); + } + } else if (typeReq == ClearPortFeature) { + if (wValue == USB_PORT_FEAT_ENABLE) + return bce_vhci_cmd_port_disable(&vhci->cq, (u8) wIndex); + if (wValue == USB_PORT_FEAT_POWER) { + status = bce_vhci_cmd_port_power_off(&vhci->cq, (u8) wIndex); + if (!status) + vhci->port_power_mask &= ~BIT(wIndex); + return status; + } + if (wValue == USB_PORT_FEAT_C_CONNECTION) + return bce_vhci_cmd_port_status(&vhci->cq, (u8) wIndex, 0x40000, &port_status); + if (wValue == USB_PORT_FEAT_C_RESET) { /* I don't think I can transfer it in any way */ + return 0; + } + if (wValue == USB_PORT_FEAT_SUSPEND) { + pr_debug("bce-vhci: Resuming port %i\n", wIndex); + return bce_vhci_cmd_port_resume(&vhci->cq, (u8) wIndex); + } + } + pr_err("bce-vhci: bce_vhci_hub_control unhandled request: %x %i %i [bufl=%i]\n", typeReq, wValue, wIndex, wLength); + dump_stack(); + return -EIO; +} + +static int bce_vhci_enable_device(struct usb_hcd *hcd, struct usb_device *udev) +{ + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + struct bce_vhci_device *vdev; + bce_vhci_device_t devid; + pr_info("bce_vhci_enable_device\n"); + + if (vhci->port_to_device[udev->portnum]) + return 0; + + /* We need to early address the device */ + if (bce_vhci_cmd_device_create(&vhci->cq, udev->portnum, &devid)) + return -EIO; + + pr_info("bce_vhci_cmd_device_create %i -> %i\n", udev->portnum, devid); + + vdev = kzalloc(sizeof(struct bce_vhci_device), GFP_KERNEL); + vhci->port_to_device[udev->portnum] = devid; + vhci->devices[devid] = vdev; + + bce_vhci_create_transfer_queue(vhci, &vdev->tq[0], &udev->ep0, devid, DMA_BIDIRECTIONAL); + udev->ep0.hcpriv = &vdev->tq[0]; + vdev->tq_mask |= BIT(0); + + bce_vhci_cmd_endpoint_create(&vhci->cq, devid, &udev->ep0.desc); + return 0; +} + +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,8,0) +static int bce_vhci_address_device(struct usb_hcd *hcd, struct usb_device *udev) +#else +static int bce_vhci_address_device(struct usb_hcd *hcd, struct usb_device *udev, unsigned int timeout_ms) //TODO: follow timeout +#endif +{ + /* This is the same as enable_device, but instead in the old scheme */ + return bce_vhci_enable_device(hcd, udev); +} + +static void bce_vhci_free_device(struct usb_hcd *hcd, struct usb_device *udev) +{ + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + int i; + bce_vhci_device_t devid; + struct bce_vhci_device *dev; + pr_info("bce_vhci_free_device %i\n", udev->portnum); + if (!vhci->port_to_device[udev->portnum]) + return; + devid = vhci->port_to_device[udev->portnum]; + dev = vhci->devices[devid]; + for (i = 0; i < 32; i++) { + if (dev->tq_mask & BIT(i)) { + bce_vhci_transfer_queue_pause(&dev->tq[i], BCE_VHCI_PAUSE_SHUTDOWN); + bce_vhci_cmd_endpoint_destroy(&vhci->cq, devid, (u8) i); + bce_vhci_destroy_transfer_queue(vhci, &dev->tq[i]); + } + } + vhci->devices[devid] = NULL; + vhci->port_to_device[udev->portnum] = 0; + bce_vhci_cmd_device_destroy(&vhci->cq, devid); + kfree(dev); +} + +static int bce_vhci_reset_device(struct bce_vhci *vhci, int index, u16 timeout) +{ + struct bce_vhci_device *dev = NULL; + bce_vhci_device_t devid; + int i; + int status; + enum dma_data_direction dir; + pr_info("bce_vhci_reset_device %i\n", index); + + devid = vhci->port_to_device[index]; + if (devid) { + dev = vhci->devices[devid]; + + for (i = 0; i < 32; i++) { + if (dev->tq_mask & BIT(i)) { + bce_vhci_transfer_queue_pause(&dev->tq[i], BCE_VHCI_PAUSE_SHUTDOWN); + bce_vhci_cmd_endpoint_destroy(&vhci->cq, devid, (u8) i); + bce_vhci_destroy_transfer_queue(vhci, &dev->tq[i]); + } + } + vhci->devices[devid] = NULL; + vhci->port_to_device[index] = 0; + bce_vhci_cmd_device_destroy(&vhci->cq, devid); + } + status = bce_vhci_cmd_port_reset(&vhci->cq, (u8) index, timeout); + + if (dev) { + if ((status = bce_vhci_cmd_device_create(&vhci->cq, index, &devid))) + return status; + vhci->devices[devid] = dev; + vhci->port_to_device[index] = devid; + + for (i = 0; i < 32; i++) { + if (dev->tq_mask & BIT(i)) { + dir = usb_endpoint_dir_in(&dev->tq[i].endp->desc) ? DMA_FROM_DEVICE : DMA_TO_DEVICE; + if (i == 0) + dir = DMA_BIDIRECTIONAL; + bce_vhci_create_transfer_queue(vhci, &dev->tq[i], dev->tq[i].endp, devid, dir); + bce_vhci_cmd_endpoint_create(&vhci->cq, devid, &dev->tq[i].endp->desc); + } + } + } + + return status; +} + +static int bce_vhci_check_bandwidth(struct usb_hcd *hcd, struct usb_device *udev) +{ + return 0; +} + +static int bce_vhci_get_frame_number(struct usb_hcd *hcd) +{ + return 0; +} + +static int bce_vhci_bus_suspend(struct usb_hcd *hcd) +{ + int i, j; + int status; + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + pr_info("bce_vhci: suspend started\n"); + + pr_info("bce_vhci: suspend endpoints\n"); + for (i = 0; i < 16; i++) { + if (!vhci->port_to_device[i]) + continue; + for (j = 0; j < 32; j++) { + if (!(vhci->devices[vhci->port_to_device[i]]->tq_mask & BIT(j))) + continue; + bce_vhci_transfer_queue_pause(&vhci->devices[vhci->port_to_device[i]]->tq[j], + BCE_VHCI_PAUSE_SUSPEND); + } + } + + pr_info("bce_vhci: suspend ports\n"); + for (i = 0; i < 16; i++) { + if (!vhci->port_to_device[i]) + continue; + bce_vhci_cmd_port_suspend(&vhci->cq, i); + } + pr_info("bce_vhci: suspend controller\n"); + if ((status = bce_vhci_cmd_controller_pause(&vhci->cq))) + return status; + + bce_vhci_event_queue_pause(&vhci->ev_commands); + bce_vhci_event_queue_pause(&vhci->ev_system); + bce_vhci_event_queue_pause(&vhci->ev_isochronous); + bce_vhci_event_queue_pause(&vhci->ev_interrupt); + bce_vhci_event_queue_pause(&vhci->ev_asynchronous); + pr_info("bce_vhci: suspend done\n"); + return 0; +} + +static int bce_vhci_bus_resume(struct usb_hcd *hcd) +{ + int i, j; + int status; + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + pr_info("bce_vhci: resume started\n"); + + bce_vhci_event_queue_resume(&vhci->ev_system); + bce_vhci_event_queue_resume(&vhci->ev_isochronous); + bce_vhci_event_queue_resume(&vhci->ev_interrupt); + bce_vhci_event_queue_resume(&vhci->ev_asynchronous); + bce_vhci_event_queue_resume(&vhci->ev_commands); + + pr_info("bce_vhci: resume controller\n"); + if ((status = bce_vhci_cmd_controller_start(&vhci->cq))) + return status; + + pr_info("bce_vhci: resume ports\n"); + for (i = 0; i < 16; i++) { + if (!vhci->port_to_device[i]) + continue; + bce_vhci_cmd_port_resume(&vhci->cq, i); + } + pr_info("bce_vhci: resume endpoints\n"); + for (i = 0; i < 16; i++) { + if (!vhci->port_to_device[i]) + continue; + for (j = 0; j < 32; j++) { + if (!(vhci->devices[vhci->port_to_device[i]]->tq_mask & BIT(j))) + continue; + bce_vhci_transfer_queue_resume(&vhci->devices[vhci->port_to_device[i]]->tq[j], + BCE_VHCI_PAUSE_SUSPEND); + } + } + + pr_info("bce_vhci: resume done\n"); + return 0; +} + +static int bce_vhci_urb_enqueue(struct usb_hcd *hcd, struct urb *urb, gfp_t mem_flags) +{ + struct bce_vhci_transfer_queue *q = urb->ep->hcpriv; + pr_debug("bce_vhci_urb_enqueue %i:%x\n", q->dev_addr, urb->ep->desc.bEndpointAddress); + if (!q) + return -ENOENT; + return bce_vhci_urb_create(q, urb); +} + +static int bce_vhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status) +{ + struct bce_vhci_transfer_queue *q = urb->ep->hcpriv; + pr_debug("bce_vhci_urb_dequeue %x\n", urb->ep->desc.bEndpointAddress); + return bce_vhci_urb_request_cancel(q, urb, status); +} + +static void bce_vhci_endpoint_reset(struct usb_hcd *hcd, struct usb_host_endpoint *ep) +{ + struct bce_vhci_transfer_queue *q = ep->hcpriv; + pr_debug("bce_vhci_endpoint_reset\n"); + if (q) + bce_vhci_transfer_queue_request_reset(q); +} + +static u8 bce_vhci_endpoint_index(u8 addr) +{ + if (addr & 0x80) + return (u8) (0x10 + (addr & 0xf)); + return (u8) (addr & 0xf); +} + +static int bce_vhci_add_endpoint(struct usb_hcd *hcd, struct usb_device *udev, struct usb_host_endpoint *endp) +{ + u8 endp_index = bce_vhci_endpoint_index(endp->desc.bEndpointAddress); + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + bce_vhci_device_t devid = vhci->port_to_device[udev->portnum]; + struct bce_vhci_device *vdev = vhci->devices[devid]; + pr_debug("bce_vhci_add_endpoint %x/%x:%x\n", udev->portnum, devid, endp_index); + + if (udev->bus->root_hub == udev) /* The USB hub */ + return 0; + if (vdev == NULL) + return -ENODEV; + if (vdev->tq_mask & BIT(endp_index)) { + endp->hcpriv = &vdev->tq[endp_index]; + return 0; + } + + bce_vhci_create_transfer_queue(vhci, &vdev->tq[endp_index], endp, devid, + usb_endpoint_dir_in(&endp->desc) ? DMA_FROM_DEVICE : DMA_TO_DEVICE); + endp->hcpriv = &vdev->tq[endp_index]; + vdev->tq_mask |= BIT(endp_index); + + bce_vhci_cmd_endpoint_create(&vhci->cq, devid, &endp->desc); + return 0; +} + +static int bce_vhci_drop_endpoint(struct usb_hcd *hcd, struct usb_device *udev, struct usb_host_endpoint *endp) +{ + u8 endp_index = bce_vhci_endpoint_index(endp->desc.bEndpointAddress); + struct bce_vhci *vhci = bce_vhci_from_hcd(hcd); + bce_vhci_device_t devid = vhci->port_to_device[udev->portnum]; + struct bce_vhci_transfer_queue *q = endp->hcpriv; + struct bce_vhci_device *vdev = vhci->devices[devid]; + pr_info("bce_vhci_drop_endpoint %x:%x\n", udev->portnum, endp_index); + if (!q) { + if (vdev && vdev->tq_mask & BIT(endp_index)) { + pr_err("something deleted the hcpriv?\n"); + q = &vdev->tq[endp_index]; + } else { + return 0; + } + } + + bce_vhci_cmd_endpoint_destroy(&vhci->cq, devid, (u8) (endp->desc.bEndpointAddress & 0x8Fu)); + vhci->devices[devid]->tq_mask &= ~BIT(endp_index); + bce_vhci_destroy_transfer_queue(vhci, q); + return 0; +} + +static int bce_vhci_create_message_queues(struct bce_vhci *vhci) +{ + if (bce_vhci_message_queue_create(vhci, &vhci->msg_commands, "VHC1HostCommands") || + bce_vhci_message_queue_create(vhci, &vhci->msg_system, "VHC1HostSystemEvents") || + bce_vhci_message_queue_create(vhci, &vhci->msg_isochronous, "VHC1HostIsochronousEvents") || + bce_vhci_message_queue_create(vhci, &vhci->msg_interrupt, "VHC1HostInterruptEvents") || + bce_vhci_message_queue_create(vhci, &vhci->msg_asynchronous, "VHC1HostAsynchronousEvents")) { + bce_vhci_destroy_message_queues(vhci); + return -EINVAL; + } + spin_lock_init(&vhci->msg_asynchronous_lock); + bce_vhci_command_queue_create(&vhci->cq, &vhci->msg_commands); + return 0; +} + +static void bce_vhci_destroy_message_queues(struct bce_vhci *vhci) +{ + bce_vhci_command_queue_destroy(&vhci->cq); + bce_vhci_message_queue_destroy(vhci, &vhci->msg_commands); + bce_vhci_message_queue_destroy(vhci, &vhci->msg_system); + bce_vhci_message_queue_destroy(vhci, &vhci->msg_isochronous); + bce_vhci_message_queue_destroy(vhci, &vhci->msg_interrupt); + bce_vhci_message_queue_destroy(vhci, &vhci->msg_asynchronous); +} + +static void bce_vhci_handle_system_event(struct bce_vhci_event_queue *q, struct bce_vhci_message *msg); +static void bce_vhci_handle_usb_event(struct bce_vhci_event_queue *q, struct bce_vhci_message *msg); + +static int bce_vhci_create_event_queues(struct bce_vhci *vhci) +{ + vhci->ev_cq = bce_create_cq(vhci->dev, 0x100); + if (!vhci->ev_cq) + return -EINVAL; +#define CREATE_EVENT_QUEUE(field, name, cb) bce_vhci_event_queue_create(vhci, &vhci->field, name, cb) + if (__bce_vhci_event_queue_create(vhci, &vhci->ev_commands, "VHC1FirmwareCommands", + bce_vhci_firmware_event_completion) || + CREATE_EVENT_QUEUE(ev_system, "VHC1FirmwareSystemEvents", bce_vhci_handle_system_event) || + CREATE_EVENT_QUEUE(ev_isochronous, "VHC1FirmwareIsochronousEvents", bce_vhci_handle_usb_event) || + CREATE_EVENT_QUEUE(ev_interrupt, "VHC1FirmwareInterruptEvents", bce_vhci_handle_usb_event) || + CREATE_EVENT_QUEUE(ev_asynchronous, "VHC1FirmwareAsynchronousEvents", bce_vhci_handle_usb_event)) { + bce_vhci_destroy_event_queues(vhci); + return -EINVAL; + } +#undef CREATE_EVENT_QUEUE + return 0; +} + +static void bce_vhci_destroy_event_queues(struct bce_vhci *vhci) +{ + bce_vhci_event_queue_destroy(vhci, &vhci->ev_commands); + bce_vhci_event_queue_destroy(vhci, &vhci->ev_system); + bce_vhci_event_queue_destroy(vhci, &vhci->ev_isochronous); + bce_vhci_event_queue_destroy(vhci, &vhci->ev_interrupt); + bce_vhci_event_queue_destroy(vhci, &vhci->ev_asynchronous); + if (vhci->ev_cq) + bce_destroy_cq(vhci->dev, vhci->ev_cq); +} + +static void bce_vhci_send_fw_event_response(struct bce_vhci *vhci, struct bce_vhci_message *req, u16 status) +{ + unsigned long timeout = 1000; + struct bce_vhci_message r = *req; + r.cmd = (u16) (req->cmd | 0x8000u); + r.status = status; + r.param1 = req->param1; + r.param2 = 0; + + if (bce_reserve_submission(vhci->msg_system.sq, &timeout)) { + pr_err("bce-vhci: Cannot reserve submision for FW event reply\n"); + return; + } + bce_vhci_message_queue_write(&vhci->msg_system, &r); +} + +static int bce_vhci_handle_firmware_event(struct bce_vhci *vhci, struct bce_vhci_message *msg) +{ + unsigned long flags; + bce_vhci_device_t devid; + u8 endp; + struct bce_vhci_device *dev; + struct bce_vhci_transfer_queue *tq; + if (msg->cmd == BCE_VHCI_CMD_ENDPOINT_REQUEST_STATE || msg->cmd == BCE_VHCI_CMD_ENDPOINT_SET_STATE) { + devid = (bce_vhci_device_t) (msg->param1 & 0xff); + endp = bce_vhci_endpoint_index((u8) ((msg->param1 >> 8) & 0xff)); + dev = vhci->devices[devid]; + if (!dev || !(dev->tq_mask & BIT(endp))) + return BCE_VHCI_BAD_ARGUMENT; + tq = &dev->tq[endp]; + } + + if (msg->cmd == BCE_VHCI_CMD_ENDPOINT_REQUEST_STATE) { + if (msg->param2 == BCE_VHCI_ENDPOINT_ACTIVE) { + bce_vhci_transfer_queue_resume(tq, BCE_VHCI_PAUSE_FIRMWARE); + return BCE_VHCI_SUCCESS; + } else if (msg->param2 == BCE_VHCI_ENDPOINT_PAUSED) { + bce_vhci_transfer_queue_pause(tq, BCE_VHCI_PAUSE_FIRMWARE); + return BCE_VHCI_SUCCESS; + } + return BCE_VHCI_BAD_ARGUMENT; + } else if (msg->cmd == BCE_VHCI_CMD_ENDPOINT_SET_STATE) { + if (msg->param2 == BCE_VHCI_ENDPOINT_STALLED) { + tq->state = msg->param2; + spin_lock_irqsave(&tq->urb_lock, flags); + tq->stalled = true; + spin_unlock_irqrestore(&tq->urb_lock, flags); + return BCE_VHCI_SUCCESS; + } + return BCE_VHCI_BAD_ARGUMENT; + } + pr_warn("bce-vhci: Unhandled firmware event: %x s=%x p1=%x p2=%llx\n", + msg->cmd, msg->status, msg->param1, msg->param2); + return BCE_VHCI_BAD_ARGUMENT; +} + +static void bce_vhci_handle_firmware_events_w(struct work_struct *ws) +{ + size_t cnt = 0; + int result; + struct bce_vhci *vhci = container_of(ws, struct bce_vhci, w_fw_events); + struct bce_queue_sq *sq = vhci->ev_commands.sq; + struct bce_sq_completion_data *cq; + struct bce_vhci_message *msg, *msg2 = NULL; + + while (true) { + if (msg2) { + msg = msg2; + msg2 = NULL; + } else if ((cq = bce_next_completion(sq))) { + if (cq->status == BCE_COMPLETION_ABORTED) { + bce_notify_submission_complete(sq); + continue; + } + msg = &vhci->ev_commands.data[sq->head]; + } else { + break; + } + + pr_debug("bce-vhci: Got fw event: %x s=%x p1=%x p2=%llx\n", msg->cmd, msg->status, msg->param1, msg->param2); + if ((cq = bce_next_completion(sq))) { + msg2 = &vhci->ev_commands.data[(sq->head + 1) % sq->el_count]; + pr_debug("bce-vhci: Got second fw event: %x s=%x p1=%x p2=%llx\n", + msg->cmd, msg->status, msg->param1, msg->param2); + if (cq->status != BCE_COMPLETION_ABORTED && + msg2->cmd == (msg->cmd | 0x4000) && msg2->param1 == msg->param1) { + /* Take two elements */ + pr_debug("bce-vhci: Cancelled\n"); + bce_vhci_send_fw_event_response(vhci, msg, BCE_VHCI_ABORT); + + bce_notify_submission_complete(sq); + bce_notify_submission_complete(sq); + msg2 = NULL; + cnt += 2; + continue; + } + + pr_warn("bce-vhci: Handle fw event - unexpected cancellation\n"); + } + + result = bce_vhci_handle_firmware_event(vhci, msg); + bce_vhci_send_fw_event_response(vhci, msg, (u16) result); + + + bce_notify_submission_complete(sq); + ++cnt; + } + bce_vhci_event_queue_submit_pending(&vhci->ev_commands, cnt); + if (atomic_read(&sq->available_commands) == sq->el_count - 1) { + pr_debug("bce-vhci: complete\n"); + complete(&vhci->ev_commands.queue_empty_completion); + } +} + +static void bce_vhci_firmware_event_completion(struct bce_queue_sq *sq) +{ + struct bce_vhci_event_queue *q = sq->userdata; + queue_work(q->vhci->tq_state_wq, &q->vhci->w_fw_events); +} + +static void bce_vhci_handle_system_event(struct bce_vhci_event_queue *q, struct bce_vhci_message *msg) +{ + if (msg->cmd & 0x8000) { + bce_vhci_command_queue_deliver_completion(&q->vhci->cq, msg); + } else { + pr_warn("bce-vhci: Unhandled system event: %x s=%x p1=%x p2=%llx\n", + msg->cmd, msg->status, msg->param1, msg->param2); + } +} + +static void bce_vhci_handle_usb_event(struct bce_vhci_event_queue *q, struct bce_vhci_message *msg) +{ + bce_vhci_device_t devid; + u8 endp; + struct bce_vhci_device *dev; + if (msg->cmd & 0x8000) { + bce_vhci_command_queue_deliver_completion(&q->vhci->cq, msg); + } else if (msg->cmd == BCE_VHCI_CMD_TRANSFER_REQUEST || msg->cmd == BCE_VHCI_CMD_CONTROL_TRANSFER_STATUS) { + devid = (bce_vhci_device_t) (msg->param1 & 0xff); + endp = bce_vhci_endpoint_index((u8) ((msg->param1 >> 8) & 0xff)); + dev = q->vhci->devices[devid]; + if (!dev || (dev->tq_mask & BIT(endp)) == 0) { + pr_err("bce-vhci: Didn't find destination for transfer queue event\n"); + return; + } + bce_vhci_transfer_queue_event(&dev->tq[endp], msg); + } else { + pr_warn("bce-vhci: Unhandled USB event: %x s=%x p1=%x p2=%llx\n", + msg->cmd, msg->status, msg->param1, msg->param2); + } +} + + + +static const struct hc_driver bce_vhci_driver = { + .description = "bce-vhci", + .product_desc = "BCE VHCI Host Controller", + .hcd_priv_size = sizeof(struct bce_vhci *), + +#if LINUX_VERSION_CODE < KERNEL_VERSION(5,4,0) + .flags = HCD_USB2, +#else + .flags = HCD_USB2 | HCD_DMA, +#endif + + .start = bce_vhci_start, + .stop = bce_vhci_stop, + .hub_status_data = bce_vhci_hub_status_data, + .hub_control = bce_vhci_hub_control, + .urb_enqueue = bce_vhci_urb_enqueue, + .urb_dequeue = bce_vhci_urb_dequeue, + .enable_device = bce_vhci_enable_device, + .free_dev = bce_vhci_free_device, + .address_device = bce_vhci_address_device, + .add_endpoint = bce_vhci_add_endpoint, + .drop_endpoint = bce_vhci_drop_endpoint, + .endpoint_reset = bce_vhci_endpoint_reset, + .check_bandwidth = bce_vhci_check_bandwidth, + .get_frame_number = bce_vhci_get_frame_number, + .bus_suspend = bce_vhci_bus_suspend, + .bus_resume = bce_vhci_bus_resume +}; + + +int __init bce_vhci_module_init(void) +{ + int result; + if ((result = alloc_chrdev_region(&bce_vhci_chrdev, 0, 1, "bce-vhci"))) + goto fail_chrdev; +#if LINUX_VERSION_CODE < KERNEL_VERSION(6,4,0) + bce_vhci_class = class_create(THIS_MODULE, "bce-vhci"); +#else + bce_vhci_class = class_create("bce-vhci"); +#endif + if (IS_ERR(bce_vhci_class)) { + result = PTR_ERR(bce_vhci_class); + goto fail_class; + } + return 0; + +fail_class: + class_destroy(bce_vhci_class); +fail_chrdev: + unregister_chrdev_region(bce_vhci_chrdev, 1); + if (!result) + result = -EINVAL; + return result; +} +void __exit bce_vhci_module_exit(void) +{ + class_destroy(bce_vhci_class); + unregister_chrdev_region(bce_vhci_chrdev, 1); +} + +module_param_named(vhci_port_mask, bce_vhci_port_mask, ushort, 0444); +MODULE_PARM_DESC(vhci_port_mask, "Specifies which VHCI ports are enabled"); diff --git a/drivers/staging/apple-bce/vhci/vhci.h b/drivers/staging/apple-bce/vhci/vhci.h new file mode 100644 index 000000000000..6c2e22622f4c --- /dev/null +++ b/drivers/staging/apple-bce/vhci/vhci.h @@ -0,0 +1,52 @@ +#ifndef BCE_VHCI_H +#define BCE_VHCI_H + +#include "queue.h" +#include "transfer.h" + +struct usb_hcd; +struct bce_queue_cq; + +struct bce_vhci_device { + struct bce_vhci_transfer_queue tq[32]; + u32 tq_mask; +}; +struct bce_vhci { + struct apple_bce_device *dev; + dev_t vdevt; + struct device *vdev; + struct usb_hcd *hcd; + struct spinlock hcd_spinlock; + struct bce_vhci_message_queue msg_commands; + struct bce_vhci_message_queue msg_system; + struct bce_vhci_message_queue msg_isochronous; + struct bce_vhci_message_queue msg_interrupt; + struct bce_vhci_message_queue msg_asynchronous; + struct spinlock msg_asynchronous_lock; + struct bce_vhci_command_queue cq; + struct bce_queue_cq *ev_cq; + struct bce_vhci_event_queue ev_commands; + struct bce_vhci_event_queue ev_system; + struct bce_vhci_event_queue ev_isochronous; + struct bce_vhci_event_queue ev_interrupt; + struct bce_vhci_event_queue ev_asynchronous; + u16 port_mask; + u8 port_count; + u16 port_power_mask; + bce_vhci_device_t port_to_device[16]; + struct bce_vhci_device *devices[16]; + struct workqueue_struct *tq_state_wq; + struct work_struct w_fw_events; +}; + +int __init bce_vhci_module_init(void); +void __exit bce_vhci_module_exit(void); + +int bce_vhci_create(struct apple_bce_device *dev, struct bce_vhci *vhci); +void bce_vhci_destroy(struct bce_vhci *vhci); +int bce_vhci_start(struct usb_hcd *hcd); +void bce_vhci_stop(struct usb_hcd *hcd); + +struct bce_vhci *bce_vhci_from_hcd(struct usb_hcd *hcd); + +#endif //BCE_VHCI_H diff --git a/include/linux/hid.h b/include/linux/hid.h index dce862cafbbd..5952d95a53e8 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -625,7 +625,9 @@ struct hid_input { enum hid_type { HID_TYPE_OTHER = 0, HID_TYPE_USBMOUSE, - HID_TYPE_USBNONE + HID_TYPE_USBNONE, + HID_TYPE_SPI_KEYBOARD, + HID_TYPE_SPI_MOUSE, }; enum hid_battery_status { @@ -786,6 +788,8 @@ struct hid_descriptor { .bus = BUS_BLUETOOTH, .vendor = (ven), .product = (prod) #define HID_I2C_DEVICE(ven, prod) \ .bus = BUS_I2C, .vendor = (ven), .product = (prod) +#define HID_SPI_DEVICE(ven, prod) \ + .bus = BUS_SPI, .vendor = (ven), .product = (prod) #define HID_REPORT_ID(rep) \ .report_type = (rep) diff --git a/include/linux/soc/apple/dockchannel.h b/include/linux/soc/apple/dockchannel.h new file mode 100644 index 000000000000..0b7093935ddf --- /dev/null +++ b/include/linux/soc/apple/dockchannel.h @@ -0,0 +1,26 @@ +/* SPDX-License-Identifier: GPL-2.0-only OR MIT */ +/* + * Apple Dockchannel devices + * Copyright (C) The Asahi Linux Contributors + */ +#ifndef _LINUX_APPLE_DOCKCHANNEL_H_ +#define _LINUX_APPLE_DOCKCHANNEL_H_ + +#include +#include +#include + +#if IS_ENABLED(CONFIG_APPLE_DOCKCHANNEL) + +struct dockchannel; + +struct dockchannel *dockchannel_init(struct platform_device *pdev); + +int dockchannel_send(struct dockchannel *dockchannel, const void *buf, size_t count); +int dockchannel_recv(struct dockchannel *dockchannel, void *buf, size_t count); +int dockchannel_await(struct dockchannel *dockchannel, + void (*callback)(void *cookie, size_t avail), + void *cookie, size_t count); + +#endif +#endif -- 2.53.0 From 848f8e594c8111928449319b2af771ca1f612abd Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Mon, 16 Feb 2026 13:51:38 +0100 Subject: [PATCH 6/7] vesa-dsc-bpp Signed-off-by: Peter Jung --- .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 16 +++ drivers/gpu/drm/drm_displayid_internal.h | 11 ++ drivers/gpu/drm/drm_edid.c | 102 +++++++++++------- include/drm/drm_connector.h | 6 ++ include/drm/drm_modes.h | 10 ++ 5 files changed, 109 insertions(+), 36 deletions(-) diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c index 75fa46f18588..8bdcc4ec5e19 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -6776,6 +6776,11 @@ static void fill_stream_properties_from_drm_display_mode( stream->output_color_space = get_output_color_space(timing_out, connector_state); stream->content_type = get_output_content_type(connector_state); + + /* DisplayID Type VII pass-through timings. */ + if (mode_in->dsc_passthrough_timings_support && info->dp_dsc_bpp_x16 != 0) { + stream->timing.dsc_fixed_bits_per_pixel_x16 = info->dp_dsc_bpp_x16; + } } static void fill_audio_info(struct audio_info *audio_info, @@ -7234,6 +7239,7 @@ create_stream_for_sink(struct drm_connector *connector, struct drm_display_mode mode; struct drm_display_mode saved_mode; struct drm_display_mode *freesync_mode = NULL; + struct drm_display_mode *dsc_passthru_mode = NULL; bool native_mode_found = false; bool recalculate_timing = false; bool scale = dm_state->scaling != RMX_OFF; @@ -7325,6 +7331,16 @@ create_stream_for_sink(struct drm_connector *connector, } } + list_for_each_entry(dsc_passthru_mode, &connector->modes, head) { + if (dsc_passthru_mode->hdisplay == mode.hdisplay && + dsc_passthru_mode->vdisplay == mode.vdisplay && + drm_mode_vrefresh(dsc_passthru_mode) == mode_refresh) { + mode.dsc_passthrough_timings_support = + dsc_passthru_mode->dsc_passthrough_timings_support; + break; + } + } + if (recalculate_timing) drm_mode_set_crtcinfo(&saved_mode, 0); diff --git a/drivers/gpu/drm/drm_displayid_internal.h b/drivers/gpu/drm/drm_displayid_internal.h index 5b1b32f73516..8f1a2f33ca1a 100644 --- a/drivers/gpu/drm/drm_displayid_internal.h +++ b/drivers/gpu/drm/drm_displayid_internal.h @@ -97,6 +97,7 @@ struct displayid_header { u8 ext_count; } __packed; +#define DISPLAYID_BLOCK_REV GENMASK(2, 0) struct displayid_block { u8 tag; u8 rev; @@ -125,6 +126,7 @@ struct displayid_detailed_timings_1 { __le16 vsw; } __packed; +#define DISPLAYID_BLOCK_PASSTHROUGH_TIMINGS_SUPPORT BIT(3) struct displayid_detailed_timing_block { struct displayid_block base; struct displayid_detailed_timings_1 timings[]; @@ -137,19 +139,28 @@ struct displayid_formula_timings_9 { u8 vrefresh; } __packed; +#define DISPLAYID_BLOCK_DESCRIPTOR_PAYLOAD_BYTES GENMASK(6, 4) struct displayid_formula_timing_block { struct displayid_block base; struct displayid_formula_timings_9 timings[]; } __packed; +#define DISPLAYID_VESA_DP_TYPE GENMASK(2, 0) #define DISPLAYID_VESA_MSO_OVERLAP GENMASK(3, 0) #define DISPLAYID_VESA_MSO_MODE GENMASK(6, 5) +#define DISPLAYID_VESA_DSC_BPP_INT GENMASK(5, 0) +#define DISPLAYID_VESA_DSC_BPP_FRACT GENMASK(3, 0) + +#define DISPLAYID_VESA_DP_TYPE_EDP 0 +#define DISPLAYID_VESA_DP_TYPE_DP 1 struct displayid_vesa_vendor_specific_block { struct displayid_block base; u8 oui[3]; u8 data_structure_type; u8 mso; + u8 dsc_bpp_int; + u8 dsc_bpp_fract; } __packed; /* diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c index 26bb7710a462..8273920f5ba4 100644 --- a/drivers/gpu/drm/drm_edid.c +++ b/drivers/gpu/drm/drm_edid.c @@ -45,6 +45,7 @@ #include #include #include +#include #include #include "drm_crtc_internal.h" @@ -6527,12 +6528,13 @@ static void drm_get_monitor_range(struct drm_connector *connector, info->monitor_range.min_vfreq, info->monitor_range.max_vfreq); } -static void drm_parse_vesa_mso_data(struct drm_connector *connector, - const struct displayid_block *block) +static void drm_parse_vesa_specific_block(struct drm_connector *connector, + const struct displayid_block *block) { struct displayid_vesa_vendor_specific_block *vesa = (struct displayid_vesa_vendor_specific_block *)block; struct drm_display_info *info = &connector->display_info; + int dp_type; if (block->num_bytes < 3) { drm_dbg_kms(connector->dev, @@ -6544,51 +6546,73 @@ static void drm_parse_vesa_mso_data(struct drm_connector *connector, if (oui(vesa->oui[0], vesa->oui[1], vesa->oui[2]) != VESA_IEEE_OUI) return; - if (sizeof(*vesa) != sizeof(*block) + block->num_bytes) { + if (block->num_bytes < 5) { drm_dbg_kms(connector->dev, "[CONNECTOR:%d:%s] Unexpected VESA vendor block size\n", connector->base.id, connector->name); return; } - switch (FIELD_GET(DISPLAYID_VESA_MSO_MODE, vesa->mso)) { - default: - drm_dbg_kms(connector->dev, "[CONNECTOR:%d:%s] Reserved MSO mode value\n", + dp_type = FIELD_GET(DISPLAYID_VESA_DP_TYPE, vesa->data_structure_type); + if (dp_type > 1) { + drm_dbg_kms(connector->dev, "[CONNECTOR:%d:%s] Reserved dp type value\n", connector->base.id, connector->name); - fallthrough; - case 0: - info->mso_stream_count = 0; - break; - case 1: - info->mso_stream_count = 2; /* 2 or 4 links */ - break; - case 2: - info->mso_stream_count = 4; /* 4 links */ - break; } - if (!info->mso_stream_count) { + /* MSO is only supported for eDP */ + if (dp_type == DISPLAYID_VESA_DP_TYPE_EDP) { + switch (FIELD_GET(DISPLAYID_VESA_MSO_MODE, vesa->mso)) { + default: + drm_dbg_kms(connector->dev, "[CONNECTOR:%d:%s] Reserved MSO mode value\n", + connector->base.id, connector->name); + fallthrough; + case 0: + info->mso_stream_count = 0; + break; + case 1: + info->mso_stream_count = 2; /* 2 or 4 links */ + break; + case 2: + info->mso_stream_count = 4; /* 4 links */ + break; + } + } + + if (info->mso_stream_count) { + info->mso_pixel_overlap = FIELD_GET(DISPLAYID_VESA_MSO_OVERLAP, vesa->mso); + if (info->mso_pixel_overlap > 8) { + drm_dbg_kms(connector->dev, + "[CONNECTOR:%d:%s] Reserved MSO pixel overlap value %u\n", + connector->base.id, connector->name, + info->mso_pixel_overlap); + info->mso_pixel_overlap = 8; + } + drm_dbg_kms(connector->dev, + "[CONNECTOR:%d:%s] MSO stream count %u, pixel overlap %u\n", + connector->base.id, connector->name, + info->mso_stream_count, info->mso_pixel_overlap); + } else { info->mso_pixel_overlap = 0; + } + + if (block->num_bytes < 7) { + /* DSC bpp is optional */ return; } - info->mso_pixel_overlap = FIELD_GET(DISPLAYID_VESA_MSO_OVERLAP, vesa->mso); - if (info->mso_pixel_overlap > 8) { + info->dp_dsc_bpp_x16 = FIELD_GET(DISPLAYID_VESA_DSC_BPP_INT, vesa->dsc_bpp_int) << 4 | + FIELD_GET(DISPLAYID_VESA_DSC_BPP_FRACT, vesa->dsc_bpp_fract); + + if (info->dp_dsc_bpp_x16 > 0) { drm_dbg_kms(connector->dev, - "[CONNECTOR:%d:%s] Reserved MSO pixel overlap value %u\n", + "[CONNECTOR:%d:%s] DSC bits per pixel " FXP_Q4_FMT "\n", connector->base.id, connector->name, - info->mso_pixel_overlap); - info->mso_pixel_overlap = 8; + FXP_Q4_ARGS(info->dp_dsc_bpp_x16)); } - - drm_dbg_kms(connector->dev, - "[CONNECTOR:%d:%s] MSO stream count %u, pixel overlap %u\n", - connector->base.id, connector->name, - info->mso_stream_count, info->mso_pixel_overlap); } -static void drm_update_mso(struct drm_connector *connector, - const struct drm_edid *drm_edid) +static void drm_update_vesa_specific_block(struct drm_connector *connector, + const struct drm_edid *drm_edid) { const struct displayid_block *block; struct displayid_iter iter; @@ -6596,7 +6620,7 @@ static void drm_update_mso(struct drm_connector *connector, displayid_iter_edid_begin(drm_edid, &iter); displayid_iter_for_each(block, &iter) { if (block->tag == DATA_BLOCK_2_VENDOR_SPECIFIC) - drm_parse_vesa_mso_data(connector, block); + drm_parse_vesa_specific_block(connector, block); } displayid_iter_end(&iter); } @@ -6633,6 +6657,7 @@ static void drm_reset_display_info(struct drm_connector *connector) info->mso_stream_count = 0; info->mso_pixel_overlap = 0; info->max_dsc_bpp = 0; + info->dp_dsc_bpp_x16 = 0; kfree(info->vics); info->vics = NULL; @@ -6756,7 +6781,7 @@ static void update_display_info(struct drm_connector *connector, if (edid->features & DRM_EDID_FEATURE_RGB_YCRCB422) info->color_formats |= DRM_COLOR_FORMAT_YCBCR422; - drm_update_mso(connector, drm_edid); + drm_update_vesa_specific_block(connector, drm_edid); out: if (drm_edid_has_internal_quirk(connector, EDID_QUIRK_NON_DESKTOP)) { @@ -6786,8 +6811,8 @@ static void update_display_info(struct drm_connector *connector, } static struct drm_display_mode *drm_mode_displayid_detailed(struct drm_device *dev, - const struct displayid_detailed_timings_1 *timings, - bool type_7) + const struct displayid_block *block, + const struct displayid_detailed_timings_1 *timings) { struct drm_display_mode *mode; unsigned int pixel_clock = (timings->pixel_clock[0] | @@ -6803,11 +6828,16 @@ static struct drm_display_mode *drm_mode_displayid_detailed(struct drm_device *d unsigned int vsync_width = le16_to_cpu(timings->vsw) + 1; bool hsync_positive = le16_to_cpu(timings->hsync) & (1 << 15); bool vsync_positive = le16_to_cpu(timings->vsync) & (1 << 15); + bool type_7 = block->tag == DATA_BLOCK_2_TYPE_7_DETAILED_TIMING; mode = drm_mode_create(dev); if (!mode) return NULL; + if (type_7 && FIELD_GET(DISPLAYID_BLOCK_REV, block->rev) >= 1) + mode->dsc_passthrough_timings_support = + block->rev & DISPLAYID_BLOCK_PASSTHROUGH_TIMINGS_SUPPORT; + /* resolution is kHz for type VII, and 10 kHz for type I */ mode->clock = type_7 ? pixel_clock : pixel_clock * 10; mode->hdisplay = hactive; @@ -6840,7 +6870,6 @@ static int add_displayid_detailed_1_modes(struct drm_connector *connector, int num_timings; struct drm_display_mode *newmode; int num_modes = 0; - bool type_7 = block->tag == DATA_BLOCK_2_TYPE_7_DETAILED_TIMING; /* blocks must be multiple of 20 bytes length */ if (block->num_bytes % 20) return 0; @@ -6849,7 +6878,7 @@ static int add_displayid_detailed_1_modes(struct drm_connector *connector, for (i = 0; i < num_timings; i++) { struct displayid_detailed_timings_1 *timings = &det->timings[i]; - newmode = drm_mode_displayid_detailed(connector->dev, timings, type_7); + newmode = drm_mode_displayid_detailed(connector->dev, block, timings); if (!newmode) continue; @@ -6896,7 +6925,8 @@ static int add_displayid_formula_modes(struct drm_connector *connector, struct drm_display_mode *newmode; int num_modes = 0; bool type_10 = block->tag == DATA_BLOCK_2_TYPE_10_FORMULA_TIMING; - int timing_size = 6 + ((formula_block->base.rev & 0x70) >> 4); + int timing_size = 6 + + FIELD_GET(DISPLAYID_BLOCK_DESCRIPTOR_PAYLOAD_BYTES, formula_block->base.rev); /* extended blocks are not supported yet */ if (timing_size != 6) diff --git a/include/drm/drm_connector.h b/include/drm/drm_connector.h index 8f34f4b8183d..7decfc288aa3 100644 --- a/include/drm/drm_connector.h +++ b/include/drm/drm_connector.h @@ -837,6 +837,12 @@ struct drm_display_info { */ u32 max_dsc_bpp; + /** + * @dp_dsc_bpp: DP Display-Stream-Compression (DSC) timing's target + * DSC bits per pixel in 6.4 fixed point format. 0 means undefined. + */ + u16 dp_dsc_bpp_x16; + /** * @vics: Array of vics_len VICs. Internal to EDID parsing. */ diff --git a/include/drm/drm_modes.h b/include/drm/drm_modes.h index b9bb92e4b029..312e5c03af9a 100644 --- a/include/drm/drm_modes.h +++ b/include/drm/drm_modes.h @@ -417,6 +417,16 @@ struct drm_display_mode { */ enum hdmi_picture_aspect picture_aspect_ratio; + /** + * @dsc_passthrough_timing_support: + * + * Indicates whether this mode timing descriptor is supported + * with specific target DSC bits per pixel only. + * + * VESA vendor-specific data block shall exist with the relevant + * DSC bits per pixel declaration when this flag is set to true. + */ + bool dsc_passthrough_timings_support; }; /** -- 2.53.0 From 8a06528807a7d8c83409a9bf276304429665c7ad Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Mon, 16 Feb 2026 13:51:49 +0100 Subject: [PATCH 7/7] vmscape Signed-off-by: Peter Jung --- Documentation/admin-guide/hw-vuln/vmscape.rst | 8 +++ .../admin-guide/kernel-parameters.txt | 4 +- arch/x86/Kconfig | 1 + arch/x86/entry/entry_64.S | 13 +++- arch/x86/include/asm/cpufeatures.h | 2 +- arch/x86/include/asm/entry-common.h | 9 ++- arch/x86/include/asm/nospec-branch.h | 11 +++- arch/x86/kernel/cpu/bugs.c | 65 +++++++++++++++---- arch/x86/kvm/x86.c | 4 +- arch/x86/net/bpf_jit_comp.c | 2 + 10 files changed, 90 insertions(+), 29 deletions(-) diff --git a/Documentation/admin-guide/hw-vuln/vmscape.rst b/Documentation/admin-guide/hw-vuln/vmscape.rst index d9b9a2b6c114..580f288ae8bf 100644 --- a/Documentation/admin-guide/hw-vuln/vmscape.rst +++ b/Documentation/admin-guide/hw-vuln/vmscape.rst @@ -86,6 +86,10 @@ The possible values in this file are: run a potentially malicious guest and issues an IBPB before the first exit to userspace after VM-exit. + * 'Mitigation: Clear BHB before exit to userspace': + + As above, conditional BHB clearing mitigation is enabled. + * 'Mitigation: IBPB on VMEXIT': IBPB is issued on every VM-exit. This occurs when other mitigations like @@ -108,3 +112,7 @@ The mitigation can be controlled via the ``vmscape=`` command line parameter: Force vulnerability detection and mitigation even on processors that are not known to be affected. + + * ``vmscape=on``: + + Choose the mitigation based on the VMSCAPE variant the CPU is affected by. diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt index 28f14d664aa3..a3e9684f63c0 100644 --- a/Documentation/admin-guide/kernel-parameters.txt +++ b/Documentation/admin-guide/kernel-parameters.txt @@ -8281,9 +8281,11 @@ Kernel parameters off - disable the mitigation ibpb - use Indirect Branch Prediction Barrier - (IBPB) mitigation (default) + (IBPB) mitigation force - force vulnerability detection even on unaffected processors + on - (default) selects IBPB or BHB clear + mitigation based on CPU vsyscall= [X86-64,EARLY] Controls the behavior of vsyscalls (i.e. calls to diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index 80527299f859..e03e35a2a6ce 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -2710,6 +2710,7 @@ config MITIGATION_TSA config MITIGATION_VMSCAPE bool "Mitigate VMSCAPE" depends on KVM + depends on HAVE_STATIC_CALL default y help Enable mitigation for VMSCAPE attacks. VMSCAPE is a hardware security diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S index f9983a1907bf..6d93602dd309 100644 --- a/arch/x86/entry/entry_64.S +++ b/arch/x86/entry/entry_64.S @@ -1529,12 +1529,19 @@ SYM_CODE_END(rewind_stack_and_make_dead) * refactored in the future if needed. The .skips are for safety, to ensure * that all RETs are in the second half of a cacheline to mitigate Indirect * Target Selection, rather than taking the slowpath via its_return_thunk. + * + * Note, callers should use a speculation barrier like LFENCE immediately after + * a call to this function to ensure BHB is cleared before indirect branches. */ SYM_FUNC_START(clear_bhb_loop) ANNOTATE_NOENDBR push %rbp mov %rsp, %rbp - movl $5, %ecx + + /* loop count differs based on BHI_CTRL, see Intel's BHI guidance */ + ALTERNATIVE "movl $5, %ecx; movl $5, %edx", \ + "movl $12, %ecx; movl $7, %edx", X86_FEATURE_BHI_CTRL + ANNOTATE_INTRA_FUNCTION_CALL call 1f jmp 5f @@ -1555,7 +1562,7 @@ SYM_FUNC_START(clear_bhb_loop) * but some Clang versions (e.g. 18) don't like this. */ .skip 32 - 18, 0xcc -2: movl $5, %eax +2: movl %edx, %eax 3: jmp 4f nop 4: sub $1, %eax @@ -1563,7 +1570,7 @@ SYM_FUNC_START(clear_bhb_loop) sub $1, %ecx jnz 1b .Lret2: RET -5: lfence +5: pop %rbp RET SYM_FUNC_END(clear_bhb_loop) diff --git a/arch/x86/include/asm/cpufeatures.h b/arch/x86/include/asm/cpufeatures.h index c3b53beb1300..aa39430476d6 100644 --- a/arch/x86/include/asm/cpufeatures.h +++ b/arch/x86/include/asm/cpufeatures.h @@ -501,7 +501,7 @@ #define X86_FEATURE_TSA_SQ_NO (21*32+11) /* AMD CPU not vulnerable to TSA-SQ */ #define X86_FEATURE_TSA_L1_NO (21*32+12) /* AMD CPU not vulnerable to TSA-L1 */ #define X86_FEATURE_CLEAR_CPU_BUF_VM (21*32+13) /* Clear CPU buffers using VERW before VMRUN */ -#define X86_FEATURE_IBPB_EXIT_TO_USER (21*32+14) /* Use IBPB on exit-to-userspace, see VMSCAPE bug */ +/* Free */ #define X86_FEATURE_ABMC (21*32+15) /* Assignable Bandwidth Monitoring Counters */ #define X86_FEATURE_MSR_IMM (21*32+16) /* MSR immediate form instructions */ #define X86_FEATURE_SGX_EUPDATESVN (21*32+17) /* Support for ENCLS[EUPDATESVN] instruction */ diff --git a/arch/x86/include/asm/entry-common.h b/arch/x86/include/asm/entry-common.h index ce3eb6d5fdf9..783e7cb50cae 100644 --- a/arch/x86/include/asm/entry-common.h +++ b/arch/x86/include/asm/entry-common.h @@ -4,6 +4,7 @@ #include #include +#include #include #include @@ -94,11 +95,9 @@ static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs, */ choose_random_kstack_offset(rdtsc()); - /* Avoid unnecessary reads of 'x86_ibpb_exit_to_user' */ - if (cpu_feature_enabled(X86_FEATURE_IBPB_EXIT_TO_USER) && - this_cpu_read(x86_ibpb_exit_to_user)) { - indirect_branch_prediction_barrier(); - this_cpu_write(x86_ibpb_exit_to_user, false); + if (unlikely(this_cpu_read(x86_predictor_flush_exit_to_user))) { + static_call_cond(vmscape_predictor_flush)(); + this_cpu_write(x86_predictor_flush_exit_to_user, false); } } #define arch_exit_to_user_mode_prepare arch_exit_to_user_mode_prepare diff --git a/arch/x86/include/asm/nospec-branch.h b/arch/x86/include/asm/nospec-branch.h index 4f4b5e8a1574..80efdb6645ba 100644 --- a/arch/x86/include/asm/nospec-branch.h +++ b/arch/x86/include/asm/nospec-branch.h @@ -331,11 +331,11 @@ #ifdef CONFIG_X86_64 .macro CLEAR_BRANCH_HISTORY - ALTERNATIVE "", "call clear_bhb_loop", X86_FEATURE_CLEAR_BHB_LOOP + ALTERNATIVE "", "call clear_bhb_loop; lfence", X86_FEATURE_CLEAR_BHB_LOOP .endm .macro CLEAR_BRANCH_HISTORY_VMEXIT - ALTERNATIVE "", "call clear_bhb_loop", X86_FEATURE_CLEAR_BHB_VMEXIT + ALTERNATIVE "", "call clear_bhb_loop; lfence", X86_FEATURE_CLEAR_BHB_VMEXIT .endm #else #define CLEAR_BRANCH_HISTORY @@ -390,6 +390,8 @@ extern void write_ibpb(void); #ifdef CONFIG_X86_64 extern void clear_bhb_loop(void); +#else +static inline void clear_bhb_loop(void) {} #endif extern void (*x86_return_thunk)(void); @@ -533,7 +535,7 @@ void alternative_msr_write(unsigned int msr, u64 val, unsigned int feature) : "memory"); } -DECLARE_PER_CPU(bool, x86_ibpb_exit_to_user); +DECLARE_PER_CPU(bool, x86_predictor_flush_exit_to_user); static inline void indirect_branch_prediction_barrier(void) { @@ -542,6 +544,9 @@ static inline void indirect_branch_prediction_barrier(void) :: "rax", "rcx", "rdx", "memory"); } +#include +DECLARE_STATIC_CALL(vmscape_predictor_flush, write_ibpb); + /* The Intel SPEC CTRL MSR base value cache */ extern u64 x86_spec_ctrl_base; DECLARE_PER_CPU(u64, x86_spec_ctrl_current); diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c index d0a2847a4bb0..2818bfcb9f9f 100644 --- a/arch/x86/kernel/cpu/bugs.c +++ b/arch/x86/kernel/cpu/bugs.c @@ -62,12 +62,11 @@ DEFINE_PER_CPU(u64, x86_spec_ctrl_current); EXPORT_PER_CPU_SYMBOL_GPL(x86_spec_ctrl_current); /* - * Set when the CPU has run a potentially malicious guest. An IBPB will - * be needed to before running userspace. That IBPB will flush the branch - * predictor content. + * Set when the CPU has run a potentially malicious guest. Indicates that a + * branch predictor flush is needed before running userspace. */ -DEFINE_PER_CPU(bool, x86_ibpb_exit_to_user); -EXPORT_PER_CPU_SYMBOL_GPL(x86_ibpb_exit_to_user); +DEFINE_PER_CPU(bool, x86_predictor_flush_exit_to_user); +EXPORT_PER_CPU_SYMBOL_GPL(x86_predictor_flush_exit_to_user); u64 x86_pred_cmd __ro_after_init = PRED_CMD_IBPB; @@ -230,6 +229,9 @@ static void x86_amd_ssb_disable(void) wrmsrq(MSR_AMD64_LS_CFG, msrval); } +DEFINE_STATIC_CALL_NULL(vmscape_predictor_flush, write_ibpb); +EXPORT_STATIC_CALL_GPL(vmscape_predictor_flush); + #undef pr_fmt #define pr_fmt(fmt) "MDS: " fmt @@ -3049,15 +3051,19 @@ static void __init srso_apply_mitigation(void) enum vmscape_mitigations { VMSCAPE_MITIGATION_NONE, VMSCAPE_MITIGATION_AUTO, + VMSCAPE_MITIGATION_ON, VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER, VMSCAPE_MITIGATION_IBPB_ON_VMEXIT, + VMSCAPE_MITIGATION_BHB_CLEAR_EXIT_TO_USER, }; static const char * const vmscape_strings[] = { - [VMSCAPE_MITIGATION_NONE] = "Vulnerable", + [VMSCAPE_MITIGATION_NONE] = "Vulnerable", /* [VMSCAPE_MITIGATION_AUTO] */ - [VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER] = "Mitigation: IBPB before exit to userspace", - [VMSCAPE_MITIGATION_IBPB_ON_VMEXIT] = "Mitigation: IBPB on VMEXIT", + /* [VMSCAPE_MITIGATION_ON] */ + [VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER] = "Mitigation: IBPB before exit to userspace", + [VMSCAPE_MITIGATION_IBPB_ON_VMEXIT] = "Mitigation: IBPB on VMEXIT", + [VMSCAPE_MITIGATION_BHB_CLEAR_EXIT_TO_USER] = "Mitigation: Clear BHB before exit to userspace", }; static enum vmscape_mitigations vmscape_mitigation __ro_after_init = @@ -3074,7 +3080,9 @@ static int __init vmscape_parse_cmdline(char *str) vmscape_mitigation = VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER; } else if (!strcmp(str, "force")) { setup_force_cpu_bug(X86_BUG_VMSCAPE); - vmscape_mitigation = VMSCAPE_MITIGATION_AUTO; + vmscape_mitigation = VMSCAPE_MITIGATION_ON; + } else if (!strcmp(str, "on")) { + vmscape_mitigation = VMSCAPE_MITIGATION_ON; } else { pr_err("Ignoring unknown vmscape=%s option.\n", str); } @@ -3085,17 +3093,42 @@ early_param("vmscape", vmscape_parse_cmdline); static void __init vmscape_select_mitigation(void) { - if (!boot_cpu_has_bug(X86_BUG_VMSCAPE) || - !boot_cpu_has(X86_FEATURE_IBPB)) { + if (!boot_cpu_has_bug(X86_BUG_VMSCAPE)) { vmscape_mitigation = VMSCAPE_MITIGATION_NONE; return; } - if (vmscape_mitigation == VMSCAPE_MITIGATION_AUTO) { - if (should_mitigate_vuln(X86_BUG_VMSCAPE)) + if ((vmscape_mitigation == VMSCAPE_MITIGATION_AUTO) && + !should_mitigate_vuln(X86_BUG_VMSCAPE)) + vmscape_mitigation = VMSCAPE_MITIGATION_NONE; + + switch (vmscape_mitigation) { + case VMSCAPE_MITIGATION_NONE: + break; + + case VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER: + if (!boot_cpu_has(X86_FEATURE_IBPB)) + vmscape_mitigation = VMSCAPE_MITIGATION_NONE; + break; + + case VMSCAPE_MITIGATION_AUTO: + case VMSCAPE_MITIGATION_ON: + /* + * CPUs with BHI_CTRL(ADL and newer) can avoid the IBPB and use BHB + * clear sequence. These CPUs are only vulnerable to the BHI variant + * of the VMSCAPE attack and does not require an IBPB flush. In + * 32-bit mode BHB clear sequence is not supported. + */ + if (boot_cpu_has(X86_FEATURE_BHI_CTRL) && IS_ENABLED(CONFIG_X86_64)) + vmscape_mitigation = VMSCAPE_MITIGATION_BHB_CLEAR_EXIT_TO_USER; + else if (boot_cpu_has(X86_FEATURE_IBPB)) vmscape_mitigation = VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER; else vmscape_mitigation = VMSCAPE_MITIGATION_NONE; + break; + + default: + break; } } @@ -3114,7 +3147,9 @@ static void __init vmscape_update_mitigation(void) static void __init vmscape_apply_mitigation(void) { if (vmscape_mitigation == VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER) - setup_force_cpu_cap(X86_FEATURE_IBPB_EXIT_TO_USER); + static_call_update(vmscape_predictor_flush, write_ibpb); + else if (vmscape_mitigation == VMSCAPE_MITIGATION_BHB_CLEAR_EXIT_TO_USER) + static_call_update(vmscape_predictor_flush, clear_bhb_loop); } #undef pr_fmt @@ -3203,9 +3238,11 @@ void cpu_bugs_smt_update(void) switch (vmscape_mitigation) { case VMSCAPE_MITIGATION_NONE: case VMSCAPE_MITIGATION_AUTO: + case VMSCAPE_MITIGATION_ON: break; case VMSCAPE_MITIGATION_IBPB_ON_VMEXIT: case VMSCAPE_MITIGATION_IBPB_EXIT_TO_USER: + case VMSCAPE_MITIGATION_BHB_CLEAR_EXIT_TO_USER: /* * Hypervisors can be attacked across-threads, warn for SMT when * STIBP is not already enabled system-wide. diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 72d37c8930ad..5b4d44a6b702 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -11437,8 +11437,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu) * set for the CPU that actually ran the guest, and not the CPU that it * may migrate to. */ - if (cpu_feature_enabled(X86_FEATURE_IBPB_EXIT_TO_USER)) - this_cpu_write(x86_ibpb_exit_to_user, true); + if (static_call_query(vmscape_predictor_flush)) + this_cpu_write(x86_predictor_flush_exit_to_user, true); /* * Consume any pending interrupts, including the possible source of diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index b0bac2a66eff..c31508be0d72 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -1620,6 +1620,8 @@ static int emit_spectre_bhb_barrier(u8 **pprog, u8 *ip, if (emit_call(&prog, func, ip)) return -EINVAL; + /* Don't speculate past this until BHB is cleared */ + EMIT_LFENCE(); EMIT1(0x59); /* pop rcx */ EMIT1(0x58); /* pop rax */ } -- 2.53.0