From 473286f8def327c647b173753bd3fbb8b3a5f000 Mon Sep 17 00:00:00 2001 From: Peter Jung Date: Fri, 27 Jun 2025 13:05:28 +0200 Subject: [PATCH 5/7] cachy Signed-off-by: Peter Jung --- .gitignore | 3 + .../ABI/testing/sysfs-driver-intel-xe-hwmon | 24 + .../admin-guide/kernel-parameters.txt | 12 + Documentation/admin-guide/sysctl/vm.rst | 72 + MAINTAINERS | 5 + Makefile | 48 +- arch/Kconfig | 19 + arch/x86/Kconfig.cpu | 70 + arch/x86/Makefile | 17 + arch/x86/include/asm/pci.h | 6 + arch/x86/pci/common.c | 7 +- block/Kconfig.iosched | 14 + block/Makefile | 8 + block/adios.c | 1382 +++++++ block/elevator.c | 12 + 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 | 14 +- drivers/gpu/drm/xe/regs/xe_pcode_regs.h | 3 + drivers/gpu/drm/xe/xe_device_types.h | 2 + drivers/gpu/drm/xe/xe_hwmon.c | 125 +- drivers/gpu/drm/xe/xe_pci.c | 4 + drivers/gpu/drm/xe/xe_pcode_api.h | 3 + drivers/input/evdev.c | 19 +- drivers/md/dm-crypt.c | 5 + drivers/media/v4l2-core/Kconfig | 5 + drivers/media/v4l2-core/Makefile | 2 + drivers/media/v4l2-core/v4l2loopback.c | 3313 +++++++++++++++++ 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 ++++++ fs/select.c | 2 +- include/linux/mm.h | 8 + include/linux/pagemap.h | 2 +- include/linux/user_namespace.h | 4 + include/linux/wait.h | 2 + init/Kconfig | 26 + kernel/Kconfig.hz | 24 + kernel/Kconfig.preempt | 2 +- kernel/fork.c | 14 + kernel/locking/rwsem.c | 4 +- kernel/sched/fair.c | 13 + kernel/sched/sched.h | 2 +- kernel/sched/wait.c | 24 + kernel/sysctl.c | 12 + kernel/user_namespace.c | 7 + 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/vmpressure.c | 4 + mm/vmscan.c | 157 +- net/ipv4/inet_connection_sock.c | 2 +- scripts/Makefile.build | 52 +- scripts/Makefile.lib | 7 +- scripts/Makefile.vmlinux_o | 16 +- scripts/Makefile.vmlinux_thinlink | 53 + scripts/head-object-list.txt | 1 + 79 files changed, 8035 insertions(+), 68 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.vmlinux_thinlink diff --git a/.gitignore b/.gitignore index f2f63e47fb88..b83a68185ef4 100644 --- a/.gitignore +++ b/.gitignore @@ -12,6 +12,7 @@ # .* *.a +*.a_thinlto_native *.asn1.[ch] *.bin *.bz2 @@ -39,6 +40,7 @@ *.mod.c *.o *.o.* +*.o_thinlto_native *.patch *.rmeta *.rpm @@ -64,6 +66,7 @@ modules.order /vmlinux /vmlinux.32 /vmlinux.map +/vmlinux.thinlink /vmlinux.symvers /vmlinux.unstripped /vmlinux-gdb.py diff --git a/Documentation/ABI/testing/sysfs-driver-intel-xe-hwmon b/Documentation/ABI/testing/sysfs-driver-intel-xe-hwmon index cb207c79680d..6fbab98fb639 100644 --- a/Documentation/ABI/testing/sysfs-driver-intel-xe-hwmon +++ b/Documentation/ABI/testing/sysfs-driver-intel-xe-hwmon @@ -124,3 +124,27 @@ Contact: intel-xe@lists.freedesktop.org Description: RO. VRAM temperature in millidegree Celsius. Only supported for particular Intel Xe graphics platforms. + +What: /sys/bus/pci/drivers/xe/.../hwmon/hwmon/fan1_input +Date: March 2025 +KernelVersion: 6.14 +Contact: intel-xe@lists.freedesktop.org +Description: RO. Fan 1 speed in RPM. + + Only supported for particular Intel Xe graphics platforms. + +What: /sys/bus/pci/drivers/xe/.../hwmon/hwmon/fan2_input +Date: March 2025 +KernelVersion: 6.14 +Contact: intel-xe@lists.freedesktop.org +Description: RO. Fan 2 speed in RPM. + + Only supported for particular Intel Xe graphics platforms. + +What: /sys/bus/pci/drivers/xe/.../hwmon/hwmon/fan3_input +Date: March 2025 +KernelVersion: 6.14 +Contact: intel-xe@lists.freedesktop.org +Description: RO. Fan 3 speed in RPM. + + Only supported for particular Intel Xe graphics platforms. diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt index 8f75ec177399..c865280efa8e 100644 --- a/Documentation/admin-guide/kernel-parameters.txt +++ b/Documentation/admin-guide/kernel-parameters.txt @@ -2318,6 +2318,9 @@ 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 @@ -4689,6 +4692,15 @@ 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 8290177b4f75..2a94faa5a4ca 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 @@ -109,6 +112,67 @@ 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 ============== @@ -973,6 +1037,14 @@ 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/MAINTAINERS b/MAINTAINERS index dd844ac8d910..4feb3b9f8f73 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -5790,6 +5790,11 @@ F: scripts/Makefile.clang F: scripts/clang-tools/ K: \b(?i:clang|llvm)\b +CLANG/LLVM THINLTO DISTRIBUTED BUILD +M: Rong Xu +S: Supported +F: scripts/Makefile.vmlinux_thinlink + CLK API M: Russell King L: linux-clk@vger.kernel.org diff --git a/Makefile b/Makefile index c1bde4eef2bf..5dca2a7b2960 100644 --- a/Makefile +++ b/Makefile @@ -298,7 +298,8 @@ no-dot-config-targets := $(clean-targets) \ outputmakefile rustavailable rustfmt rustfmtcheck no-sync-config-targets := $(no-dot-config-targets) %install modules_sign kernelrelease \ image_name -single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %/ +single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %.o_thinlto_native \ + %.a_thinlto_native %.o.thinlto.bc %/ config-build := mixed-build := @@ -857,11 +858,19 @@ 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 += $(call cc-option,-fmodulo-sched -fmodulo-sched-allow-regmoves -fivopts -fmodulo-sched) + # Always set `debug-assertions` and `overflow-checks` because their default # depends on `opt-level` and `debug-assertions`, respectively. KBUILD_RUSTFLAGS += -Cdebug-assertions=$(if $(CONFIG_RUST_DEBUG_ASSERTIONS),y,n) @@ -991,10 +1000,10 @@ 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 # for CONFIG_LTO_CLANG_THIN or CONFIG_LTO_CLANG_THIN_DIST +CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit endif CC_FLAGS_LTO += -fvisibility=hidden @@ -1213,8 +1222,34 @@ vmlinux.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE $(call if_changed,ar_vmlinux.a) PHONY += vmlinux_o +ifdef CONFIG_LTO_CLANG_THIN_DIST +vmlinux.thinlink: vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE + $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_thinlink +targets += vmlinux.thinlink + +vmlinux.a_thinlto_native := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_OBJS)) +quiet_cmd_ar_vmlinux.a_thinlto_native = AR $@ + cmd_ar_vmlinux.a_thinlto_native = \ + rm -f $@; \ + $(AR) cDPrST $@ $(vmlinux.a_thinlto_native); \ + $(AR) mPiT $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt) + +define rule_gen_vmlinux.a_thinlto_native + +$(Q)$(MAKE) $(build)=. need-builtin=1 thinlto_final_pass=1 need-modorder=1 built-in.a_thinlto_native + $(call cmd_and_savecmd,ar_vmlinux.a_thinlto_native) +endef + +vmlinux.a_thinlto_native: vmlinux.thinlink scripts/head-object-list.txt FORCE + $(call if_changed_rule,gen_vmlinux.a_thinlto_native) + +targets += vmlinux.a_thinlto_native + +vmlinux_o: vmlinux.a_thinlto_native + $(Q)$(MAKE) thinlto_final_pass=1 -f $(srctree)/scripts/Makefile.vmlinux_o +else vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS) $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_o +endif vmlinux.o modules.builtin.modinfo modules.builtin: vmlinux_o @: @@ -1572,7 +1607,8 @@ CLEAN_FILES += vmlinux.symvers modules-only.symvers \ modules.builtin.ranges vmlinux.o.map vmlinux.unstripped \ compile_commands.json rust/test \ rust-project.json .vmlinux.objs .vmlinux.export.c \ - .builtin-dtbs-list .builtin-dtb.S + .builtin-dtbs-list .builtin-dtb.S \ + .vmlinux_thinlto_bc_files vmlinux.thinlink # Directories & files removed with 'make mrproper' MRPROPER_FILES += include/config include/generated \ @@ -2023,6 +2059,8 @@ clean: $(clean-dirs) -o -name '*.symtypes' -o -name 'modules.order' \ -o -name '*.c.[012]*.*' \ -o -name '*.ll' \ + -o -name '*.a_thinlto_native' -o -name '*.o_thinlto_native' \ + -o -name '*.o.thinlto.bc' \ -o -name '*.gcno' \ \) -type f -print \ -o -name '.tmp_*' -print \ diff --git a/arch/Kconfig b/arch/Kconfig index b0adb665041f..30dccda07c67 100644 --- a/arch/Kconfig +++ b/arch/Kconfig @@ -810,6 +810,25 @@ 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 .o_thinlto_native 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 753b8763abae..d4ce964d9713 100644 --- a/arch/x86/Kconfig.cpu +++ b/arch/x86/Kconfig.cpu @@ -245,6 +245,76 @@ config MATOM endchoice +config CC_HAS_MARCH_NATIVE + # This flag might not be available in cross-compilers: + def_bool $(cc-option, -march=native) + # LLVM 18 has an easily triggered internal compiler error in core + # networking code with '-march=native' on certain systems: + # https://github.com/llvm/llvm-project/issues/72026 + # LLVM 19 introduces an optimization that resolves some high stack + # 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 + depends on CC_HAS_MARCH_NATIVE + help + Optimize for the current CPU used to compile the kernel. + Use this option if you intend to build the kernel for your + local machine. + + Note that such a kernel might not work optimally on a + different x86 machine. + + 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 594723005d95..cbd234e9e743 100644 --- a/arch/x86/Makefile +++ b/arch/x86/Makefile @@ -173,8 +173,25 @@ else # Use -mskip-rax-setup if supported. KBUILD_CFLAGS += $(call cc-option,-mskip-rax-setup) +ifdef CONFIG_X86_NATIVE_CPU + KBUILD_CFLAGS += -march=native + KBUILD_RUSTFLAGS += -Ctarget-cpu=native +endif + +ifdef CONFIG_MZEN4 + KBUILD_CFLAGS += -march=znver4 + KBUILD_RUSTFLAGS += -Ctarget-cpu=znver4 +endif + +ifdef CONFIG_GENERIC_CPU +ifeq ($(CONFIG_X86_64_VERSION),1) KBUILD_CFLAGS += -march=x86-64 -mtune=generic KBUILD_RUSTFLAGS += -Ctarget-cpu=x86-64 -Ztune-cpu=generic +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/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 3a941dc0d27f..94e3bf608bd2 100644 --- a/block/Makefile +++ b/block/Makefile @@ -23,6 +23,7 @@ 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 @@ -37,3 +38,10 @@ 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..6f6fb237df69 --- /dev/null +++ b/block/adios.c @@ -0,0 +1,1382 @@ +// 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 "elevator.h" +#include "blk.h" +#include "blk-mq.h" +#include "blk-mq-sched.h" + +#define ADIOS_VERSION "2.0.0" + +// Define operation types supported by ADIOS +enum adios_op_type { + ADIOS_READ = 0, + ADIOS_WRITE = 1, + ADIOS_DISCARD = 2, + ADIOS_OTHER = 3, + ADIOS_OPTYPES = 4, +}; + +// Global variable to control the latency +static u64 default_global_latency_window = 32000000ULL; +// Ratio below which batch queues should be refilled +static u8 default_bq_refill_below_ratio = 25; + +// 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; + +// Latency targets for each operation type +static u64 default_latency_target[ADIOS_OPTYPES] = { + [ADIOS_READ] = 1ULL * 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] = 24, + [ADIOS_WRITE] = 48, + [ADIOS_DISCARD] = 1, + [ADIOS_OTHER] = 1, +}; + +static u32 default_dl_prio[2] = {7, 0}; + +// 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 + +// Structure to hold latency bucket data for small requests +struct latency_bucket_small { + u64 sum_latency; + u32 count; +}; + +// Structure to hold latency bucket data for large requests +struct latency_bucket_large { + u64 sum_latency; + u64 sum_block_size; + u32 count; +}; + +// New structure to hold per-cpu buckets, improving data locality and code clarity. +struct per_cpu_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 the latency model context data +struct latency_model { + seqlock_t lock; + u64 base; + u64 slope; + u64 small_sum_delay; + u64 small_count; + u64 large_sum_delay; + u64 large_sum_bsize; + u64 last_update_jiffies; + + // Per-CPU buckets to avoid lock contention on the completion path + struct per_cpu_lm_buckets __percpu *pcpu_buckets; + + u32 lm_shrink_at_kreqs; + u32 lm_shrink_at_gbytes; + u8 lm_shrink_resist; +}; + +#define ADIOS_BQ_PAGES 2 + +// Adios scheduler data +struct adios_data { + spinlock_t pq_lock; + struct list_head prio_queue; + + struct rb_root_cached dl_tree[2]; + spinlock_t lock; + u8 dl_queued; + s64 dl_bias; + s32 dl_prio[2]; + + u64 global_latency_window; + 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; + u8 bq_refill_below_ratio; + + u8 bq_page; + bool more_bq_ready; + struct list_head batch_queue[ADIOS_BQ_PAGES][ADIOS_OPTYPES]; + u32 batch_count[ADIOS_BQ_PAGES][ADIOS_OPTYPES]; + spinlock_t bq_lock; + + struct per_cpu_lm_buckets *aggr_buckets; + + struct latency_model latency_model[ADIOS_OPTYPES]; + struct timer_list update_timer; + + atomic64_t total_pred_lat; + + 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; +} __attribute__((aligned(64))); + +static const int adios_prio_to_weight[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, +}; + +// Count the number of entries in aggregated small buckets +static u32 lm_count_small_entries(struct latency_bucket_small *buckets) { + u32 total_count = 0; + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) + total_count += buckets[i].count; + return total_count; +} + +// Update the small buckets in the latency model from aggregated data +static bool lm_update_small_buckets(struct latency_model *model, + struct latency_bucket_small *buckets, + u32 total_count, bool count_all) { + u64 sum_latency = 0; + u32 sum_count = 0; + u32 cumulative_count = 0, threshold_count = 0; + u8 outlier_threshold_bucket = 0; + u8 outlier_percentile = LM_OUTLIER_PERCENTILE; + u8 reduction; + + if (count_all) + outlier_percentile = 100; + + // Calculate the threshold count for outlier detection + threshold_count = (total_count * outlier_percentile) / 100; + + // Identify the bucket that corresponds to the outlier threshold + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) { + cumulative_count += buckets[i].count; + if (cumulative_count >= threshold_count) { + 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->sum_latency; + sum_count += bucket->count; + } else { + // The threshold bucket's contribution is proportional + u64 remaining_count = + threshold_count - (cumulative_count - bucket->count); + if (bucket->count > 0) { + sum_latency += + div_u64((bucket->sum_latency * remaining_count), bucket->count); + sum_count += remaining_count; + } + } + } + + // Shrink the model if it reaches at the readjustment threshold + if (model->small_count >= 1000ULL * model->lm_shrink_at_kreqs) { + reduction = model->lm_shrink_resist; + if (model->small_count >> reduction) { + model->small_sum_delay -= model->small_sum_delay >> reduction; + model->small_count -= model->small_count >> reduction; + } + } + + // Accumulate the average latency into the statistics + model->small_sum_delay += sum_latency; + model->small_count += sum_count; + + return true; +} + +// Count the number of entries in aggregated large buckets +static u32 lm_count_large_entries(struct latency_bucket_large *buckets) { + u32 total_count = 0; + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) + total_count += buckets[i].count; + return total_count; +} + +// Update the large buckets in the latency model from aggregated data +static bool lm_update_large_buckets(struct latency_model *model, + struct latency_bucket_large *buckets, + u32 total_count, bool count_all) { + s64 sum_latency = 0; + u64 sum_block_size = 0, intercept; + u32 cumulative_count = 0, threshold_count = 0; + u8 outlier_threshold_bucket = 0; + u8 outlier_percentile = LM_OUTLIER_PERCENTILE; + u8 reduction; + + if (count_all) + outlier_percentile = 100; + + // Calculate the threshold count for outlier detection + threshold_count = (total_count * outlier_percentile) / 100; + + // Identify the bucket that corresponds to the outlier threshold + for (u8 i = 0; i < LM_LAT_BUCKET_COUNT; i++) { + cumulative_count += buckets[i].count; + if (cumulative_count >= threshold_count) { + 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->sum_latency; + sum_block_size += bucket->sum_block_size; + } else { + // The threshold bucket's contribution is proportional + u64 remaining_count = + threshold_count - (cumulative_count - bucket->count); + if (bucket->count > 0) { + sum_latency += + div_u64((bucket->sum_latency * remaining_count), bucket->count); + sum_block_size += + div_u64((bucket->sum_block_size * remaining_count), bucket->count); + } + } + } + + // Shrink the model if it reaches at the readjustment threshold + if (model->large_sum_bsize >= 0x40000000ULL * model->lm_shrink_at_gbytes) { + reduction = model->lm_shrink_resist; + if (model->large_sum_bsize >> reduction) { + model->large_sum_delay -= model->large_sum_delay >> reduction; + model->large_sum_bsize -= model->large_sum_bsize >> reduction; + } + } + + // Accumulate the average delay into the statistics + intercept = model->base * threshold_count; + if (sum_latency > intercept) + sum_latency -= intercept; + + model->large_sum_delay += sum_latency; + model->large_sum_bsize += sum_block_size; + + return true; +} + +// Update the latency model parameters and statistics +static void latency_model_update( + struct adios_data *ad, struct latency_model *model) { + u64 now; + u32 small_count, large_count; + bool time_elapsed; + bool small_processed = false, large_processed = false; + struct per_cpu_lm_buckets *aggr = ad->aggr_buckets; + struct latency_bucket_small *asb; + struct latency_bucket_large *alb; + struct per_cpu_lm_buckets *pcpu_b; + unsigned long flags; + int cpu; + + memset(aggr, 0, sizeof(*aggr)); + + write_seqlock_irqsave(&model->lock, flags); + + // 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].count) { + asb = &aggr->small_bucket[i]; + asb->count += pcpu_b->small_bucket[i].count; + asb->sum_latency += pcpu_b->small_bucket[i].sum_latency; + } + if (pcpu_b->large_bucket[i].count) { + alb = &aggr->large_bucket[i]; + alb->count += pcpu_b->large_bucket[i].count; + alb->sum_latency += pcpu_b->large_bucket[i].sum_latency; + alb->sum_block_size += pcpu_b->large_bucket[i].sum_block_size; + } + } + // Reset per-cpu buckets after aggregating + memset(pcpu_b, 0, sizeof(*pcpu_b)); + } + + // Whether enough time has elapsed since the last update + now = jiffies; + time_elapsed = unlikely(!model->base) || model->last_update_jiffies + + msecs_to_jiffies(LM_INTERVAL_THRESHOLD) <= now; + + // Count the number of entries in aggregated buckets + small_count = lm_count_small_entries(aggr->small_bucket); + large_count = lm_count_large_entries(aggr->large_bucket); + + // Update small buckets + if (small_count && (time_elapsed || + LM_SAMPLES_THRESHOLD <= small_count || !model->base)) + small_processed = lm_update_small_buckets( + model, aggr->small_bucket, small_count, !model->base); + // Update large buckets + if (large_count && (time_elapsed || + LM_SAMPLES_THRESHOLD <= large_count || !model->slope)) + large_processed = lm_update_large_buckets( + model, aggr->large_bucket, large_count, !model->slope); + + // Update the base parameter if small bucket was processed + if (small_processed && likely(model->small_count)) + model->base = div_u64(model->small_sum_delay, model->small_count); + + // Update the slope parameter if large bucket was processed + if (large_processed && likely(model->large_sum_bsize)) + model->slope = div_u64(model->large_sum_delay, + DIV_ROUND_UP_ULL(model->large_sum_bsize, 1024)); + + // Reset statistics and update last updated jiffies if time has elapsed + if (time_elapsed) + model->last_update_jiffies = now; + + write_sequnlock_irqrestore(&model->lock, flags); +} + +// 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) { + u8 bucket_index; + struct per_cpu_lm_buckets *buckets; + + buckets = this_cpu_ptr(model->pcpu_buckets); + + if (block_size <= LM_BLOCK_SIZE_THRESHOLD) { + // Handle small requests + bucket_index = lm_input_bucket_index(latency, model->base ?: 1); + + if (bucket_index >= LM_LAT_BUCKET_COUNT) + bucket_index = LM_LAT_BUCKET_COUNT - 1; + + buckets->small_bucket[bucket_index].count++; + buckets->small_bucket[bucket_index].sum_latency += latency; + + if (unlikely(!model->base)) { + latency_model_update(ad, model); + return; + } + } else { + // Handle large requests + if (!model->base || !pred_lat) + 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].count++; + buckets->large_bucket[bucket_index].sum_latency += latency; + buckets->large_bucket[bucket_index].sum_block_size += block_size; + } +} + +// 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, base, slope; + unsigned int seq; + + do { + seq = read_seqbegin(&model->lock); + base = model->base; + slope = model->slope; + } while (read_seqretry(&model->lock, seq)); + + result = base; + if (block_size > LM_BLOCK_SIZE_THRESHOLD) + result += slope * + DIV_ROUND_UP_ULL(block_size - LM_BLOCK_SIZE_THRESHOLD, 1024); + + 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]; +} + +// 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; + + rd->block_size = blk_rq_bytes(rq); + u8 optype = adios_optype(rq); + rd->pred_lat = + latency_model_predict(&ad->latency_model[optype], rd->block_size); + rd->deadline = + rq->start_time_ns + ad->latency_target[optype] + rd->pred_lat; + + while (*link) { + dlg = rb_entry(*link, struct dl_group, node); + s64 diff = rd->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 != rd->deadline) { + dlg = kmem_cache_zalloc(ad->dl_group_pool, GFP_ATOMIC); + if (!dlg) + return; + dlg->deadline = rd->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; + ad->dl_queued |= 1 << dl_idx; +} + +// 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)) + ad->dl_queued &= ~(1 << dl_idx); +} + +// 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; +} + +// 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); +} + +// Update async_depth when the number of requests in the queue changes +static void adios_depth_updated(struct blk_mq_hw_ctx *hctx) { + struct request_queue *q = hctx->queue; + struct adios_data *ad = q->elevator->elevator_data; + struct blk_mq_tags *tags = hctx->sched_tags; + + ad->async_depth = q->nr_requests; + + sbitmap_queue_min_shallow_depth(&tags->bitmap_tags, 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) { + struct adios_data *ad = q->elevator->elevator_data; + struct request *free = NULL; + bool ret; + + scoped_guard(spinlock_irqsave, &ad->lock) + ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free); + + if (free) + blk_mq_free_request(free); + + return ret; +} + +// Insert a request into the scheduler +static void insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq, + blk_insert_t insert_flags, struct list_head *free) { + bool dl_idx = adios_optype_not_read(rq); + struct request_queue *q = hctx->queue; + struct adios_data *ad = q->elevator->elevator_data; + + if (insert_flags & BLK_MQ_INSERT_AT_HEAD) { + scoped_guard(spinlock_irqsave, &ad->pq_lock) + list_add_tail(&rq->queuelist, &ad->prio_queue); + return; + } + + guard(spinlock_irqsave)(&ad->lock); + + if (blk_mq_sched_try_insert_merge(q, rq, free)) + return; + + 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; + } +} + +// 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 *rq; + LIST_HEAD(free); + + while (!list_empty(list)) { + rq = list_first_entry(list, struct request, queuelist); + list_del_init(&rq->queuelist); + insert_request(hctx, rq, insert_flags, &free); + } + + 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; + + 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); +} + +// Select the next request to dispatch from the deadline-sorted red-black tree +static struct request *next_request(struct adios_data *ad) { + struct adios_rq_data *rd; + bool dl_idx, bias_idx, reduce_bias; + + if (!ad->dl_queued) + return NULL; + + dl_idx = ad->dl_queued >> 1; + rd = get_dl_first_rd(ad, dl_idx); + + bias_idx = ad->dl_bias < 0; + reduce_bias = (bias_idx == dl_idx); + + if (ad->dl_queued == 0x3) { + struct adios_rq_data *trd[2]; + trd[0] = get_dl_first_rd(ad, 0); + trd[1] = rd; + + rd = trd[bias_idx]; + + reduce_bias = + (trd[bias_idx]->deadline > trd[((u8)bias_idx + 1) % 2]->deadline); + } + + if (reduce_bias) { + s64 sign = ((int)bias_idx << 1) - 1; + if (unlikely(!rd->pred_lat)) + ad->dl_bias = sign; + else { + ad->dl_bias += sign * (s64)((rd->pred_lat * + adios_prio_to_weight[ad->dl_prio[bias_idx] + 20]) >> 10); + } + } + + return rd->rq; +} + +// Reset the batch queue counts for a given page +static void reset_batch_counts(struct adios_data *ad, u8 page) { + memset(&ad->batch_count[page], 0, sizeof(ad->batch_count[page])); +} + +// Initialize all batch queues +static void init_batch_queues(struct adios_data *ad) { + for (u8 page = 0; page < ADIOS_BQ_PAGES; page++) { + reset_batch_counts(ad, page); + + for (u8 optype = 0; optype < ADIOS_OPTYPES; optype++) + INIT_LIST_HEAD(&ad->batch_queue[page][optype]); + } +} + +// Fill the batch queues with requests from the deadline-sorted red-black tree +static bool fill_batch_queues(struct adios_data *ad, u64 current_lat) { + u32 count = 0; + u32 optype_count[ADIOS_OPTYPES] = {0}; + u8 page = (ad->bq_page + 1) % ADIOS_BQ_PAGES; + + reset_batch_counts(ad, page); + + scoped_guard(spinlock_irqsave, &ad->lock) + while (true) { + struct request *rq = next_request(ad); + if (!rq) + break; + + struct adios_rq_data *rd = get_rq_data(rq); + u8 optype = adios_optype(rq); + current_lat += rd->pred_lat; + + // Check batch size and total predicted latency + if (count && (!ad->latency_model[optype].base || + ad->batch_count[page][optype] >= ad->batch_limit[optype] || + current_lat > ad->global_latency_window)) { + break; + } + + remove_request(ad, rq); + + // Add request to the corresponding batch queue + list_add_tail(&rq->queuelist, &ad->batch_queue[page][optype]); + ad->batch_count[page][optype]++; + atomic64_add(rd->pred_lat, &ad->total_pred_lat); + optype_count[optype]++; + count++; + } + + if (count) { + ad->more_bq_ready = true; + for (u8 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->more_bq_ready = false; + ad->bq_page = (ad->bq_page + 1) % ADIOS_BQ_PAGES; +} + +// Dispatch a request from the batch queues +static struct request *dispatch_from_bq(struct adios_data *ad) { + struct request *rq = NULL; + u64 tpl; + + guard(spinlock_irqsave)(&ad->bq_lock); + + tpl = atomic64_read(&ad->total_pred_lat); + + if (!ad->more_bq_ready && (!tpl || + tpl < div_u64(ad->global_latency_window * ad->bq_refill_below_ratio, 100))) + fill_batch_queues(ad, tpl); + +again: + // Check if there are any requests in the batch queues + for (u8 i = 0; i < ADIOS_OPTYPES; i++) { + if (!list_empty(&ad->batch_queue[ad->bq_page][i])) { + rq = list_first_entry(&ad->batch_queue[ad->bq_page][i], + struct request, queuelist); + list_del_init(&rq->queuelist); + return rq; + } + } + + // If there's more batch queue page available, flip to it and retry + if (ad->more_bq_ready) { + flip_bq_page(ad); + goto again; + } + + 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); + + if (!list_empty(&ad->prio_queue)) { + rq = list_first_entry(&ad->prio_queue, struct request, queuelist); + list_del_init(&rq->queuelist); + } + return rq; +} + +// 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; + + rq = dispatch_from_pq(ad); + if (rq) goto found; + rq = dispatch_from_bq(ad); + if (!rq) return NULL; +found: + 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 = from_timer(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); + + atomic64_sub(rd->pred_lat, &ad->total_pred_lat); + + if (!rq->io_start_time_ns || !rd->block_size) + return; + u64 latency = now - rq->io_start_time_ns; + u8 optype = adios_optype(rq); + latency_model_input(ad, &ad->latency_model[optype], + rd->block_size, latency, rd->pred_lat); + 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; + } +} + +static inline bool pq_has_work(struct adios_data *ad) { + guard(spinlock_irqsave)(&ad->pq_lock); + return !list_empty(&ad->prio_queue); +} + +static inline bool bq_has_work(struct adios_data *ad) { + guard(spinlock_irqsave)(&ad->bq_lock); + + for (u8 i = 0; i < ADIOS_OPTYPES; i++) + if (!list_empty(&ad->batch_queue[ad->bq_page][i])) + return true; + + return ad->more_bq_ready; +} + +static inline bool dl_tree_has_work(struct adios_data *ad) { + guard(spinlock_irqsave)(&ad->lock); + return ad->dl_queued; +} + +// 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 pq_has_work(ad) || bq_has_work(ad) || dl_tree_has_work(ad); +} + +// Initialize the scheduler-specific data for a hardware queue +static int adios_init_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) { + adios_depth_updated(hctx); + return 0; +} + +// Initialize the scheduler-specific data when initializing the request queue +static int adios_init_sched(struct request_queue *q, struct elevator_type *e) { + struct adios_data *ad; + struct elevator_queue *eq; + int ret = -ENOMEM; + int cpu = 0; + + eq = elevator_alloc(q, e); + if (!eq) + return ret; + + ad = kzalloc_node(sizeof(*ad), GFP_KERNEL, q->node); + if (!ad) + goto put_eq; + + // 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; + } + + eq->elevator_data = ad; + + ad->global_latency_window = default_global_latency_window; + ad->bq_refill_below_ratio = default_bq_refill_below_ratio; + + INIT_LIST_HEAD(&ad->prio_queue); + for (u8 i = 0; i < 2; i++) + ad->dl_tree[i] = RB_ROOT_CACHED; + ad->dl_bias = 0; + ad->dl_queued = 0x0; + for (u8 i = 0; i < 2; i++) + ad->dl_prio[i] = default_dl_prio[i]; + + 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 (cpu = 0; cpu < ADIOS_OPTYPES; cpu++) { + struct latency_model *model = &ad->latency_model[cpu]; + seqlock_init(&model->lock); + + model->pcpu_buckets = alloc_percpu(struct per_cpu_lm_buckets); + if (!model->pcpu_buckets) + goto free_buckets; + + model->last_update_jiffies = jiffies; + 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; + + ad->latency_target[cpu] = default_latency_target[cpu]; + ad->batch_limit[cpu] = default_batch_limit[cpu]; + } + timer_setup(&ad->update_timer, update_timer_callback, 0); + init_batch_queues(ad); + + spin_lock_init(&ad->lock); + spin_lock_init(&ad->pq_lock); + spin_lock_init(&ad->bq_lock); + + /* 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; + return 0; + +free_buckets: + pr_err("adios: Failed to allocate per-cpu buckets\n"); + while (--cpu >= 0) { + struct latency_model *prev_model = &ad->latency_model[cpu]; + 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->prio_queue)); + + for (u8 i = 0; i < ADIOS_OPTYPES; i++) { + struct latency_model *model = &ad->latency_model[i]; + free_percpu(model->pcpu_buckets); + } + 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); +} + +// Define sysfs attributes for read operation latency model +#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]; \ + ssize_t len = 0; \ + u64 base, slope; \ + unsigned int seq; \ + do { \ + seq = read_seqbegin(&model->lock); \ + base = model->base; \ + slope = model->slope; \ + } while (read_seqretry(&model->lock, seq)); \ + 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_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; \ + ad->latency_model[optype].base = 0ULL; \ + ad->latency_target[optype] = nsec; \ + 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_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; \ +} \ +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]); \ +} + +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); +} + +// Set the global latency window +static ssize_t adios_global_latency_window_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; + + ad->global_latency_window = nsec; + + return count; +} + +// Show the global latency window +static ssize_t adios_global_latency_window_show( + struct elevator_queue *e, char *page) { + struct adios_data *ad = e->elevator_data; + return sprintf(page, "%llu\n", ad->global_latency_window); +} + +// Show the bq_refill_below_ratio +static ssize_t adios_bq_refill_below_ratio_show( + struct elevator_queue *e, char *page) { + struct adios_data *ad = e->elevator_data; + return sprintf(page, "%d\n", ad->bq_refill_below_ratio); +} + +// Set the bq_refill_below_ratio +static ssize_t adios_bq_refill_below_ratio_store( + struct elevator_queue *e, const char *page, size_t count) { + struct adios_data *ad = e->elevator_data; + int ratio; + int ret; + + ret = kstrtoint(page, 10, &ratio); + if (ret || ratio < 0 || ratio > 100) + return -EINVAL; + + ad->bq_refill_below_ratio = ratio; + + return count; +} + +// 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 +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; + unsigned long val; + int ret; + + ret = kstrtoul(page, 10, &val); + if (ret || val != 1) + return -EINVAL; + + for (u8 i = 0; i < ADIOS_OPTYPES; i++) { + struct latency_model *model = &ad->latency_model[i]; + write_seqlock_bh(&model->lock); + model->base = 0ULL; + model->slope = 0ULL; + model->small_sum_delay = 0ULL; + model->small_count = 0ULL; + model->large_sum_delay = 0ULL; + model->large_sum_bsize = 0ULL; + write_sequnlock_bh(&model->lock); + } + + 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]; \ + write_seqlock_bh(&model->lock); \ + model->model_field = val; \ + write_sequnlock_bh(&model->lock); \ + } \ + 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 int seq; \ + struct latency_model *model = &ad->latency_model[0]; \ + do { \ + seq = read_seqbegin(&model->lock); \ + val = model->model_field; \ + } while (read_seqretry(&model->lock, seq)); \ + 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(batch_limit_read), + AD_ATTR_RW(batch_limit_write), + AD_ATTR_RW(batch_limit_discard), + + AD_ATTR_RO(lat_model_read), + AD_ATTR_RO(lat_model_write), + AD_ATTR_RO(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_hctx = adios_init_hctx, + .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 dc4cadef728e..6757ed11bf60 100644 --- a/block/elevator.c +++ b/block/elevator.c @@ -556,11 +556,23 @@ static struct elevator_type *elevator_get_default(struct request_queue *q) if (q->tag_set->flags & BLK_MQ_F_NO_SCHED_BY_DEFAULT) return NULL; +#ifdef CONFIG_MQ_IOSCHED_DEFAULT_ADIOS + return elevator_find_get("adios"); +#endif // CONFIG_MQ_IOSCHED_DEFAULT_ADIOS + if (q->nr_hw_queues != 1 && !blk_mq_is_shared_tags(q->tag_set->flags)) +#if defined(CONFIG_CACHY) + return elevator_find_get("mq-deadline"); +#else return NULL; +#endif +#if defined(CONFIG_CACHY) && defined(CONFIG_IOSCHED_BFQ) + return elevator_find_get("bfq"); +#else return elevator_find_get("mq-deadline"); +#endif } /* diff --git a/drivers/Makefile b/drivers/Makefile index b5749cf67044..5beba9f57254 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 3fc65df9ccb8..eab47951eb31 100644 --- a/drivers/ata/ahci.c +++ b/drivers/ata/ahci.c @@ -1662,7 +1662,7 @@ 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; @@ -1675,7 +1675,7 @@ 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++) { @@ -1690,18 +1690,11 @@ 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) @@ -1945,7 +1938,9 @@ 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 ba9bf06f1c77..b8ceac594acd 100644 --- a/drivers/cpufreq/intel_pstate.c +++ b/drivers/cpufreq/intel_pstate.c @@ -3828,6 +3828,8 @@ 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 c3641331d4de..ce1072fe4921 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -161,6 +161,7 @@ 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 72c807f5822e..6ef3fedc1233 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -143,6 +143,7 @@ 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 */ @@ -262,6 +263,15 @@ 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 96118a0e1ffe..c6364925b65d 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -4574,7 +4574,7 @@ 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 ebabfe3a512f..4d3ebcaacca1 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 @@ -97,7 +97,7 @@ 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 e8bdd7f0c460..b4aafe6103bc 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 @@ -481,7 +481,7 @@ 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 @@ -563,7 +563,7 @@ 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 @@ -742,7 +742,7 @@ 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 3e0f45f1711c..f645cb7831a0 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 @@ -1601,7 +1601,7 @@ 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) @@ -1792,7 +1792,7 @@ 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 @@ -1888,7 +1888,7 @@ 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); #endif /* Create (reset) the plane state */ diff --git a/drivers/gpu/drm/amd/pm/amdgpu_pm.c b/drivers/gpu/drm/amd/pm/amdgpu_pm.c index 922def51685b..da76611edb10 100644 --- a/drivers/gpu/drm/amd/pm/amdgpu_pm.c +++ b/drivers/gpu/drm/amd/pm/amdgpu_pm.c @@ -3055,6 +3055,9 @@ 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 46cce1d2aaf3..e3adb7aea969 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c +++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c @@ -2854,7 +2854,10 @@ 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; @@ -2878,7 +2881,14 @@ static int smu_set_power_limit(void *handle, uint32_t limit) if (smu->ppt_funcs->set_power_limit) return smu->ppt_funcs->set_power_limit(smu, limit_type, 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/gpu/drm/xe/regs/xe_pcode_regs.h b/drivers/gpu/drm/xe/regs/xe_pcode_regs.h index 8846eb9ce2a4..c7d5d782e3f9 100644 --- a/drivers/gpu/drm/xe/regs/xe_pcode_regs.h +++ b/drivers/gpu/drm/xe/regs/xe_pcode_regs.h @@ -21,6 +21,9 @@ #define BMG_PACKAGE_POWER_SKU XE_REG(0x138098) #define BMG_PACKAGE_POWER_SKU_UNIT XE_REG(0x1380dc) #define BMG_PACKAGE_ENERGY_STATUS XE_REG(0x138120) +#define BMG_FAN_1_SPEED XE_REG(0x138140) +#define BMG_FAN_2_SPEED XE_REG(0x138170) +#define BMG_FAN_3_SPEED XE_REG(0x1381a0) #define BMG_VRAM_TEMPERATURE XE_REG(0x1382c0) #define BMG_PACKAGE_TEMPERATURE XE_REG(0x138434) #define BMG_PACKAGE_RAPL_LIMIT XE_REG(0x138440) diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h index 0482f26aa480..0564164935c6 100644 --- a/drivers/gpu/drm/xe/xe_device_types.h +++ b/drivers/gpu/drm/xe/xe_device_types.h @@ -314,6 +314,8 @@ struct xe_device { u8 has_atomic_enable_pte_bit:1; /** @info.has_device_atomics_on_smem: Supports device atomics on SMEM */ u8 has_device_atomics_on_smem:1; + /** @info.has_fan_control: Device supports fan control */ + u8 has_fan_control:1; /** @info.has_flat_ccs: Whether flat CCS metadata is used */ u8 has_flat_ccs:1; /** @info.has_heci_cscfi: device has heci cscfi */ diff --git a/drivers/gpu/drm/xe/xe_hwmon.c b/drivers/gpu/drm/xe/xe_hwmon.c index 48d80ffdf7bb..eb293aec36a0 100644 --- a/drivers/gpu/drm/xe/xe_hwmon.c +++ b/drivers/gpu/drm/xe/xe_hwmon.c @@ -5,6 +5,7 @@ #include #include +#include #include #include @@ -27,6 +28,7 @@ enum xe_hwmon_reg { REG_PKG_POWER_SKU_UNIT, REG_GT_PERF_STATUS, REG_PKG_ENERGY_STATUS, + REG_FAN_SPEED, }; enum xe_hwmon_reg_operation { @@ -42,6 +44,13 @@ enum xe_hwmon_channel { CHANNEL_MAX, }; +enum xe_fan_channel { + FAN_1, + FAN_2, + FAN_3, + FAN_MAX, +}; + /* * SF_* - scale factors for particular quantities according to hwmon spec. */ @@ -61,6 +70,16 @@ struct xe_hwmon_energy_info { long accum_energy; }; +/** + * struct xe_hwmon_fan_info - to cache previous fan reading + */ +struct xe_hwmon_fan_info { + /** @reg_val_prev: previous fan reg val */ + u32 reg_val_prev; + /** @time_prev: previous timestamp */ + u64 time_prev; +}; + /** * struct xe_hwmon - xe hwmon data structure */ @@ -79,6 +98,8 @@ struct xe_hwmon { int scl_shift_time; /** @ei: Energy info for energyN_input */ struct xe_hwmon_energy_info ei[CHANNEL_MAX]; + /** @fi: Fan info for fanN_input */ + struct xe_hwmon_fan_info fi[FAN_MAX]; }; static struct xe_reg xe_hwmon_get_reg(struct xe_hwmon *hwmon, enum xe_hwmon_reg hwmon_reg, @@ -144,6 +165,14 @@ static struct xe_reg xe_hwmon_get_reg(struct xe_hwmon *hwmon, enum xe_hwmon_reg return PCU_CR_PACKAGE_ENERGY_STATUS; } break; + case REG_FAN_SPEED: + if (channel == FAN_1) + return BMG_FAN_1_SPEED; + else if (channel == FAN_2) + return BMG_FAN_2_SPEED; + else if (channel == FAN_3) + return BMG_FAN_3_SPEED; + break; default: drm_warn(&xe->drm, "Unknown xe hwmon reg id: %d\n", hwmon_reg); break; @@ -454,6 +483,7 @@ static const struct hwmon_channel_info * const hwmon_info[] = { HWMON_CHANNEL_INFO(curr, HWMON_C_LABEL, HWMON_C_CRIT | HWMON_C_LABEL), HWMON_CHANNEL_INFO(in, HWMON_I_INPUT | HWMON_I_LABEL, HWMON_I_INPUT | HWMON_I_LABEL), HWMON_CHANNEL_INFO(energy, HWMON_E_INPUT | HWMON_E_LABEL, HWMON_E_INPUT | HWMON_E_LABEL), + HWMON_CHANNEL_INFO(fan, HWMON_F_INPUT, HWMON_F_INPUT, HWMON_F_INPUT), NULL }; @@ -480,6 +510,19 @@ static int xe_hwmon_pcode_write_i1(const struct xe_hwmon *hwmon, u32 uval) (uval & POWER_SETUP_I1_DATA_MASK)); } +static int xe_hwmon_pcode_read_fan_control(const struct xe_hwmon *hwmon, u32 subcmd, u32 *uval) +{ + struct xe_tile *root_tile = xe_device_get_root_tile(hwmon->xe); + + /* Platforms that don't return correct value */ + if (hwmon->xe->info.platform == XE_DG2 && subcmd == FSC_READ_NUM_FANS) { + *uval = 2; + return 0; + } + + return xe_pcode_read(root_tile, PCODE_MBOX(FAN_SPEED_CONTROL, subcmd, 0), uval, NULL); +} + static int xe_hwmon_power_curr_crit_read(struct xe_hwmon *hwmon, int channel, long *value, u32 scale_factor) { @@ -705,6 +748,75 @@ xe_hwmon_energy_read(struct xe_hwmon *hwmon, u32 attr, int channel, long *val) } } +static umode_t +xe_hwmon_fan_is_visible(struct xe_hwmon *hwmon, u32 attr, int channel) +{ + u32 uval; + + if (!hwmon->xe->info.has_fan_control) + return 0; + + switch (attr) { + case hwmon_fan_input: + if (xe_hwmon_pcode_read_fan_control(hwmon, FSC_READ_NUM_FANS, &uval)) + return 0; + + return channel < uval ? 0444 : 0; + default: + return 0; + } +} + +static int +xe_hwmon_fan_input_read(struct xe_hwmon *hwmon, int channel, long *val) +{ + struct xe_mmio *mmio = xe_root_tile_mmio(hwmon->xe); + struct xe_hwmon_fan_info *fi = &hwmon->fi[channel]; + u64 rotations, time_now, time; + u32 reg_val; + int ret = 0; + + mutex_lock(&hwmon->hwmon_lock); + + reg_val = xe_mmio_read32(mmio, xe_hwmon_get_reg(hwmon, REG_FAN_SPEED, channel)); + time_now = get_jiffies_64(); + + /* + * HW register value is accumulated count of pulses from PWM fan with the scale + * of 2 pulses per rotation. + */ + rotations = (reg_val - fi->reg_val_prev) / 2; + + time = jiffies_delta_to_msecs(time_now - fi->time_prev); + if (unlikely(!time)) { + ret = -EAGAIN; + goto unlock; + } + + /* + * Calculate fan speed in RPM by time averaging two subsequent readings in minutes. + * RPM = number of rotations * msecs per minute / time in msecs + */ + *val = DIV_ROUND_UP_ULL(rotations * (MSEC_PER_SEC * 60), time); + + fi->reg_val_prev = reg_val; + fi->time_prev = time_now; +unlock: + mutex_unlock(&hwmon->hwmon_lock); + return ret; +} + +static int +xe_hwmon_fan_read(struct xe_hwmon *hwmon, u32 attr, int channel, long *val) +{ + switch (attr) { + case hwmon_fan_input: + return xe_hwmon_fan_input_read(hwmon, channel, val); + default: + return -EOPNOTSUPP; + } +} + static umode_t xe_hwmon_is_visible(const void *drvdata, enum hwmon_sensor_types type, u32 attr, int channel) @@ -730,6 +842,9 @@ xe_hwmon_is_visible(const void *drvdata, enum hwmon_sensor_types type, case hwmon_energy: ret = xe_hwmon_energy_is_visible(hwmon, attr, channel); break; + case hwmon_fan: + ret = xe_hwmon_fan_is_visible(hwmon, attr, channel); + break; default: ret = 0; break; @@ -765,6 +880,9 @@ xe_hwmon_read(struct device *dev, enum hwmon_sensor_types type, u32 attr, case hwmon_energy: ret = xe_hwmon_energy_read(hwmon, attr, channel, val); break; + case hwmon_fan: + ret = xe_hwmon_fan_read(hwmon, attr, channel, val); + break; default: ret = -EOPNOTSUPP; break; @@ -842,7 +960,7 @@ static void xe_hwmon_get_preregistration_info(struct xe_hwmon *hwmon) { struct xe_mmio *mmio = xe_root_tile_mmio(hwmon->xe); - long energy; + long energy, fan_speed; u64 val_sku_unit = 0; int channel; struct xe_reg pkg_power_sku_unit; @@ -866,6 +984,11 @@ xe_hwmon_get_preregistration_info(struct xe_hwmon *hwmon) for (channel = 0; channel < CHANNEL_MAX; channel++) if (xe_hwmon_is_visible(hwmon, hwmon_energy, hwmon_energy_input, channel)) xe_hwmon_energy_get(hwmon, channel, &energy); + + /* Initialize 'struct xe_hwmon_fan_info' with initial fan register reading. */ + for (channel = 0; channel < FAN_MAX; channel++) + if (xe_hwmon_is_visible(hwmon, hwmon_fan, hwmon_fan_input, channel)) + xe_hwmon_fan_input_read(hwmon, channel, &fan_speed); } static void xe_hwmon_mutex_destroy(void *arg) diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c index 30f7ce06c896..57c607337848 100644 --- a/drivers/gpu/drm/xe/xe_pci.c +++ b/drivers/gpu/drm/xe/xe_pci.c @@ -62,6 +62,7 @@ struct xe_device_desc { u8 is_dgfx:1; u8 has_display:1; + u8 has_fan_control:1; u8 has_heci_gscfi:1; u8 has_heci_cscfi:1; u8 has_llc:1; @@ -303,6 +304,7 @@ static const struct xe_device_desc dg2_desc = { DG2_FEATURES, .has_display = true, + .has_fan_control = true, }; static const __maybe_unused struct xe_device_desc pvc_desc = { @@ -337,6 +339,7 @@ static const struct xe_device_desc bmg_desc = { PLATFORM(BATTLEMAGE), .dma_mask_size = 46, .has_display = true, + .has_fan_control = true, .has_heci_cscfi = 1, }; @@ -576,6 +579,7 @@ static int xe_info_init_early(struct xe_device *xe, xe->info.dma_mask_size = desc->dma_mask_size; xe->info.is_dgfx = desc->is_dgfx; + xe->info.has_fan_control = desc->has_fan_control; xe->info.has_heci_gscfi = desc->has_heci_gscfi; xe->info.has_heci_cscfi = desc->has_heci_cscfi; xe->info.has_llc = desc->has_llc; diff --git a/drivers/gpu/drm/xe/xe_pcode_api.h b/drivers/gpu/drm/xe/xe_pcode_api.h index 2bae9afdbd35..e622ae17f08d 100644 --- a/drivers/gpu/drm/xe/xe_pcode_api.h +++ b/drivers/gpu/drm/xe/xe_pcode_api.h @@ -49,6 +49,9 @@ /* Domain IDs (param2) */ #define PCODE_MBOX_DOMAIN_HBM 0x2 +#define FAN_SPEED_CONTROL 0x7D +#define FSC_READ_NUM_FANS 0x4 + #define PCODE_SCRATCH(x) XE_REG(0x138320 + ((x) * 4)) /* PCODE_SCRATCH0 */ #define AUXINFO_REG_OFFSET REG_GENMASK(17, 15) diff --git a/drivers/input/evdev.c b/drivers/input/evdev.c index b5cbb57ee5f6..a0f7fa1518c6 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/md/dm-crypt.c b/drivers/md/dm-crypt.c index 9dfdb63220d7..91aeee72a15e 100644 --- a/drivers/md/dm-crypt.c +++ b/drivers/md/dm-crypt.c @@ -3284,6 +3284,11 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) goto bad; } +#ifdef CONFIG_CACHY + set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); + set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); +#endif + ret = crypt_ctr_cipher(ti, argv[0], argv[1]); if (ret < 0) goto bad; diff --git a/drivers/media/v4l2-core/Kconfig b/drivers/media/v4l2-core/Kconfig index 331b8e535e5b..80dabeebf580 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 2177b9d63a8f..c179507cedc4 100644 --- a/drivers/media/v4l2-core/Makefile +++ b/drivers/media/v4l2-core/Makefile @@ -33,5 +33,7 @@ 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..74bd8125caa3 --- /dev/null +++ b/drivers/media/v4l2-core/v4l2loopback.c @@ -0,0 +1,3313 @@ +/* -*- 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 "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) && defined(from_timer) +#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 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; +}; + +#define fh_to_opener(ptr) container_of((ptr), struct v4l2_loopback_opener, 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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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 = fh_to_opener(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); + 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 = fh_to_opener(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); + 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 = fh_to_opener(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 = from_timer(dev, t, 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 = from_timer(dev, t, 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: + video_device_release(dev->vdev); +out_free_handler: + v4l2_ctrl_handler_free(&dev->ctrl_handler); +out_unregister: + video_set_drvdata(dev->vdev, NULL); + 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..e48e0ce5949d --- /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 0 + +/* /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 038ccbd9e3ba..de5e4f5145af 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 d0f7b749b9a6..f5aaa23b254a 100644 --- a/drivers/pci/quirks.c +++ b/drivers/pci/quirks.c @@ -3747,6 +3747,106 @@ 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. @@ -5194,6 +5294,7 @@ 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 5a3c670aec27..831ffcf7c730 100644 --- a/drivers/scsi/Kconfig +++ b/drivers/scsi/Kconfig @@ -1521,4 +1521,6 @@ 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..2d7524b66199 --- /dev/null +++ b/drivers/scsi/vhba/Makefile @@ -0,0 +1,4 @@ +VHBA_VERSION := 20240917 + +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..878a3be0ba2b --- /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(4, 0, 0) && LINUX_VERSION_CODE < KERNEL_VERSION(6, 14, 0) + .tag_alloc_policy = BLK_TAG_ALLOC_RR, +#else + .tag_alloc_policy_rr = true, +#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/fs/select.c b/fs/select.c index 7da531b1cf6b..0eaf3522abe9 100644 --- a/fs/select.c +++ b/fs/select.c @@ -857,7 +857,7 @@ static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait, int fd = pollfd->fd; __poll_t mask, filter; - if (fd < 0) + if (unlikely(fd < 0)) return 0; CLASS(fd, f)(fd); diff --git a/include/linux/mm.h b/include/linux/mm.h index e51dba8398f7..ece96e871a3b 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -193,6 +193,14 @@ 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 26baa78f1ca7..8fa2ff2682bc 100644 --- a/include/linux/pagemap.h +++ b/include/linux/pagemap.h @@ -1341,7 +1341,7 @@ 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/include/linux/user_namespace.h b/include/linux/user_namespace.h index a0bb6d012137..93129fea552e 100644 --- a/include/linux/user_namespace.h +++ b/include/linux/user_namespace.h @@ -168,6 +168,8 @@ static inline void set_userns_rlimit_max(struct user_namespace *ns, #ifdef CONFIG_USER_NS +extern int unprivileged_userns_clone; + static inline struct user_namespace *get_user_ns(struct user_namespace *ns) { if (ns) @@ -201,6 +203,8 @@ extern bool current_in_userns(const struct user_namespace *target_ns); struct ns_common *ns_get_owner(struct ns_common *ns); #else +#define unprivileged_userns_clone 0 + static inline struct user_namespace *get_user_ns(struct user_namespace *ns) { return &init_user_ns; diff --git a/include/linux/wait.h b/include/linux/wait.h index 965a19809c7e..3d442267a256 100644 --- a/include/linux/wait.h +++ b/include/linux/wait.h @@ -163,6 +163,7 @@ static inline bool wq_has_sleeper(struct wait_queue_head *wq_head) extern void add_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry); extern void add_wait_queue_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry); +extern void add_wait_queue_exclusive_lifo(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry); extern void add_wait_queue_priority(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry); extern void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry); @@ -1195,6 +1196,7 @@ do { \ */ void prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state); bool prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state); +void prepare_to_wait_exclusive_lifo(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state); long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state); void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry); long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout); diff --git a/init/Kconfig b/init/Kconfig index bf3a920064be..c524858118d2 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -163,6 +163,10 @@ config THREAD_INFO_IN_TASK menu "General setup" +config CACHY + bool "Some kernel tweaks by CachyOS" + default y + config BROKEN bool @@ -1330,6 +1334,22 @@ 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 @@ -1479,6 +1499,12 @@ 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 54ea59ff8fbe..18f87e0dd137 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 168681fc4b25..39e47bfaea58 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -106,6 +106,10 @@ #include #include +#ifdef CONFIG_USER_NS +#include +#endif + #include #include #include @@ -2194,6 +2198,10 @@ __latent_entropy struct task_struct *copy_process( if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS)) return ERR_PTR(-EINVAL); + if ((clone_flags & CLONE_NEWUSER) && !unprivileged_userns_clone) + if (!capable(CAP_SYS_ADMIN)) + return ERR_PTR(-EPERM); + /* * Thread groups must share signals as well, and detached threads * can only be started up within the thread group. @@ -3354,6 +3362,12 @@ int ksys_unshare(unsigned long unshare_flags) if (unshare_flags & CLONE_NEWNS) unshare_flags |= CLONE_FS; + if ((unshare_flags & CLONE_NEWUSER) && !unprivileged_userns_clone) { + err = -EPERM; + if (!capable(CAP_SYS_ADMIN)) + goto bad_unshare_out; + } + err = check_unshare_flags(unshare_flags); if (err) goto bad_unshare_out; diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c index 2ddb827e3bea..464049c4af3f 100644 --- a/kernel/locking/rwsem.c +++ b/kernel/locking/rwsem.c @@ -747,6 +747,7 @@ 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(); @@ -783,7 +784,8 @@ 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 1a1caae2b2f7..ff9070239624 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) { @@ -124,8 +133,12 @@ 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. */ diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 47972f34ea70..0892942cf913 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2790,7 +2790,7 @@ 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/kernel/sched/wait.c b/kernel/sched/wait.c index 51e38f5f4701..c5cc616484ba 100644 --- a/kernel/sched/wait.c +++ b/kernel/sched/wait.c @@ -47,6 +47,17 @@ void add_wait_queue_priority(struct wait_queue_head *wq_head, struct wait_queue_ } EXPORT_SYMBOL_GPL(add_wait_queue_priority); +void add_wait_queue_exclusive_lifo(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry) +{ + unsigned long flags; + + wq_entry->flags |= WQ_FLAG_EXCLUSIVE; + spin_lock_irqsave(&wq_head->lock, flags); + __add_wait_queue(wq_head, wq_entry); + spin_unlock_irqrestore(&wq_head->lock, flags); +} +EXPORT_SYMBOL(add_wait_queue_exclusive_lifo); + void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry) { unsigned long flags; @@ -258,6 +269,19 @@ prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_ent } EXPORT_SYMBOL(prepare_to_wait_exclusive); +void prepare_to_wait_exclusive_lifo(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state) +{ + unsigned long flags; + + wq_entry->flags |= WQ_FLAG_EXCLUSIVE; + spin_lock_irqsave(&wq_head->lock, flags); + if (list_empty(&wq_entry->entry)) + __add_wait_queue(wq_head, wq_entry); + set_current_state(state); + spin_unlock_irqrestore(&wq_head->lock, flags); +} +EXPORT_SYMBOL(prepare_to_wait_exclusive_lifo); + void init_wait_entry(struct wait_queue_entry *wq_entry, int flags) { wq_entry->flags = flags; diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 3b7a7308e35b..fe7bec454345 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -70,6 +70,9 @@ #ifdef CONFIG_RT_MUTEXES #include #endif +#ifdef CONFIG_USER_NS +#include +#endif /* shared constants to be used in various sysctls */ const int sysctl_vals[] = { 0, 1, 2, 3, 4, 100, 200, 1000, 3000, INT_MAX, 65535, -1 }; @@ -1595,6 +1598,15 @@ static const struct ctl_table kern_table[] = { .mode = 0644, .proc_handler = proc_dointvec, }, +#ifdef CONFIG_USER_NS + { + .procname = "unprivileged_userns_clone", + .data = &unprivileged_userns_clone, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = proc_dointvec, + }, +#endif #ifdef CONFIG_PROC_SYSCTL { .procname = "tainted", diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c index 682f40d5632d..434a25f7b2ed 100644 --- a/kernel/user_namespace.c +++ b/kernel/user_namespace.c @@ -22,6 +22,13 @@ #include #include +/* sysctl */ +#ifdef CONFIG_USER_NS_UNPRIVILEGED +int unprivileged_userns_clone = 1; +#else +int unprivileged_userns_clone; +#endif + static struct kmem_cache *user_ns_cachep __ro_after_init; static DEFINE_MUTEX(userns_state_mutex); diff --git a/mm/Kconfig b/mm/Kconfig index e113f713b493..825b5932debc 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -462,6 +462,69 @@ 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 @@ -654,7 +717,7 @@ 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 ca71fd3c3181..2070b1e5106f 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -1923,7 +1923,11 @@ 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 47d76d03ce30..48615051a8cf 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -64,7 +64,11 @@ unsigned long transparent_hugepage_flags __read_mostly = #ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE (1<> (20 - PAGE_SHIFT); /* Use a smaller cluster for small-memory machines */ @@ -1098,6 +1102,7 @@ void __init swap_setup(void) page_cluster = 2; else page_cluster = 3; +#endif /* CONFIG_CACHY */ /* * Right now other parts of the system means that we * _really_ don't want to cluster much more diff --git a/mm/util.c b/mm/util.c index 448117da071f..b9334775c51d 100644 --- a/mm/util.c +++ b/mm/util.c @@ -857,6 +857,40 @@ static const struct ctl_table util_sysctl_table[] = { .mode = 0644, .proc_handler = proc_doulongvec_minmax, }, + { + .procname = "workingset_protection", + .data = &sysctl_workingset_protection, + .maxlen = sizeof(bool), + .mode = 0644, + .proc_handler = &proc_dobool, + }, + { + .procname = "anon_min_ratio", + .data = &sysctl_anon_min_ratio, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = &vm_workingset_protection_update_handler, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE_HUNDRED, + }, + { + .procname = "clean_low_ratio", + .data = &sysctl_clean_low_ratio, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = &vm_workingset_protection_update_handler, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE_HUNDRED, + }, + { + .procname = "clean_min_ratio", + .data = &sysctl_clean_min_ratio, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = &vm_workingset_protection_update_handler, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE_HUNDRED, + }, }; static int __init init_vm_util_sysctls(void) diff --git a/mm/vmpressure.c b/mm/vmpressure.c index bd5183dfd879..3a410f53a07c 100644 --- a/mm/vmpressure.c +++ b/mm/vmpressure.c @@ -43,7 +43,11 @@ static const unsigned long vmpressure_win = SWAP_CLUSTER_MAX * 16; * essence, they are percents: the higher the value, the more number * unsuccessful reclaims there were. */ +#ifdef CONFIG_CACHY +static const unsigned int vmpressure_level_med = 65; +#else static const unsigned int vmpressure_level_med = 60; +#endif static const unsigned int vmpressure_level_critical = 95; /* diff --git a/mm/vmscan.c b/mm/vmscan.c index 3783e45bfc92..75f48ea2b336 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -148,6 +148,15 @@ struct scan_control { /* The file folios on the current node are dangerously low */ unsigned int file_is_tiny:1; + /* The anonymous pages on the current node are below vm.anon_min_ratio */ + unsigned int anon_below_min:1; + + /* The clean file pages on the current node are below vm.clean_low_ratio */ + unsigned int clean_below_low:1; + + /* The clean file pages on the current node are below vm.clean_min_ratio */ + unsigned int clean_below_min:1; + /* Always discard instead of demoting to lower tier memory */ unsigned int no_demotion:1; @@ -197,10 +206,23 @@ struct scan_control { #define prefetchw_prev_lru_folio(_folio, _base, _field) do { } while (0) #endif +bool sysctl_workingset_protection __read_mostly = true; +u8 sysctl_anon_min_ratio __read_mostly = CONFIG_ANON_MIN_RATIO; +u8 sysctl_clean_low_ratio __read_mostly = CONFIG_CLEAN_LOW_RATIO; +u8 sysctl_clean_min_ratio __read_mostly = CONFIG_CLEAN_MIN_RATIO; +static u64 sysctl_anon_min_ratio_kb __read_mostly = 0; +static u64 sysctl_clean_low_ratio_kb __read_mostly = 0; +static u64 sysctl_clean_min_ratio_kb __read_mostly = 0; +static u64 workingset_protection_prev_totalram __read_mostly = 0; + /* * From 0 .. MAX_SWAPPINESS. Higher means more swappy. */ +#ifdef CONFIG_CACHY +int vm_swappiness = 100; +#else int vm_swappiness = 60; +#endif #ifdef CONFIG_MEMCG @@ -1147,6 +1169,10 @@ static unsigned int shrink_folio_list(struct list_head *folio_list, if (!sc->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 @@ -2521,6 +2547,15 @@ 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. @@ -2638,6 +2673,14 @@ 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; } } @@ -2657,6 +2700,96 @@ static bool can_age_anon_pages(struct pglist_data *pgdat, return can_demote(pgdat->node_id, sc); } +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 @@ -4623,11 +4756,21 @@ 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) @@ -4646,7 +4789,7 @@ static int isolate_folios(struct lruvec *lruvec, struct scan_control *sc, int sw 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; @@ -4889,6 +5032,12 @@ 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; @@ -6027,6 +6176,8 @@ 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/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c index dd5cf8914a28..b216088a6abd 100644 --- a/net/ipv4/inet_connection_sock.c +++ b/net/ipv4/inet_connection_sock.c @@ -632,7 +632,7 @@ static int inet_csk_wait_for_connect(struct sock *sk, long timeo) * having to remove and re-insert us on the wait queue. */ for (;;) { - prepare_to_wait_exclusive(sk_sleep(sk), &wait, + prepare_to_wait_exclusive_lifo(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); release_sock(sk); if (reqsk_queue_empty(&icsk->icsk_accept_queue)) diff --git a/scripts/Makefile.build b/scripts/Makefile.build index 13dcd86e74ca..338e1aec0eaa 100644 --- a/scripts/Makefile.build +++ b/scripts/Makefile.build @@ -50,18 +50,23 @@ endif # =========================================================================== +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a) +ifeq ($(thinlto_final_pass),1) +builtin_suffix :=.a_thinlto_native +endif + # subdir-builtin and subdir-modorder may contain duplications. Use $(sort ...) -subdir-builtin := $(sort $(filter %/built-in.a, $(real-obj-y))) +subdir-builtin := $(sort $(filter %/built-in$(builtin_suffix), $(real-obj-y))) subdir-modorder := $(sort $(filter %/modules.order, $(obj-m))) targets-for-builtin := $(extra-y) ifneq ($(strip $(lib-y) $(lib-m) $(lib-)),) -targets-for-builtin += $(obj)/lib.a +targets-for-builtin += $(obj)/lib$(builtin_suffix) endif ifdef need-builtin -targets-for-builtin += $(obj)/built-in.a +targets-for-builtin += $(obj)/built-in$(builtin_suffix) endif targets-for-modules := $(foreach x, o mod, \ @@ -337,6 +342,10 @@ $(obj)/%.o: $(obj)/%.S FORCE targets += $(filter-out $(subdir-builtin), $(real-obj-y)) targets += $(filter-out $(subdir-modorder), $(real-obj-m)) targets += $(lib-y) $(always-y) +ifeq ($(builtin_suffix),.a_thinlto_native) +native_targets = $(patsubst,%.o,%.o_thinlto_native,$(targets)) +targets += $(native_targets) +endif # Linker scripts preprocessor (.lds.S -> .lds) # --------------------------------------------------------------------------- @@ -347,6 +356,24 @@ quiet_cmd_cpp_lds_S = LDS $@ $(obj)/%.lds: $(src)/%.lds.S FORCE $(call if_changed_dep,cpp_lds_S) +ifdef CONFIG_LTO_CLANG_THIN_DIST +# Generate .o_thinlto_native (obj) from .o (bitcode) file +# --------------------------------------------------------------------------- +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@ + +cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \ + $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \ + -flto=thin, $(c_flags)) \ + -Wno-unused-command-line-argument \ + -x ir -fthinlto-index=$<.thinlto.bc -c -o $@ \ + $(if $(findstring ../,$<), \ + $$(realpath --relative-to=$(srcroot) $<), $<), \ + cp $< $@) + +$(obj)/%.o_thinlto_native: $(obj)/%.o FORCE + $(call if_changed,cc_o_bc) +endif + # ASN.1 grammar # --------------------------------------------------------------------------- quiet_cmd_asn1_compiler = ASN.1 $(basename $@).[ch] @@ -360,7 +387,7 @@ $(obj)/%.asn1.c $(obj)/%.asn1.h: $(src)/%.asn1 $(objtree)/scripts/asn1_compiler # --------------------------------------------------------------------------- # To build objects in subdirs, we need to descend into the directories -$(subdir-builtin): $(obj)/%/built-in.a: $(obj)/% ; +$(subdir-builtin): $(obj)/%/built-in$(builtin_suffix): $(obj)/% ; $(subdir-modorder): $(obj)/%/modules.order: $(obj)/% ; # @@ -377,6 +404,12 @@ quiet_cmd_ar_builtin = AR $@ $(obj)/built-in.a: $(real-obj-y) FORCE $(call if_changed,ar_builtin) +ifdef CONFIG_LTO_CLANG_THIN_DIST +# Rule to compile a set of .o_thinlto_native files into one .a_thinlto_native file. +$(obj)/built-in.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(real-obj-y)) FORCE + $(call if_changed,ar_builtin) +endif + # This is a list of build artifacts from the current Makefile and its # sub-directories. The timestamp should be updated when any of the member files. @@ -394,6 +427,14 @@ $(obj)/modules.order: $(obj-m) FORCE $(obj)/lib.a: $(lib-y) FORCE $(call if_changed,ar) +ifdef CONFIG_LTO_CLANG_THIN_DIST +quiet_cmd_ar_native = AR $@ + cmd_ar_native = rm -f $@; $(AR) cDPrsT $@ $(patsubst %.o,%.o_thinlto_native,$(real-prereqs)) + +$(obj)/lib.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(lib-y)) FORCE + $(call if_changed,ar_native) +endif + quiet_cmd_ld_multi_m = LD [M] $@ cmd_ld_multi_m = $(LD) $(ld_flags) -r -o $@ @$< $(cmd_objtool) @@ -459,7 +500,8 @@ $(single-subdir-goals): $(single-subdirs) PHONY += $(subdir-ym) $(subdir-ym): $(Q)$(MAKE) $(build)=$@ \ - need-builtin=$(if $(filter $@/built-in.a, $(subdir-builtin)),1) \ + need-builtin=$(if $(filter $@/built-in$(builtin_suffix), $(subdir-builtin)),1) \ + thinlto_final_pass=$(if $(filter .a_thinlto_native, $(builtin_suffix)),1) \ need-modorder=$(if $(filter $@/modules.order, $(subdir-modorder)),1) \ $(filter $@/%, $(single-subdir-goals)) diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib index 2fe73cda0bdd..9cfd23590334 100644 --- a/scripts/Makefile.lib +++ b/scripts/Makefile.lib @@ -34,8 +34,13 @@ else obj-m := $(filter-out %/, $(obj-m)) endif +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a) +ifeq ($(thinlto_final_pass),1) + builtin_suffix :=.a_thinlto_native +endif + ifdef need-builtin -obj-y := $(patsubst %/, %/built-in.a, $(obj-y)) +obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y)) else obj-y := $(filter-out %/, $(obj-y)) endif diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o index b024ffb3e201..f9abc45a68b3 100644 --- a/scripts/Makefile.vmlinux_o +++ b/scripts/Makefile.vmlinux_o @@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include # for objtool include $(srctree)/scripts/Makefile.lib +ifeq ($(thinlto_final_pass),1) +vmlinux_a := vmlinux.a_thinlto_native +vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS)) +else +vmlinux_a := vmlinux.a +vmlinux_libs := $(KBUILD_VMLINUX_LIBS) +endif + # Generate a linker script to ensure correct ordering of initcalls for Clang LTO # --------------------------------------------------------------------------- @@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@ $(PERL) $(real-prereqs) > $@ .tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \ - vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE + $(vmlinux_a) $(vmlinux_libs) FORCE $(call if_changed,gen_initcalls_lds) targets := .tmp_initcalls.lds @@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@ $(LD) ${KBUILD_LDFLAGS} -r -o $@ \ $(vmlinux-o-ld-args-y) \ $(addprefix -T , $(initcalls-lds)) \ - --whole-archive vmlinux.a --no-whole-archive \ - --start-group $(KBUILD_VMLINUX_LIBS) --end-group \ + --whole-archive $(vmlinux_a) --no-whole-archive \ + --start-group $(vmlinux_libs) --end-group \ $(cmd_objtool) define rule_ld_vmlinux.o @@ -68,7 +76,7 @@ define rule_ld_vmlinux.o $(call cmd,gen_objtooldep) endef -vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE +vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE $(call if_changed_rule,ld_vmlinux.o) targets += vmlinux.o diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink new file mode 100644 index 000000000000..13e4026c7d45 --- /dev/null +++ b/scripts/Makefile.vmlinux_thinlink @@ -0,0 +1,53 @@ +# SPDX-License-Identifier: GPL-2.0-only + +PHONY := __default +__default: vmlinux.thinlink + +include include/config/auto.conf +include $(srctree)/scripts/Kbuild.include + + +# Generate a linker script to ensure correct ordering of initcalls for Clang LTO +# --------------------------------------------------------------------------- + +quiet_cmd_gen_initcalls_lds = GEN $@ + cmd_gen_initcalls_lds = \ + $(PYTHON3) $(srctree)/scripts/jobserver-exec \ + $(PERL) $(real-prereqs) > $@ + +.tmp_initcalls_thinlink.lds: $(srctree)/scripts/generate_initcall_order.pl \ + vmlinux.a FORCE + $(call if_changed,gen_initcalls_lds) + +targets := .tmp_initcalls_thinlink.lds + +initcalls-lds := .tmp_initcalls_thinlink.lds + +quiet_cmd_ld_vmlinux.thinlink = LD $@ + cmd_ld_vmlinux.thinlink = \ + $(AR) t vmlinux.a > .vmlinux_thinlto_bc_files; \ + $(LD) ${KBUILD_LDFLAGS} -r $(addprefix -T , $(initcalls-lds)) \ + --thinlto-index-only @.vmlinux_thinlto_bc_files; \ + touch vmlinux.thinlink + +vmlinux.thinlink: vmlinux.a $(initcalls-lds) FORCE + $(call if_changed,ld_vmlinux.thinlink) + +targets += vmlinux.thinlink + +# 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/head-object-list.txt b/scripts/head-object-list.txt index 7274dfc65af6..90710b87a387 100644 --- a/scripts/head-object-list.txt +++ b/scripts/head-object-list.txt @@ -18,6 +18,7 @@ arch/arm/kernel/head.o arch/csky/kernel/head.o arch/hexagon/kernel/head.o arch/loongarch/kernel/head.o +arch/loongarch/kernel/head.o_thinlto_native arch/m68k/68000/head.o arch/m68k/coldfire/head.o arch/m68k/kernel/head.o -- 2.50.0.131.gcf6f63ea6b