From 8208f7e9c7aca26f27d1253c8b758a8cd71675a6 Mon Sep 17 00:00:00 2001 From: Piotr Gorski Date: Thu, 21 Aug 2025 07:43:47 +0200 Subject: [PATCH] bore Signed-off-by: Piotr Gorski --- include/linux/sched.h | 20 ++ include/linux/sched/bore.h | 44 ++++ init/Kconfig | 17 ++ kernel/Kconfig.hz | 17 ++ kernel/fork.c | 8 + kernel/futex/waitwake.c | 11 + kernel/sched/Makefile | 1 + kernel/sched/bore.c | 440 +++++++++++++++++++++++++++++++++++++ kernel/sched/core.c | 12 + kernel/sched/debug.c | 61 ++++- kernel/sched/fair.c | 110 +++++++++- kernel/sched/features.h | 3 + kernel/sched/sched.h | 9 + 13 files changed, 741 insertions(+), 12 deletions(-) create mode 100644 include/linux/sched/bore.h create mode 100644 kernel/sched/bore.c diff --git a/include/linux/sched.h b/include/linux/sched.h index aa9c5be7a..5b3e79027 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -565,6 +565,14 @@ struct sched_statistics { #endif /* CONFIG_SCHEDSTATS */ } ____cacheline_aligned; +#ifdef CONFIG_SCHED_BORE +struct sched_burst_cache { + u32 value; + u32 count; + u64 timestamp; +}; +#endif // CONFIG_SCHED_BORE + struct sched_entity { /* For load-balancing: */ struct load_weight load; @@ -584,6 +592,18 @@ struct sched_entity { u64 sum_exec_runtime; u64 prev_sum_exec_runtime; u64 vruntime; +#ifdef CONFIG_SCHED_BORE + u64 burst_time; + u32 prev_burst_penalty; + u32 curr_burst_penalty; + u32 burst_penalty; + u8 burst_score; + u8 damper; + bool stop_burst_update; + bool waiting_for_futex; + struct sched_burst_cache child_burst; + struct sched_burst_cache group_burst; +#endif // CONFIG_SCHED_BORE s64 vlag; u64 slice; diff --git a/include/linux/sched/bore.h b/include/linux/sched/bore.h new file mode 100644 index 000000000..c2819edb1 --- /dev/null +++ b/include/linux/sched/bore.h @@ -0,0 +1,44 @@ + +#include +#include + +#ifndef _LINUX_SCHED_BORE_H +#define _LINUX_SCHED_BORE_H +#define SCHED_BORE_AUTHOR "Masahito Suzuki" +#define SCHED_BORE_PROGNAME "BORE CPU Scheduler modification" + +#define SCHED_BORE_VERSION "6.4.0" + +#ifdef CONFIG_SCHED_BORE +extern u8 __read_mostly sched_bore; +extern u8 __read_mostly sched_burst_exclude_kthreads; +extern u8 __read_mostly sched_burst_min_smooth; +extern u8 __read_mostly sched_burst_max_damper; +extern u8 __read_mostly sched_burst_fork_atavistic; +extern u8 __read_mostly sched_burst_penalty_offset; +extern u8 __read_mostly sched_burst_futex_boost; +extern uint __read_mostly sched_burst_penalty_scale; +extern uint __read_mostly sched_burst_cache_stop_count; +extern uint __read_mostly sched_burst_cache_lifetime; +extern uint __read_mostly sched_deadline_boost_mask; + +extern u8 effective_prio_bore(struct task_struct *p); +extern void update_burst_score(struct sched_entity *se); +extern void update_curr_bore(u64 delta_exec, struct sched_entity *se); + +extern void restart_burst(struct sched_entity *se); +extern void restart_burst_rescale_deadline(struct sched_entity *se); + +extern int sched_bore_update_handler(const struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, loff_t *ppos); + +extern void sched_clone_bore( + struct task_struct *p, struct task_struct *parent, u64 clone_flags, u64 now); + +extern void reset_task_bore(struct task_struct *p); +extern void sched_bore_init(void); + +extern void reweight_entity( + struct cfs_rq *cfs_rq, struct sched_entity *se, unsigned long weight); +#endif // CONFIG_SCHED_BORE +#endif // _LINUX_SCHED_BORE_H diff --git a/init/Kconfig b/init/Kconfig index 2e15b4a84..62792c19c 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1381,6 +1381,23 @@ config CHECKPOINT_RESTORE If unsure, say N here. +config SCHED_BORE + bool "Burst-Oriented Response Enhancer" + default y + help + In Desktop and Mobile computing, one might prefer interactive + tasks to keep responsive no matter what they run in the background. + + Enabling this kernel feature modifies the scheduler to discriminate + tasks by their burst time (runtime since it last went sleeping or + yielding state) and prioritize those that run less bursty. + Such tasks usually include window compositor, widgets backend, + terminal emulator, video playback, games and so on. + With a little impact to scheduling fairness, it may improve + responsiveness especially under heavy background workload. + + If unsure, say Y here. + config SCHED_AUTOGROUP bool "Automatic process group scheduling" select CGROUPS diff --git a/kernel/Kconfig.hz b/kernel/Kconfig.hz index ce1435cb0..9eee2005e 100644 --- a/kernel/Kconfig.hz +++ b/kernel/Kconfig.hz @@ -57,3 +57,20 @@ config HZ config SCHED_HRTICK def_bool HIGH_RES_TIMERS + +config MIN_BASE_SLICE_NS + int "Default value for min_base_slice_ns" + default 2000000 + help + The BORE Scheduler automatically calculates the optimal base + slice for the configured HZ using the following equation: + + base_slice_ns = + 1000000000/HZ * DIV_ROUNDUP(min_base_slice_ns, 1000000000/HZ) + + This option sets the default lower bound limit of the base slice + to prevent the loss of task throughput due to overscheduling. + + Setting this value too high can cause the system to boot with + an unnecessarily large base slice, resulting in high scheduling + latency and poor system responsiveness. diff --git a/kernel/fork.c b/kernel/fork.c index 1ee8eb11f..2eaaaf9e8 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -115,6 +115,10 @@ /* For dup_mmap(). */ #include "../mm/internal.h" +#ifdef CONFIG_SCHED_BORE +#include +#endif // CONFIG_SCHED_BORE + #include #define CREATE_TRACE_POINTS @@ -2313,6 +2317,10 @@ __latent_entropy struct task_struct *copy_process( * Need tasklist lock for parent etc handling! */ write_lock_irq(&tasklist_lock); +#ifdef CONFIG_SCHED_BORE + if (likely(p->pid)) + sched_clone_bore(p, current, clone_flags, p->start_time); +#endif // CONFIG_SCHED_BORE /* CLONE_PARENT re-uses the old parent */ if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) { diff --git a/kernel/futex/waitwake.c b/kernel/futex/waitwake.c index e2bbe5509..37e7b7aa0 100644 --- a/kernel/futex/waitwake.c +++ b/kernel/futex/waitwake.c @@ -4,6 +4,9 @@ #include #include #include +#ifdef CONFIG_SCHED_BORE +#include +#endif // CONFIG_SCHED_BORE #include "futex.h" @@ -355,7 +358,15 @@ void futex_do_wait(struct futex_q *q, struct hrtimer_sleeper *timeout) * is no timeout, or if it has yet to expire. */ if (!timeout || timeout->task) +#ifdef CONFIG_SCHED_BORE + { + current->se.waiting_for_futex = true; +#endif // CONFIG_SCHED_BORE schedule(); +#ifdef CONFIG_SCHED_BORE + current->se.waiting_for_futex = false; + } +#endif // CONFIG_SCHED_BORE } __set_current_state(TASK_RUNNING); } diff --git a/kernel/sched/Makefile b/kernel/sched/Makefile index 8ae86371d..b688084bc 100644 --- a/kernel/sched/Makefile +++ b/kernel/sched/Makefile @@ -37,3 +37,4 @@ obj-y += core.o obj-y += fair.o obj-y += build_policy.o obj-y += build_utility.o +obj-$(CONFIG_SCHED_BORE) += bore.o diff --git a/kernel/sched/bore.c b/kernel/sched/bore.c new file mode 100644 index 000000000..de6f1526e --- /dev/null +++ b/kernel/sched/bore.c @@ -0,0 +1,440 @@ +/* + * Burst-Oriented Response Enhancer (BORE) CPU Scheduler + * Copyright (C) 2021-2025 Masahito Suzuki + */ +#include +#include +#include +#include "sched.h" + +#ifdef CONFIG_SCHED_BORE +u8 __read_mostly sched_bore = 1; +u8 __read_mostly sched_burst_exclude_kthreads = 1; +u8 __read_mostly sched_burst_min_smooth = 6; +u8 __read_mostly sched_burst_max_damper = 40; +u8 __read_mostly sched_burst_fork_atavistic = 1; +u8 __read_mostly sched_burst_penalty_offset = 24; +u8 __read_mostly sched_burst_futex_boost = 1; +uint __read_mostly sched_burst_penalty_scale = 3180; +uint __read_mostly sched_burst_cache_stop_count = 64; +uint __read_mostly sched_burst_cache_lifetime = 75000000; +uint __read_mostly sched_deadline_boost_mask = ENQUEUE_INITIAL + | ENQUEUE_WAKEUP; +static int __maybe_unused maxval_prio = 39; +static int __maybe_unused maxval_6_bits = 63; +static int __maybe_unused maxval_8_bits = 255; +static int __maybe_unused maxval_12_bits = 4095; + +#define BURST_PENALTY_SHIFT 12 +#define MAX_BURST_PENALTY ((40U << BURST_PENALTY_SHIFT) - 1) + +static u32 log2p1_u64_u32fp(u64 v, u8 fp) { + if (!v) return 0; + u32 exponent = fls64(v); + u32 mantissa = (u32)(v << (64 - exponent) << 1 >> (64 - fp)); + return exponent << fp | mantissa; +} + +static inline u32 calc_burst_penalty(u64 burst_time) { + u32 greed, tolerance, penalty, scaled_penalty; + + greed = log2p1_u64_u32fp(burst_time, BURST_PENALTY_SHIFT); + tolerance = sched_burst_penalty_offset << BURST_PENALTY_SHIFT; + penalty = max(0, (s32)(greed - tolerance)); + scaled_penalty = penalty * sched_burst_penalty_scale >> 10; + + return min(MAX_BURST_PENALTY, scaled_penalty); +} + +static inline u64 __scale_slice(u64 delta, u8 score) +{return mul_u64_u32_shr(delta, sched_prio_to_wmult[score], 22);} + +static inline u64 __unscale_slice(u64 delta, u8 score) +{return mul_u64_u32_shr(delta, sched_prio_to_weight[score], 10);} + +static void reweight_task_by_prio(struct task_struct *p, int prio) { + struct sched_entity *se = &p->se; + unsigned long weight = scale_load(sched_prio_to_weight[prio]); + + se->stop_burst_update = true; + reweight_entity(cfs_rq_of(se), se, weight); + se->stop_burst_update = false; + se->load.inv_weight = sched_prio_to_wmult[prio]; +} + +inline u8 effective_prio_bore(struct task_struct *p) { + int prio = p->static_prio - MAX_RT_PRIO; + if (likely(sched_bore)) + prio += p->se.burst_score; + return (u8)clamp(prio, 0, maxval_prio); +} + +void update_burst_score(struct sched_entity *se) { + if (!entity_is_task(se)) return; + struct task_struct *p = task_of(se); + u8 prev_prio = effective_prio_bore(p); + + u8 burst_score = 0; + if (!((p->flags & PF_KTHREAD) && likely(sched_burst_exclude_kthreads))) + burst_score = se->burst_penalty >> BURST_PENALTY_SHIFT; + se->burst_score = burst_score; + + u8 new_prio = effective_prio_bore(p); + if (new_prio != prev_prio) + reweight_task_by_prio(p, new_prio); +} + +void update_curr_bore(u64 delta_exec, struct sched_entity *se) { + if (!entity_is_task(se) || se->stop_burst_update) return; + + u32 prev_penalty = se->prev_burst_penalty; + u32 curr_penalty = calc_burst_penalty(se->burst_time); + + se->burst_time += delta_exec; + se->curr_burst_penalty = curr_penalty; + if (curr_penalty > prev_penalty) { + u8 damper = se->damper; + u32 excess = curr_penalty - prev_penalty; + u32 soft_excess = excess / damper + !!(excess % damper); + se->burst_penalty = prev_penalty + soft_excess; + } + update_burst_score(se); +} + +static inline u32 binary_smooth(s32 new, s32 old, u8 damper) { + return old + ((new - old) / damper) + !!((new - old) % damper); +} + +static void __restart_burst(struct sched_entity *se) { + se->prev_burst_penalty = binary_smooth(se->curr_burst_penalty, + se->prev_burst_penalty, max(se->damper, sched_burst_min_smooth)); + se->burst_time = se->curr_burst_penalty = 0; + + u8 max_damper = sched_burst_max_damper; + if (se->damper < max_damper) + se->damper++; + else if (unlikely(se->damper > max_damper)) + se->damper = max_damper; +} + +inline void restart_burst(struct sched_entity *se) { + __restart_burst(se); + se->burst_penalty = se->prev_burst_penalty; + update_burst_score(se); +} + +void restart_burst_rescale_deadline(struct sched_entity *se) { + s64 vscaled, wremain, vremain = se->deadline - se->vruntime; + struct task_struct *p = task_of(se); + u8 prev_prio = effective_prio_bore(p); + restart_burst(se); + u8 new_prio = effective_prio_bore(p); + if (prev_prio > new_prio) { + wremain = __unscale_slice(abs(vremain), prev_prio); + vscaled = __scale_slice(wremain, new_prio); + if (unlikely(vremain < 0)) + vscaled = -vscaled; + se->deadline = se->vruntime + vscaled; + } +} + +static inline bool task_is_bore_eligible(struct task_struct *p) +{return p && p->sched_class == &fair_sched_class && !p->exit_state;} + +static inline void reset_task_weights_bore(void) { + struct task_struct *task; + struct rq *rq; + struct rq_flags rf; + + write_lock_irq(&tasklist_lock); + for_each_process(task) { + if (!task_is_bore_eligible(task)) continue; + rq = task_rq_lock(task, &rf); + update_rq_clock(rq); + reweight_task_by_prio(task, effective_prio_bore(task)); + task_rq_unlock(rq, task, &rf); + } + write_unlock_irq(&tasklist_lock); +} + +int sched_bore_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; + + reset_task_weights_bore(); + + return 0; +} + +#define for_each_child(p, t) \ + list_for_each_entry(t, &(p)->children, sibling) + +static inline u32 count_entries_upto2(struct list_head *head) { + struct list_head *next = head->next; + return (next != head) + (next->next != head); +} + +static inline bool burst_cache_expired(struct sched_burst_cache *bc, u64 now) +{return (s64)(bc->timestamp + sched_burst_cache_lifetime - now) < 0;} + +static void update_burst_cache(struct sched_burst_cache *bc, + struct task_struct *p, u32 cnt, u32 sum, u64 now) { + u32 avg = cnt ? sum / cnt : 0; + bc->value = max(avg, p->se.burst_penalty); + bc->count = cnt; + bc->timestamp = now; +} + +static inline void update_child_burst_direct(struct task_struct *p, u64 now) { + u32 cnt = 0, sum = 0; + struct task_struct *child; + + for_each_child(p, child) { + if (!task_is_bore_eligible(child)) continue; + cnt++; + sum += child->se.burst_penalty; + } + + update_burst_cache(&p->se.child_burst, p, cnt, sum, now); +} + +static inline u32 inherit_burst_direct( + struct task_struct *p, u64 now, u64 clone_flags) { + struct task_struct *parent = p; + struct sched_burst_cache *bc; + + if (clone_flags & CLONE_PARENT) + parent = parent->real_parent; + + bc = &parent->se.child_burst; + if (burst_cache_expired(bc, now)) + update_child_burst_direct(parent, now); + + return bc->value; +} + +static void update_child_burst_topological( + struct task_struct *p, u64 now, u32 depth, u32 *acnt, u32 *asum) { + u32 cnt = 0, dcnt = 0, sum = 0; + struct task_struct *child, *dec; + struct sched_burst_cache *bc __maybe_unused; + + for_each_child(p, child) { + dec = child; + while ((dcnt = count_entries_upto2(&dec->children)) == 1) + dec = list_first_entry(&dec->children, struct task_struct, sibling); + + if (!dcnt || !depth) { + if (!task_is_bore_eligible(dec)) continue; + cnt++; + sum += dec->se.burst_penalty; + continue; + } + bc = &dec->se.child_burst; + if (!burst_cache_expired(bc, now)) { + cnt += bc->count; + sum += bc->value * bc->count; + if (sched_burst_cache_stop_count <= cnt) break; + continue; + } + update_child_burst_topological(dec, now, depth - 1, &cnt, &sum); + } + + update_burst_cache(&p->se.child_burst, p, cnt, sum, now); + *acnt += cnt; + *asum += sum; +} + +static inline u32 inherit_burst_topological( + struct task_struct *p, u64 now, u64 clone_flags) { + struct task_struct *anc = p; + struct sched_burst_cache *bc; + u32 cnt = 0, sum = 0; + u32 base_child_cnt = 0; + + if (clone_flags & CLONE_PARENT) { + anc = anc->real_parent; + base_child_cnt = 1; + } + + for (struct task_struct *next; + anc != (next = anc->real_parent) && + count_entries_upto2(&anc->children) <= base_child_cnt;) { + anc = next; + base_child_cnt = 1; + } + + bc = &anc->se.child_burst; + if (burst_cache_expired(bc, now)) + update_child_burst_topological( + anc, now, sched_burst_fork_atavistic - 1, &cnt, &sum); + + return bc->value; +} + +static inline void update_tg_burst(struct task_struct *p, u64 now) { + struct task_struct *task; + u32 cnt = 0, sum = 0; + + for_each_thread(p, task) { + if (!task_is_bore_eligible(task)) continue; + cnt++; + sum += task->se.burst_penalty; + } + + update_burst_cache(&p->se.group_burst, p, cnt, sum, now); +} + +static inline u32 inherit_burst_tg(struct task_struct *p, u64 now) { + struct task_struct *parent = p->group_leader; + struct sched_burst_cache *bc = &parent->se.group_burst; + if (burst_cache_expired(bc, now)) + update_tg_burst(parent, now); + + return bc->value; +} + +void sched_clone_bore(struct task_struct *p, + struct task_struct *parent, u64 clone_flags, u64 now) { + struct sched_entity *se = &p->se; + u32 penalty; + + if (!task_is_bore_eligible(p)) return; + + penalty = (clone_flags & CLONE_THREAD)? + inherit_burst_tg(parent, now): + (likely(sched_burst_fork_atavistic)? + inherit_burst_topological(parent, now, clone_flags): + inherit_burst_direct(parent, now, clone_flags)); + + __restart_burst(se); + se->burst_penalty = se->prev_burst_penalty = + max(se->prev_burst_penalty, penalty); + se->damper = 1; + se->child_burst.timestamp = 0; + se->group_burst.timestamp = 0; +} + +void reset_task_bore(struct task_struct *p) { + p->se.burst_time = 0; + p->se.prev_burst_penalty = 0; + p->se.curr_burst_penalty = 0; + p->se.burst_penalty = 0; + p->se.burst_score = 0; + p->se.damper = 1; + memset(&p->se.child_burst, 0, sizeof(struct sched_burst_cache)); + memset(&p->se.group_burst, 0, sizeof(struct sched_burst_cache)); +} + +void __init sched_bore_init(void) { + printk(KERN_INFO "%s %s by %s\n", + SCHED_BORE_PROGNAME, SCHED_BORE_VERSION, SCHED_BORE_AUTHOR); + reset_task_bore(&init_task); +} + +#ifdef CONFIG_SYSCTL +static struct ctl_table sched_bore_sysctls[] = { + { + .procname = "sched_bore", + .data = &sched_bore, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = sched_bore_update_handler, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_exclude_kthreads", + .data = &sched_burst_exclude_kthreads, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_min_smooth", + .data = &sched_burst_min_smooth, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ONE, + .extra2 = &maxval_8_bits, + }, + { + .procname = "sched_burst_max_damper", + .data = &sched_burst_max_damper, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ONE, + .extra2 = &maxval_8_bits, + }, + { + .procname = "sched_burst_fork_atavistic", + .data = &sched_burst_fork_atavistic, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_THREE, + }, + { + .procname = "sched_burst_penalty_offset", + .data = &sched_burst_penalty_offset, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &maxval_6_bits, + }, + { + .procname = "sched_burst_futex_boost", + .data = &sched_burst_futex_boost, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_penalty_scale", + .data = &sched_burst_penalty_scale, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &maxval_12_bits, + }, + { + .procname = "sched_burst_cache_stop_count", + .data = &sched_burst_cache_stop_count, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec, + }, + { + .procname = "sched_burst_cache_lifetime", + .data = &sched_burst_cache_lifetime, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec, + }, + { + .procname = "sched_deadline_boost_mask", + .data = &sched_deadline_boost_mask, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec, + }, +}; + +static int __init sched_bore_sysctl_init(void) { + register_sysctl_init("kernel", sched_bore_sysctls); + return 0; +} +late_initcall(sched_bore_sysctl_init); +#endif // CONFIG_SYSCTL +#endif // CONFIG_SCHED_BORE diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 81c6df746..b6ec8ce9b 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -97,6 +97,10 @@ #include "../../io_uring/io-wq.h" #include "../smpboot.h" +#ifdef CONFIG_SCHED_BORE +#include +#endif // CONFIG_SCHED_BORE + EXPORT_TRACEPOINT_SYMBOL_GPL(ipi_send_cpu); EXPORT_TRACEPOINT_SYMBOL_GPL(ipi_send_cpumask); @@ -1423,7 +1427,11 @@ int tg_nop(struct task_group *tg, void *data) void set_load_weight(struct task_struct *p, bool update_load) { +#ifdef CONFIG_SCHED_BORE + int prio = effective_prio_bore(p); +#else // !CONFIG_SCHED_BORE int prio = p->static_prio - MAX_RT_PRIO; +#endif // CONFIG_SCHED_BORE struct load_weight lw; if (task_has_idle_policy(p)) { @@ -8523,6 +8531,10 @@ void __init sched_init(void) BUG_ON(!sched_class_above(&ext_sched_class, &idle_sched_class)); #endif +#ifdef CONFIG_SCHED_BORE + sched_bore_init(); +#endif // CONFIG_SCHED_BORE + wait_bit_init(); #ifdef CONFIG_FAIR_GROUP_SCHED diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c index 557246880..c1f6219f2 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -167,7 +167,53 @@ static const struct file_operations sched_feat_fops = { }; #ifdef CONFIG_SMP +#ifdef CONFIG_SCHED_BORE +#define DEFINE_SYSCTL_SCHED_FUNC(name, update_func) \ +static ssize_t sched_##name##_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos) \ +{ \ + char buf[16]; \ + unsigned int value; \ +\ + if (cnt > 15) \ + cnt = 15; \ +\ + if (copy_from_user(&buf, ubuf, cnt)) \ + return -EFAULT; \ + buf[cnt] = '\0'; \ +\ + if (kstrtouint(buf, 10, &value)) \ + return -EINVAL; \ +\ + sysctl_sched_##name = value; \ + sched_update_##update_func(); \ +\ + *ppos += cnt; \ + return cnt; \ +} \ +\ +static int sched_##name##_show(struct seq_file *m, void *v) \ +{ \ + seq_printf(m, "%d\n", sysctl_sched_##name); \ + return 0; \ +} \ +\ +static int sched_##name##_open(struct inode *inode, struct file *filp) \ +{ \ + return single_open(filp, sched_##name##_show, NULL); \ +} \ +\ +static const struct file_operations sched_##name##_fops = { \ + .open = sched_##name##_open, \ + .write = sched_##name##_write, \ + .read = seq_read, \ + .llseek = seq_lseek, \ + .release = single_release, \ +}; +DEFINE_SYSCTL_SCHED_FUNC(min_base_slice, min_base_slice) + +#undef DEFINE_SYSCTL_SCHED_FUNC +#else // !CONFIG_SCHED_BORE static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos) { @@ -213,7 +259,7 @@ static const struct file_operations sched_scaling_fops = { .llseek = seq_lseek, .release = single_release, }; - +#endif // CONFIG_SCHED_BORE #endif /* SMP */ #ifdef CONFIG_PREEMPT_DYNAMIC @@ -507,13 +553,20 @@ static __init int sched_init_debug(void) debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops); #endif +#ifdef CONFIG_SCHED_BORE + debugfs_create_file("min_base_slice_ns", 0644, debugfs_sched, NULL, &sched_min_base_slice_fops); + debugfs_create_u32("base_slice_ns", 0444, debugfs_sched, &sysctl_sched_base_slice); +#else // !CONFIG_SCHED_BORE debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice); +#endif // CONFIG_SCHED_BORE debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms); debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once); #ifdef CONFIG_SMP +#if !defined(CONFIG_SCHED_BORE) debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops); +#endif // CONFIG_SCHED_BORE debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); @@ -762,6 +815,9 @@ print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)), SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime))); +#ifdef CONFIG_SCHED_BORE + SEQ_printf(m, " %2d", p->se.burst_score); +#endif // CONFIG_SCHED_BORE #ifdef CONFIG_NUMA_BALANCING SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); #endif @@ -1248,6 +1304,9 @@ void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, P(se.load.weight); #ifdef CONFIG_SMP +#ifdef CONFIG_SCHED_BORE + P(se.burst_score); +#endif // CONFIG_SCHED_BORE P(se.avg.load_sum); P(se.avg.runnable_sum); P(se.avg.util_sum); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 7a14da539..e8db636d9 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -58,6 +58,10 @@ #include "stats.h" #include "autogroup.h" +#ifdef CONFIG_SCHED_BORE +#include +#endif // CONFIG_SCHED_BORE + /* * The initial- and re-scaling of tunables is configurable * @@ -67,17 +71,30 @@ * SCHED_TUNABLESCALING_LOG - scaled logarithmically, *1+ilog(ncpus) * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus * - * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus)) + * BORE : default SCHED_TUNABLESCALING_NONE = *1 constant + * EEVDF: default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus)) */ +#ifdef CONFIG_SCHED_BORE +unsigned int sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_NONE; +#else // !CONFIG_SCHED_BORE unsigned int sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG; +#endif // CONFIG_SCHED_BORE /* * Minimal preemption granularity for CPU-bound tasks: * - * (default: 0.70 msec * (1 + ilog(ncpus)), units: nanoseconds) + * BORE : base_slice = minimum multiple of nsecs_per_tick >= min_base_slice + * (default min_base_slice = 2000000 constant, units: nanoseconds) + * EEVDF: default 0.70 msec * (1 + ilog(ncpus)), units: nanoseconds */ +#ifdef CONFIG_SCHED_BORE +static const unsigned int nsecs_per_tick = 1000000000ULL / HZ; +unsigned int sysctl_sched_min_base_slice = CONFIG_MIN_BASE_SLICE_NS; +__read_mostly uint sysctl_sched_base_slice = nsecs_per_tick; +#else // !CONFIG_SCHED_BORE unsigned int sysctl_sched_base_slice = 700000ULL; static unsigned int normalized_sysctl_sched_base_slice = 700000ULL; +#endif // CONFIG_SCHED_BORE __read_mostly unsigned int sysctl_sched_migration_cost = 500000UL; @@ -191,6 +208,13 @@ static inline void update_load_set(struct load_weight *lw, unsigned long w) * * This idea comes from the SD scheduler of Con Kolivas: */ +#ifdef CONFIG_SCHED_BORE +static void update_sysctl(void) { + sysctl_sched_base_slice = nsecs_per_tick * + max(1UL, DIV_ROUND_UP(sysctl_sched_min_base_slice, nsecs_per_tick)); +} +void sched_update_min_base_slice(void) { update_sysctl(); } +#else // !CONFIG_SCHED_BORE static unsigned int get_update_sysctl_factor(void) { unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8); @@ -221,6 +245,7 @@ static void update_sysctl(void) SET_SYSCTL(sched_base_slice); #undef SET_SYSCTL } +#endif // CONFIG_SCHED_BORE void __init sched_init_granularity(void) { @@ -700,6 +725,9 @@ static void update_entity_lag(struct cfs_rq *cfs_rq, struct sched_entity *se) vlag = avg_vruntime(cfs_rq) - se->vruntime; limit = calc_delta_fair(max_t(u64, 2*se->slice, TICK_NSEC), se); +#ifdef CONFIG_SCHED_BORE + limit >>= !!sched_bore; +#endif // CONFIG_SCHED_BORE se->vlag = clamp(vlag, -limit, limit); } @@ -939,7 +967,14 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq) if (curr && (!curr->on_rq || !entity_eligible(cfs_rq, curr))) curr = NULL; +#if !defined(CONFIG_SCHED_BORE) if (sched_feat(RUN_TO_PARITY) && curr && protect_slice(curr)) +#else // CONFIG_SCHED_BORE + bool run_to_parity = likely(sched_bore) ? + sched_feat(RUN_TO_PARITY_BORE) : sched_feat(RUN_TO_PARITY); + if (run_to_parity && curr && protect_slice(curr) && + (!curr->waiting_for_futex || unlikely(!sched_bore))) +#endif // CONFIG_SCHED_BORE return curr; /* Pick the leftmost entity if it's eligible */ @@ -997,6 +1032,7 @@ struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) * Scheduling class statistics methods: */ #ifdef CONFIG_SMP +#if !defined(CONFIG_SCHED_BORE) int sched_update_scaling(void) { unsigned int factor = get_update_sysctl_factor(); @@ -1008,6 +1044,7 @@ int sched_update_scaling(void) return 0; } +#endif // CONFIG_SCHED_BORE #endif static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se); @@ -1237,6 +1274,9 @@ static void update_curr(struct cfs_rq *cfs_rq) if (unlikely(delta_exec <= 0)) return; +#ifdef CONFIG_SCHED_BORE + update_curr_bore(delta_exec, curr); +#endif // CONFIG_SCHED_BORE curr->vruntime += calc_delta_fair(delta_exec, curr); resched = update_deadline(cfs_rq, curr); update_min_vruntime(cfs_rq); @@ -3794,7 +3834,7 @@ dequeue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { } static void place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags); -static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, +void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, unsigned long weight) { bool curr = cfs_rq->curr == se; @@ -5205,12 +5245,11 @@ void __setparam_fair(struct task_struct *p, const struct sched_attr *attr) static void place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) { - u64 vslice, vruntime = avg_vruntime(cfs_rq); + u64 vslice = 0, vruntime = avg_vruntime(cfs_rq); s64 lag = 0; if (!se->custom_slice) se->slice = sysctl_sched_base_slice; - vslice = calc_delta_fair(se->slice, se); /* * Due to how V is constructed as the weighted average of entities, @@ -5295,7 +5334,16 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) se->rel_deadline = 0; return; } - +#ifdef CONFIG_SCHED_BORE + if (likely(sched_bore) && sched_burst_futex_boost && se->waiting_for_futex) + goto vslice_found; +#endif // !CONFIG_SCHED_BORE + vslice = calc_delta_fair(se->slice, se); +#ifdef CONFIG_SCHED_BORE + if (likely(sched_bore)) + vslice >>= !!(flags & sched_deadline_boost_mask); + else +#endif // CONFIG_SCHED_BORE /* * When joining the competition; the existing tasks will be, * on average, halfway through their slice, as such start tasks @@ -5304,6 +5352,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) if (sched_feat(PLACE_DEADLINE_INITIAL) && (flags & ENQUEUE_INITIAL)) vslice /= 2; +#ifdef CONFIG_SCHED_BORE +vslice_found: +#endif // CONFIG_SCHED_BORE /* * EEVDF: vd_i = ve_i + r_i/w_i */ @@ -5314,7 +5365,7 @@ static void check_enqueue_throttle(struct cfs_rq *cfs_rq); static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq); static void -requeue_delayed_entity(struct sched_entity *se); +requeue_delayed_entity(struct sched_entity *se, int flags); static void enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) @@ -5478,6 +5529,10 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) if (sched_feat(DELAY_DEQUEUE) && delay && !entity_eligible(cfs_rq, se)) { update_load_avg(cfs_rq, se, 0); +#ifdef CONFIG_SCHED_BORE + if (sched_feat(DELAY_ZERO) && likely(sched_bore)) + update_entity_lag(cfs_rq, se); +#endif // CONFIG_SCHED_BORE set_delayed(se); return false; } @@ -6894,7 +6949,7 @@ static int sched_idle_cpu(int cpu) #endif static void -requeue_delayed_entity(struct sched_entity *se) +requeue_delayed_entity(struct sched_entity *se, int flags) { struct cfs_rq *cfs_rq = cfs_rq_of(se); @@ -6907,13 +6962,22 @@ requeue_delayed_entity(struct sched_entity *se) WARN_ON_ONCE(!se->on_rq); if (sched_feat(DELAY_ZERO)) { +#ifdef CONFIG_SCHED_BORE + if (likely(sched_bore)) + flags |= ENQUEUE_WAKEUP; + else { +#endif // CONFIG_SCHED_BORE + flags = 0; update_entity_lag(cfs_rq, se); +#ifdef CONFIG_SCHED_BORE + } +#endif // CONFIG_SCHED_BORE if (se->vlag > 0) { cfs_rq->nr_queued--; if (se != cfs_rq->curr) __dequeue_entity(cfs_rq, se); se->vlag = 0; - place_entity(cfs_rq, se, 0); + place_entity(cfs_rq, se, flags); if (se != cfs_rq->curr) __enqueue_entity(cfs_rq, se); cfs_rq->nr_queued++; @@ -6950,7 +7014,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags) util_est_enqueue(&rq->cfs, p); if (flags & ENQUEUE_DELAYED) { - requeue_delayed_entity(se); + requeue_delayed_entity(se, flags); return; } @@ -6968,7 +7032,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags) for_each_sched_entity(se) { if (se->on_rq) { if (se->sched_delayed) - requeue_delayed_entity(se); + requeue_delayed_entity(se, flags); break; } cfs_rq = cfs_rq_of(se); @@ -7190,6 +7254,15 @@ static bool dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) util_est_dequeue(&rq->cfs, p); util_est_update(&rq->cfs, p, flags & DEQUEUE_SLEEP); +#ifdef CONFIG_SCHED_BORE + struct cfs_rq *cfs_rq = &rq->cfs; + struct sched_entity *se = &p->se; + if (flags & DEQUEUE_SLEEP && entity_is_task(se)) { + if (cfs_rq->curr == se) + update_curr(cfs_rq); + restart_burst(se); + } +#endif // CONFIG_SCHED_BORE if (dequeue_entities(rq, &p->se, flags) < 0) return false; @@ -9019,16 +9092,25 @@ static void yield_task_fair(struct rq *rq) /* * Are we the only task in the tree? */ +#if !defined(CONFIG_SCHED_BORE) if (unlikely(rq->nr_running == 1)) return; clear_buddies(cfs_rq, se); +#endif // CONFIG_SCHED_BORE update_rq_clock(rq); /* * Update run-time statistics of the 'current'. */ update_curr(cfs_rq); +#ifdef CONFIG_SCHED_BORE + restart_burst_rescale_deadline(se); + if (unlikely(rq->nr_running == 1)) + return; + + clear_buddies(cfs_rq, se); +#endif // CONFIG_SCHED_BORE /* * Tell update_rq_clock() that we've just updated, * so we don't do microscopic update in schedule() @@ -13142,6 +13224,9 @@ static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) static void task_fork_fair(struct task_struct *p) { set_task_max_allowed_capacity(p); +#ifdef CONFIG_SCHED_BORE + update_burst_score(&p->se); +#endif // CONFIG_SCHED_BORE } /* @@ -13259,6 +13344,9 @@ static void switched_to_fair(struct rq *rq, struct task_struct *p) { WARN_ON_ONCE(p->se.sched_delayed); +#ifdef CONFIG_SCHED_BORE + reset_task_bore(p); +#endif // CONFIG_SCHED_BORE attach_task_cfs_rq(p); set_task_max_allowed_capacity(p); diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 3c12d9f93..b87171053 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -18,6 +18,9 @@ SCHED_FEAT(PLACE_REL_DEADLINE, true) * 0-lag point or until is has exhausted it's slice. */ SCHED_FEAT(RUN_TO_PARITY, true) +#ifdef CONFIG_SCHED_BORE +SCHED_FEAT(RUN_TO_PARITY_BORE, false) +#endif // CONFIG_SCHED_BORE /* * Allow wakeup of tasks with a shorter slice to cancel RUN_TO_PARITY for * current. diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 83e3aa917..ef5d684df 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -2119,7 +2119,11 @@ extern int group_balance_cpu(struct sched_group *sg); extern void update_sched_domain_debugfs(void); extern void dirty_sched_domain_sysctl(int cpu); +#ifdef CONFIG_SCHED_BORE +extern void sched_update_min_base_slice(void); +#else // !CONFIG_SCHED_BORE extern int sched_update_scaling(void); +#endif // CONFIG_SCHED_BORE static inline const struct cpumask *task_user_cpus(struct task_struct *p) { @@ -2825,7 +2829,12 @@ extern void wakeup_preempt(struct rq *rq, struct task_struct *p, int flags); extern __read_mostly unsigned int sysctl_sched_nr_migrate; extern __read_mostly unsigned int sysctl_sched_migration_cost; +#ifdef CONFIG_SCHED_BORE +extern unsigned int sysctl_sched_min_base_slice; +extern __read_mostly uint sysctl_sched_base_slice; +#else // !CONFIG_SCHED_BORE extern unsigned int sysctl_sched_base_slice; +#endif // CONFIG_SCHED_BORE extern int sysctl_resched_latency_warn_ms; extern int sysctl_resched_latency_warn_once; -- 2.50.1