From b81d0a14e0383fb596314320f9853e9c416545bb Mon Sep 17 00:00:00 2001 From: Masahito S Date: Thu, 28 Mar 2024 00:38:44 +0900 Subject: [PATCH] linux6.1.y-bore5.1.0 --- include/linux/sched.h | 10 +++ init/Kconfig | 17 ++++ kernel/sched/core.c | 143 ++++++++++++++++++++++++++++++++ kernel/sched/debug.c | 3 + kernel/sched/fair.c | 187 ++++++++++++++++++++++++++++++++++++++---- 5 files changed, 346 insertions(+), 14 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index 0cac69902e..159733cf83 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -555,6 +555,16 @@ struct sched_entity { u64 sum_exec_runtime; u64 vruntime; u64 prev_sum_exec_runtime; +#ifdef CONFIG_SCHED_BORE + u64 burst_time; + u8 prev_burst_penalty; + u8 curr_burst_penalty; + u8 burst_penalty; + u8 burst_score; + u8 child_burst; + u32 child_burst_cnt; + u64 child_burst_last_cached; +#endif // CONFIG_SCHED_BORE u64 nr_migrations; diff --git a/init/Kconfig b/init/Kconfig index 1487046402..31d5bf4f5e 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1280,6 +1280,23 @@ 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/sched/core.c b/kernel/sched/core.c index 18a4f8f28a..c9270d4a77 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4368,6 +4368,138 @@ int wake_up_state(struct task_struct *p, unsigned int state) return try_to_wake_up(p, state, 0); } +#ifdef CONFIG_SCHED_BORE +extern u8 sched_burst_fork_atavistic; +extern uint sched_burst_cache_lifetime; + +static void __init sched_init_bore(void) { + init_task.se.burst_time = 0; + init_task.se.prev_burst_penalty = 0; + init_task.se.curr_burst_penalty = 0; + init_task.se.burst_penalty = 0; + init_task.se.burst_score = 0; + init_task.se.child_burst_last_cached = 0; +} + +void inline sched_fork_bore(struct task_struct *p) { + p->se.burst_time = 0; + p->se.curr_burst_penalty = 0; + p->se.burst_score = 0; + p->se.child_burst_last_cached = 0; +} + +static u32 count_child_tasks(struct task_struct *p) { + struct task_struct *child; + u32 cnt = 0; + list_for_each_entry(child, &p->children, sibling) {cnt++;} + return cnt; +} + +static inline bool task_is_inheritable(struct task_struct *p) { + return (p->sched_class == &fair_sched_class); +} + +static inline bool child_burst_cache_expired(struct task_struct *p, u64 now) { + u64 expiration_time = + p->se.child_burst_last_cached + sched_burst_cache_lifetime; + return ((s64)(expiration_time - now) < 0); +} + +static void __update_child_burst_cache( + struct task_struct *p, u32 cnt, u32 sum, u64 now) { + u8 avg = 0; + if (cnt) avg = sum / cnt; + p->se.child_burst = max(avg, p->se.burst_penalty); + p->se.child_burst_cnt = cnt; + p->se.child_burst_last_cached = now; +} + +static inline void update_child_burst_direct(struct task_struct *p, u64 now) { + struct task_struct *child; + u32 cnt = 0; + u32 sum = 0; + + list_for_each_entry(child, &p->children, sibling) { + if (!task_is_inheritable(child)) continue; + cnt++; + sum += child->se.burst_penalty; + } + + __update_child_burst_cache(p, cnt, sum, now); +} + +static inline u8 __inherit_burst_direct(struct task_struct *p, u64 now) { + struct task_struct *parent = p->real_parent; + if (child_burst_cache_expired(parent, now)) + update_child_burst_direct(parent, now); + + return parent->se.child_burst; +} + +static void update_child_burst_topological( + struct task_struct *p, u64 now, u32 depth, u32 *acnt, u32 *asum) { + struct task_struct *child, *dec; + u32 cnt = 0, dcnt = 0; + u32 sum = 0; + + list_for_each_entry(child, &p->children, sibling) { + dec = child; + while ((dcnt = count_child_tasks(dec)) == 1) + dec = list_first_entry(&dec->children, struct task_struct, sibling); + + if (!dcnt || !depth) { + if (!task_is_inheritable(dec)) continue; + cnt++; + sum += dec->se.burst_penalty; + continue; + } + if (!child_burst_cache_expired(dec, now)) { + cnt += dec->se.child_burst_cnt; + sum += (u32)dec->se.child_burst * dec->se.child_burst_cnt; + continue; + } + update_child_burst_topological(dec, now, depth - 1, &cnt, &sum); + } + + __update_child_burst_cache(p, cnt, sum, now); + *acnt += cnt; + *asum += sum; +} + +static inline u8 __inherit_burst_topological(struct task_struct *p, u64 now) { + struct task_struct *anc = p->real_parent; + u32 cnt = 0, sum = 0; + + while (anc->real_parent != anc && count_child_tasks(anc) == 1) + anc = anc->real_parent; + + if (child_burst_cache_expired(anc, now)) + update_child_burst_topological( + anc, now, sched_burst_fork_atavistic - 1, &cnt, &sum); + + return anc->se.child_burst; +} + +static inline void inherit_burst(struct task_struct *p) { + u8 burst_cache; + u64 now = ktime_get_ns(); + + read_lock(&tasklist_lock); + burst_cache = likely(sched_burst_fork_atavistic)? + __inherit_burst_topological(p, now): + __inherit_burst_direct(p, now); + read_unlock(&tasklist_lock); + + p->se.prev_burst_penalty = max(p->se.prev_burst_penalty, burst_cache); +} + +static void sched_post_fork_bore(struct task_struct *p) { + if (p->sched_class == &fair_sched_class) + inherit_burst(p); + p->se.burst_penalty = p->se.prev_burst_penalty; +} +#endif // CONFIG_SCHED_BORE + /* * Perform scheduler related setup for a newly forked process p. * p is forked by current. @@ -4384,6 +4516,9 @@ static void __sched_fork(unsigned long clone_flags, struct task_struct *p) p->se.prev_sum_exec_runtime = 0; p->se.nr_migrations = 0; p->se.vruntime = 0; +#ifdef CONFIG_SCHED_BORE + sched_fork_bore(p); +#endif // CONFIG_SCHED_BORE INIT_LIST_HEAD(&p->se.group_node); #ifdef CONFIG_FAIR_GROUP_SCHED @@ -4689,6 +4824,9 @@ void sched_cgroup_fork(struct task_struct *p, struct kernel_clone_args *kargs) void sched_post_fork(struct task_struct *p) { +#ifdef CONFIG_SCHED_BORE + sched_post_fork_bore(p); +#endif // CONFIG_SCHED_BORE uclamp_post_fork(p); } @@ -9675,6 +9813,11 @@ void __init sched_init(void) BUG_ON(&dl_sched_class != &stop_sched_class + 1); #endif +#ifdef CONFIG_SCHED_BORE + sched_init_bore(); + printk(KERN_INFO "BORE (Burst-Oriented Response Enhancer) CPU Scheduler modification 5.1.0 by Masahito Suzuki"); +#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 1637b65ba0..a0157ac1be 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -547,6 +547,9 @@ 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 diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 2558ab9033..138e1901ea 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -19,6 +19,9 @@ * * Adaptive scheduling granularity, math enhancements by Peter Zijlstra * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra + * + * Burst-Oriented Response Enhancer (BORE) CPU Scheduler + * Copyright (C) 2021-2024 Masahito Suzuki */ #include #include @@ -126,6 +129,76 @@ static unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL; const_debug unsigned int sysctl_sched_migration_cost = 500000UL; +#ifdef CONFIG_SCHED_BORE +u8 __read_mostly sched_bore = 1; +u8 __read_mostly sched_burst_smoothness_long = 1; +u8 __read_mostly sched_burst_smoothness_short = 0; +u8 __read_mostly sched_burst_fork_atavistic = 2; +u8 __read_mostly sched_burst_penalty_offset = 22; +uint __read_mostly sched_burst_penalty_scale = 1280; +uint __read_mostly sched_burst_cache_lifetime = 60000000; +static int __maybe_unused sixty_four = 64; +static int __maybe_unused maxval_12_bits = 4095; + +#define MAX_BURST_PENALTY (39U <<2) + +static inline u32 log2plus1_u64_u32f8(u64 v) { + u32 msb = fls64(v); + s32 excess_bits = msb - 9; + u8 fractional = (0 <= excess_bits)? v >> excess_bits: v << -excess_bits; + return msb << 8 | fractional; +} + +static inline u32 calc_burst_penalty(u64 burst_time) { + u32 greed, tolerance, penalty, scaled_penalty; + + greed = log2plus1_u64_u32f8(burst_time); + tolerance = sched_burst_penalty_offset << 8; + penalty = max(0, (s32)greed - (s32)tolerance); + scaled_penalty = penalty * sched_burst_penalty_scale >> 16; + + return min(MAX_BURST_PENALTY, scaled_penalty); +} + +static inline u64 scale_slice(u64 delta, struct sched_entity *se) { + return mul_u64_u32_shr(delta, sched_prio_to_wmult[se->burst_score], 22); +} + +static void update_burst_score(struct sched_entity *se) { + if (!entity_is_task(se)) return; + struct task_struct *p = task_of(se); + u8 prio = p->static_prio - MAX_RT_PRIO; + u8 prev_prio = min(39, prio + se->burst_score); + + se->burst_score = se->burst_penalty >> 2; + + u8 new_prio = min(39, prio + se->burst_score); + if (new_prio != prev_prio) + reweight_task(p, new_prio); +} + +static void update_burst_penalty(struct sched_entity *se) { + se->curr_burst_penalty = calc_burst_penalty(se->burst_time); + se->burst_penalty = max(se->prev_burst_penalty, se->curr_burst_penalty); + update_burst_score(se); +} + +static inline u32 binary_smooth(u32 new, u32 old) { + int increment = new - old; + return (0 <= increment)? + old + ( increment >> (int)sched_burst_smoothness_long): + old - (-increment >> (int)sched_burst_smoothness_short); +} + +static void restart_burst(struct sched_entity *se) { + se->burst_penalty = se->prev_burst_penalty = + binary_smooth(se->curr_burst_penalty, se->prev_burst_penalty); + se->curr_burst_penalty = 0; + se->burst_time = 0; + update_burst_score(se); +} +#endif // CONFIG_SCHED_BORE + int sched_thermal_decay_shift; static int __init setup_sched_thermal_decay_shift(char *str) { @@ -180,6 +253,69 @@ static unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL; #ifdef CONFIG_SYSCTL static struct ctl_table sched_fair_sysctls[] = { +#ifdef CONFIG_SCHED_BORE + { + .procname = "sched_bore", + .data = &sched_bore, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ONE, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_smoothness_long", + .data = &sched_burst_smoothness_long, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_smoothness_short", + .data = &sched_burst_smoothness_short, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .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 = &sixty_four, + }, + { + .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_lifetime", + .data = &sched_burst_cache_lifetime, + .maxlen = sizeof(uint), + .mode = 0644, + .proc_handler = proc_douintvec, + }, +#endif // CONFIG_SCHED_BORE { .procname = "sched_child_runs_first", .data = &sysctl_sched_child_runs_first, @@ -695,7 +831,6 @@ static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se) { if (unlikely(se->load.weight != NICE_0_LOAD)) delta = __calc_delta(delta, NICE_0_LOAD, &se->load); - return delta; } @@ -905,7 +1040,13 @@ static void update_curr(struct cfs_rq *cfs_rq) curr->sum_exec_runtime += delta_exec; schedstat_add(cfs_rq->exec_clock, delta_exec); +#ifdef CONFIG_SCHED_BORE + curr->burst_time += delta_exec; + update_burst_penalty(curr); + curr->vruntime += max(1ULL, calc_delta_fair(delta_exec, curr)); +#else // !CONFIG_SCHED_BORE curr->vruntime += calc_delta_fair(delta_exec, curr); +#endif // CONFIG_SCHED_BORE update_min_vruntime(cfs_rq); if (entity_is_task(curr)) { @@ -6171,6 +6312,14 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags) bool was_sched_idle = sched_idle_rq(rq); util_est_dequeue(&rq->cfs, p); +#ifdef CONFIG_SCHED_BORE + if (task_sleep) { + cfs_rq = cfs_rq_of(se); + if (cfs_rq->curr == se) + update_curr(cfs_rq); + restart_burst(se); + } +#endif // CONFIG_SCHED_BORE for_each_sched_entity(se) { cfs_rq = cfs_rq_of(se); @@ -7939,24 +8088,31 @@ 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 - if (curr->policy != SCHED_BATCH) { - update_rq_clock(rq); - /* - * Update run-time statistics of the 'current'. - */ - update_curr(cfs_rq); - /* - * Tell update_rq_clock() that we've just updated, - * so we don't do microscopic update in schedule() - * and double the fastpath cost. - */ - rq_clock_skip_update(rq); - } + update_rq_clock(rq); + /* + * Update run-time statistics of the 'current'. + */ + update_curr(cfs_rq); +#ifdef CONFIG_SCHED_BORE + restart_burst(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() + * and double the fastpath cost. + */ + rq_clock_skip_update(rq); set_skip_buddy(se); } @@ -11869,6 +12025,9 @@ static void task_fork_fair(struct task_struct *p) update_curr(cfs_rq); se->vruntime = curr->vruntime; } +#ifdef CONFIG_SCHED_BORE + update_burst_score(se); +#endif // CONFIG_SCHED_BORE place_entity(cfs_rq, se, 1); if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) { -- 2.34.1