From 1ab81cfa061f454316364a32761ce45a7479e616 Mon Sep 17 00:00:00 2001 From: Piotr Gorski Date: Thu, 7 Mar 2024 22:28:47 +0100 Subject: [PATCH] bore-cachy Signed-off-by: Piotr Gorski --- include/linux/sched.h | 12 ++ init/Kconfig | 19 +++ kernel/sched/core.c | 148 +++++++++++++++++++ kernel/sched/debug.c | 61 +++++++- kernel/sched/fair.c | 319 ++++++++++++++++++++++++++++++++++++---- kernel/sched/features.h | 4 + kernel/sched/sched.h | 7 + 7 files changed, 542 insertions(+), 28 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index ffe8f618a..7ac6163f9 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -547,6 +547,18 @@ struct sched_entity { u64 sum_exec_runtime; u64 prev_sum_exec_runtime; u64 vruntime; +#ifdef CONFIG_SCHED_BORE + u64 burst_time; + u8 prev_burst_penalty; + u8 curr_burst_penalty; + u8 burst_penalty; + u8 burst_score; + u32 burst_load; + bool on_cfs_rq; + u8 child_burst; + u32 child_burst_cnt; + u64 child_burst_last_cached; +#endif // CONFIG_SCHED_BORE s64 vlag; u64 slice; diff --git a/init/Kconfig b/init/Kconfig index 47671886d..c99132cf6 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1299,6 +1299,25 @@ 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. + + You can turn it off by setting the sysctl kernel.sched_bore = 0. + + 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 9116bcc90..43e4311db 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -4507,6 +4507,143 @@ int wake_up_state(struct task_struct *p, unsigned int state) return try_to_wake_up(p, state, 0); } +#ifdef CONFIG_SCHED_BORE +extern bool 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.on_cfs_rq = false; + init_task.se.child_burst_last_cached = 0; + init_task.se.burst_load = 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.on_cfs_rq = false; + p->se.child_burst_last_cached = 0; + p->se.burst_load = 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 && likely(sched_bore)) + 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. @@ -4523,6 +4660,9 @@ 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 p->se.vlag = 0; p->se.slice = sysctl_sched_base_slice; INIT_LIST_HEAD(&p->se.group_node); @@ -4839,6 +4979,9 @@ 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); } @@ -9910,6 +10053,11 @@ 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 4.5.2 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 8d5d98a58..a565363fd 100644 --- a/kernel/sched/debug.c +++ b/kernel/sched/debug.c @@ -167,7 +167,52 @@ static const struct file_operations sched_feat_fops = { }; #ifdef CONFIG_SMP +#ifdef CONFIG_SCHED_BORE +static ssize_t sched_min_base_slice_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; + if (!value) + return -EINVAL; + + sysctl_sched_min_base_slice = value; + sched_update_min_base_slice(); + + *ppos += cnt; + return cnt; +} + +static int sched_min_base_slice_show(struct seq_file *m, void *v) +{ + seq_printf(m, "%d\n", sysctl_sched_min_base_slice); + return 0; +} + +static int sched_min_base_slice_open(struct inode *inode, struct file *filp) +{ + return single_open(filp, sched_min_base_slice_show, NULL); +} + +static const struct file_operations sched_min_base_slice_fops = { + .open = sched_min_base_slice_open, + .write = sched_min_base_slice_write, + .read = seq_read, + .llseek = seq_lseek, + .release = single_release, +}; +#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 +258,7 @@ static const struct file_operations sched_scaling_fops = { .llseek = seq_lseek, .release = single_release, }; - +#endif // CONFIG_SCHED_BORE #endif /* SMP */ #ifdef CONFIG_PREEMPT_DYNAMIC @@ -347,13 +392,20 @@ 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", 0400, 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); @@ -595,6 +647,9 @@ 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 @@ -1068,6 +1123,10 @@ 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_load); + 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 fc0a9de42..3ee4e7e70 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 @@ -64,28 +67,128 @@ * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *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: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds) + * (BORE default: max(1 sec / HZ, min_base_slice) constant, units: nanoseconds) + * (EEVDF default: 0.75 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 +#ifdef CONFIG_SCHED_BORE +unsigned int sysctl_sched_base_slice = 1000000000ULL / HZ; +static unsigned int configured_sched_base_slice = 1000000000ULL / HZ; +unsigned int sysctl_sched_min_base_slice = 2000000ULL; +#else // !CONFIG_SCHED_BORE unsigned int sysctl_sched_base_slice = 750000ULL; static unsigned int normalized_sysctl_sched_base_slice = 750000ULL; -#endif +#endif // CONFIG_SCHED_BORE -#ifdef CONFIG_CACHY -const_debug unsigned int sysctl_sched_migration_cost = 300000UL; -#else const_debug unsigned int sysctl_sched_migration_cost = 500000UL; -#endif + +#ifdef CONFIG_SCHED_BORE +u8 __read_mostly sched_bore = 1; +u8 __read_mostly sched_burst_score_rounding = 0; +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; +u8 __read_mostly sched_vlag_deviation_limit = 11; +static int __maybe_unused thirty_two = 32; +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 inline u64 __unscale_slice(u64 delta, u8 score) { + return mul_u64_u32_shr(delta, sched_prio_to_weight[score], 10); +} + +static inline u64 unscale_slice(u64 delta, struct sched_entity *se) { + return __unscale_slice(delta, se->burst_score); +} + +static void avg_vruntime_add(struct cfs_rq *cfs_rq, struct sched_entity *se); +static void avg_vruntime_sub(struct cfs_rq *cfs_rq, struct sched_entity *se); + +static void update_burst_score(struct sched_entity *se) { + struct cfs_rq *cfs_rq = cfs_rq_of(se); + u8 prev_score = se->burst_score; + u32 penalty = se->burst_penalty; + if (sched_burst_score_rounding) penalty += 0x2U; + se->burst_score = penalty >> 2; + + if ((se->burst_score != prev_score) && se->on_cfs_rq) { + avg_vruntime_sub(cfs_rq, se); + avg_vruntime_add(cfs_rq, se); + } +} + +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); +} + +static void restart_burst_rescale_deadline(struct sched_entity *se) { + s64 vscaled, wremain, vremain = se->deadline - se->vruntime; + u8 prev_score = se->burst_score; + restart_burst(se); + if (prev_score > se->burst_score) { + wremain = __unscale_slice(abs(vremain), prev_score); + vscaled = scale_slice(wremain, se); + if (unlikely(vremain < 0)) + vscaled = -vscaled; + se->deadline = se->vruntime + vscaled; + } +} +#endif // CONFIG_SCHED_BORE int sched_thermal_decay_shift; static int __init setup_sched_thermal_decay_shift(char *str) @@ -136,12 +239,8 @@ 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. */ @@ -150,6 +249,87 @@ static unsigned int sysctl_numa_balancing_promote_rate_limit = 65536; #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_ZERO, + .extra2 = SYSCTL_ONE, + }, + { + .procname = "sched_burst_score_rounding", + .data = &sched_burst_score_rounding, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .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, + }, + { + .procname = "sched_vlag_deviation_limit", + .data = &sched_vlag_deviation_limit, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &thirty_two, + }, +#endif // CONFIG_SCHED_BORE #ifdef CONFIG_CFS_BANDWIDTH { .procname = "sched_cfs_bandwidth_slice_us", @@ -208,6 +388,13 @@ 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 = + max(sysctl_sched_min_base_slice, configured_sched_base_slice); +} +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); @@ -238,6 +425,7 @@ static void update_sysctl(void) SET_SYSCTL(sched_base_slice); #undef SET_SYSCTL } +#endif // CONFIG_SCHED_BORE void __init sched_init_granularity(void) { @@ -311,6 +499,9 @@ 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); +#ifdef CONFIG_SCHED_BORE + if (likely(sched_bore)) delta = scale_slice(delta, se); +#endif // CONFIG_SCHED_BORE return delta; } @@ -637,10 +828,26 @@ static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) * * As measured, the max (key * weight) value was ~44 bits for a kernel build. */ +#if !defined(CONFIG_SCHED_BORE) +#define entity_weight(se) scale_load_down(se->load.weight) +#else // CONFIG_SCHED_BORE +static unsigned long entity_weight(struct sched_entity *se) { + unsigned long weight = se->load.weight; + if (likely(sched_bore)) weight = unscale_slice(weight, se); +#ifdef CONFIG_64BIT + weight >>= SCHED_FIXEDPOINT_SHIFT - 3; +#endif // CONFIG_64BIT + return weight; +} +#endif // CONFIG_SCHED_BORE + static void avg_vruntime_add(struct cfs_rq *cfs_rq, struct sched_entity *se) { - unsigned long weight = scale_load_down(se->load.weight); + unsigned long weight = entity_weight(se); +#ifdef CONFIG_SCHED_BORE + se->burst_load = weight; +#endif // CONFIG_SCHED_BORE s64 key = entity_key(cfs_rq, se); cfs_rq->avg_vruntime += key * weight; @@ -650,7 +857,12 @@ avg_vruntime_add(struct cfs_rq *cfs_rq, struct sched_entity *se) static void avg_vruntime_sub(struct cfs_rq *cfs_rq, struct sched_entity *se) { - unsigned long weight = scale_load_down(se->load.weight); +#if !defined(CONFIG_SCHED_BORE) + unsigned long weight = entity_weight(se); +#else // CONFIG_SCHED_BORE + unsigned long weight = se->burst_load; + se->burst_load = 0; +#endif // CONFIG_SCHED_BORE s64 key = entity_key(cfs_rq, se); cfs_rq->avg_vruntime -= key * weight; @@ -670,14 +882,14 @@ void avg_vruntime_update(struct cfs_rq *cfs_rq, s64 delta) * Specifically: avg_runtime() + 0 must result in entity_eligible() := true * For this to be so, the result of this function must have a left bias. */ -u64 avg_vruntime(struct cfs_rq *cfs_rq) +static u64 avg_key(struct cfs_rq *cfs_rq) { struct sched_entity *curr = cfs_rq->curr; s64 avg = cfs_rq->avg_vruntime; long load = cfs_rq->avg_load; if (curr && curr->on_rq) { - unsigned long weight = scale_load_down(curr->load.weight); + unsigned long weight = entity_weight(curr); avg += entity_key(cfs_rq, curr) * weight; load += weight; @@ -687,12 +899,15 @@ u64 avg_vruntime(struct cfs_rq *cfs_rq) /* sign flips effective floor / ceil */ if (avg < 0) avg -= (load - 1); - avg = div_s64(avg, load); + avg = div64_s64(avg, load); } - return cfs_rq->min_vruntime + avg; + return avg; } +u64 avg_vruntime(struct cfs_rq *cfs_rq) { + return cfs_rq->min_vruntime + avg_key(cfs_rq); +} /* * lag_i = S - s_i = w_i * (V - v_i) * @@ -717,6 +932,9 @@ static void update_entity_lag(struct cfs_rq *cfs_rq, struct sched_entity *se) lag = avg_vruntime(cfs_rq) - se->vruntime; limit = calc_delta_fair(max_t(u64, 2*se->slice, TICK_NSEC), se); +#ifdef CONFIG_SCHED_BORE + if (likely(sched_bore)) limit >>= 1; +#endif // CONFIG_SCHED_BORE se->vlag = clamp(lag, -limit, limit); } @@ -744,7 +962,7 @@ static int vruntime_eligible(struct cfs_rq *cfs_rq, u64 vruntime) long load = cfs_rq->avg_load; if (curr && curr->on_rq) { - unsigned long weight = scale_load_down(curr->load.weight); + unsigned long weight = entity_weight(curr); avg += entity_key(cfs_rq, curr) * weight; load += weight; @@ -840,10 +1058,16 @@ static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) se->min_vruntime = se->vruntime; rb_add_augmented_cached(&se->run_node, &cfs_rq->tasks_timeline, __entity_less, &min_vruntime_cb); +#ifdef CONFIG_SCHED_BORE + se->on_cfs_rq = true; +#endif // CONFIG_SCHED_BORE } static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) { +#ifdef CONFIG_SCHED_BORE + se->on_cfs_rq = false; +#endif // CONFIG_SCHED_BORE rb_erase_augmented_cached(&se->run_node, &cfs_rq->tasks_timeline, &min_vruntime_cb); avg_vruntime_sub(cfs_rq, se); @@ -968,6 +1192,7 @@ 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(); @@ -979,6 +1204,7 @@ int sched_update_scaling(void) return 0; } +#endif // CONFIG_SCHED_BORE #endif #endif @@ -1178,7 +1404,13 @@ static void update_curr(struct cfs_rq *cfs_rq) if (unlikely(delta_exec <= 0)) return; +#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_deadline(cfs_rq, curr); update_min_vruntime(cfs_rq); @@ -5170,8 +5402,8 @@ static inline void update_misfit_status(struct task_struct *p, struct rq *rq) {} static void place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) { - u64 vslice, vruntime = avg_vruntime(cfs_rq); - s64 lag = 0; + s64 lag = 0, key = avg_key(cfs_rq); + u64 vslice, vruntime = cfs_rq->min_vruntime + key; se->slice = sysctl_sched_base_slice; vslice = calc_delta_fair(se->slice, se); @@ -5184,6 +5416,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) * * EEVDF: placement strategy #1 / #2 */ +#ifdef CONFIG_SCHED_BORE + if (unlikely(!sched_bore) || se->vlag) +#endif // CONFIG_SCHED_BORE if (sched_feat(PLACE_LAG) && cfs_rq->nr_running) { struct sched_entity *curr = cfs_rq->curr; unsigned long load; @@ -5244,12 +5479,22 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) */ load = cfs_rq->avg_load; if (curr && curr->on_rq) - load += scale_load_down(curr->load.weight); + load += entity_weight(curr); - lag *= load + scale_load_down(se->load.weight); + lag *= load + entity_weight(se); +#if !defined(CONFIG_SCHED_BORE) if (WARN_ON_ONCE(!load)) +#else // CONFIG_SCHED_BORE + if (unlikely(!load)) +#endif // CONFIG_SCHED_BORE load = 1; - lag = div_s64(lag, load); + lag = div64_s64(lag, load); +#ifdef CONFIG_SCHED_BORE + if (likely(sched_bore)) { + s64 limit = vslice << sched_vlag_deviation_limit; + lag = clamp(lag, -limit, limit); + } +#endif // CONFIG_SCHED_BORE } se->vruntime = vruntime - lag; @@ -6816,6 +7061,14 @@ 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); @@ -8565,16 +8818,25 @@ 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() @@ -12664,6 +12926,9 @@ static void task_fork_fair(struct task_struct *p) curr = cfs_rq->curr; if (curr) update_curr(cfs_rq); +#ifdef CONFIG_SCHED_BORE + update_burst_score(se); +#endif // CONFIG_SCHED_BORE place_entity(cfs_rq, se, ENQUEUE_INITIAL); rq_unlock(rq, &rf); } diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 143f55df8..3f0fe409f 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -6,7 +6,11 @@ */ SCHED_FEAT(PLACE_LAG, true) SCHED_FEAT(PLACE_DEADLINE_INITIAL, true) +#ifdef CONFIG_SCHED_BORE +SCHED_FEAT(RUN_TO_PARITY, false) +#else // !CONFIG_SCHED_BORE SCHED_FEAT(RUN_TO_PARITY, true) +#endif // CONFIG_SCHED_BORE /* * Prefer to schedule the task we woke last (assuming it failed diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index ed5c758c7..9d62372ae 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -1965,7 +1965,11 @@ static inline void dirty_sched_domain_sysctl(int cpu) } #endif +#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) { @@ -2552,6 +2556,9 @@ extern const_debug unsigned int sysctl_sched_nr_migrate; extern const_debug unsigned int sysctl_sched_migration_cost; extern unsigned int sysctl_sched_base_slice; +#ifdef CONFIG_SCHED_BORE +extern unsigned int sysctl_sched_min_base_slice; +#endif // CONFIG_SCHED_BORE #ifdef CONFIG_SCHED_DEBUG extern int sysctl_resched_latency_warn_ms; -- 2.43.0.232.ge79552d197