2 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
4 * Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
6 * Interactivity improvements by Mike Galbraith
7 * (C) 2007 Mike Galbraith <efault@gmx.de>
9 * Various enhancements by Dmitry Adamushko.
10 * (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
12 * Group scheduling enhancements by Srivatsa Vaddagiri
13 * Copyright IBM Corporation, 2007
14 * Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
16 * Scaled math optimizations by Thomas Gleixner
17 * Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
19 * Adaptive scheduling granularity, math enhancements by Peter Zijlstra
20 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
23 #include <linux/latencytop.h>
24 #include <linux/sched.h>
25 #include <linux/cpumask.h>
26 #include <linux/slab.h>
27 #include <linux/profile.h>
28 #include <linux/interrupt.h>
29 #include <linux/mempolicy.h>
30 #include <linux/migrate.h>
31 #include <linux/task_work.h>
33 #include <trace/events/sched.h>
38 * Targeted preemption latency for CPU-bound tasks:
39 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
41 * NOTE: this latency value is not the same as the concept of
42 * 'timeslice length' - timeslices in CFS are of variable length
43 * and have no persistent notion like in traditional, time-slice
44 * based scheduling concepts.
46 * (to see the precise effective timeslice length of your workload,
47 * run vmstat and monitor the context-switches (cs) field)
49 unsigned int sysctl_sched_latency = 6000000ULL;
50 unsigned int normalized_sysctl_sched_latency = 6000000ULL;
53 * The initial- and re-scaling of tunables is configurable
54 * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
57 * SCHED_TUNABLESCALING_NONE - unscaled, always *1
58 * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
59 * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
61 enum sched_tunable_scaling sysctl_sched_tunable_scaling
62 = SCHED_TUNABLESCALING_LOG;
65 * Minimal preemption granularity for CPU-bound tasks:
66 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
68 unsigned int sysctl_sched_min_granularity = 750000ULL;
69 unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
72 * is kept at sysctl_sched_latency / sysctl_sched_min_granularity
74 static unsigned int sched_nr_latency = 8;
77 * After fork, child runs first. If set to 0 (default) then
78 * parent will (try to) run first.
80 unsigned int sysctl_sched_child_runs_first __read_mostly;
83 * SCHED_OTHER wake-up granularity.
84 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
86 * This option delays the preemption effects of decoupled workloads
87 * and reduces their over-scheduling. Synchronous workloads will still
88 * have immediate wakeup/sleep latencies.
90 unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
91 unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL;
93 const_debug unsigned int sysctl_sched_migration_cost = 500000UL;
96 * The exponential sliding window over which load is averaged for shares
100 unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;
102 #ifdef CONFIG_CFS_BANDWIDTH
104 * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
105 * each time a cfs_rq requests quota.
107 * Note: in the case that the slice exceeds the runtime remaining (either due
108 * to consumption or the quota being specified to be smaller than the slice)
109 * we will always only issue the remaining available time.
111 * default: 5 msec, units: microseconds
113 unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL;
117 * Increase the granularity value when there are more CPUs,
118 * because with more CPUs the 'effective latency' as visible
119 * to users decreases. But the relationship is not linear,
120 * so pick a second-best guess by going with the log2 of the
123 * This idea comes from the SD scheduler of Con Kolivas:
125 static int get_update_sysctl_factor(void)
127 unsigned int cpus = min_t(int, num_online_cpus(), 8);
130 switch (sysctl_sched_tunable_scaling) {
131 case SCHED_TUNABLESCALING_NONE:
134 case SCHED_TUNABLESCALING_LINEAR:
137 case SCHED_TUNABLESCALING_LOG:
139 factor = 1 + ilog2(cpus);
146 static void update_sysctl(void)
148 unsigned int factor = get_update_sysctl_factor();
150 #define SET_SYSCTL(name) \
151 (sysctl_##name = (factor) * normalized_sysctl_##name)
152 SET_SYSCTL(sched_min_granularity);
153 SET_SYSCTL(sched_latency);
154 SET_SYSCTL(sched_wakeup_granularity);
158 void sched_init_granularity(void)
163 #if BITS_PER_LONG == 32
164 # define WMULT_CONST (~0UL)
166 # define WMULT_CONST (1UL << 32)
169 #define WMULT_SHIFT 32
172 * Shift right and round:
174 #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
177 * delta *= weight / lw
180 calc_delta_mine(unsigned long delta_exec, unsigned long weight,
181 struct load_weight *lw)
186 * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched
187 * entities since MIN_SHARES = 2. Treat weight as 1 if less than
188 * 2^SCHED_LOAD_RESOLUTION.
190 if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION)))
191 tmp = (u64)delta_exec * scale_load_down(weight);
193 tmp = (u64)delta_exec;
195 if (!lw->inv_weight) {
196 unsigned long w = scale_load_down(lw->weight);
198 if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
200 else if (unlikely(!w))
201 lw->inv_weight = WMULT_CONST;
203 lw->inv_weight = WMULT_CONST / w;
207 * Check whether we'd overflow the 64-bit multiplication:
209 if (unlikely(tmp > WMULT_CONST))
210 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
213 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
215 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
219 const struct sched_class fair_sched_class;
221 /**************************************************************
222 * CFS operations on generic schedulable entities:
225 #ifdef CONFIG_FAIR_GROUP_SCHED
227 /* cpu runqueue to which this cfs_rq is attached */
228 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
233 /* An entity is a task if it doesn't "own" a runqueue */
234 #define entity_is_task(se) (!se->my_q)
236 static inline struct task_struct *task_of(struct sched_entity *se)
238 #ifdef CONFIG_SCHED_DEBUG
239 WARN_ON_ONCE(!entity_is_task(se));
241 return container_of(se, struct task_struct, se);
244 /* Walk up scheduling entities hierarchy */
245 #define for_each_sched_entity(se) \
246 for (; se; se = se->parent)
248 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
253 /* runqueue on which this entity is (to be) queued */
254 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
259 /* runqueue "owned" by this group */
260 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
265 static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
268 static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
270 if (!cfs_rq->on_list) {
272 * Ensure we either appear before our parent (if already
273 * enqueued) or force our parent to appear after us when it is
274 * enqueued. The fact that we always enqueue bottom-up
275 * reduces this to two cases.
277 if (cfs_rq->tg->parent &&
278 cfs_rq->tg->parent->cfs_rq[cpu_of(rq_of(cfs_rq))]->on_list) {
279 list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
280 &rq_of(cfs_rq)->leaf_cfs_rq_list);
282 list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
283 &rq_of(cfs_rq)->leaf_cfs_rq_list);
287 /* We should have no load, but we need to update last_decay. */
288 update_cfs_rq_blocked_load(cfs_rq, 0);
292 static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
294 if (cfs_rq->on_list) {
295 list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
300 /* Iterate thr' all leaf cfs_rq's on a runqueue */
301 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
302 list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)
304 /* Do the two (enqueued) entities belong to the same group ? */
306 is_same_group(struct sched_entity *se, struct sched_entity *pse)
308 if (se->cfs_rq == pse->cfs_rq)
314 static inline struct sched_entity *parent_entity(struct sched_entity *se)
319 /* return depth at which a sched entity is present in the hierarchy */
320 static inline int depth_se(struct sched_entity *se)
324 for_each_sched_entity(se)
331 find_matching_se(struct sched_entity **se, struct sched_entity **pse)
333 int se_depth, pse_depth;
336 * preemption test can be made between sibling entities who are in the
337 * same cfs_rq i.e who have a common parent. Walk up the hierarchy of
338 * both tasks until we find their ancestors who are siblings of common
342 /* First walk up until both entities are at same depth */
343 se_depth = depth_se(*se);
344 pse_depth = depth_se(*pse);
346 while (se_depth > pse_depth) {
348 *se = parent_entity(*se);
351 while (pse_depth > se_depth) {
353 *pse = parent_entity(*pse);
356 while (!is_same_group(*se, *pse)) {
357 *se = parent_entity(*se);
358 *pse = parent_entity(*pse);
362 #else /* !CONFIG_FAIR_GROUP_SCHED */
364 static inline struct task_struct *task_of(struct sched_entity *se)
366 return container_of(se, struct task_struct, se);
369 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
371 return container_of(cfs_rq, struct rq, cfs);
374 #define entity_is_task(se) 1
376 #define for_each_sched_entity(se) \
377 for (; se; se = NULL)
379 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
381 return &task_rq(p)->cfs;
384 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
386 struct task_struct *p = task_of(se);
387 struct rq *rq = task_rq(p);
392 /* runqueue "owned" by this group */
393 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
398 static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
402 static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
406 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
407 for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)
410 is_same_group(struct sched_entity *se, struct sched_entity *pse)
415 static inline struct sched_entity *parent_entity(struct sched_entity *se)
421 find_matching_se(struct sched_entity **se, struct sched_entity **pse)
425 #endif /* CONFIG_FAIR_GROUP_SCHED */
427 static __always_inline
428 void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec);
430 /**************************************************************
431 * Scheduling class tree data structure manipulation methods:
434 static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
436 s64 delta = (s64)(vruntime - max_vruntime);
438 max_vruntime = vruntime;
443 static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
445 s64 delta = (s64)(vruntime - min_vruntime);
447 min_vruntime = vruntime;
452 static inline int entity_before(struct sched_entity *a,
453 struct sched_entity *b)
455 return (s64)(a->vruntime - b->vruntime) < 0;
458 static void update_min_vruntime(struct cfs_rq *cfs_rq)
460 u64 vruntime = cfs_rq->min_vruntime;
463 vruntime = cfs_rq->curr->vruntime;
465 if (cfs_rq->rb_leftmost) {
466 struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost,
471 vruntime = se->vruntime;
473 vruntime = min_vruntime(vruntime, se->vruntime);
476 /* ensure we never gain time by being placed backwards. */
477 cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
480 cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
485 * Enqueue an entity into the rb-tree:
487 static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
489 struct rb_node **link = &cfs_rq->tasks_timeline.rb_node;
490 struct rb_node *parent = NULL;
491 struct sched_entity *entry;
495 * Find the right place in the rbtree:
499 entry = rb_entry(parent, struct sched_entity, run_node);
501 * We dont care about collisions. Nodes with
502 * the same key stay together.
504 if (entity_before(se, entry)) {
505 link = &parent->rb_left;
507 link = &parent->rb_right;
513 * Maintain a cache of leftmost tree entries (it is frequently
517 cfs_rq->rb_leftmost = &se->run_node;
519 rb_link_node(&se->run_node, parent, link);
520 rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline);
523 static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
525 if (cfs_rq->rb_leftmost == &se->run_node) {
526 struct rb_node *next_node;
528 next_node = rb_next(&se->run_node);
529 cfs_rq->rb_leftmost = next_node;
532 rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
535 struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
537 struct rb_node *left = cfs_rq->rb_leftmost;
542 return rb_entry(left, struct sched_entity, run_node);
545 static struct sched_entity *__pick_next_entity(struct sched_entity *se)
547 struct rb_node *next = rb_next(&se->run_node);
552 return rb_entry(next, struct sched_entity, run_node);
555 #ifdef CONFIG_SCHED_DEBUG
556 struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
558 struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
563 return rb_entry(last, struct sched_entity, run_node);
566 /**************************************************************
567 * Scheduling class statistics methods:
570 int sched_proc_update_handler(struct ctl_table *table, int write,
571 void __user *buffer, size_t *lenp,
574 int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
575 int factor = get_update_sysctl_factor();
580 sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
581 sysctl_sched_min_granularity);
583 #define WRT_SYSCTL(name) \
584 (normalized_sysctl_##name = sysctl_##name / (factor))
585 WRT_SYSCTL(sched_min_granularity);
586 WRT_SYSCTL(sched_latency);
587 WRT_SYSCTL(sched_wakeup_granularity);
597 static inline unsigned long
598 calc_delta_fair(unsigned long delta, struct sched_entity *se)
600 if (unlikely(se->load.weight != NICE_0_LOAD))
601 delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load);
607 * The idea is to set a period in which each task runs once.
609 * When there are too many tasks (sched_nr_latency) we have to stretch
610 * this period because otherwise the slices get too small.
612 * p = (nr <= nl) ? l : l*nr/nl
614 static u64 __sched_period(unsigned long nr_running)
616 u64 period = sysctl_sched_latency;
617 unsigned long nr_latency = sched_nr_latency;
619 if (unlikely(nr_running > nr_latency)) {
620 period = sysctl_sched_min_granularity;
621 period *= nr_running;
628 * We calculate the wall-time slice from the period by taking a part
629 * proportional to the weight.
633 static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
635 u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
637 for_each_sched_entity(se) {
638 struct load_weight *load;
639 struct load_weight lw;
641 cfs_rq = cfs_rq_of(se);
642 load = &cfs_rq->load;
644 if (unlikely(!se->on_rq)) {
647 update_load_add(&lw, se->load.weight);
650 slice = calc_delta_mine(slice, se->load.weight, load);
656 * We calculate the vruntime slice of a to-be-inserted task.
660 static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
662 return calc_delta_fair(sched_slice(cfs_rq, se), se);
666 * Update the current task's runtime statistics. Skip current tasks that
667 * are not in our scheduling class.
670 __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
671 unsigned long delta_exec)
673 unsigned long delta_exec_weighted;
675 schedstat_set(curr->statistics.exec_max,
676 max((u64)delta_exec, curr->statistics.exec_max));
678 curr->sum_exec_runtime += delta_exec;
679 schedstat_add(cfs_rq, exec_clock, delta_exec);
680 delta_exec_weighted = calc_delta_fair(delta_exec, curr);
682 curr->vruntime += delta_exec_weighted;
683 update_min_vruntime(cfs_rq);
686 static void update_curr(struct cfs_rq *cfs_rq)
688 struct sched_entity *curr = cfs_rq->curr;
689 u64 now = rq_of(cfs_rq)->clock_task;
690 unsigned long delta_exec;
696 * Get the amount of time the current task was running
697 * since the last time we changed load (this cannot
698 * overflow on 32 bits):
700 delta_exec = (unsigned long)(now - curr->exec_start);
704 __update_curr(cfs_rq, curr, delta_exec);
705 curr->exec_start = now;
707 if (entity_is_task(curr)) {
708 struct task_struct *curtask = task_of(curr);
710 trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
711 cpuacct_charge(curtask, delta_exec);
712 account_group_exec_runtime(curtask, delta_exec);
715 account_cfs_rq_runtime(cfs_rq, delta_exec);
719 update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
721 schedstat_set(se->statistics.wait_start, rq_of(cfs_rq)->clock);
725 * Task is being enqueued - update stats:
727 static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
730 * Are we enqueueing a waiting task? (for current tasks
731 * a dequeue/enqueue event is a NOP)
733 if (se != cfs_rq->curr)
734 update_stats_wait_start(cfs_rq, se);
738 update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
740 schedstat_set(se->statistics.wait_max, max(se->statistics.wait_max,
741 rq_of(cfs_rq)->clock - se->statistics.wait_start));
742 schedstat_set(se->statistics.wait_count, se->statistics.wait_count + 1);
743 schedstat_set(se->statistics.wait_sum, se->statistics.wait_sum +
744 rq_of(cfs_rq)->clock - se->statistics.wait_start);
745 #ifdef CONFIG_SCHEDSTATS
746 if (entity_is_task(se)) {
747 trace_sched_stat_wait(task_of(se),
748 rq_of(cfs_rq)->clock - se->statistics.wait_start);
751 schedstat_set(se->statistics.wait_start, 0);
755 update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
758 * Mark the end of the wait period if dequeueing a
761 if (se != cfs_rq->curr)
762 update_stats_wait_end(cfs_rq, se);
766 * We are picking a new current task - update its stats:
769 update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
772 * We are starting a new run period:
774 se->exec_start = rq_of(cfs_rq)->clock_task;
777 /**************************************************
778 * Scheduling class queueing methods:
781 #ifdef CONFIG_NUMA_BALANCING
783 * numa task sample period in ms
785 unsigned int sysctl_numa_balancing_scan_period_min = 100;
786 unsigned int sysctl_numa_balancing_scan_period_max = 100*50;
787 unsigned int sysctl_numa_balancing_scan_period_reset = 100*600;
789 /* Portion of address space to scan in MB */
790 unsigned int sysctl_numa_balancing_scan_size = 256;
792 /* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
793 unsigned int sysctl_numa_balancing_scan_delay = 1000;
795 static void task_numa_placement(struct task_struct *p)
799 if (!p->mm) /* for example, ksmd faulting in a user's mm */
801 seq = ACCESS_ONCE(p->mm->numa_scan_seq);
802 if (p->numa_scan_seq == seq)
804 p->numa_scan_seq = seq;
806 /* FIXME: Scheduling placement policy hints go here */
810 * Got a PROT_NONE fault for a page on @node.
812 void task_numa_fault(int node, int pages, bool migrated)
814 struct task_struct *p = current;
816 if (!sched_feat_numa(NUMA))
819 /* FIXME: Allocate task-specific structure for placement policy here */
822 * If pages are properly placed (did not migrate) then scan slower.
823 * This is reset periodically in case of phase changes
826 p->numa_scan_period = min(sysctl_numa_balancing_scan_period_max,
827 p->numa_scan_period + jiffies_to_msecs(10));
829 task_numa_placement(p);
832 static void reset_ptenuma_scan(struct task_struct *p)
834 ACCESS_ONCE(p->mm->numa_scan_seq)++;
835 p->mm->numa_scan_offset = 0;
839 * The expensive part of numa migration is done from task_work context.
840 * Triggered from task_tick_numa().
842 void task_numa_work(struct callback_head *work)
844 unsigned long migrate, next_scan, now = jiffies;
845 struct task_struct *p = current;
846 struct mm_struct *mm = p->mm;
847 struct vm_area_struct *vma;
848 unsigned long start, end;
851 WARN_ON_ONCE(p != container_of(work, struct task_struct, numa_work));
853 work->next = work; /* protect against double add */
855 * Who cares about NUMA placement when they're dying.
857 * NOTE: make sure not to dereference p->mm before this check,
858 * exit_task_work() happens _after_ exit_mm() so we could be called
859 * without p->mm even though we still had it when we enqueued this
862 if (p->flags & PF_EXITING)
866 * We do not care about task placement until a task runs on a node
867 * other than the first one used by the address space. This is
868 * largely because migrations are driven by what CPU the task
869 * is running on. If it's never scheduled on another node, it'll
870 * not migrate so why bother trapping the fault.
872 if (mm->first_nid == NUMA_PTE_SCAN_INIT)
873 mm->first_nid = numa_node_id();
874 if (mm->first_nid != NUMA_PTE_SCAN_ACTIVE) {
875 /* Are we running on a new node yet? */
876 if (numa_node_id() == mm->first_nid &&
877 !sched_feat_numa(NUMA_FORCE))
880 mm->first_nid = NUMA_PTE_SCAN_ACTIVE;
884 * Reset the scan period if enough time has gone by. Objective is that
885 * scanning will be reduced if pages are properly placed. As tasks
886 * can enter different phases this needs to be re-examined. Lacking
887 * proper tracking of reference behaviour, this blunt hammer is used.
889 migrate = mm->numa_next_reset;
890 if (time_after(now, migrate)) {
891 p->numa_scan_period = sysctl_numa_balancing_scan_period_min;
892 next_scan = now + msecs_to_jiffies(sysctl_numa_balancing_scan_period_reset);
893 xchg(&mm->numa_next_reset, next_scan);
897 * Enforce maximal scan/migration frequency..
899 migrate = mm->numa_next_scan;
900 if (time_before(now, migrate))
903 if (p->numa_scan_period == 0)
904 p->numa_scan_period = sysctl_numa_balancing_scan_period_min;
906 next_scan = now + msecs_to_jiffies(p->numa_scan_period);
907 if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
911 * Do not set pte_numa if the current running node is rate-limited.
912 * This loses statistics on the fault but if we are unwilling to
913 * migrate to this node, it is less likely we can do useful work
915 if (migrate_ratelimited(numa_node_id()))
918 start = mm->numa_scan_offset;
919 pages = sysctl_numa_balancing_scan_size;
920 pages <<= 20 - PAGE_SHIFT; /* MB in pages */
924 down_read(&mm->mmap_sem);
925 vma = find_vma(mm, start);
927 reset_ptenuma_scan(p);
931 for (; vma; vma = vma->vm_next) {
932 if (!vma_migratable(vma))
935 /* Skip small VMAs. They are not likely to be of relevance */
936 if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
940 start = max(start, vma->vm_start);
941 end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
942 end = min(end, vma->vm_end);
943 pages -= change_prot_numa(vma, start, end);
948 } while (end != vma->vm_end);
953 * It is possible to reach the end of the VMA list but the last few VMAs are
954 * not guaranteed to the vma_migratable. If they are not, we would find the
955 * !migratable VMA on the next scan but not reset the scanner to the start
959 mm->numa_scan_offset = start;
961 reset_ptenuma_scan(p);
962 up_read(&mm->mmap_sem);
966 * Drive the periodic memory faults..
968 void task_tick_numa(struct rq *rq, struct task_struct *curr)
970 struct callback_head *work = &curr->numa_work;
974 * We don't care about NUMA placement if we don't have memory.
976 if (!curr->mm || (curr->flags & PF_EXITING) || work->next != work)
980 * Using runtime rather than walltime has the dual advantage that
981 * we (mostly) drive the selection from busy threads and that the
982 * task needs to have done some actual work before we bother with
985 now = curr->se.sum_exec_runtime;
986 period = (u64)curr->numa_scan_period * NSEC_PER_MSEC;
988 if (now - curr->node_stamp > period) {
989 if (!curr->node_stamp)
990 curr->numa_scan_period = sysctl_numa_balancing_scan_period_min;
991 curr->node_stamp = now;
993 if (!time_before(jiffies, curr->mm->numa_next_scan)) {
994 init_task_work(work, task_numa_work); /* TODO: move this into sched_fork() */
995 task_work_add(curr, work, true);
1000 static void task_tick_numa(struct rq *rq, struct task_struct *curr)
1003 #endif /* CONFIG_NUMA_BALANCING */
1006 account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
1008 update_load_add(&cfs_rq->load, se->load.weight);
1009 if (!parent_entity(se))
1010 update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
1012 if (entity_is_task(se))
1013 list_add(&se->group_node, &rq_of(cfs_rq)->cfs_tasks);
1015 cfs_rq->nr_running++;
1019 account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
1021 update_load_sub(&cfs_rq->load, se->load.weight);
1022 if (!parent_entity(se))
1023 update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
1024 if (entity_is_task(se))
1025 list_del_init(&se->group_node);
1026 cfs_rq->nr_running--;
1029 #ifdef CONFIG_FAIR_GROUP_SCHED
1031 static inline long calc_tg_weight(struct task_group *tg, struct cfs_rq *cfs_rq)
1036 * Use this CPU's actual weight instead of the last load_contribution
1037 * to gain a more accurate current total weight. See
1038 * update_cfs_rq_load_contribution().
1040 tg_weight = atomic64_read(&tg->load_avg);
1041 tg_weight -= cfs_rq->tg_load_contrib;
1042 tg_weight += cfs_rq->load.weight;
1047 static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
1049 long tg_weight, load, shares;
1051 tg_weight = calc_tg_weight(tg, cfs_rq);
1052 load = cfs_rq->load.weight;
1054 shares = (tg->shares * load);
1056 shares /= tg_weight;
1058 if (shares < MIN_SHARES)
1059 shares = MIN_SHARES;
1060 if (shares > tg->shares)
1061 shares = tg->shares;
1065 # else /* CONFIG_SMP */
1066 static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
1070 # endif /* CONFIG_SMP */
1071 static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
1072 unsigned long weight)
1075 /* commit outstanding execution time */
1076 if (cfs_rq->curr == se)
1077 update_curr(cfs_rq);
1078 account_entity_dequeue(cfs_rq, se);
1081 update_load_set(&se->load, weight);
1084 account_entity_enqueue(cfs_rq, se);
1087 static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);
1089 static void update_cfs_shares(struct cfs_rq *cfs_rq)
1091 struct task_group *tg;
1092 struct sched_entity *se;
1096 se = tg->se[cpu_of(rq_of(cfs_rq))];
1097 if (!se || throttled_hierarchy(cfs_rq))
1100 if (likely(se->load.weight == tg->shares))
1103 shares = calc_cfs_shares(cfs_rq, tg);
1105 reweight_entity(cfs_rq_of(se), se, shares);
1107 #else /* CONFIG_FAIR_GROUP_SCHED */
1108 static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
1111 #endif /* CONFIG_FAIR_GROUP_SCHED */
1113 /* Only depends on SMP, FAIR_GROUP_SCHED may be removed when useful in lb */
1114 #if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)
1116 * We choose a half-life close to 1 scheduling period.
1117 * Note: The tables below are dependent on this value.
1119 #define LOAD_AVG_PERIOD 32
1120 #define LOAD_AVG_MAX 47742 /* maximum possible load avg */
1121 #define LOAD_AVG_MAX_N 345 /* number of full periods to produce LOAD_MAX_AVG */
1123 /* Precomputed fixed inverse multiplies for multiplication by y^n */
1124 static const u32 runnable_avg_yN_inv[] = {
1125 0xffffffff, 0xfa83b2da, 0xf5257d14, 0xefe4b99a, 0xeac0c6e6, 0xe5b906e6,
1126 0xe0ccdeeb, 0xdbfbb796, 0xd744fcc9, 0xd2a81d91, 0xce248c14, 0xc9b9bd85,
1127 0xc5672a10, 0xc12c4cc9, 0xbd08a39e, 0xb8fbaf46, 0xb504f333, 0xb123f581,
1128 0xad583ee9, 0xa9a15ab4, 0xa5fed6a9, 0xa2704302, 0x9ef5325f, 0x9b8d39b9,
1129 0x9837f050, 0x94f4efa8, 0x91c3d373, 0x8ea4398a, 0x8b95c1e3, 0x88980e80,
1130 0x85aac367, 0x82cd8698,
1134 * Precomputed \Sum y^k { 1<=k<=n }. These are floor(true_value) to prevent
1135 * over-estimates when re-combining.
1137 static const u32 runnable_avg_yN_sum[] = {
1138 0, 1002, 1982, 2941, 3880, 4798, 5697, 6576, 7437, 8279, 9103,
1139 9909,10698,11470,12226,12966,13690,14398,15091,15769,16433,17082,
1140 17718,18340,18949,19545,20128,20698,21256,21802,22336,22859,23371,
1145 * val * y^n, where y^32 ~= 0.5 (~1 scheduling period)
1147 static __always_inline u64 decay_load(u64 val, u64 n)
1149 unsigned int local_n;
1153 else if (unlikely(n > LOAD_AVG_PERIOD * 63))
1156 /* after bounds checking we can collapse to 32-bit */
1160 * As y^PERIOD = 1/2, we can combine
1161 * y^n = 1/2^(n/PERIOD) * k^(n%PERIOD)
1162 * With a look-up table which covers k^n (n<PERIOD)
1164 * To achieve constant time decay_load.
1166 if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
1167 val >>= local_n / LOAD_AVG_PERIOD;
1168 local_n %= LOAD_AVG_PERIOD;
1171 val *= runnable_avg_yN_inv[local_n];
1172 /* We don't use SRR here since we always want to round down. */
1177 * For updates fully spanning n periods, the contribution to runnable
1178 * average will be: \Sum 1024*y^n
1180 * We can compute this reasonably efficiently by combining:
1181 * y^PERIOD = 1/2 with precomputed \Sum 1024*y^n {for n <PERIOD}
1183 static u32 __compute_runnable_contrib(u64 n)
1187 if (likely(n <= LOAD_AVG_PERIOD))
1188 return runnable_avg_yN_sum[n];
1189 else if (unlikely(n >= LOAD_AVG_MAX_N))
1190 return LOAD_AVG_MAX;
1192 /* Compute \Sum k^n combining precomputed values for k^i, \Sum k^j */
1194 contrib /= 2; /* y^LOAD_AVG_PERIOD = 1/2 */
1195 contrib += runnable_avg_yN_sum[LOAD_AVG_PERIOD];
1197 n -= LOAD_AVG_PERIOD;
1198 } while (n > LOAD_AVG_PERIOD);
1200 contrib = decay_load(contrib, n);
1201 return contrib + runnable_avg_yN_sum[n];
1205 * We can represent the historical contribution to runnable average as the
1206 * coefficients of a geometric series. To do this we sub-divide our runnable
1207 * history into segments of approximately 1ms (1024us); label the segment that
1208 * occurred N-ms ago p_N, with p_0 corresponding to the current period, e.g.
1210 * [<- 1024us ->|<- 1024us ->|<- 1024us ->| ...
1212 * (now) (~1ms ago) (~2ms ago)
1214 * Let u_i denote the fraction of p_i that the entity was runnable.
1216 * We then designate the fractions u_i as our co-efficients, yielding the
1217 * following representation of historical load:
1218 * u_0 + u_1*y + u_2*y^2 + u_3*y^3 + ...
1220 * We choose y based on the with of a reasonably scheduling period, fixing:
1223 * This means that the contribution to load ~32ms ago (u_32) will be weighted
1224 * approximately half as much as the contribution to load within the last ms
1227 * When a period "rolls over" and we have new u_0`, multiplying the previous
1228 * sum again by y is sufficient to update:
1229 * load_avg = u_0` + y*(u_0 + u_1*y + u_2*y^2 + ... )
1230 * = u_0 + u_1*y + u_2*y^2 + ... [re-labeling u_i --> u_{i+1}]
1232 static __always_inline int __update_entity_runnable_avg(u64 now,
1233 struct sched_avg *sa,
1237 u32 runnable_contrib;
1238 int delta_w, decayed = 0;
1240 delta = now - sa->last_runnable_update;
1242 * This should only happen when time goes backwards, which it
1243 * unfortunately does during sched clock init when we swap over to TSC.
1245 if ((s64)delta < 0) {
1246 sa->last_runnable_update = now;
1251 * Use 1024ns as the unit of measurement since it's a reasonable
1252 * approximation of 1us and fast to compute.
1257 sa->last_runnable_update = now;
1259 /* delta_w is the amount already accumulated against our next period */
1260 delta_w = sa->runnable_avg_period % 1024;
1261 if (delta + delta_w >= 1024) {
1262 /* period roll-over */
1266 * Now that we know we're crossing a period boundary, figure
1267 * out how much from delta we need to complete the current
1268 * period and accrue it.
1270 delta_w = 1024 - delta_w;
1272 sa->runnable_avg_sum += delta_w;
1273 sa->runnable_avg_period += delta_w;
1277 /* Figure out how many additional periods this update spans */
1278 periods = delta / 1024;
1281 sa->runnable_avg_sum = decay_load(sa->runnable_avg_sum,
1283 sa->runnable_avg_period = decay_load(sa->runnable_avg_period,
1286 /* Efficiently calculate \sum (1..n_period) 1024*y^i */
1287 runnable_contrib = __compute_runnable_contrib(periods);
1289 sa->runnable_avg_sum += runnable_contrib;
1290 sa->runnable_avg_period += runnable_contrib;
1293 /* Remainder of delta accrued against u_0` */
1295 sa->runnable_avg_sum += delta;
1296 sa->runnable_avg_period += delta;
1301 /* Synchronize an entity's decay with its parenting cfs_rq.*/
1302 static inline u64 __synchronize_entity_decay(struct sched_entity *se)
1304 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1305 u64 decays = atomic64_read(&cfs_rq->decay_counter);
1307 decays -= se->avg.decay_count;
1311 se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
1312 se->avg.decay_count = 0;
1317 #ifdef CONFIG_FAIR_GROUP_SCHED
1318 static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
1321 struct task_group *tg = cfs_rq->tg;
1324 tg_contrib = cfs_rq->runnable_load_avg + cfs_rq->blocked_load_avg;
1325 tg_contrib -= cfs_rq->tg_load_contrib;
1327 if (force_update || abs64(tg_contrib) > cfs_rq->tg_load_contrib / 8) {
1328 atomic64_add(tg_contrib, &tg->load_avg);
1329 cfs_rq->tg_load_contrib += tg_contrib;
1334 * Aggregate cfs_rq runnable averages into an equivalent task_group
1335 * representation for computing load contributions.
1337 static inline void __update_tg_runnable_avg(struct sched_avg *sa,
1338 struct cfs_rq *cfs_rq)
1340 struct task_group *tg = cfs_rq->tg;
1343 /* The fraction of a cpu used by this cfs_rq */
1344 contrib = div_u64(sa->runnable_avg_sum << NICE_0_SHIFT,
1345 sa->runnable_avg_period + 1);
1346 contrib -= cfs_rq->tg_runnable_contrib;
1348 if (abs(contrib) > cfs_rq->tg_runnable_contrib / 64) {
1349 atomic_add(contrib, &tg->runnable_avg);
1350 cfs_rq->tg_runnable_contrib += contrib;
1354 static inline void __update_group_entity_contrib(struct sched_entity *se)
1356 struct cfs_rq *cfs_rq = group_cfs_rq(se);
1357 struct task_group *tg = cfs_rq->tg;
1362 contrib = cfs_rq->tg_load_contrib * tg->shares;
1363 se->avg.load_avg_contrib = div64_u64(contrib,
1364 atomic64_read(&tg->load_avg) + 1);
1367 * For group entities we need to compute a correction term in the case
1368 * that they are consuming <1 cpu so that we would contribute the same
1369 * load as a task of equal weight.
1371 * Explicitly co-ordinating this measurement would be expensive, but
1372 * fortunately the sum of each cpus contribution forms a usable
1373 * lower-bound on the true value.
1375 * Consider the aggregate of 2 contributions. Either they are disjoint
1376 * (and the sum represents true value) or they are disjoint and we are
1377 * understating by the aggregate of their overlap.
1379 * Extending this to N cpus, for a given overlap, the maximum amount we
1380 * understand is then n_i(n_i+1)/2 * w_i where n_i is the number of
1381 * cpus that overlap for this interval and w_i is the interval width.
1383 * On a small machine; the first term is well-bounded which bounds the
1384 * total error since w_i is a subset of the period. Whereas on a
1385 * larger machine, while this first term can be larger, if w_i is the
1386 * of consequential size guaranteed to see n_i*w_i quickly converge to
1387 * our upper bound of 1-cpu.
1389 runnable_avg = atomic_read(&tg->runnable_avg);
1390 if (runnable_avg < NICE_0_LOAD) {
1391 se->avg.load_avg_contrib *= runnable_avg;
1392 se->avg.load_avg_contrib >>= NICE_0_SHIFT;
1396 static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
1397 int force_update) {}
1398 static inline void __update_tg_runnable_avg(struct sched_avg *sa,
1399 struct cfs_rq *cfs_rq) {}
1400 static inline void __update_group_entity_contrib(struct sched_entity *se) {}
1403 static inline void __update_task_entity_contrib(struct sched_entity *se)
1407 /* avoid overflowing a 32-bit type w/ SCHED_LOAD_SCALE */
1408 contrib = se->avg.runnable_avg_sum * scale_load_down(se->load.weight);
1409 contrib /= (se->avg.runnable_avg_period + 1);
1410 se->avg.load_avg_contrib = scale_load(contrib);
1413 /* Compute the current contribution to load_avg by se, return any delta */
1414 static long __update_entity_load_avg_contrib(struct sched_entity *se)
1416 long old_contrib = se->avg.load_avg_contrib;
1418 if (entity_is_task(se)) {
1419 __update_task_entity_contrib(se);
1421 __update_tg_runnable_avg(&se->avg, group_cfs_rq(se));
1422 __update_group_entity_contrib(se);
1425 return se->avg.load_avg_contrib - old_contrib;
1428 static inline void subtract_blocked_load_contrib(struct cfs_rq *cfs_rq,
1431 if (likely(load_contrib < cfs_rq->blocked_load_avg))
1432 cfs_rq->blocked_load_avg -= load_contrib;
1434 cfs_rq->blocked_load_avg = 0;
1437 static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
1439 /* Update a sched_entity's runnable average */
1440 static inline void update_entity_load_avg(struct sched_entity *se,
1443 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1448 * For a group entity we need to use their owned cfs_rq_clock_task() in
1449 * case they are the parent of a throttled hierarchy.
1451 if (entity_is_task(se))
1452 now = cfs_rq_clock_task(cfs_rq);
1454 now = cfs_rq_clock_task(group_cfs_rq(se));
1456 if (!__update_entity_runnable_avg(now, &se->avg, se->on_rq))
1459 contrib_delta = __update_entity_load_avg_contrib(se);
1465 cfs_rq->runnable_load_avg += contrib_delta;
1467 subtract_blocked_load_contrib(cfs_rq, -contrib_delta);
1471 * Decay the load contributed by all blocked children and account this so that
1472 * their contribution may appropriately discounted when they wake up.
1474 static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq, int force_update)
1476 u64 now = cfs_rq_clock_task(cfs_rq) >> 20;
1479 decays = now - cfs_rq->last_decay;
1480 if (!decays && !force_update)
1483 if (atomic64_read(&cfs_rq->removed_load)) {
1484 u64 removed_load = atomic64_xchg(&cfs_rq->removed_load, 0);
1485 subtract_blocked_load_contrib(cfs_rq, removed_load);
1489 cfs_rq->blocked_load_avg = decay_load(cfs_rq->blocked_load_avg,
1491 atomic64_add(decays, &cfs_rq->decay_counter);
1492 cfs_rq->last_decay = now;
1495 __update_cfs_rq_tg_load_contrib(cfs_rq, force_update);
1498 static inline void update_rq_runnable_avg(struct rq *rq, int runnable)
1500 __update_entity_runnable_avg(rq->clock_task, &rq->avg, runnable);
1501 __update_tg_runnable_avg(&rq->avg, &rq->cfs);
1504 /* Add the load generated by se into cfs_rq's child load-average */
1505 static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
1506 struct sched_entity *se,
1510 * We track migrations using entity decay_count <= 0, on a wake-up
1511 * migration we use a negative decay count to track the remote decays
1512 * accumulated while sleeping.
1514 if (unlikely(se->avg.decay_count <= 0)) {
1515 se->avg.last_runnable_update = rq_of(cfs_rq)->clock_task;
1516 if (se->avg.decay_count) {
1518 * In a wake-up migration we have to approximate the
1519 * time sleeping. This is because we can't synchronize
1520 * clock_task between the two cpus, and it is not
1521 * guaranteed to be read-safe. Instead, we can
1522 * approximate this using our carried decays, which are
1523 * explicitly atomically readable.
1525 se->avg.last_runnable_update -= (-se->avg.decay_count)
1527 update_entity_load_avg(se, 0);
1528 /* Indicate that we're now synchronized and on-rq */
1529 se->avg.decay_count = 0;
1533 __synchronize_entity_decay(se);
1536 /* migrated tasks did not contribute to our blocked load */
1538 subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
1539 update_entity_load_avg(se, 0);
1542 cfs_rq->runnable_load_avg += se->avg.load_avg_contrib;
1543 /* we force update consideration on load-balancer moves */
1544 update_cfs_rq_blocked_load(cfs_rq, !wakeup);
1548 * Remove se's load from this cfs_rq child load-average, if the entity is
1549 * transitioning to a blocked state we track its projected decay using
1552 static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
1553 struct sched_entity *se,
1556 update_entity_load_avg(se, 1);
1557 /* we force update consideration on load-balancer moves */
1558 update_cfs_rq_blocked_load(cfs_rq, !sleep);
1560 cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib;
1562 cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
1563 se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
1564 } /* migrations, e.g. sleep=0 leave decay_count == 0 */
1568 * Update the rq's load with the elapsed running time before entering
1569 * idle. if the last scheduled task is not a CFS task, idle_enter will
1570 * be the only way to update the runnable statistic.
1572 void idle_enter_fair(struct rq *this_rq)
1574 update_rq_runnable_avg(this_rq, 1);
1578 * Update the rq's load with the elapsed idle time before a task is
1579 * scheduled. if the newly scheduled task is not a CFS task, idle_exit will
1580 * be the only way to update the runnable statistic.
1582 void idle_exit_fair(struct rq *this_rq)
1584 update_rq_runnable_avg(this_rq, 0);
1588 static inline void update_entity_load_avg(struct sched_entity *se,
1589 int update_cfs_rq) {}
1590 static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
1591 static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
1592 struct sched_entity *se,
1594 static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
1595 struct sched_entity *se,
1597 static inline void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
1598 int force_update) {}
1601 static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
1603 #ifdef CONFIG_SCHEDSTATS
1604 struct task_struct *tsk = NULL;
1606 if (entity_is_task(se))
1609 if (se->statistics.sleep_start) {
1610 u64 delta = rq_of(cfs_rq)->clock - se->statistics.sleep_start;
1615 if (unlikely(delta > se->statistics.sleep_max))
1616 se->statistics.sleep_max = delta;
1618 se->statistics.sleep_start = 0;
1619 se->statistics.sum_sleep_runtime += delta;
1622 account_scheduler_latency(tsk, delta >> 10, 1);
1623 trace_sched_stat_sleep(tsk, delta);
1626 if (se->statistics.block_start) {
1627 u64 delta = rq_of(cfs_rq)->clock - se->statistics.block_start;
1632 if (unlikely(delta > se->statistics.block_max))
1633 se->statistics.block_max = delta;
1635 se->statistics.block_start = 0;
1636 se->statistics.sum_sleep_runtime += delta;
1639 if (tsk->in_iowait) {
1640 se->statistics.iowait_sum += delta;
1641 se->statistics.iowait_count++;
1642 trace_sched_stat_iowait(tsk, delta);
1645 trace_sched_stat_blocked(tsk, delta);
1648 * Blocking time is in units of nanosecs, so shift by
1649 * 20 to get a milliseconds-range estimation of the
1650 * amount of time that the task spent sleeping:
1652 if (unlikely(prof_on == SLEEP_PROFILING)) {
1653 profile_hits(SLEEP_PROFILING,
1654 (void *)get_wchan(tsk),
1657 account_scheduler_latency(tsk, delta >> 10, 0);
1663 static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se)
1665 #ifdef CONFIG_SCHED_DEBUG
1666 s64 d = se->vruntime - cfs_rq->min_vruntime;
1671 if (d > 3*sysctl_sched_latency)
1672 schedstat_inc(cfs_rq, nr_spread_over);
1677 place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
1679 u64 vruntime = cfs_rq->min_vruntime;
1682 * The 'current' period is already promised to the current tasks,
1683 * however the extra weight of the new task will slow them down a
1684 * little, place the new task so that it fits in the slot that
1685 * stays open at the end.
1687 if (initial && sched_feat(START_DEBIT))
1688 vruntime += sched_vslice(cfs_rq, se);
1690 /* sleeps up to a single latency don't count. */
1692 unsigned long thresh = sysctl_sched_latency;
1695 * Halve their sleep time's effect, to allow
1696 * for a gentler effect of sleepers:
1698 if (sched_feat(GENTLE_FAIR_SLEEPERS))
1704 /* ensure we never gain time by being placed backwards. */
1705 se->vruntime = max_vruntime(se->vruntime, vruntime);
1708 static void check_enqueue_throttle(struct cfs_rq *cfs_rq);
1711 enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
1714 * Update the normalized vruntime before updating min_vruntime
1715 * through callig update_curr().
1717 if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
1718 se->vruntime += cfs_rq->min_vruntime;
1721 * Update run-time statistics of the 'current'.
1723 update_curr(cfs_rq);
1724 enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP);
1725 account_entity_enqueue(cfs_rq, se);
1726 update_cfs_shares(cfs_rq);
1728 if (flags & ENQUEUE_WAKEUP) {
1729 place_entity(cfs_rq, se, 0);
1730 enqueue_sleeper(cfs_rq, se);
1733 update_stats_enqueue(cfs_rq, se);
1734 check_spread(cfs_rq, se);
1735 if (se != cfs_rq->curr)
1736 __enqueue_entity(cfs_rq, se);
1739 if (cfs_rq->nr_running == 1) {
1740 list_add_leaf_cfs_rq(cfs_rq);
1741 check_enqueue_throttle(cfs_rq);
1745 static void __clear_buddies_last(struct sched_entity *se)
1747 for_each_sched_entity(se) {
1748 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1749 if (cfs_rq->last == se)
1750 cfs_rq->last = NULL;
1756 static void __clear_buddies_next(struct sched_entity *se)
1758 for_each_sched_entity(se) {
1759 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1760 if (cfs_rq->next == se)
1761 cfs_rq->next = NULL;
1767 static void __clear_buddies_skip(struct sched_entity *se)
1769 for_each_sched_entity(se) {
1770 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1771 if (cfs_rq->skip == se)
1772 cfs_rq->skip = NULL;
1778 static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
1780 if (cfs_rq->last == se)
1781 __clear_buddies_last(se);
1783 if (cfs_rq->next == se)
1784 __clear_buddies_next(se);
1786 if (cfs_rq->skip == se)
1787 __clear_buddies_skip(se);
1790 static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
1793 dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
1796 * Update run-time statistics of the 'current'.
1798 update_curr(cfs_rq);
1799 dequeue_entity_load_avg(cfs_rq, se, flags & DEQUEUE_SLEEP);
1801 update_stats_dequeue(cfs_rq, se);
1802 if (flags & DEQUEUE_SLEEP) {
1803 #ifdef CONFIG_SCHEDSTATS
1804 if (entity_is_task(se)) {
1805 struct task_struct *tsk = task_of(se);
1807 if (tsk->state & TASK_INTERRUPTIBLE)
1808 se->statistics.sleep_start = rq_of(cfs_rq)->clock;
1809 if (tsk->state & TASK_UNINTERRUPTIBLE)
1810 se->statistics.block_start = rq_of(cfs_rq)->clock;
1815 clear_buddies(cfs_rq, se);
1817 if (se != cfs_rq->curr)
1818 __dequeue_entity(cfs_rq, se);
1820 account_entity_dequeue(cfs_rq, se);
1823 * Normalize the entity after updating the min_vruntime because the
1824 * update can refer to the ->curr item and we need to reflect this
1825 * movement in our normalized position.
1827 if (!(flags & DEQUEUE_SLEEP))
1828 se->vruntime -= cfs_rq->min_vruntime;
1830 /* return excess runtime on last dequeue */
1831 return_cfs_rq_runtime(cfs_rq);
1833 update_min_vruntime(cfs_rq);
1834 update_cfs_shares(cfs_rq);
1838 * Preempt the current task with a newly woken task if needed:
1841 check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
1843 unsigned long ideal_runtime, delta_exec;
1844 struct sched_entity *se;
1847 ideal_runtime = sched_slice(cfs_rq, curr);
1848 delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
1849 if (delta_exec > ideal_runtime) {
1850 resched_task(rq_of(cfs_rq)->curr);
1852 * The current task ran long enough, ensure it doesn't get
1853 * re-elected due to buddy favours.
1855 clear_buddies(cfs_rq, curr);
1860 * Ensure that a task that missed wakeup preemption by a
1861 * narrow margin doesn't have to wait for a full slice.
1862 * This also mitigates buddy induced latencies under load.
1864 if (delta_exec < sysctl_sched_min_granularity)
1867 se = __pick_first_entity(cfs_rq);
1868 delta = curr->vruntime - se->vruntime;
1873 if (delta > ideal_runtime)
1874 resched_task(rq_of(cfs_rq)->curr);
1878 set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
1880 /* 'current' is not kept within the tree. */
1883 * Any task has to be enqueued before it get to execute on
1884 * a CPU. So account for the time it spent waiting on the
1887 update_stats_wait_end(cfs_rq, se);
1888 __dequeue_entity(cfs_rq, se);
1891 update_stats_curr_start(cfs_rq, se);
1893 #ifdef CONFIG_SCHEDSTATS
1895 * Track our maximum slice length, if the CPU's load is at
1896 * least twice that of our own weight (i.e. dont track it
1897 * when there are only lesser-weight tasks around):
1899 if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
1900 se->statistics.slice_max = max(se->statistics.slice_max,
1901 se->sum_exec_runtime - se->prev_sum_exec_runtime);
1904 se->prev_sum_exec_runtime = se->sum_exec_runtime;
1908 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);
1911 * Pick the next process, keeping these things in mind, in this order:
1912 * 1) keep things fair between processes/task groups
1913 * 2) pick the "next" process, since someone really wants that to run
1914 * 3) pick the "last" process, for cache locality
1915 * 4) do not run the "skip" process, if something else is available
1917 static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
1919 struct sched_entity *se = __pick_first_entity(cfs_rq);
1920 struct sched_entity *left = se;
1923 * Avoid running the skip buddy, if running something else can
1924 * be done without getting too unfair.
1926 if (cfs_rq->skip == se) {
1927 struct sched_entity *second = __pick_next_entity(se);
1928 if (second && wakeup_preempt_entity(second, left) < 1)
1933 * Prefer last buddy, try to return the CPU to a preempted task.
1935 if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, left) < 1)
1939 * Someone really wants this to run. If it's not unfair, run it.
1941 if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, left) < 1)
1944 clear_buddies(cfs_rq, se);
1949 static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
1951 static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
1954 * If still on the runqueue then deactivate_task()
1955 * was not called and update_curr() has to be done:
1958 update_curr(cfs_rq);
1960 /* throttle cfs_rqs exceeding runtime */
1961 check_cfs_rq_runtime(cfs_rq);
1963 check_spread(cfs_rq, prev);
1965 update_stats_wait_start(cfs_rq, prev);
1966 /* Put 'current' back into the tree. */
1967 __enqueue_entity(cfs_rq, prev);
1968 /* in !on_rq case, update occurred at dequeue */
1969 update_entity_load_avg(prev, 1);
1971 cfs_rq->curr = NULL;
1975 entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
1978 * Update run-time statistics of the 'current'.
1980 update_curr(cfs_rq);
1983 * Ensure that runnable average is periodically updated.
1985 update_entity_load_avg(curr, 1);
1986 update_cfs_rq_blocked_load(cfs_rq, 1);
1988 #ifdef CONFIG_SCHED_HRTICK
1990 * queued ticks are scheduled to match the slice, so don't bother
1991 * validating it and just reschedule.
1994 resched_task(rq_of(cfs_rq)->curr);
1998 * don't let the period tick interfere with the hrtick preemption
2000 if (!sched_feat(DOUBLE_TICK) &&
2001 hrtimer_active(&rq_of(cfs_rq)->hrtick_timer))
2005 if (cfs_rq->nr_running > 1)
2006 check_preempt_tick(cfs_rq, curr);
2010 /**************************************************
2011 * CFS bandwidth control machinery
2014 #ifdef CONFIG_CFS_BANDWIDTH
2016 #ifdef HAVE_JUMP_LABEL
2017 static struct static_key __cfs_bandwidth_used;
2019 static inline bool cfs_bandwidth_used(void)
2021 return static_key_false(&__cfs_bandwidth_used);
2024 void account_cfs_bandwidth_used(int enabled, int was_enabled)
2026 /* only need to count groups transitioning between enabled/!enabled */
2027 if (enabled && !was_enabled)
2028 static_key_slow_inc(&__cfs_bandwidth_used);
2029 else if (!enabled && was_enabled)
2030 static_key_slow_dec(&__cfs_bandwidth_used);
2032 #else /* HAVE_JUMP_LABEL */
2033 static bool cfs_bandwidth_used(void)
2038 void account_cfs_bandwidth_used(int enabled, int was_enabled) {}
2039 #endif /* HAVE_JUMP_LABEL */
2042 * default period for cfs group bandwidth.
2043 * default: 0.1s, units: nanoseconds
2045 static inline u64 default_cfs_period(void)
2047 return 100000000ULL;
2050 static inline u64 sched_cfs_bandwidth_slice(void)
2052 return (u64)sysctl_sched_cfs_bandwidth_slice * NSEC_PER_USEC;
2056 * Replenish runtime according to assigned quota and update expiration time.
2057 * We use sched_clock_cpu directly instead of rq->clock to avoid adding
2058 * additional synchronization around rq->lock.
2060 * requires cfs_b->lock
2062 void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
2066 if (cfs_b->quota == RUNTIME_INF)
2069 now = sched_clock_cpu(smp_processor_id());
2070 cfs_b->runtime = cfs_b->quota;
2071 cfs_b->runtime_expires = now + ktime_to_ns(cfs_b->period);
2074 static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
2076 return &tg->cfs_bandwidth;
2079 /* rq->task_clock normalized against any time this cfs_rq has spent throttled */
2080 static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
2082 if (unlikely(cfs_rq->throttle_count))
2083 return cfs_rq->throttled_clock_task;
2085 return rq_of(cfs_rq)->clock_task - cfs_rq->throttled_clock_task_time;
2088 /* returns 0 on failure to allocate runtime */
2089 static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2091 struct task_group *tg = cfs_rq->tg;
2092 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
2093 u64 amount = 0, min_amount, expires;
2095 /* note: this is a positive sum as runtime_remaining <= 0 */
2096 min_amount = sched_cfs_bandwidth_slice() - cfs_rq->runtime_remaining;
2098 raw_spin_lock(&cfs_b->lock);
2099 if (cfs_b->quota == RUNTIME_INF)
2100 amount = min_amount;
2103 * If the bandwidth pool has become inactive, then at least one
2104 * period must have elapsed since the last consumption.
2105 * Refresh the global state and ensure bandwidth timer becomes
2108 if (!cfs_b->timer_active) {
2109 __refill_cfs_bandwidth_runtime(cfs_b);
2110 __start_cfs_bandwidth(cfs_b);
2113 if (cfs_b->runtime > 0) {
2114 amount = min(cfs_b->runtime, min_amount);
2115 cfs_b->runtime -= amount;
2119 expires = cfs_b->runtime_expires;
2120 raw_spin_unlock(&cfs_b->lock);
2122 cfs_rq->runtime_remaining += amount;
2124 * we may have advanced our local expiration to account for allowed
2125 * spread between our sched_clock and the one on which runtime was
2128 if ((s64)(expires - cfs_rq->runtime_expires) > 0)
2129 cfs_rq->runtime_expires = expires;
2131 return cfs_rq->runtime_remaining > 0;
2135 * Note: This depends on the synchronization provided by sched_clock and the
2136 * fact that rq->clock snapshots this value.
2138 static void expire_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2140 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
2141 struct rq *rq = rq_of(cfs_rq);
2143 /* if the deadline is ahead of our clock, nothing to do */
2144 if (likely((s64)(rq->clock - cfs_rq->runtime_expires) < 0))
2147 if (cfs_rq->runtime_remaining < 0)
2151 * If the local deadline has passed we have to consider the
2152 * possibility that our sched_clock is 'fast' and the global deadline
2153 * has not truly expired.
2155 * Fortunately we can check determine whether this the case by checking
2156 * whether the global deadline has advanced.
2159 if ((s64)(cfs_rq->runtime_expires - cfs_b->runtime_expires) >= 0) {
2160 /* extend local deadline, drift is bounded above by 2 ticks */
2161 cfs_rq->runtime_expires += TICK_NSEC;
2163 /* global deadline is ahead, expiration has passed */
2164 cfs_rq->runtime_remaining = 0;
2168 static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
2169 unsigned long delta_exec)
2171 /* dock delta_exec before expiring quota (as it could span periods) */
2172 cfs_rq->runtime_remaining -= delta_exec;
2173 expire_cfs_rq_runtime(cfs_rq);
2175 if (likely(cfs_rq->runtime_remaining > 0))
2179 * if we're unable to extend our runtime we resched so that the active
2180 * hierarchy can be throttled
2182 if (!assign_cfs_rq_runtime(cfs_rq) && likely(cfs_rq->curr))
2183 resched_task(rq_of(cfs_rq)->curr);
2186 static __always_inline
2187 void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec)
2189 if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
2192 __account_cfs_rq_runtime(cfs_rq, delta_exec);
2195 static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
2197 return cfs_bandwidth_used() && cfs_rq->throttled;
2200 /* check whether cfs_rq, or any parent, is throttled */
2201 static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
2203 return cfs_bandwidth_used() && cfs_rq->throttle_count;
2207 * Ensure that neither of the group entities corresponding to src_cpu or
2208 * dest_cpu are members of a throttled hierarchy when performing group
2209 * load-balance operations.
2211 static inline int throttled_lb_pair(struct task_group *tg,
2212 int src_cpu, int dest_cpu)
2214 struct cfs_rq *src_cfs_rq, *dest_cfs_rq;
2216 src_cfs_rq = tg->cfs_rq[src_cpu];
2217 dest_cfs_rq = tg->cfs_rq[dest_cpu];
2219 return throttled_hierarchy(src_cfs_rq) ||
2220 throttled_hierarchy(dest_cfs_rq);
2223 /* updated child weight may affect parent so we have to do this bottom up */
2224 static int tg_unthrottle_up(struct task_group *tg, void *data)
2226 struct rq *rq = data;
2227 struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
2229 cfs_rq->throttle_count--;
2231 if (!cfs_rq->throttle_count) {
2232 /* adjust cfs_rq_clock_task() */
2233 cfs_rq->throttled_clock_task_time += rq->clock_task -
2234 cfs_rq->throttled_clock_task;
2241 static int tg_throttle_down(struct task_group *tg, void *data)
2243 struct rq *rq = data;
2244 struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
2246 /* group is entering throttled state, stop time */
2247 if (!cfs_rq->throttle_count)
2248 cfs_rq->throttled_clock_task = rq->clock_task;
2249 cfs_rq->throttle_count++;
2254 static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
2256 struct rq *rq = rq_of(cfs_rq);
2257 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
2258 struct sched_entity *se;
2259 long task_delta, dequeue = 1;
2261 se = cfs_rq->tg->se[cpu_of(rq_of(cfs_rq))];
2263 /* freeze hierarchy runnable averages while throttled */
2265 walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
2268 task_delta = cfs_rq->h_nr_running;
2269 for_each_sched_entity(se) {
2270 struct cfs_rq *qcfs_rq = cfs_rq_of(se);
2271 /* throttled entity or throttle-on-deactivate */
2276 dequeue_entity(qcfs_rq, se, DEQUEUE_SLEEP);
2277 qcfs_rq->h_nr_running -= task_delta;
2279 if (qcfs_rq->load.weight)
2284 rq->nr_running -= task_delta;
2286 cfs_rq->throttled = 1;
2287 cfs_rq->throttled_clock = rq->clock;
2288 raw_spin_lock(&cfs_b->lock);
2289 list_add_tail_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
2290 raw_spin_unlock(&cfs_b->lock);
2293 void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
2295 struct rq *rq = rq_of(cfs_rq);
2296 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
2297 struct sched_entity *se;
2301 se = cfs_rq->tg->se[cpu_of(rq_of(cfs_rq))];
2303 cfs_rq->throttled = 0;
2304 raw_spin_lock(&cfs_b->lock);
2305 cfs_b->throttled_time += rq->clock - cfs_rq->throttled_clock;
2306 list_del_rcu(&cfs_rq->throttled_list);
2307 raw_spin_unlock(&cfs_b->lock);
2309 update_rq_clock(rq);
2310 /* update hierarchical throttle state */
2311 walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);
2313 if (!cfs_rq->load.weight)
2316 task_delta = cfs_rq->h_nr_running;
2317 for_each_sched_entity(se) {
2321 cfs_rq = cfs_rq_of(se);
2323 enqueue_entity(cfs_rq, se, ENQUEUE_WAKEUP);
2324 cfs_rq->h_nr_running += task_delta;
2326 if (cfs_rq_throttled(cfs_rq))
2331 rq->nr_running += task_delta;
2333 /* determine whether we need to wake up potentially idle cpu */
2334 if (rq->curr == rq->idle && rq->cfs.nr_running)
2335 resched_task(rq->curr);
2338 static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
2339 u64 remaining, u64 expires)
2341 struct cfs_rq *cfs_rq;
2342 u64 runtime = remaining;
2345 list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
2347 struct rq *rq = rq_of(cfs_rq);
2349 raw_spin_lock(&rq->lock);
2350 if (!cfs_rq_throttled(cfs_rq))
2353 runtime = -cfs_rq->runtime_remaining + 1;
2354 if (runtime > remaining)
2355 runtime = remaining;
2356 remaining -= runtime;
2358 cfs_rq->runtime_remaining += runtime;
2359 cfs_rq->runtime_expires = expires;
2361 /* we check whether we're throttled above */
2362 if (cfs_rq->runtime_remaining > 0)
2363 unthrottle_cfs_rq(cfs_rq);
2366 raw_spin_unlock(&rq->lock);
2377 * Responsible for refilling a task_group's bandwidth and unthrottling its
2378 * cfs_rqs as appropriate. If there has been no activity within the last
2379 * period the timer is deactivated until scheduling resumes; cfs_b->idle is
2380 * used to track this state.
2382 static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun)
2384 u64 runtime, runtime_expires;
2385 int idle = 1, throttled;
2387 raw_spin_lock(&cfs_b->lock);
2388 /* no need to continue the timer with no bandwidth constraint */
2389 if (cfs_b->quota == RUNTIME_INF)
2392 throttled = !list_empty(&cfs_b->throttled_cfs_rq);
2393 /* idle depends on !throttled (for the case of a large deficit) */
2394 idle = cfs_b->idle && !throttled;
2395 cfs_b->nr_periods += overrun;
2397 /* if we're going inactive then everything else can be deferred */
2401 __refill_cfs_bandwidth_runtime(cfs_b);
2404 /* mark as potentially idle for the upcoming period */
2409 /* account preceding periods in which throttling occurred */
2410 cfs_b->nr_throttled += overrun;
2413 * There are throttled entities so we must first use the new bandwidth
2414 * to unthrottle them before making it generally available. This
2415 * ensures that all existing debts will be paid before a new cfs_rq is
2418 runtime = cfs_b->runtime;
2419 runtime_expires = cfs_b->runtime_expires;
2423 * This check is repeated as we are holding onto the new bandwidth
2424 * while we unthrottle. This can potentially race with an unthrottled
2425 * group trying to acquire new bandwidth from the global pool.
2427 while (throttled && runtime > 0) {
2428 raw_spin_unlock(&cfs_b->lock);
2429 /* we can't nest cfs_b->lock while distributing bandwidth */
2430 runtime = distribute_cfs_runtime(cfs_b, runtime,
2432 raw_spin_lock(&cfs_b->lock);
2434 throttled = !list_empty(&cfs_b->throttled_cfs_rq);
2437 /* return (any) remaining runtime */
2438 cfs_b->runtime = runtime;
2440 * While we are ensured activity in the period following an
2441 * unthrottle, this also covers the case in which the new bandwidth is
2442 * insufficient to cover the existing bandwidth deficit. (Forcing the
2443 * timer to remain active while there are any throttled entities.)
2448 cfs_b->timer_active = 0;
2449 raw_spin_unlock(&cfs_b->lock);
2454 /* a cfs_rq won't donate quota below this amount */
2455 static const u64 min_cfs_rq_runtime = 1 * NSEC_PER_MSEC;
2456 /* minimum remaining period time to redistribute slack quota */
2457 static const u64 min_bandwidth_expiration = 2 * NSEC_PER_MSEC;
2458 /* how long we wait to gather additional slack before distributing */
2459 static const u64 cfs_bandwidth_slack_period = 5 * NSEC_PER_MSEC;
2461 /* are we near the end of the current quota period? */
2462 static int runtime_refresh_within(struct cfs_bandwidth *cfs_b, u64 min_expire)
2464 struct hrtimer *refresh_timer = &cfs_b->period_timer;
2467 /* if the call-back is running a quota refresh is already occurring */
2468 if (hrtimer_callback_running(refresh_timer))
2471 /* is a quota refresh about to occur? */
2472 remaining = ktime_to_ns(hrtimer_expires_remaining(refresh_timer));
2473 if (remaining < min_expire)
2479 static void start_cfs_slack_bandwidth(struct cfs_bandwidth *cfs_b)
2481 u64 min_left = cfs_bandwidth_slack_period + min_bandwidth_expiration;
2483 /* if there's a quota refresh soon don't bother with slack */
2484 if (runtime_refresh_within(cfs_b, min_left))
2487 start_bandwidth_timer(&cfs_b->slack_timer,
2488 ns_to_ktime(cfs_bandwidth_slack_period));
2491 /* we know any runtime found here is valid as update_curr() precedes return */
2492 static void __return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2494 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
2495 s64 slack_runtime = cfs_rq->runtime_remaining - min_cfs_rq_runtime;
2497 if (slack_runtime <= 0)
2500 raw_spin_lock(&cfs_b->lock);
2501 if (cfs_b->quota != RUNTIME_INF &&
2502 cfs_rq->runtime_expires == cfs_b->runtime_expires) {
2503 cfs_b->runtime += slack_runtime;
2505 /* we are under rq->lock, defer unthrottling using a timer */
2506 if (cfs_b->runtime > sched_cfs_bandwidth_slice() &&
2507 !list_empty(&cfs_b->throttled_cfs_rq))
2508 start_cfs_slack_bandwidth(cfs_b);
2510 raw_spin_unlock(&cfs_b->lock);
2512 /* even if it's not valid for return we don't want to try again */
2513 cfs_rq->runtime_remaining -= slack_runtime;
2516 static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2518 if (!cfs_bandwidth_used())
2521 if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
2524 __return_cfs_rq_runtime(cfs_rq);
2528 * This is done with a timer (instead of inline with bandwidth return) since
2529 * it's necessary to juggle rq->locks to unthrottle their respective cfs_rqs.
2531 static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b)
2533 u64 runtime = 0, slice = sched_cfs_bandwidth_slice();
2536 /* confirm we're still not at a refresh boundary */
2537 if (runtime_refresh_within(cfs_b, min_bandwidth_expiration))
2540 raw_spin_lock(&cfs_b->lock);
2541 if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice) {
2542 runtime = cfs_b->runtime;
2545 expires = cfs_b->runtime_expires;
2546 raw_spin_unlock(&cfs_b->lock);
2551 runtime = distribute_cfs_runtime(cfs_b, runtime, expires);
2553 raw_spin_lock(&cfs_b->lock);
2554 if (expires == cfs_b->runtime_expires)
2555 cfs_b->runtime = runtime;
2556 raw_spin_unlock(&cfs_b->lock);
2560 * When a group wakes up we want to make sure that its quota is not already
2561 * expired/exceeded, otherwise it may be allowed to steal additional ticks of
2562 * runtime as update_curr() throttling can not not trigger until it's on-rq.
2564 static void check_enqueue_throttle(struct cfs_rq *cfs_rq)
2566 if (!cfs_bandwidth_used())
2569 /* an active group must be handled by the update_curr()->put() path */
2570 if (!cfs_rq->runtime_enabled || cfs_rq->curr)
2573 /* ensure the group is not already throttled */
2574 if (cfs_rq_throttled(cfs_rq))
2577 /* update runtime allocation */
2578 account_cfs_rq_runtime(cfs_rq, 0);
2579 if (cfs_rq->runtime_remaining <= 0)
2580 throttle_cfs_rq(cfs_rq);
2583 /* conditionally throttle active cfs_rq's from put_prev_entity() */
2584 static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2586 if (!cfs_bandwidth_used())
2589 if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
2593 * it's possible for a throttled entity to be forced into a running
2594 * state (e.g. set_curr_task), in this case we're finished.
2596 if (cfs_rq_throttled(cfs_rq))
2599 throttle_cfs_rq(cfs_rq);
2602 static inline u64 default_cfs_period(void);
2603 static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun);
2604 static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b);
2606 static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
2608 struct cfs_bandwidth *cfs_b =
2609 container_of(timer, struct cfs_bandwidth, slack_timer);
2610 do_sched_cfs_slack_timer(cfs_b);
2612 return HRTIMER_NORESTART;
2615 static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
2617 struct cfs_bandwidth *cfs_b =
2618 container_of(timer, struct cfs_bandwidth, period_timer);
2624 now = hrtimer_cb_get_time(timer);
2625 overrun = hrtimer_forward(timer, now, cfs_b->period);
2630 idle = do_sched_cfs_period_timer(cfs_b, overrun);
2633 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
2636 void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
2638 raw_spin_lock_init(&cfs_b->lock);
2640 cfs_b->quota = RUNTIME_INF;
2641 cfs_b->period = ns_to_ktime(default_cfs_period());
2643 INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
2644 hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2645 cfs_b->period_timer.function = sched_cfs_period_timer;
2646 hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2647 cfs_b->slack_timer.function = sched_cfs_slack_timer;
2650 static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2652 cfs_rq->runtime_enabled = 0;
2653 INIT_LIST_HEAD(&cfs_rq->throttled_list);
2656 /* requires cfs_b->lock, may release to reprogram timer */
2657 void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
2660 * The timer may be active because we're trying to set a new bandwidth
2661 * period or because we're racing with the tear-down path
2662 * (timer_active==0 becomes visible before the hrtimer call-back
2663 * terminates). In either case we ensure that it's re-programmed
2665 while (unlikely(hrtimer_active(&cfs_b->period_timer))) {
2666 raw_spin_unlock(&cfs_b->lock);
2667 /* ensure cfs_b->lock is available while we wait */
2668 hrtimer_cancel(&cfs_b->period_timer);
2670 raw_spin_lock(&cfs_b->lock);
2671 /* if someone else restarted the timer then we're done */
2672 if (cfs_b->timer_active)
2676 cfs_b->timer_active = 1;
2677 start_bandwidth_timer(&cfs_b->period_timer, cfs_b->period);
2680 static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
2682 hrtimer_cancel(&cfs_b->period_timer);
2683 hrtimer_cancel(&cfs_b->slack_timer);
2686 static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
2688 struct cfs_rq *cfs_rq;
2690 for_each_leaf_cfs_rq(rq, cfs_rq) {
2691 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
2693 if (!cfs_rq->runtime_enabled)
2697 * clock_task is not advancing so we just need to make sure
2698 * there's some valid quota amount
2700 cfs_rq->runtime_remaining = cfs_b->quota;
2701 if (cfs_rq_throttled(cfs_rq))
2702 unthrottle_cfs_rq(cfs_rq);
2706 #else /* CONFIG_CFS_BANDWIDTH */
2707 static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
2709 return rq_of(cfs_rq)->clock_task;
2712 static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
2713 unsigned long delta_exec) {}
2714 static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
2715 static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
2716 static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
2718 static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
2723 static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
2728 static inline int throttled_lb_pair(struct task_group *tg,
2729 int src_cpu, int dest_cpu)
2734 void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}
2736 #ifdef CONFIG_FAIR_GROUP_SCHED
2737 static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
2740 static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
2744 static inline void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}
2745 static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
2747 #endif /* CONFIG_CFS_BANDWIDTH */
2749 /**************************************************
2750 * CFS operations on tasks:
2753 #ifdef CONFIG_SCHED_HRTICK
2754 static void hrtick_start_fair(struct rq *rq, struct task_struct *p)
2756 struct sched_entity *se = &p->se;
2757 struct cfs_rq *cfs_rq = cfs_rq_of(se);
2759 WARN_ON(task_rq(p) != rq);
2761 if (cfs_rq->nr_running > 1) {
2762 u64 slice = sched_slice(cfs_rq, se);
2763 u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime;
2764 s64 delta = slice - ran;
2773 * Don't schedule slices shorter than 10000ns, that just
2774 * doesn't make sense. Rely on vruntime for fairness.
2777 delta = max_t(s64, 10000LL, delta);
2779 hrtick_start(rq, delta);
2784 * called from enqueue/dequeue and updates the hrtick when the
2785 * current task is from our class and nr_running is low enough
2788 static void hrtick_update(struct rq *rq)
2790 struct task_struct *curr = rq->curr;
2792 if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
2795 if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
2796 hrtick_start_fair(rq, curr);
2798 #else /* !CONFIG_SCHED_HRTICK */
2800 hrtick_start_fair(struct rq *rq, struct task_struct *p)
2804 static inline void hrtick_update(struct rq *rq)
2810 * The enqueue_task method is called before nr_running is
2811 * increased. Here we update the fair scheduling stats and
2812 * then put the task into the rbtree:
2815 enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
2817 struct cfs_rq *cfs_rq;
2818 struct sched_entity *se = &p->se;
2820 for_each_sched_entity(se) {
2823 cfs_rq = cfs_rq_of(se);
2824 enqueue_entity(cfs_rq, se, flags);
2827 * end evaluation on encountering a throttled cfs_rq
2829 * note: in the case of encountering a throttled cfs_rq we will
2830 * post the final h_nr_running increment below.
2832 if (cfs_rq_throttled(cfs_rq))
2834 cfs_rq->h_nr_running++;
2836 flags = ENQUEUE_WAKEUP;
2839 for_each_sched_entity(se) {
2840 cfs_rq = cfs_rq_of(se);
2841 cfs_rq->h_nr_running++;
2843 if (cfs_rq_throttled(cfs_rq))
2846 update_cfs_shares(cfs_rq);
2847 update_entity_load_avg(se, 1);
2851 update_rq_runnable_avg(rq, rq->nr_running);
2857 static void set_next_buddy(struct sched_entity *se);
2860 * The dequeue_task method is called before nr_running is
2861 * decreased. We remove the task from the rbtree and
2862 * update the fair scheduling stats:
2864 static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
2866 struct cfs_rq *cfs_rq;
2867 struct sched_entity *se = &p->se;
2868 int task_sleep = flags & DEQUEUE_SLEEP;
2870 for_each_sched_entity(se) {
2871 cfs_rq = cfs_rq_of(se);
2872 dequeue_entity(cfs_rq, se, flags);
2875 * end evaluation on encountering a throttled cfs_rq
2877 * note: in the case of encountering a throttled cfs_rq we will
2878 * post the final h_nr_running decrement below.
2880 if (cfs_rq_throttled(cfs_rq))
2882 cfs_rq->h_nr_running--;
2884 /* Don't dequeue parent if it has other entities besides us */
2885 if (cfs_rq->load.weight) {
2887 * Bias pick_next to pick a task from this cfs_rq, as
2888 * p is sleeping when it is within its sched_slice.
2890 if (task_sleep && parent_entity(se))
2891 set_next_buddy(parent_entity(se));
2893 /* avoid re-evaluating load for this entity */
2894 se = parent_entity(se);
2897 flags |= DEQUEUE_SLEEP;
2900 for_each_sched_entity(se) {
2901 cfs_rq = cfs_rq_of(se);
2902 cfs_rq->h_nr_running--;
2904 if (cfs_rq_throttled(cfs_rq))
2907 update_cfs_shares(cfs_rq);
2908 update_entity_load_avg(se, 1);
2913 update_rq_runnable_avg(rq, 1);
2919 /* Used instead of source_load when we know the type == 0 */
2920 static unsigned long weighted_cpuload(const int cpu)
2922 return cpu_rq(cpu)->load.weight;
2926 * Return a low guess at the load of a migration-source cpu weighted
2927 * according to the scheduling class and "nice" value.
2929 * We want to under-estimate the load of migration sources, to
2930 * balance conservatively.
2932 static unsigned long source_load(int cpu, int type)
2934 struct rq *rq = cpu_rq(cpu);
2935 unsigned long total = weighted_cpuload(cpu);
2937 if (type == 0 || !sched_feat(LB_BIAS))
2940 return min(rq->cpu_load[type-1], total);
2944 * Return a high guess at the load of a migration-target cpu weighted
2945 * according to the scheduling class and "nice" value.
2947 static unsigned long target_load(int cpu, int type)
2949 struct rq *rq = cpu_rq(cpu);
2950 unsigned long total = weighted_cpuload(cpu);
2952 if (type == 0 || !sched_feat(LB_BIAS))
2955 return max(rq->cpu_load[type-1], total);
2958 static unsigned long power_of(int cpu)
2960 return cpu_rq(cpu)->cpu_power;
2963 static unsigned long cpu_avg_load_per_task(int cpu)
2965 struct rq *rq = cpu_rq(cpu);
2966 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
2969 return rq->load.weight / nr_running;
2975 static void task_waking_fair(struct task_struct *p)
2977 struct sched_entity *se = &p->se;
2978 struct cfs_rq *cfs_rq = cfs_rq_of(se);
2981 #ifndef CONFIG_64BIT
2982 u64 min_vruntime_copy;
2985 min_vruntime_copy = cfs_rq->min_vruntime_copy;
2987 min_vruntime = cfs_rq->min_vruntime;
2988 } while (min_vruntime != min_vruntime_copy);
2990 min_vruntime = cfs_rq->min_vruntime;
2993 se->vruntime -= min_vruntime;
2996 #ifdef CONFIG_FAIR_GROUP_SCHED
2998 * effective_load() calculates the load change as seen from the root_task_group
3000 * Adding load to a group doesn't make a group heavier, but can cause movement
3001 * of group shares between cpus. Assuming the shares were perfectly aligned one
3002 * can calculate the shift in shares.
3004 * Calculate the effective load difference if @wl is added (subtracted) to @tg
3005 * on this @cpu and results in a total addition (subtraction) of @wg to the
3006 * total group weight.
3008 * Given a runqueue weight distribution (rw_i) we can compute a shares
3009 * distribution (s_i) using:
3011 * s_i = rw_i / \Sum rw_j (1)
3013 * Suppose we have 4 CPUs and our @tg is a direct child of the root group and
3014 * has 7 equal weight tasks, distributed as below (rw_i), with the resulting
3015 * shares distribution (s_i):
3017 * rw_i = { 2, 4, 1, 0 }
3018 * s_i = { 2/7, 4/7, 1/7, 0 }
3020 * As per wake_affine() we're interested in the load of two CPUs (the CPU the
3021 * task used to run on and the CPU the waker is running on), we need to
3022 * compute the effect of waking a task on either CPU and, in case of a sync
3023 * wakeup, compute the effect of the current task going to sleep.
3025 * So for a change of @wl to the local @cpu with an overall group weight change
3026 * of @wl we can compute the new shares distribution (s'_i) using:
3028 * s'_i = (rw_i + @wl) / (@wg + \Sum rw_j) (2)
3030 * Suppose we're interested in CPUs 0 and 1, and want to compute the load
3031 * differences in waking a task to CPU 0. The additional task changes the
3032 * weight and shares distributions like:
3034 * rw'_i = { 3, 4, 1, 0 }
3035 * s'_i = { 3/8, 4/8, 1/8, 0 }
3037 * We can then compute the difference in effective weight by using:
3039 * dw_i = S * (s'_i - s_i) (3)
3041 * Where 'S' is the group weight as seen by its parent.
3043 * Therefore the effective change in loads on CPU 0 would be 5/56 (3/8 - 2/7)
3044 * times the weight of the group. The effect on CPU 1 would be -4/56 (4/8 -
3045 * 4/7) times the weight of the group.
3047 static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
3049 struct sched_entity *se = tg->se[cpu];
3051 if (!tg->parent) /* the trivial, non-cgroup case */
3054 for_each_sched_entity(se) {
3060 * W = @wg + \Sum rw_j
3062 W = wg + calc_tg_weight(tg, se->my_q);
3067 w = se->my_q->load.weight + wl;
3070 * wl = S * s'_i; see (2)
3073 wl = (w * tg->shares) / W;
3078 * Per the above, wl is the new se->load.weight value; since
3079 * those are clipped to [MIN_SHARES, ...) do so now. See
3080 * calc_cfs_shares().
3082 if (wl < MIN_SHARES)
3086 * wl = dw_i = S * (s'_i - s_i); see (3)
3088 wl -= se->load.weight;
3091 * Recursively apply this logic to all parent groups to compute
3092 * the final effective load change on the root group. Since
3093 * only the @tg group gets extra weight, all parent groups can
3094 * only redistribute existing shares. @wl is the shift in shares
3095 * resulting from this level per the above.
3104 static inline unsigned long effective_load(struct task_group *tg, int cpu,
3105 unsigned long wl, unsigned long wg)
3112 static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
3114 s64 this_load, load;
3115 int idx, this_cpu, prev_cpu;
3116 unsigned long tl_per_task;
3117 struct task_group *tg;
3118 unsigned long weight;
3122 this_cpu = smp_processor_id();
3123 prev_cpu = task_cpu(p);
3124 load = source_load(prev_cpu, idx);
3125 this_load = target_load(this_cpu, idx);
3128 * If sync wakeup then subtract the (maximum possible)
3129 * effect of the currently running task from the load
3130 * of the current CPU:
3133 tg = task_group(current);
3134 weight = current->se.load.weight;
3136 this_load += effective_load(tg, this_cpu, -weight, -weight);
3137 load += effective_load(tg, prev_cpu, 0, -weight);
3141 weight = p->se.load.weight;
3144 * In low-load situations, where prev_cpu is idle and this_cpu is idle
3145 * due to the sync cause above having dropped this_load to 0, we'll
3146 * always have an imbalance, but there's really nothing you can do
3147 * about that, so that's good too.
3149 * Otherwise check if either cpus are near enough in load to allow this
3150 * task to be woken on this_cpu.
3152 if (this_load > 0) {
3153 s64 this_eff_load, prev_eff_load;
3155 this_eff_load = 100;
3156 this_eff_load *= power_of(prev_cpu);
3157 this_eff_load *= this_load +
3158 effective_load(tg, this_cpu, weight, weight);
3160 prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
3161 prev_eff_load *= power_of(this_cpu);
3162 prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
3164 balanced = this_eff_load <= prev_eff_load;
3169 * If the currently running task will sleep within
3170 * a reasonable amount of time then attract this newly
3173 if (sync && balanced)
3176 schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
3177 tl_per_task = cpu_avg_load_per_task(this_cpu);
3180 (this_load <= load &&
3181 this_load + target_load(prev_cpu, idx) <= tl_per_task)) {
3183 * This domain has SD_WAKE_AFFINE and
3184 * p is cache cold in this domain, and
3185 * there is no bad imbalance.
3187 schedstat_inc(sd, ttwu_move_affine);
3188 schedstat_inc(p, se.statistics.nr_wakeups_affine);
3196 * find_idlest_group finds and returns the least busy CPU group within the
3199 static struct sched_group *
3200 find_idlest_group(struct sched_domain *sd, struct task_struct *p,
3201 int this_cpu, int load_idx)
3203 struct sched_group *idlest = NULL, *group = sd->groups;
3204 unsigned long min_load = ULONG_MAX, this_load = 0;
3205 int imbalance = 100 + (sd->imbalance_pct-100)/2;
3208 unsigned long load, avg_load;
3212 /* Skip over this group if it has no CPUs allowed */
3213 if (!cpumask_intersects(sched_group_cpus(group),
3214 tsk_cpus_allowed(p)))
3217 local_group = cpumask_test_cpu(this_cpu,
3218 sched_group_cpus(group));
3220 /* Tally up the load of all CPUs in the group */
3223 for_each_cpu(i, sched_group_cpus(group)) {
3224 /* Bias balancing toward cpus of our domain */
3226 load = source_load(i, load_idx);
3228 load = target_load(i, load_idx);
3233 /* Adjust by relative CPU power of the group */
3234 avg_load = (avg_load * SCHED_POWER_SCALE) / group->sgp->power;
3237 this_load = avg_load;
3238 } else if (avg_load < min_load) {
3239 min_load = avg_load;
3242 } while (group = group->next, group != sd->groups);
3244 if (!idlest || 100*this_load < imbalance*min_load)
3250 * find_idlest_cpu - find the idlest cpu among the cpus in group.
3253 find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
3255 unsigned long load, min_load = ULONG_MAX;
3259 /* Traverse only the allowed CPUs */
3260 for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
3261 load = weighted_cpuload(i);
3263 if (load < min_load || (load == min_load && i == this_cpu)) {
3273 * Try and locate an idle CPU in the sched_domain.
3275 static int select_idle_sibling(struct task_struct *p, int target)
3277 struct sched_domain *sd;
3278 struct sched_group *sg;
3279 int i = task_cpu(p);
3281 if (idle_cpu(target))
3285 * If the prevous cpu is cache affine and idle, don't be stupid.
3287 if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
3291 * Otherwise, iterate the domains and find an elegible idle cpu.
3293 sd = rcu_dereference(per_cpu(sd_llc, target));
3294 for_each_lower_domain(sd) {
3297 if (!cpumask_intersects(sched_group_cpus(sg),
3298 tsk_cpus_allowed(p)))
3301 for_each_cpu(i, sched_group_cpus(sg)) {
3302 if (i == target || !idle_cpu(i))
3306 target = cpumask_first_and(sched_group_cpus(sg),
3307 tsk_cpus_allowed(p));
3311 } while (sg != sd->groups);
3318 * sched_balance_self: balance the current task (running on cpu) in domains
3319 * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
3322 * Balance, ie. select the least loaded group.
3324 * Returns the target CPU number, or the same CPU if no balancing is needed.
3326 * preempt must be disabled.
3329 select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags)
3331 struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
3332 int cpu = smp_processor_id();
3333 int prev_cpu = task_cpu(p);
3335 int want_affine = 0;
3336 int sync = wake_flags & WF_SYNC;
3338 if (p->nr_cpus_allowed == 1)
3341 if (sd_flag & SD_BALANCE_WAKE) {
3342 if (cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
3348 for_each_domain(cpu, tmp) {
3349 if (!(tmp->flags & SD_LOAD_BALANCE))
3353 * If both cpu and prev_cpu are part of this domain,
3354 * cpu is a valid SD_WAKE_AFFINE target.
3356 if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
3357 cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
3362 if (tmp->flags & sd_flag)
3367 if (cpu != prev_cpu && wake_affine(affine_sd, p, sync))
3370 new_cpu = select_idle_sibling(p, prev_cpu);
3375 int load_idx = sd->forkexec_idx;
3376 struct sched_group *group;
3379 if (!(sd->flags & sd_flag)) {
3384 if (sd_flag & SD_BALANCE_WAKE)
3385 load_idx = sd->wake_idx;
3387 group = find_idlest_group(sd, p, cpu, load_idx);
3393 new_cpu = find_idlest_cpu(group, p, cpu);
3394 if (new_cpu == -1 || new_cpu == cpu) {
3395 /* Now try balancing at a lower domain level of cpu */
3400 /* Now try balancing at a lower domain level of new_cpu */
3402 weight = sd->span_weight;
3404 for_each_domain(cpu, tmp) {
3405 if (weight <= tmp->span_weight)
3407 if (tmp->flags & sd_flag)
3410 /* while loop will break here if sd == NULL */
3419 * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
3420 * removed when useful for applications beyond shares distribution (e.g.
3423 #ifdef CONFIG_FAIR_GROUP_SCHED
3425 * Called immediately before a task is migrated to a new cpu; task_cpu(p) and
3426 * cfs_rq_of(p) references at time of call are still valid and identify the
3427 * previous cpu. However, the caller only guarantees p->pi_lock is held; no
3428 * other assumptions, including the state of rq->lock, should be made.
3431 migrate_task_rq_fair(struct task_struct *p, int next_cpu)
3433 struct sched_entity *se = &p->se;
3434 struct cfs_rq *cfs_rq = cfs_rq_of(se);
3437 * Load tracking: accumulate removed load so that it can be processed
3438 * when we next update owning cfs_rq under rq->lock. Tasks contribute
3439 * to blocked load iff they have a positive decay-count. It can never
3440 * be negative here since on-rq tasks have decay-count == 0.
3442 if (se->avg.decay_count) {
3443 se->avg.decay_count = -__synchronize_entity_decay(se);
3444 atomic64_add(se->avg.load_avg_contrib, &cfs_rq->removed_load);
3448 #endif /* CONFIG_SMP */
3450 static unsigned long
3451 wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
3453 unsigned long gran = sysctl_sched_wakeup_granularity;
3456 * Since its curr running now, convert the gran from real-time
3457 * to virtual-time in his units.
3459 * By using 'se' instead of 'curr' we penalize light tasks, so
3460 * they get preempted easier. That is, if 'se' < 'curr' then
3461 * the resulting gran will be larger, therefore penalizing the
3462 * lighter, if otoh 'se' > 'curr' then the resulting gran will
3463 * be smaller, again penalizing the lighter task.
3465 * This is especially important for buddies when the leftmost
3466 * task is higher priority than the buddy.
3468 return calc_delta_fair(gran, se);
3472 * Should 'se' preempt 'curr'.
3486 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
3488 s64 gran, vdiff = curr->vruntime - se->vruntime;
3493 gran = wakeup_gran(curr, se);
3500 static void set_last_buddy(struct sched_entity *se)
3502 if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
3505 for_each_sched_entity(se)
3506 cfs_rq_of(se)->last = se;
3509 static void set_next_buddy(struct sched_entity *se)
3511 if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
3514 for_each_sched_entity(se)
3515 cfs_rq_of(se)->next = se;
3518 static void set_skip_buddy(struct sched_entity *se)
3520 for_each_sched_entity(se)
3521 cfs_rq_of(se)->skip = se;
3525 * Preempt the current task with a newly woken task if needed:
3527 static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
3529 struct task_struct *curr = rq->curr;
3530 struct sched_entity *se = &curr->se, *pse = &p->se;
3531 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
3532 int scale = cfs_rq->nr_running >= sched_nr_latency;
3533 int next_buddy_marked = 0;
3535 if (unlikely(se == pse))
3539 * This is possible from callers such as move_task(), in which we
3540 * unconditionally check_prempt_curr() after an enqueue (which may have
3541 * lead to a throttle). This both saves work and prevents false
3542 * next-buddy nomination below.
3544 if (unlikely(throttled_hierarchy(cfs_rq_of(pse))))
3547 if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
3548 set_next_buddy(pse);
3549 next_buddy_marked = 1;
3553 * We can come here with TIF_NEED_RESCHED already set from new task
3556 * Note: this also catches the edge-case of curr being in a throttled
3557 * group (e.g. via set_curr_task), since update_curr() (in the
3558 * enqueue of curr) will have resulted in resched being set. This
3559 * prevents us from potentially nominating it as a false LAST_BUDDY
3562 if (test_tsk_need_resched(curr))
3565 /* Idle tasks are by definition preempted by non-idle tasks. */
3566 if (unlikely(curr->policy == SCHED_IDLE) &&
3567 likely(p->policy != SCHED_IDLE))
3571 * Batch and idle tasks do not preempt non-idle tasks (their preemption
3572 * is driven by the tick):
3574 if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
3577 find_matching_se(&se, &pse);
3578 update_curr(cfs_rq_of(se));
3580 if (wakeup_preempt_entity(se, pse) == 1) {
3582 * Bias pick_next to pick the sched entity that is
3583 * triggering this preemption.
3585 if (!next_buddy_marked)
3586 set_next_buddy(pse);
3595 * Only set the backward buddy when the current task is still
3596 * on the rq. This can happen when a wakeup gets interleaved
3597 * with schedule on the ->pre_schedule() or idle_balance()
3598 * point, either of which can * drop the rq lock.
3600 * Also, during early boot the idle thread is in the fair class,
3601 * for obvious reasons its a bad idea to schedule back to it.
3603 if (unlikely(!se->on_rq || curr == rq->idle))
3606 if (sched_feat(LAST_BUDDY) && scale && entity_is_task(se))
3610 static struct task_struct *pick_next_task_fair(struct rq *rq)
3612 struct task_struct *p;
3613 struct cfs_rq *cfs_rq = &rq->cfs;
3614 struct sched_entity *se;
3616 if (!cfs_rq->nr_running)
3620 se = pick_next_entity(cfs_rq);
3621 set_next_entity(cfs_rq, se);
3622 cfs_rq = group_cfs_rq(se);
3626 if (hrtick_enabled(rq))
3627 hrtick_start_fair(rq, p);
3633 * Account for a descheduled task:
3635 static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
3637 struct sched_entity *se = &prev->se;
3638 struct cfs_rq *cfs_rq;
3640 for_each_sched_entity(se) {
3641 cfs_rq = cfs_rq_of(se);
3642 put_prev_entity(cfs_rq, se);
3647 * sched_yield() is very simple
3649 * The magic of dealing with the ->skip buddy is in pick_next_entity.
3651 static void yield_task_fair(struct rq *rq)
3653 struct task_struct *curr = rq->curr;
3654 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
3655 struct sched_entity *se = &curr->se;
3658 * Are we the only task in the tree?
3660 if (unlikely(rq->nr_running == 1))
3663 clear_buddies(cfs_rq, se);
3665 if (curr->policy != SCHED_BATCH) {
3666 update_rq_clock(rq);
3668 * Update run-time statistics of the 'current'.
3670 update_curr(cfs_rq);
3672 * Tell update_rq_clock() that we've just updated,
3673 * so we don't do microscopic update in schedule()
3674 * and double the fastpath cost.
3676 rq->skip_clock_update = 1;
3682 static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
3684 struct sched_entity *se = &p->se;
3686 /* throttled hierarchies are not runnable */
3687 if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
3690 /* Tell the scheduler that we'd really like pse to run next. */
3693 yield_task_fair(rq);
3699 /**************************************************
3700 * Fair scheduling class load-balancing methods.
3704 * The purpose of load-balancing is to achieve the same basic fairness the
3705 * per-cpu scheduler provides, namely provide a proportional amount of compute
3706 * time to each task. This is expressed in the following equation:
3708 * W_i,n/P_i == W_j,n/P_j for all i,j (1)
3710 * Where W_i,n is the n-th weight average for cpu i. The instantaneous weight
3711 * W_i,0 is defined as:
3713 * W_i,0 = \Sum_j w_i,j (2)
3715 * Where w_i,j is the weight of the j-th runnable task on cpu i. This weight
3716 * is derived from the nice value as per prio_to_weight[].
3718 * The weight average is an exponential decay average of the instantaneous
3721 * W'_i,n = (2^n - 1) / 2^n * W_i,n + 1 / 2^n * W_i,0 (3)
3723 * P_i is the cpu power (or compute capacity) of cpu i, typically it is the
3724 * fraction of 'recent' time available for SCHED_OTHER task execution. But it
3725 * can also include other factors [XXX].
3727 * To achieve this balance we define a measure of imbalance which follows
3728 * directly from (1):
3730 * imb_i,j = max{ avg(W/P), W_i/P_i } - min{ avg(W/P), W_j/P_j } (4)
3732 * We them move tasks around to minimize the imbalance. In the continuous
3733 * function space it is obvious this converges, in the discrete case we get
3734 * a few fun cases generally called infeasible weight scenarios.
3737 * - infeasible weights;
3738 * - local vs global optima in the discrete case. ]
3743 * In order to solve the imbalance equation (4), and avoid the obvious O(n^2)
3744 * for all i,j solution, we create a tree of cpus that follows the hardware
3745 * topology where each level pairs two lower groups (or better). This results
3746 * in O(log n) layers. Furthermore we reduce the number of cpus going up the
3747 * tree to only the first of the previous level and we decrease the frequency
3748 * of load-balance at each level inv. proportional to the number of cpus in
3754 * \Sum { --- * --- * 2^i } = O(n) (5)
3756 * `- size of each group
3757 * | | `- number of cpus doing load-balance
3759 * `- sum over all levels
3761 * Coupled with a limit on how many tasks we can migrate every balance pass,
3762 * this makes (5) the runtime complexity of the balancer.
3764 * An important property here is that each CPU is still (indirectly) connected
3765 * to every other cpu in at most O(log n) steps:
3767 * The adjacency matrix of the resulting graph is given by:
3770 * A_i,j = \Union (i % 2^k == 0) && i / 2^(k+1) == j / 2^(k+1) (6)
3773 * And you'll find that:
3775 * A^(log_2 n)_i,j != 0 for all i,j (7)
3777 * Showing there's indeed a path between every cpu in at most O(log n) steps.
3778 * The task movement gives a factor of O(m), giving a convergence complexity
3781 * O(nm log n), n := nr_cpus, m := nr_tasks (8)
3786 * In order to avoid CPUs going idle while there's still work to do, new idle
3787 * balancing is more aggressive and has the newly idle cpu iterate up the domain
3788 * tree itself instead of relying on other CPUs to bring it work.
3790 * This adds some complexity to both (5) and (8) but it reduces the total idle
3798 * Cgroups make a horror show out of (2), instead of a simple sum we get:
3801 * W_i,0 = \Sum_j \Prod_k w_k * ----- (9)
3806 * s_k,i = \Sum_j w_i,j,k and S_k = \Sum_i s_k,i (10)
3808 * w_i,j,k is the weight of the j-th runnable task in the k-th cgroup on cpu i.
3810 * The big problem is S_k, its a global sum needed to compute a local (W_i)
3813 * [XXX write more on how we solve this.. _after_ merging pjt's patches that
3814 * rewrite all of this once again.]
3817 static unsigned long __read_mostly max_load_balance_interval = HZ/10;
3819 #define LBF_ALL_PINNED 0x01
3820 #define LBF_NEED_BREAK 0x02
3821 #define LBF_SOME_PINNED 0x04
3824 struct sched_domain *sd;
3832 struct cpumask *dst_grpmask;
3834 enum cpu_idle_type idle;
3836 /* The set of CPUs under consideration for load-balancing */
3837 struct cpumask *cpus;
3842 unsigned int loop_break;
3843 unsigned int loop_max;
3847 * move_task - move a task from one runqueue to another runqueue.
3848 * Both runqueues must be locked.
3850 static void move_task(struct task_struct *p, struct lb_env *env)
3852 deactivate_task(env->src_rq, p, 0);
3853 set_task_cpu(p, env->dst_cpu);
3854 activate_task(env->dst_rq, p, 0);
3855 check_preempt_curr(env->dst_rq, p, 0);
3859 * Is this task likely cache-hot:
3862 task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
3866 if (p->sched_class != &fair_sched_class)
3869 if (unlikely(p->policy == SCHED_IDLE))
3873 * Buddy candidates are cache hot:
3875 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
3876 (&p->se == cfs_rq_of(&p->se)->next ||
3877 &p->se == cfs_rq_of(&p->se)->last))
3880 if (sysctl_sched_migration_cost == -1)
3882 if (sysctl_sched_migration_cost == 0)
3885 delta = now - p->se.exec_start;
3887 return delta < (s64)sysctl_sched_migration_cost;
3891 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
3894 int can_migrate_task(struct task_struct *p, struct lb_env *env)
3896 int tsk_cache_hot = 0;
3898 * We do not migrate tasks that are:
3899 * 1) throttled_lb_pair, or
3900 * 2) cannot be migrated to this CPU due to cpus_allowed, or
3901 * 3) running (obviously), or
3902 * 4) are cache-hot on their current CPU.
3904 if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
3907 if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
3910 schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
3913 * Remember if this task can be migrated to any other cpu in
3914 * our sched_group. We may want to revisit it if we couldn't
3915 * meet load balance goals by pulling other tasks on src_cpu.
3917 * Also avoid computing new_dst_cpu if we have already computed
3918 * one in current iteration.
3920 if (!env->dst_grpmask || (env->flags & LBF_SOME_PINNED))
3923 /* Prevent to re-select dst_cpu via env's cpus */
3924 for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
3925 if (cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) {
3926 env->flags |= LBF_SOME_PINNED;
3927 env->new_dst_cpu = cpu;
3935 /* Record that we found atleast one task that could run on dst_cpu */
3936 env->flags &= ~LBF_ALL_PINNED;
3938 if (task_running(env->src_rq, p)) {
3939 schedstat_inc(p, se.statistics.nr_failed_migrations_running);
3944 * Aggressive migration if:
3945 * 1) task is cache cold, or
3946 * 2) too many balance attempts have failed.
3949 tsk_cache_hot = task_hot(p, env->src_rq->clock_task, env->sd);
3950 if (!tsk_cache_hot ||
3951 env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
3953 if (tsk_cache_hot) {
3954 schedstat_inc(env->sd, lb_hot_gained[env->idle]);
3955 schedstat_inc(p, se.statistics.nr_forced_migrations);
3961 schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
3966 * move_one_task tries to move exactly one task from busiest to this_rq, as
3967 * part of active balancing operations within "domain".
3968 * Returns 1 if successful and 0 otherwise.
3970 * Called with both runqueues locked.
3972 static int move_one_task(struct lb_env *env)
3974 struct task_struct *p, *n;
3976 list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
3977 if (!can_migrate_task(p, env))
3982 * Right now, this is only the second place move_task()
3983 * is called, so we can safely collect move_task()
3984 * stats here rather than inside move_task().
3986 schedstat_inc(env->sd, lb_gained[env->idle]);
3992 static unsigned long task_h_load(struct task_struct *p);
3994 static const unsigned int sched_nr_migrate_break = 32;
3997 * move_tasks tries to move up to imbalance weighted load from busiest to
3998 * this_rq, as part of a balancing operation within domain "sd".
3999 * Returns 1 if successful and 0 otherwise.
4001 * Called with both runqueues locked.
4003 static int move_tasks(struct lb_env *env)
4005 struct list_head *tasks = &env->src_rq->cfs_tasks;
4006 struct task_struct *p;
4010 if (env->imbalance <= 0)
4013 while (!list_empty(tasks)) {
4014 p = list_first_entry(tasks, struct task_struct, se.group_node);
4017 /* We've more or less seen every task there is, call it quits */
4018 if (env->loop > env->loop_max)
4021 /* take a breather every nr_migrate tasks */
4022 if (env->loop > env->loop_break) {
4023 env->loop_break += sched_nr_migrate_break;
4024 env->flags |= LBF_NEED_BREAK;
4028 if (!can_migrate_task(p, env))
4031 load = task_h_load(p);
4033 if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
4036 if ((load / 2) > env->imbalance)
4041 env->imbalance -= load;
4043 #ifdef CONFIG_PREEMPT
4045 * NEWIDLE balancing is a source of latency, so preemptible
4046 * kernels will stop after the first task is pulled to minimize
4047 * the critical section.
4049 if (env->idle == CPU_NEWLY_IDLE)
4054 * We only want to steal up to the prescribed amount of
4057 if (env->imbalance <= 0)
4062 list_move_tail(&p->se.group_node, tasks);
4066 * Right now, this is one of only two places move_task() is called,
4067 * so we can safely collect move_task() stats here rather than
4068 * inside move_task().
4070 schedstat_add(env->sd, lb_gained[env->idle], pulled);
4075 #ifdef CONFIG_FAIR_GROUP_SCHED
4077 * update tg->load_weight by folding this cpu's load_avg
4079 static void __update_blocked_averages_cpu(struct task_group *tg, int cpu)
4081 struct sched_entity *se = tg->se[cpu];
4082 struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
4084 /* throttled entities do not contribute to load */
4085 if (throttled_hierarchy(cfs_rq))
4088 update_cfs_rq_blocked_load(cfs_rq, 1);
4091 update_entity_load_avg(se, 1);
4093 * We pivot on our runnable average having decayed to zero for
4094 * list removal. This generally implies that all our children
4095 * have also been removed (modulo rounding error or bandwidth
4096 * control); however, such cases are rare and we can fix these
4099 * TODO: fix up out-of-order children on enqueue.
4101 if (!se->avg.runnable_avg_sum && !cfs_rq->nr_running)
4102 list_del_leaf_cfs_rq(cfs_rq);
4104 struct rq *rq = rq_of(cfs_rq);
4105 update_rq_runnable_avg(rq, rq->nr_running);
4109 static void update_blocked_averages(int cpu)
4111 struct rq *rq = cpu_rq(cpu);
4112 struct cfs_rq *cfs_rq;
4113 unsigned long flags;
4115 raw_spin_lock_irqsave(&rq->lock, flags);
4116 update_rq_clock(rq);
4118 * Iterates the task_group tree in a bottom up fashion, see
4119 * list_add_leaf_cfs_rq() for details.
4121 for_each_leaf_cfs_rq(rq, cfs_rq) {
4123 * Note: We may want to consider periodically releasing
4124 * rq->lock about these updates so that creating many task
4125 * groups does not result in continually extending hold time.
4127 __update_blocked_averages_cpu(cfs_rq->tg, rq->cpu);
4130 raw_spin_unlock_irqrestore(&rq->lock, flags);
4134 * Compute the cpu's hierarchical load factor for each task group.
4135 * This needs to be done in a top-down fashion because the load of a child
4136 * group is a fraction of its parents load.
4138 static int tg_load_down(struct task_group *tg, void *data)
4141 long cpu = (long)data;
4144 load = cpu_rq(cpu)->load.weight;
4146 load = tg->parent->cfs_rq[cpu]->h_load;
4147 load *= tg->se[cpu]->load.weight;
4148 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
4151 tg->cfs_rq[cpu]->h_load = load;
4156 static void update_h_load(long cpu)
4158 struct rq *rq = cpu_rq(cpu);
4159 unsigned long now = jiffies;
4161 if (rq->h_load_throttle == now)
4164 rq->h_load_throttle = now;
4167 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
4171 static unsigned long task_h_load(struct task_struct *p)
4173 struct cfs_rq *cfs_rq = task_cfs_rq(p);
4176 load = p->se.load.weight;
4177 load = div_u64(load * cfs_rq->h_load, cfs_rq->load.weight + 1);
4182 static inline void update_blocked_averages(int cpu)
4186 static inline void update_h_load(long cpu)
4190 static unsigned long task_h_load(struct task_struct *p)
4192 return p->se.load.weight;
4196 /********** Helpers for find_busiest_group ************************/
4198 * sd_lb_stats - Structure to store the statistics of a sched_domain
4199 * during load balancing.
4201 struct sd_lb_stats {
4202 struct sched_group *busiest; /* Busiest group in this sd */
4203 struct sched_group *this; /* Local group in this sd */
4204 unsigned long total_load; /* Total load of all groups in sd */
4205 unsigned long total_pwr; /* Total power of all groups in sd */
4206 unsigned long avg_load; /* Average load across all groups in sd */
4208 /** Statistics of this group */
4209 unsigned long this_load;
4210 unsigned long this_load_per_task;
4211 unsigned long this_nr_running;
4212 unsigned long this_has_capacity;
4213 unsigned int this_idle_cpus;
4215 /* Statistics of the busiest group */
4216 unsigned int busiest_idle_cpus;
4217 unsigned long max_load;
4218 unsigned long busiest_load_per_task;
4219 unsigned long busiest_nr_running;
4220 unsigned long busiest_group_capacity;
4221 unsigned long busiest_has_capacity;
4222 unsigned int busiest_group_weight;
4224 int group_imb; /* Is there imbalance in this sd */
4228 * sg_lb_stats - stats of a sched_group required for load_balancing
4230 struct sg_lb_stats {
4231 unsigned long avg_load; /*Avg load across the CPUs of the group */
4232 unsigned long group_load; /* Total load over the CPUs of the group */
4233 unsigned long sum_nr_running; /* Nr tasks running in the group */
4234 unsigned long sum_weighted_load; /* Weighted load of group's tasks */
4235 unsigned long group_capacity;
4236 unsigned long idle_cpus;
4237 unsigned long group_weight;
4238 int group_imb; /* Is there an imbalance in the group ? */
4239 int group_has_capacity; /* Is there extra capacity in the group? */
4243 * get_sd_load_idx - Obtain the load index for a given sched domain.
4244 * @sd: The sched_domain whose load_idx is to be obtained.
4245 * @idle: The Idle status of the CPU for whose sd load_icx is obtained.
4247 static inline int get_sd_load_idx(struct sched_domain *sd,
4248 enum cpu_idle_type idle)
4254 load_idx = sd->busy_idx;
4257 case CPU_NEWLY_IDLE:
4258 load_idx = sd->newidle_idx;
4261 load_idx = sd->idle_idx;
4268 static unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu)
4270 return SCHED_POWER_SCALE;
4273 unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu)
4275 return default_scale_freq_power(sd, cpu);
4278 static unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu)
4280 unsigned long weight = sd->span_weight;
4281 unsigned long smt_gain = sd->smt_gain;
4288 unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu)
4290 return default_scale_smt_power(sd, cpu);
4293 static unsigned long scale_rt_power(int cpu)
4295 struct rq *rq = cpu_rq(cpu);
4296 u64 total, available, age_stamp, avg;
4299 * Since we're reading these variables without serialization make sure
4300 * we read them once before doing sanity checks on them.
4302 age_stamp = ACCESS_ONCE(rq->age_stamp);
4303 avg = ACCESS_ONCE(rq->rt_avg);
4305 total = sched_avg_period() + (rq->clock - age_stamp);
4307 if (unlikely(total < avg)) {
4308 /* Ensures that power won't end up being negative */
4311 available = total - avg;
4314 if (unlikely((s64)total < SCHED_POWER_SCALE))
4315 total = SCHED_POWER_SCALE;
4317 total >>= SCHED_POWER_SHIFT;
4319 return div_u64(available, total);
4322 static void update_cpu_power(struct sched_domain *sd, int cpu)
4324 unsigned long weight = sd->span_weight;
4325 unsigned long power = SCHED_POWER_SCALE;
4326 struct sched_group *sdg = sd->groups;
4328 if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
4329 if (sched_feat(ARCH_POWER))
4330 power *= arch_scale_smt_power(sd, cpu);
4332 power *= default_scale_smt_power(sd, cpu);
4334 power >>= SCHED_POWER_SHIFT;
4337 sdg->sgp->power_orig = power;
4339 if (sched_feat(ARCH_POWER))
4340 power *= arch_scale_freq_power(sd, cpu);
4342 power *= default_scale_freq_power(sd, cpu);
4344 power >>= SCHED_POWER_SHIFT;
4346 power *= scale_rt_power(cpu);
4347 power >>= SCHED_POWER_SHIFT;
4352 cpu_rq(cpu)->cpu_power = power;
4353 sdg->sgp->power = power;
4356 void update_group_power(struct sched_domain *sd, int cpu)
4358 struct sched_domain *child = sd->child;
4359 struct sched_group *group, *sdg = sd->groups;
4360 unsigned long power;
4361 unsigned long interval;
4363 interval = msecs_to_jiffies(sd->balance_interval);
4364 interval = clamp(interval, 1UL, max_load_balance_interval);
4365 sdg->sgp->next_update = jiffies + interval;
4368 update_cpu_power(sd, cpu);
4374 if (child->flags & SD_OVERLAP) {
4376 * SD_OVERLAP domains cannot assume that child groups
4377 * span the current group.
4380 for_each_cpu(cpu, sched_group_cpus(sdg))
4381 power += power_of(cpu);
4384 * !SD_OVERLAP domains can assume that child groups
4385 * span the current group.
4388 group = child->groups;
4390 power += group->sgp->power;
4391 group = group->next;
4392 } while (group != child->groups);
4395 sdg->sgp->power_orig = sdg->sgp->power = power;
4399 * Try and fix up capacity for tiny siblings, this is needed when
4400 * things like SD_ASYM_PACKING need f_b_g to select another sibling
4401 * which on its own isn't powerful enough.
4403 * See update_sd_pick_busiest() and check_asym_packing().
4406 fix_small_capacity(struct sched_domain *sd, struct sched_group *group)
4409 * Only siblings can have significantly less than SCHED_POWER_SCALE
4411 if (!(sd->flags & SD_SHARE_CPUPOWER))
4415 * If ~90% of the cpu_power is still there, we're good.
4417 if (group->sgp->power * 32 > group->sgp->power_orig * 29)
4424 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
4425 * @env: The load balancing environment.
4426 * @group: sched_group whose statistics are to be updated.
4427 * @load_idx: Load index of sched_domain of this_cpu for load calc.
4428 * @local_group: Does group contain this_cpu.
4429 * @balance: Should we balance.
4430 * @sgs: variable to hold the statistics for this group.
4432 static inline void update_sg_lb_stats(struct lb_env *env,
4433 struct sched_group *group, int load_idx,
4434 int local_group, int *balance, struct sg_lb_stats *sgs)
4436 unsigned long nr_running, max_nr_running, min_nr_running;
4437 unsigned long load, max_cpu_load, min_cpu_load;
4438 unsigned int balance_cpu = -1, first_idle_cpu = 0;
4439 unsigned long avg_load_per_task = 0;
4443 balance_cpu = group_balance_cpu(group);
4445 /* Tally up the load of all CPUs in the group */
4447 min_cpu_load = ~0UL;
4449 min_nr_running = ~0UL;
4451 for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
4452 struct rq *rq = cpu_rq(i);
4454 nr_running = rq->nr_running;
4456 /* Bias balancing toward cpus of our domain */
4458 if (idle_cpu(i) && !first_idle_cpu &&
4459 cpumask_test_cpu(i, sched_group_mask(group))) {
4464 load = target_load(i, load_idx);
4466 load = source_load(i, load_idx);
4467 if (load > max_cpu_load)
4468 max_cpu_load = load;
4469 if (min_cpu_load > load)
4470 min_cpu_load = load;
4472 if (nr_running > max_nr_running)
4473 max_nr_running = nr_running;
4474 if (min_nr_running > nr_running)
4475 min_nr_running = nr_running;
4478 sgs->group_load += load;
4479 sgs->sum_nr_running += nr_running;
4480 sgs->sum_weighted_load += weighted_cpuload(i);
4486 * First idle cpu or the first cpu(busiest) in this sched group
4487 * is eligible for doing load balancing at this and above
4488 * domains. In the newly idle case, we will allow all the cpu's
4489 * to do the newly idle load balance.
4492 if (env->idle != CPU_NEWLY_IDLE) {
4493 if (balance_cpu != env->dst_cpu) {
4497 update_group_power(env->sd, env->dst_cpu);
4498 } else if (time_after_eq(jiffies, group->sgp->next_update))
4499 update_group_power(env->sd, env->dst_cpu);
4502 /* Adjust by relative CPU power of the group */
4503 sgs->avg_load = (sgs->group_load*SCHED_POWER_SCALE) / group->sgp->power;
4506 * Consider the group unbalanced when the imbalance is larger
4507 * than the average weight of a task.
4509 * APZ: with cgroup the avg task weight can vary wildly and
4510 * might not be a suitable number - should we keep a
4511 * normalized nr_running number somewhere that negates
4514 if (sgs->sum_nr_running)
4515 avg_load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
4517 if ((max_cpu_load - min_cpu_load) >= avg_load_per_task &&
4518 (max_nr_running - min_nr_running) > 1)
4521 sgs->group_capacity = DIV_ROUND_CLOSEST(group->sgp->power,
4523 if (!sgs->group_capacity)
4524 sgs->group_capacity = fix_small_capacity(env->sd, group);
4525 sgs->group_weight = group->group_weight;
4527 if (sgs->group_capacity > sgs->sum_nr_running)
4528 sgs->group_has_capacity = 1;
4532 * update_sd_pick_busiest - return 1 on busiest group
4533 * @env: The load balancing environment.
4534 * @sds: sched_domain statistics
4535 * @sg: sched_group candidate to be checked for being the busiest
4536 * @sgs: sched_group statistics
4538 * Determine if @sg is a busier group than the previously selected
4541 static bool update_sd_pick_busiest(struct lb_env *env,
4542 struct sd_lb_stats *sds,
4543 struct sched_group *sg,
4544 struct sg_lb_stats *sgs)
4546 if (sgs->avg_load <= sds->max_load)
4549 if (sgs->sum_nr_running > sgs->group_capacity)
4556 * ASYM_PACKING needs to move all the work to the lowest
4557 * numbered CPUs in the group, therefore mark all groups
4558 * higher than ourself as busy.
4560 if ((env->sd->flags & SD_ASYM_PACKING) && sgs->sum_nr_running &&
4561 env->dst_cpu < group_first_cpu(sg)) {
4565 if (group_first_cpu(sds->busiest) > group_first_cpu(sg))
4573 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
4574 * @env: The load balancing environment.
4575 * @balance: Should we balance.
4576 * @sds: variable to hold the statistics for this sched_domain.
4578 static inline void update_sd_lb_stats(struct lb_env *env,
4579 int *balance, struct sd_lb_stats *sds)
4581 struct sched_domain *child = env->sd->child;
4582 struct sched_group *sg = env->sd->groups;
4583 struct sg_lb_stats sgs;
4584 int load_idx, prefer_sibling = 0;
4586 if (child && child->flags & SD_PREFER_SIBLING)
4589 load_idx = get_sd_load_idx(env->sd, env->idle);
4594 local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
4595 memset(&sgs, 0, sizeof(sgs));
4596 update_sg_lb_stats(env, sg, load_idx, local_group, balance, &sgs);
4598 if (local_group && !(*balance))
4601 sds->total_load += sgs.group_load;
4602 sds->total_pwr += sg->sgp->power;
4605 * In case the child domain prefers tasks go to siblings
4606 * first, lower the sg capacity to one so that we'll try
4607 * and move all the excess tasks away. We lower the capacity
4608 * of a group only if the local group has the capacity to fit
4609 * these excess tasks, i.e. nr_running < group_capacity. The
4610 * extra check prevents the case where you always pull from the
4611 * heaviest group when it is already under-utilized (possible
4612 * with a large weight task outweighs the tasks on the system).
4614 if (prefer_sibling && !local_group && sds->this_has_capacity)
4615 sgs.group_capacity = min(sgs.group_capacity, 1UL);
4618 sds->this_load = sgs.avg_load;
4620 sds->this_nr_running = sgs.sum_nr_running;
4621 sds->this_load_per_task = sgs.sum_weighted_load;
4622 sds->this_has_capacity = sgs.group_has_capacity;
4623 sds->this_idle_cpus = sgs.idle_cpus;
4624 } else if (update_sd_pick_busiest(env, sds, sg, &sgs)) {
4625 sds->max_load = sgs.avg_load;
4627 sds->busiest_nr_running = sgs.sum_nr_running;
4628 sds->busiest_idle_cpus = sgs.idle_cpus;
4629 sds->busiest_group_capacity = sgs.group_capacity;
4630 sds->busiest_load_per_task = sgs.sum_weighted_load;
4631 sds->busiest_has_capacity = sgs.group_has_capacity;
4632 sds->busiest_group_weight = sgs.group_weight;
4633 sds->group_imb = sgs.group_imb;
4637 } while (sg != env->sd->groups);
4641 * check_asym_packing - Check to see if the group is packed into the
4644 * This is primarily intended to used at the sibling level. Some
4645 * cores like POWER7 prefer to use lower numbered SMT threads. In the
4646 * case of POWER7, it can move to lower SMT modes only when higher
4647 * threads are idle. When in lower SMT modes, the threads will
4648 * perform better since they share less core resources. Hence when we
4649 * have idle threads, we want them to be the higher ones.
4651 * This packing function is run on idle threads. It checks to see if
4652 * the busiest CPU in this domain (core in the P7 case) has a higher
4653 * CPU number than the packing function is being run on. Here we are
4654 * assuming lower CPU number will be equivalent to lower a SMT thread
4657 * Returns 1 when packing is required and a task should be moved to
4658 * this CPU. The amount of the imbalance is returned in *imbalance.
4660 * @env: The load balancing environment.
4661 * @sds: Statistics of the sched_domain which is to be packed
4663 static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
4667 if (!(env->sd->flags & SD_ASYM_PACKING))
4673 busiest_cpu = group_first_cpu(sds->busiest);
4674 if (env->dst_cpu > busiest_cpu)
4677 env->imbalance = DIV_ROUND_CLOSEST(
4678 sds->max_load * sds->busiest->sgp->power, SCHED_POWER_SCALE);
4684 * fix_small_imbalance - Calculate the minor imbalance that exists
4685 * amongst the groups of a sched_domain, during
4687 * @env: The load balancing environment.
4688 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
4691 void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
4693 unsigned long tmp, pwr_now = 0, pwr_move = 0;
4694 unsigned int imbn = 2;
4695 unsigned long scaled_busy_load_per_task;
4697 if (sds->this_nr_running) {
4698 sds->this_load_per_task /= sds->this_nr_running;
4699 if (sds->busiest_load_per_task >
4700 sds->this_load_per_task)
4703 sds->this_load_per_task =
4704 cpu_avg_load_per_task(env->dst_cpu);
4707 scaled_busy_load_per_task = sds->busiest_load_per_task
4708 * SCHED_POWER_SCALE;
4709 scaled_busy_load_per_task /= sds->busiest->sgp->power;
4711 if (sds->max_load - sds->this_load + scaled_busy_load_per_task >=
4712 (scaled_busy_load_per_task * imbn)) {
4713 env->imbalance = sds->busiest_load_per_task;
4718 * OK, we don't have enough imbalance to justify moving tasks,
4719 * however we may be able to increase total CPU power used by
4723 pwr_now += sds->busiest->sgp->power *
4724 min(sds->busiest_load_per_task, sds->max_load);
4725 pwr_now += sds->this->sgp->power *
4726 min(sds->this_load_per_task, sds->this_load);
4727 pwr_now /= SCHED_POWER_SCALE;
4729 /* Amount of load we'd subtract */
4730 tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE) /
4731 sds->busiest->sgp->power;
4732 if (sds->max_load > tmp)
4733 pwr_move += sds->busiest->sgp->power *
4734 min(sds->busiest_load_per_task, sds->max_load - tmp);
4736 /* Amount of load we'd add */
4737 if (sds->max_load * sds->busiest->sgp->power <
4738 sds->busiest_load_per_task * SCHED_POWER_SCALE)
4739 tmp = (sds->max_load * sds->busiest->sgp->power) /
4740 sds->this->sgp->power;
4742 tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE) /
4743 sds->this->sgp->power;
4744 pwr_move += sds->this->sgp->power *
4745 min(sds->this_load_per_task, sds->this_load + tmp);
4746 pwr_move /= SCHED_POWER_SCALE;
4748 /* Move if we gain throughput */
4749 if (pwr_move > pwr_now)
4750 env->imbalance = sds->busiest_load_per_task;
4754 * calculate_imbalance - Calculate the amount of imbalance present within the
4755 * groups of a given sched_domain during load balance.
4756 * @env: load balance environment
4757 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
4759 static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
4761 unsigned long max_pull, load_above_capacity = ~0UL;
4763 sds->busiest_load_per_task /= sds->busiest_nr_running;
4764 if (sds->group_imb) {
4765 sds->busiest_load_per_task =
4766 min(sds->busiest_load_per_task, sds->avg_load);
4770 * In the presence of smp nice balancing, certain scenarios can have
4771 * max load less than avg load(as we skip the groups at or below
4772 * its cpu_power, while calculating max_load..)
4774 if (sds->max_load < sds->avg_load) {
4776 return fix_small_imbalance(env, sds);
4779 if (!sds->group_imb) {
4781 * Don't want to pull so many tasks that a group would go idle.
4783 load_above_capacity = (sds->busiest_nr_running -
4784 sds->busiest_group_capacity);
4786 load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_POWER_SCALE);
4788 load_above_capacity /= sds->busiest->sgp->power;
4792 * We're trying to get all the cpus to the average_load, so we don't
4793 * want to push ourselves above the average load, nor do we wish to
4794 * reduce the max loaded cpu below the average load. At the same time,
4795 * we also don't want to reduce the group load below the group capacity
4796 * (so that we can implement power-savings policies etc). Thus we look
4797 * for the minimum possible imbalance.
4798 * Be careful of negative numbers as they'll appear as very large values
4799 * with unsigned longs.
4801 max_pull = min(sds->max_load - sds->avg_load, load_above_capacity);
4803 /* How much load to actually move to equalise the imbalance */
4804 env->imbalance = min(max_pull * sds->busiest->sgp->power,
4805 (sds->avg_load - sds->this_load) * sds->this->sgp->power)
4806 / SCHED_POWER_SCALE;
4809 * if *imbalance is less than the average load per runnable task
4810 * there is no guarantee that any tasks will be moved so we'll have
4811 * a think about bumping its value to force at least one task to be
4814 if (env->imbalance < sds->busiest_load_per_task)
4815 return fix_small_imbalance(env, sds);
4819 /******* find_busiest_group() helpers end here *********************/
4822 * find_busiest_group - Returns the busiest group within the sched_domain
4823 * if there is an imbalance. If there isn't an imbalance, and
4824 * the user has opted for power-savings, it returns a group whose
4825 * CPUs can be put to idle by rebalancing those tasks elsewhere, if
4826 * such a group exists.
4828 * Also calculates the amount of weighted load which should be moved
4829 * to restore balance.
4831 * @env: The load balancing environment.
4832 * @balance: Pointer to a variable indicating if this_cpu
4833 * is the appropriate cpu to perform load balancing at this_level.
4835 * Returns: - the busiest group if imbalance exists.
4836 * - If no imbalance and user has opted for power-savings balance,
4837 * return the least loaded group whose CPUs can be
4838 * put to idle by rebalancing its tasks onto our group.
4840 static struct sched_group *
4841 find_busiest_group(struct lb_env *env, int *balance)
4843 struct sd_lb_stats sds;
4845 memset(&sds, 0, sizeof(sds));
4848 * Compute the various statistics relavent for load balancing at
4851 update_sd_lb_stats(env, balance, &sds);
4854 * this_cpu is not the appropriate cpu to perform load balancing at
4860 if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
4861 check_asym_packing(env, &sds))
4864 /* There is no busy sibling group to pull tasks from */
4865 if (!sds.busiest || sds.busiest_nr_running == 0)
4868 sds.avg_load = (SCHED_POWER_SCALE * sds.total_load) / sds.total_pwr;
4871 * If the busiest group is imbalanced the below checks don't
4872 * work because they assumes all things are equal, which typically
4873 * isn't true due to cpus_allowed constraints and the like.
4878 /* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
4879 if (env->idle == CPU_NEWLY_IDLE && sds.this_has_capacity &&
4880 !sds.busiest_has_capacity)
4884 * If the local group is more busy than the selected busiest group
4885 * don't try and pull any tasks.
4887 if (sds.this_load >= sds.max_load)
4891 * Don't pull any tasks if this group is already above the domain
4894 if (sds.this_load >= sds.avg_load)
4897 if (env->idle == CPU_IDLE) {
4899 * This cpu is idle. If the busiest group load doesn't
4900 * have more tasks than the number of available cpu's and
4901 * there is no imbalance between this and busiest group
4902 * wrt to idle cpu's, it is balanced.
4904 if ((sds.this_idle_cpus <= sds.busiest_idle_cpus + 1) &&
4905 sds.busiest_nr_running <= sds.busiest_group_weight)
4909 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
4910 * imbalance_pct to be conservative.
4912 if (100 * sds.max_load <= env->sd->imbalance_pct * sds.this_load)
4917 /* Looks like there is an imbalance. Compute it */
4918 calculate_imbalance(env, &sds);
4928 * find_busiest_queue - find the busiest runqueue among the cpus in group.
4930 static struct rq *find_busiest_queue(struct lb_env *env,
4931 struct sched_group *group)
4933 struct rq *busiest = NULL, *rq;
4934 unsigned long max_load = 0;
4937 for_each_cpu(i, sched_group_cpus(group)) {
4938 unsigned long power = power_of(i);
4939 unsigned long capacity = DIV_ROUND_CLOSEST(power,
4944 capacity = fix_small_capacity(env->sd, group);
4946 if (!cpumask_test_cpu(i, env->cpus))
4950 wl = weighted_cpuload(i);
4953 * When comparing with imbalance, use weighted_cpuload()
4954 * which is not scaled with the cpu power.
4956 if (capacity && rq->nr_running == 1 && wl > env->imbalance)
4960 * For the load comparisons with the other cpu's, consider
4961 * the weighted_cpuload() scaled with the cpu power, so that
4962 * the load can be moved away from the cpu that is potentially
4963 * running at a lower capacity.
4965 wl = (wl * SCHED_POWER_SCALE) / power;
4967 if (wl > max_load) {
4977 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
4978 * so long as it is large enough.
4980 #define MAX_PINNED_INTERVAL 512
4982 /* Working cpumask for load_balance and load_balance_newidle. */
4983 DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
4985 static int need_active_balance(struct lb_env *env)
4987 struct sched_domain *sd = env->sd;
4989 if (env->idle == CPU_NEWLY_IDLE) {
4992 * ASYM_PACKING needs to force migrate tasks from busy but
4993 * higher numbered CPUs in order to pack all tasks in the
4994 * lowest numbered CPUs.
4996 if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
5000 return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
5003 static int active_load_balance_cpu_stop(void *data);
5006 * Check this_cpu to ensure it is balanced within domain. Attempt to move
5007 * tasks if there is an imbalance.
5009 static int load_balance(int this_cpu, struct rq *this_rq,
5010 struct sched_domain *sd, enum cpu_idle_type idle,
5013 int ld_moved, cur_ld_moved, active_balance = 0;
5014 struct sched_group *group;
5016 unsigned long flags;
5017 struct cpumask *cpus = __get_cpu_var(load_balance_mask);
5019 struct lb_env env = {
5021 .dst_cpu = this_cpu,
5023 .dst_grpmask = sched_group_cpus(sd->groups),
5025 .loop_break = sched_nr_migrate_break,
5030 * For NEWLY_IDLE load_balancing, we don't need to consider
5031 * other cpus in our group
5033 if (idle == CPU_NEWLY_IDLE)
5034 env.dst_grpmask = NULL;
5036 cpumask_copy(cpus, cpu_active_mask);
5038 schedstat_inc(sd, lb_count[idle]);
5041 group = find_busiest_group(&env, balance);
5047 schedstat_inc(sd, lb_nobusyg[idle]);
5051 busiest = find_busiest_queue(&env, group);
5053 schedstat_inc(sd, lb_nobusyq[idle]);
5057 BUG_ON(busiest == env.dst_rq);
5059 schedstat_add(sd, lb_imbalance[idle], env.imbalance);
5062 if (busiest->nr_running > 1) {
5064 * Attempt to move tasks. If find_busiest_group has found
5065 * an imbalance but busiest->nr_running <= 1, the group is
5066 * still unbalanced. ld_moved simply stays zero, so it is
5067 * correctly treated as an imbalance.
5069 env.flags |= LBF_ALL_PINNED;
5070 env.src_cpu = busiest->cpu;
5071 env.src_rq = busiest;
5072 env.loop_max = min(sysctl_sched_nr_migrate, busiest->nr_running);
5074 update_h_load(env.src_cpu);
5076 local_irq_save(flags);
5077 double_rq_lock(env.dst_rq, busiest);
5080 * cur_ld_moved - load moved in current iteration
5081 * ld_moved - cumulative load moved across iterations
5083 cur_ld_moved = move_tasks(&env);
5084 ld_moved += cur_ld_moved;
5085 double_rq_unlock(env.dst_rq, busiest);
5086 local_irq_restore(flags);
5089 * some other cpu did the load balance for us.
5091 if (cur_ld_moved && env.dst_cpu != smp_processor_id())
5092 resched_cpu(env.dst_cpu);
5094 if (env.flags & LBF_NEED_BREAK) {
5095 env.flags &= ~LBF_NEED_BREAK;
5100 * Revisit (affine) tasks on src_cpu that couldn't be moved to
5101 * us and move them to an alternate dst_cpu in our sched_group
5102 * where they can run. The upper limit on how many times we
5103 * iterate on same src_cpu is dependent on number of cpus in our
5106 * This changes load balance semantics a bit on who can move
5107 * load to a given_cpu. In addition to the given_cpu itself
5108 * (or a ilb_cpu acting on its behalf where given_cpu is
5109 * nohz-idle), we now have balance_cpu in a position to move
5110 * load to given_cpu. In rare situations, this may cause
5111 * conflicts (balance_cpu and given_cpu/ilb_cpu deciding
5112 * _independently_ and at _same_ time to move some load to
5113 * given_cpu) causing exceess load to be moved to given_cpu.
5114 * This however should not happen so much in practice and
5115 * moreover subsequent load balance cycles should correct the
5116 * excess load moved.
5118 if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0) {
5120 env.dst_rq = cpu_rq(env.new_dst_cpu);
5121 env.dst_cpu = env.new_dst_cpu;
5122 env.flags &= ~LBF_SOME_PINNED;
5124 env.loop_break = sched_nr_migrate_break;
5126 /* Prevent to re-select dst_cpu via env's cpus */
5127 cpumask_clear_cpu(env.dst_cpu, env.cpus);
5130 * Go back to "more_balance" rather than "redo" since we
5131 * need to continue with same src_cpu.
5136 /* All tasks on this runqueue were pinned by CPU affinity */
5137 if (unlikely(env.flags & LBF_ALL_PINNED)) {
5138 cpumask_clear_cpu(cpu_of(busiest), cpus);
5139 if (!cpumask_empty(cpus)) {
5141 env.loop_break = sched_nr_migrate_break;
5149 schedstat_inc(sd, lb_failed[idle]);
5151 * Increment the failure counter only on periodic balance.
5152 * We do not want newidle balance, which can be very
5153 * frequent, pollute the failure counter causing
5154 * excessive cache_hot migrations and active balances.
5156 if (idle != CPU_NEWLY_IDLE)
5157 sd->nr_balance_failed++;
5159 if (need_active_balance(&env)) {
5160 raw_spin_lock_irqsave(&busiest->lock, flags);
5162 /* don't kick the active_load_balance_cpu_stop,
5163 * if the curr task on busiest cpu can't be
5166 if (!cpumask_test_cpu(this_cpu,
5167 tsk_cpus_allowed(busiest->curr))) {
5168 raw_spin_unlock_irqrestore(&busiest->lock,
5170 env.flags |= LBF_ALL_PINNED;
5171 goto out_one_pinned;
5175 * ->active_balance synchronizes accesses to
5176 * ->active_balance_work. Once set, it's cleared
5177 * only after active load balance is finished.
5179 if (!busiest->active_balance) {
5180 busiest->active_balance = 1;
5181 busiest->push_cpu = this_cpu;
5184 raw_spin_unlock_irqrestore(&busiest->lock, flags);
5186 if (active_balance) {
5187 stop_one_cpu_nowait(cpu_of(busiest),
5188 active_load_balance_cpu_stop, busiest,
5189 &busiest->active_balance_work);
5193 * We've kicked active balancing, reset the failure
5196 sd->nr_balance_failed = sd->cache_nice_tries+1;
5199 sd->nr_balance_failed = 0;
5201 if (likely(!active_balance)) {
5202 /* We were unbalanced, so reset the balancing interval */
5203 sd->balance_interval = sd->min_interval;
5206 * If we've begun active balancing, start to back off. This
5207 * case may not be covered by the all_pinned logic if there
5208 * is only 1 task on the busy runqueue (because we don't call
5211 if (sd->balance_interval < sd->max_interval)
5212 sd->balance_interval *= 2;
5218 schedstat_inc(sd, lb_balanced[idle]);
5220 sd->nr_balance_failed = 0;
5223 /* tune up the balancing interval */
5224 if (((env.flags & LBF_ALL_PINNED) &&
5225 sd->balance_interval < MAX_PINNED_INTERVAL) ||
5226 (sd->balance_interval < sd->max_interval))
5227 sd->balance_interval *= 2;
5235 * idle_balance is called by schedule() if this_cpu is about to become
5236 * idle. Attempts to pull tasks from other CPUs.
5238 void idle_balance(int this_cpu, struct rq *this_rq)
5240 struct sched_domain *sd;
5241 int pulled_task = 0;
5242 unsigned long next_balance = jiffies + HZ;
5244 this_rq->idle_stamp = this_rq->clock;
5246 if (this_rq->avg_idle < sysctl_sched_migration_cost)
5250 * Drop the rq->lock, but keep IRQ/preempt disabled.
5252 raw_spin_unlock(&this_rq->lock);
5254 update_blocked_averages(this_cpu);
5256 for_each_domain(this_cpu, sd) {
5257 unsigned long interval;
5260 if (!(sd->flags & SD_LOAD_BALANCE))
5263 if (sd->flags & SD_BALANCE_NEWIDLE) {
5264 /* If we've pulled tasks over stop searching: */
5265 pulled_task = load_balance(this_cpu, this_rq,
5266 sd, CPU_NEWLY_IDLE, &balance);
5269 interval = msecs_to_jiffies(sd->balance_interval);
5270 if (time_after(next_balance, sd->last_balance + interval))
5271 next_balance = sd->last_balance + interval;
5273 this_rq->idle_stamp = 0;
5279 raw_spin_lock(&this_rq->lock);
5281 if (pulled_task || time_after(jiffies, this_rq->next_balance)) {
5283 * We are going idle. next_balance may be set based on
5284 * a busy processor. So reset next_balance.
5286 this_rq->next_balance = next_balance;
5291 * active_load_balance_cpu_stop is run by cpu stopper. It pushes
5292 * running tasks off the busiest CPU onto idle CPUs. It requires at
5293 * least 1 task to be running on each physical CPU where possible, and
5294 * avoids physical / logical imbalances.
5296 static int active_load_balance_cpu_stop(void *data)
5298 struct rq *busiest_rq = data;
5299 int busiest_cpu = cpu_of(busiest_rq);
5300 int target_cpu = busiest_rq->push_cpu;
5301 struct rq *target_rq = cpu_rq(target_cpu);
5302 struct sched_domain *sd;
5304 raw_spin_lock_irq(&busiest_rq->lock);
5306 /* make sure the requested cpu hasn't gone down in the meantime */
5307 if (unlikely(busiest_cpu != smp_processor_id() ||
5308 !busiest_rq->active_balance))
5311 /* Is there any task to move? */
5312 if (busiest_rq->nr_running <= 1)
5316 * This condition is "impossible", if it occurs
5317 * we need to fix it. Originally reported by
5318 * Bjorn Helgaas on a 128-cpu setup.
5320 BUG_ON(busiest_rq == target_rq);
5322 /* move a task from busiest_rq to target_rq */
5323 double_lock_balance(busiest_rq, target_rq);
5325 /* Search for an sd spanning us and the target CPU. */
5327 for_each_domain(target_cpu, sd) {
5328 if ((sd->flags & SD_LOAD_BALANCE) &&
5329 cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
5334 struct lb_env env = {
5336 .dst_cpu = target_cpu,
5337 .dst_rq = target_rq,
5338 .src_cpu = busiest_rq->cpu,
5339 .src_rq = busiest_rq,
5343 schedstat_inc(sd, alb_count);
5345 if (move_one_task(&env))
5346 schedstat_inc(sd, alb_pushed);
5348 schedstat_inc(sd, alb_failed);
5351 double_unlock_balance(busiest_rq, target_rq);
5353 busiest_rq->active_balance = 0;
5354 raw_spin_unlock_irq(&busiest_rq->lock);
5360 * idle load balancing details
5361 * - When one of the busy CPUs notice that there may be an idle rebalancing
5362 * needed, they will kick the idle load balancer, which then does idle
5363 * load balancing for all the idle CPUs.
5366 cpumask_var_t idle_cpus_mask;
5368 unsigned long next_balance; /* in jiffy units */
5369 } nohz ____cacheline_aligned;
5371 static inline int find_new_ilb(int call_cpu)
5373 int ilb = cpumask_first(nohz.idle_cpus_mask);
5375 if (ilb < nr_cpu_ids && idle_cpu(ilb))
5382 * Kick a CPU to do the nohz balancing, if it is time for it. We pick the
5383 * nohz_load_balancer CPU (if there is one) otherwise fallback to any idle
5384 * CPU (if there is one).
5386 static void nohz_balancer_kick(int cpu)
5390 nohz.next_balance++;
5392 ilb_cpu = find_new_ilb(cpu);
5394 if (ilb_cpu >= nr_cpu_ids)
5397 if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
5400 * Use smp_send_reschedule() instead of resched_cpu().
5401 * This way we generate a sched IPI on the target cpu which
5402 * is idle. And the softirq performing nohz idle load balance
5403 * will be run before returning from the IPI.
5405 smp_send_reschedule(ilb_cpu);
5409 static inline void nohz_balance_exit_idle(int cpu)
5411 if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
5412 cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
5413 atomic_dec(&nohz.nr_cpus);
5414 clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
5418 static inline void set_cpu_sd_state_busy(void)
5420 struct sched_domain *sd;
5421 int cpu = smp_processor_id();
5424 sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd);
5426 if (!sd || !sd->nohz_idle)
5430 for (; sd; sd = sd->parent)
5431 atomic_inc(&sd->groups->sgp->nr_busy_cpus);
5436 void set_cpu_sd_state_idle(void)
5438 struct sched_domain *sd;
5439 int cpu = smp_processor_id();
5442 sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd);
5444 if (!sd || sd->nohz_idle)
5448 for (; sd; sd = sd->parent)
5449 atomic_dec(&sd->groups->sgp->nr_busy_cpus);
5455 * This routine will record that the cpu is going idle with tick stopped.
5456 * This info will be used in performing idle load balancing in the future.
5458 void nohz_balance_enter_idle(int cpu)
5461 * If this cpu is going down, then nothing needs to be done.
5463 if (!cpu_active(cpu))
5466 if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
5469 cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
5470 atomic_inc(&nohz.nr_cpus);
5471 set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
5474 static int __cpuinit sched_ilb_notifier(struct notifier_block *nfb,
5475 unsigned long action, void *hcpu)
5477 switch (action & ~CPU_TASKS_FROZEN) {
5479 nohz_balance_exit_idle(smp_processor_id());
5487 static DEFINE_SPINLOCK(balancing);
5490 * Scale the max load_balance interval with the number of CPUs in the system.
5491 * This trades load-balance latency on larger machines for less cross talk.
5493 void update_max_interval(void)
5495 max_load_balance_interval = HZ*num_online_cpus()/10;
5499 * It checks each scheduling domain to see if it is due to be balanced,
5500 * and initiates a balancing operation if so.
5502 * Balancing parameters are set up in init_sched_domains.
5504 static void rebalance_domains(int cpu, enum cpu_idle_type idle)
5507 struct rq *rq = cpu_rq(cpu);
5508 unsigned long interval;
5509 struct sched_domain *sd;
5510 /* Earliest time when we have to do rebalance again */
5511 unsigned long next_balance = jiffies + 60*HZ;
5512 int update_next_balance = 0;
5515 update_blocked_averages(cpu);
5518 for_each_domain(cpu, sd) {
5519 if (!(sd->flags & SD_LOAD_BALANCE))
5522 interval = sd->balance_interval;
5523 if (idle != CPU_IDLE)
5524 interval *= sd->busy_factor;
5526 /* scale ms to jiffies */
5527 interval = msecs_to_jiffies(interval);
5528 interval = clamp(interval, 1UL, max_load_balance_interval);
5530 need_serialize = sd->flags & SD_SERIALIZE;
5532 if (need_serialize) {
5533 if (!spin_trylock(&balancing))
5537 if (time_after_eq(jiffies, sd->last_balance + interval)) {
5538 if (load_balance(cpu, rq, sd, idle, &balance)) {
5540 * The LBF_SOME_PINNED logic could have changed
5541 * env->dst_cpu, so we can't know our idle
5542 * state even if we migrated tasks. Update it.
5544 idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
5546 sd->last_balance = jiffies;
5549 spin_unlock(&balancing);
5551 if (time_after(next_balance, sd->last_balance + interval)) {
5552 next_balance = sd->last_balance + interval;
5553 update_next_balance = 1;
5557 * Stop the load balance at this level. There is another
5558 * CPU in our sched group which is doing load balancing more
5567 * next_balance will be updated only when there is a need.
5568 * When the cpu is attached to null domain for ex, it will not be
5571 if (likely(update_next_balance))
5572 rq->next_balance = next_balance;
5577 * In CONFIG_NO_HZ case, the idle balance kickee will do the
5578 * rebalancing for all the cpus for whom scheduler ticks are stopped.
5580 static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle)
5582 struct rq *this_rq = cpu_rq(this_cpu);
5586 if (idle != CPU_IDLE ||
5587 !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
5590 for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
5591 if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
5595 * If this cpu gets work to do, stop the load balancing
5596 * work being done for other cpus. Next load
5597 * balancing owner will pick it up.
5602 rq = cpu_rq(balance_cpu);
5604 raw_spin_lock_irq(&rq->lock);
5605 update_rq_clock(rq);
5606 update_idle_cpu_load(rq);
5607 raw_spin_unlock_irq(&rq->lock);
5609 rebalance_domains(balance_cpu, CPU_IDLE);
5611 if (time_after(this_rq->next_balance, rq->next_balance))
5612 this_rq->next_balance = rq->next_balance;
5614 nohz.next_balance = this_rq->next_balance;
5616 clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
5620 * Current heuristic for kicking the idle load balancer in the presence
5621 * of an idle cpu is the system.
5622 * - This rq has more than one task.
5623 * - At any scheduler domain level, this cpu's scheduler group has multiple
5624 * busy cpu's exceeding the group's power.
5625 * - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
5626 * domain span are idle.
5628 static inline int nohz_kick_needed(struct rq *rq, int cpu)
5630 unsigned long now = jiffies;
5631 struct sched_domain *sd;
5633 if (unlikely(idle_cpu(cpu)))
5637 * We may be recently in ticked or tickless idle mode. At the first
5638 * busy tick after returning from idle, we will update the busy stats.
5640 set_cpu_sd_state_busy();
5641 nohz_balance_exit_idle(cpu);
5644 * None are in tickless mode and hence no need for NOHZ idle load
5647 if (likely(!atomic_read(&nohz.nr_cpus)))
5650 if (time_before(now, nohz.next_balance))
5653 if (rq->nr_running >= 2)
5657 for_each_domain(cpu, sd) {
5658 struct sched_group *sg = sd->groups;
5659 struct sched_group_power *sgp = sg->sgp;
5660 int nr_busy = atomic_read(&sgp->nr_busy_cpus);
5662 if (sd->flags & SD_SHARE_PKG_RESOURCES && nr_busy > 1)
5663 goto need_kick_unlock;
5665 if (sd->flags & SD_ASYM_PACKING && nr_busy != sg->group_weight
5666 && (cpumask_first_and(nohz.idle_cpus_mask,
5667 sched_domain_span(sd)) < cpu))
5668 goto need_kick_unlock;
5670 if (!(sd->flags & (SD_SHARE_PKG_RESOURCES | SD_ASYM_PACKING)))
5682 static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle) { }
5686 * run_rebalance_domains is triggered when needed from the scheduler tick.
5687 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
5689 static void run_rebalance_domains(struct softirq_action *h)
5691 int this_cpu = smp_processor_id();
5692 struct rq *this_rq = cpu_rq(this_cpu);
5693 enum cpu_idle_type idle = this_rq->idle_balance ?
5694 CPU_IDLE : CPU_NOT_IDLE;
5696 rebalance_domains(this_cpu, idle);
5699 * If this cpu has a pending nohz_balance_kick, then do the
5700 * balancing on behalf of the other idle cpus whose ticks are
5703 nohz_idle_balance(this_cpu, idle);
5706 static inline int on_null_domain(int cpu)
5708 return !rcu_dereference_sched(cpu_rq(cpu)->sd);
5712 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
5714 void trigger_load_balance(struct rq *rq, int cpu)
5716 /* Don't need to rebalance while attached to NULL domain */
5717 if (time_after_eq(jiffies, rq->next_balance) &&
5718 likely(!on_null_domain(cpu)))
5719 raise_softirq(SCHED_SOFTIRQ);
5721 if (nohz_kick_needed(rq, cpu) && likely(!on_null_domain(cpu)))
5722 nohz_balancer_kick(cpu);
5726 static void rq_online_fair(struct rq *rq)
5731 static void rq_offline_fair(struct rq *rq)
5735 /* Ensure any throttled groups are reachable by pick_next_task */
5736 unthrottle_offline_cfs_rqs(rq);
5739 #endif /* CONFIG_SMP */
5742 * scheduler tick hitting a task of our scheduling class:
5744 static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
5746 struct cfs_rq *cfs_rq;
5747 struct sched_entity *se = &curr->se;
5749 for_each_sched_entity(se) {
5750 cfs_rq = cfs_rq_of(se);
5751 entity_tick(cfs_rq, se, queued);
5754 if (sched_feat_numa(NUMA))
5755 task_tick_numa(rq, curr);
5757 update_rq_runnable_avg(rq, 1);
5761 * called on fork with the child task as argument from the parent's context
5762 * - child not yet on the tasklist
5763 * - preemption disabled
5765 static void task_fork_fair(struct task_struct *p)
5767 struct cfs_rq *cfs_rq;
5768 struct sched_entity *se = &p->se, *curr;
5769 int this_cpu = smp_processor_id();
5770 struct rq *rq = this_rq();
5771 unsigned long flags;
5773 raw_spin_lock_irqsave(&rq->lock, flags);
5775 update_rq_clock(rq);
5777 cfs_rq = task_cfs_rq(current);
5778 curr = cfs_rq->curr;
5780 if (unlikely(task_cpu(p) != this_cpu)) {
5782 __set_task_cpu(p, this_cpu);
5786 update_curr(cfs_rq);
5789 se->vruntime = curr->vruntime;
5790 place_entity(cfs_rq, se, 1);
5792 if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
5794 * Upon rescheduling, sched_class::put_prev_task() will place
5795 * 'current' within the tree based on its new key value.
5797 swap(curr->vruntime, se->vruntime);
5798 resched_task(rq->curr);
5801 se->vruntime -= cfs_rq->min_vruntime;
5803 raw_spin_unlock_irqrestore(&rq->lock, flags);
5807 * Priority of the task has changed. Check to see if we preempt
5811 prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
5817 * Reschedule if we are currently running on this runqueue and
5818 * our priority decreased, or if we are not currently running on
5819 * this runqueue and our priority is higher than the current's
5821 if (rq->curr == p) {
5822 if (p->prio > oldprio)
5823 resched_task(rq->curr);
5825 check_preempt_curr(rq, p, 0);
5828 static void switched_from_fair(struct rq *rq, struct task_struct *p)
5830 struct sched_entity *se = &p->se;
5831 struct cfs_rq *cfs_rq = cfs_rq_of(se);
5834 * Ensure the task's vruntime is normalized, so that when its
5835 * switched back to the fair class the enqueue_entity(.flags=0) will
5836 * do the right thing.
5838 * If it was on_rq, then the dequeue_entity(.flags=0) will already
5839 * have normalized the vruntime, if it was !on_rq, then only when
5840 * the task is sleeping will it still have non-normalized vruntime.
5842 if (!se->on_rq && p->state != TASK_RUNNING) {
5844 * Fix up our vruntime so that the current sleep doesn't
5845 * cause 'unlimited' sleep bonus.
5847 place_entity(cfs_rq, se, 0);
5848 se->vruntime -= cfs_rq->min_vruntime;
5851 #if defined(CONFIG_FAIR_GROUP_SCHED) && defined(CONFIG_SMP)
5853 * Remove our load from contribution when we leave sched_fair
5854 * and ensure we don't carry in an old decay_count if we
5857 if (p->se.avg.decay_count) {
5858 struct cfs_rq *cfs_rq = cfs_rq_of(&p->se);
5859 __synchronize_entity_decay(&p->se);
5860 subtract_blocked_load_contrib(cfs_rq,
5861 p->se.avg.load_avg_contrib);
5867 * We switched to the sched_fair class.
5869 static void switched_to_fair(struct rq *rq, struct task_struct *p)
5875 * We were most likely switched from sched_rt, so
5876 * kick off the schedule if running, otherwise just see
5877 * if we can still preempt the current task.
5880 resched_task(rq->curr);
5882 check_preempt_curr(rq, p, 0);
5885 /* Account for a task changing its policy or group.
5887 * This routine is mostly called to set cfs_rq->curr field when a task
5888 * migrates between groups/classes.
5890 static void set_curr_task_fair(struct rq *rq)
5892 struct sched_entity *se = &rq->curr->se;
5894 for_each_sched_entity(se) {
5895 struct cfs_rq *cfs_rq = cfs_rq_of(se);
5897 set_next_entity(cfs_rq, se);
5898 /* ensure bandwidth has been allocated on our new cfs_rq */
5899 account_cfs_rq_runtime(cfs_rq, 0);
5903 void init_cfs_rq(struct cfs_rq *cfs_rq)
5905 cfs_rq->tasks_timeline = RB_ROOT;
5906 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
5907 #ifndef CONFIG_64BIT
5908 cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
5910 #if defined(CONFIG_FAIR_GROUP_SCHED) && defined(CONFIG_SMP)
5911 atomic64_set(&cfs_rq->decay_counter, 1);
5912 atomic64_set(&cfs_rq->removed_load, 0);
5916 #ifdef CONFIG_FAIR_GROUP_SCHED
5917 static void task_move_group_fair(struct task_struct *p, int on_rq)
5919 struct cfs_rq *cfs_rq;
5921 * If the task was not on the rq at the time of this cgroup movement
5922 * it must have been asleep, sleeping tasks keep their ->vruntime
5923 * absolute on their old rq until wakeup (needed for the fair sleeper
5924 * bonus in place_entity()).
5926 * If it was on the rq, we've just 'preempted' it, which does convert
5927 * ->vruntime to a relative base.
5929 * Make sure both cases convert their relative position when migrating
5930 * to another cgroup's rq. This does somewhat interfere with the
5931 * fair sleeper stuff for the first placement, but who cares.
5934 * When !on_rq, vruntime of the task has usually NOT been normalized.
5935 * But there are some cases where it has already been normalized:
5937 * - Moving a forked child which is waiting for being woken up by
5938 * wake_up_new_task().
5939 * - Moving a task which has been woken up by try_to_wake_up() and
5940 * waiting for actually being woken up by sched_ttwu_pending().
5942 * To prevent boost or penalty in the new cfs_rq caused by delta
5943 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
5945 if (!on_rq && (!p->se.sum_exec_runtime || p->state == TASK_WAKING))
5949 p->se.vruntime -= cfs_rq_of(&p->se)->min_vruntime;
5950 set_task_rq(p, task_cpu(p));
5952 cfs_rq = cfs_rq_of(&p->se);
5953 p->se.vruntime += cfs_rq->min_vruntime;
5956 * migrate_task_rq_fair() will have removed our previous
5957 * contribution, but we must synchronize for ongoing future
5960 p->se.avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
5961 cfs_rq->blocked_load_avg += p->se.avg.load_avg_contrib;
5966 void free_fair_sched_group(struct task_group *tg)
5970 destroy_cfs_bandwidth(tg_cfs_bandwidth(tg));
5972 for_each_possible_cpu(i) {
5974 kfree(tg->cfs_rq[i]);
5983 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
5985 struct cfs_rq *cfs_rq;
5986 struct sched_entity *se;
5989 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
5992 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
5996 tg->shares = NICE_0_LOAD;
5998 init_cfs_bandwidth(tg_cfs_bandwidth(tg));
6000 for_each_possible_cpu(i) {
6001 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
6002 GFP_KERNEL, cpu_to_node(i));
6006 se = kzalloc_node(sizeof(struct sched_entity),
6007 GFP_KERNEL, cpu_to_node(i));
6011 init_cfs_rq(cfs_rq);
6012 init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]);
6023 void unregister_fair_sched_group(struct task_group *tg, int cpu)
6025 struct rq *rq = cpu_rq(cpu);
6026 unsigned long flags;
6029 * Only empty task groups can be destroyed; so we can speculatively
6030 * check on_list without danger of it being re-added.
6032 if (!tg->cfs_rq[cpu]->on_list)
6035 raw_spin_lock_irqsave(&rq->lock, flags);
6036 list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
6037 raw_spin_unlock_irqrestore(&rq->lock, flags);
6040 void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
6041 struct sched_entity *se, int cpu,
6042 struct sched_entity *parent)
6044 struct rq *rq = cpu_rq(cpu);
6048 init_cfs_rq_runtime(cfs_rq);
6050 tg->cfs_rq[cpu] = cfs_rq;
6053 /* se could be NULL for root_task_group */
6058 se->cfs_rq = &rq->cfs;
6060 se->cfs_rq = parent->my_q;
6063 update_load_set(&se->load, 0);
6064 se->parent = parent;
6067 static DEFINE_MUTEX(shares_mutex);
6069 int sched_group_set_shares(struct task_group *tg, unsigned long shares)
6072 unsigned long flags;
6075 * We can't change the weight of the root cgroup.
6080 shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));
6082 mutex_lock(&shares_mutex);
6083 if (tg->shares == shares)
6086 tg->shares = shares;
6087 for_each_possible_cpu(i) {
6088 struct rq *rq = cpu_rq(i);
6089 struct sched_entity *se;
6092 /* Propagate contribution to hierarchy */
6093 raw_spin_lock_irqsave(&rq->lock, flags);
6094 for_each_sched_entity(se)
6095 update_cfs_shares(group_cfs_rq(se));
6096 raw_spin_unlock_irqrestore(&rq->lock, flags);
6100 mutex_unlock(&shares_mutex);
6103 #else /* CONFIG_FAIR_GROUP_SCHED */
6105 void free_fair_sched_group(struct task_group *tg) { }
6107 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
6112 void unregister_fair_sched_group(struct task_group *tg, int cpu) { }
6114 #endif /* CONFIG_FAIR_GROUP_SCHED */
6117 static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
6119 struct sched_entity *se = &task->se;
6120 unsigned int rr_interval = 0;
6123 * Time slice is 0 for SCHED_OTHER tasks that are on an otherwise
6126 if (rq->cfs.load.weight)
6127 rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
6133 * All the scheduling class methods:
6135 const struct sched_class fair_sched_class = {
6136 .next = &idle_sched_class,
6137 .enqueue_task = enqueue_task_fair,
6138 .dequeue_task = dequeue_task_fair,
6139 .yield_task = yield_task_fair,
6140 .yield_to_task = yield_to_task_fair,
6142 .check_preempt_curr = check_preempt_wakeup,
6144 .pick_next_task = pick_next_task_fair,
6145 .put_prev_task = put_prev_task_fair,
6148 .select_task_rq = select_task_rq_fair,
6149 #ifdef CONFIG_FAIR_GROUP_SCHED
6150 .migrate_task_rq = migrate_task_rq_fair,
6152 .rq_online = rq_online_fair,
6153 .rq_offline = rq_offline_fair,
6155 .task_waking = task_waking_fair,
6158 .set_curr_task = set_curr_task_fair,
6159 .task_tick = task_tick_fair,
6160 .task_fork = task_fork_fair,
6162 .prio_changed = prio_changed_fair,
6163 .switched_from = switched_from_fair,
6164 .switched_to = switched_to_fair,
6166 .get_rr_interval = get_rr_interval_fair,
6168 #ifdef CONFIG_FAIR_GROUP_SCHED
6169 .task_move_group = task_move_group_fair,
6173 #ifdef CONFIG_SCHED_DEBUG
6174 void print_cfs_stats(struct seq_file *m, int cpu)
6176 struct cfs_rq *cfs_rq;
6179 for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
6180 print_cfs_rq(m, cpu, cfs_rq);
6185 __init void init_sched_fair_class(void)
6188 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
6191 nohz.next_balance = jiffies;
6192 zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
6193 cpu_notifier(sched_ilb_notifier, 0);