sched: backward looking buddy
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / sched_fair.c
bloba6b1db8a0bd8754dd9bed5dd5b1f72cfa709acbf
1 /*
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>
26 * Targeted preemption latency for CPU-bound tasks:
27 * (default: 20ms * (1 + ilog(ncpus)), units: nanoseconds)
29 * NOTE: this latency value is not the same as the concept of
30 * 'timeslice length' - timeslices in CFS are of variable length
31 * and have no persistent notion like in traditional, time-slice
32 * based scheduling concepts.
34 * (to see the precise effective timeslice length of your workload,
35 * run vmstat and monitor the context-switches (cs) field)
37 unsigned int sysctl_sched_latency = 20000000ULL;
40 * Minimal preemption granularity for CPU-bound tasks:
41 * (default: 4 msec * (1 + ilog(ncpus)), units: nanoseconds)
43 unsigned int sysctl_sched_min_granularity = 4000000ULL;
46 * is kept at sysctl_sched_latency / sysctl_sched_min_granularity
48 static unsigned int sched_nr_latency = 5;
51 * After fork, child runs first. (default) If set to 0 then
52 * parent will (try to) run first.
54 const_debug unsigned int sysctl_sched_child_runs_first = 1;
57 * sys_sched_yield() compat mode
59 * This option switches the agressive yield implementation of the
60 * old scheduler back on.
62 unsigned int __read_mostly sysctl_sched_compat_yield;
65 * SCHED_OTHER wake-up granularity.
66 * (default: 5 msec * (1 + ilog(ncpus)), units: nanoseconds)
68 * This option delays the preemption effects of decoupled workloads
69 * and reduces their over-scheduling. Synchronous workloads will still
70 * have immediate wakeup/sleep latencies.
72 unsigned int sysctl_sched_wakeup_granularity = 5000000UL;
74 const_debug unsigned int sysctl_sched_migration_cost = 500000UL;
76 static const struct sched_class fair_sched_class;
78 /**************************************************************
79 * CFS operations on generic schedulable entities:
82 static inline struct task_struct *task_of(struct sched_entity *se)
84 return container_of(se, struct task_struct, se);
87 #ifdef CONFIG_FAIR_GROUP_SCHED
89 /* cpu runqueue to which this cfs_rq is attached */
90 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
92 return cfs_rq->rq;
95 /* An entity is a task if it doesn't "own" a runqueue */
96 #define entity_is_task(se) (!se->my_q)
98 /* Walk up scheduling entities hierarchy */
99 #define for_each_sched_entity(se) \
100 for (; se; se = se->parent)
102 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
104 return p->se.cfs_rq;
107 /* runqueue on which this entity is (to be) queued */
108 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
110 return se->cfs_rq;
113 /* runqueue "owned" by this group */
114 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
116 return grp->my_q;
119 /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on
120 * another cpu ('this_cpu')
122 static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu)
124 return cfs_rq->tg->cfs_rq[this_cpu];
127 /* Iterate thr' all leaf cfs_rq's on a runqueue */
128 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
129 list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)
131 /* Do the two (enqueued) entities belong to the same group ? */
132 static inline int
133 is_same_group(struct sched_entity *se, struct sched_entity *pse)
135 if (se->cfs_rq == pse->cfs_rq)
136 return 1;
138 return 0;
141 static inline struct sched_entity *parent_entity(struct sched_entity *se)
143 return se->parent;
146 /* return depth at which a sched entity is present in the hierarchy */
147 static inline int depth_se(struct sched_entity *se)
149 int depth = 0;
151 for_each_sched_entity(se)
152 depth++;
154 return depth;
157 static void
158 find_matching_se(struct sched_entity **se, struct sched_entity **pse)
160 int se_depth, pse_depth;
163 * preemption test can be made between sibling entities who are in the
164 * same cfs_rq i.e who have a common parent. Walk up the hierarchy of
165 * both tasks until we find their ancestors who are siblings of common
166 * parent.
169 /* First walk up until both entities are at same depth */
170 se_depth = depth_se(*se);
171 pse_depth = depth_se(*pse);
173 while (se_depth > pse_depth) {
174 se_depth--;
175 *se = parent_entity(*se);
178 while (pse_depth > se_depth) {
179 pse_depth--;
180 *pse = parent_entity(*pse);
183 while (!is_same_group(*se, *pse)) {
184 *se = parent_entity(*se);
185 *pse = parent_entity(*pse);
189 #else /* CONFIG_FAIR_GROUP_SCHED */
191 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
193 return container_of(cfs_rq, struct rq, cfs);
196 #define entity_is_task(se) 1
198 #define for_each_sched_entity(se) \
199 for (; se; se = NULL)
201 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
203 return &task_rq(p)->cfs;
206 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
208 struct task_struct *p = task_of(se);
209 struct rq *rq = task_rq(p);
211 return &rq->cfs;
214 /* runqueue "owned" by this group */
215 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
217 return NULL;
220 static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu)
222 return &cpu_rq(this_cpu)->cfs;
225 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
226 for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)
228 static inline int
229 is_same_group(struct sched_entity *se, struct sched_entity *pse)
231 return 1;
234 static inline struct sched_entity *parent_entity(struct sched_entity *se)
236 return NULL;
239 static inline void
240 find_matching_se(struct sched_entity **se, struct sched_entity **pse)
244 #endif /* CONFIG_FAIR_GROUP_SCHED */
247 /**************************************************************
248 * Scheduling class tree data structure manipulation methods:
251 static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime)
253 s64 delta = (s64)(vruntime - min_vruntime);
254 if (delta > 0)
255 min_vruntime = vruntime;
257 return min_vruntime;
260 static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
262 s64 delta = (s64)(vruntime - min_vruntime);
263 if (delta < 0)
264 min_vruntime = vruntime;
266 return min_vruntime;
269 static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se)
271 return se->vruntime - cfs_rq->min_vruntime;
274 static void update_min_vruntime(struct cfs_rq *cfs_rq)
276 u64 vruntime = cfs_rq->min_vruntime;
278 if (cfs_rq->curr)
279 vruntime = cfs_rq->curr->vruntime;
281 if (cfs_rq->rb_leftmost) {
282 struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost,
283 struct sched_entity,
284 run_node);
286 if (vruntime == cfs_rq->min_vruntime)
287 vruntime = se->vruntime;
288 else
289 vruntime = min_vruntime(vruntime, se->vruntime);
292 cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
296 * Enqueue an entity into the rb-tree:
298 static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
300 struct rb_node **link = &cfs_rq->tasks_timeline.rb_node;
301 struct rb_node *parent = NULL;
302 struct sched_entity *entry;
303 s64 key = entity_key(cfs_rq, se);
304 int leftmost = 1;
307 * Find the right place in the rbtree:
309 while (*link) {
310 parent = *link;
311 entry = rb_entry(parent, struct sched_entity, run_node);
313 * We dont care about collisions. Nodes with
314 * the same key stay together.
316 if (key < entity_key(cfs_rq, entry)) {
317 link = &parent->rb_left;
318 } else {
319 link = &parent->rb_right;
320 leftmost = 0;
325 * Maintain a cache of leftmost tree entries (it is frequently
326 * used):
328 if (leftmost)
329 cfs_rq->rb_leftmost = &se->run_node;
331 rb_link_node(&se->run_node, parent, link);
332 rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline);
335 static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
337 if (cfs_rq->rb_leftmost == &se->run_node) {
338 struct rb_node *next_node;
340 next_node = rb_next(&se->run_node);
341 cfs_rq->rb_leftmost = next_node;
344 rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
347 static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq)
349 struct rb_node *left = cfs_rq->rb_leftmost;
351 if (!left)
352 return NULL;
354 return rb_entry(left, struct sched_entity, run_node);
357 static struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
359 struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
361 if (!last)
362 return NULL;
364 return rb_entry(last, struct sched_entity, run_node);
367 /**************************************************************
368 * Scheduling class statistics methods:
371 #ifdef CONFIG_SCHED_DEBUG
372 int sched_nr_latency_handler(struct ctl_table *table, int write,
373 struct file *filp, void __user *buffer, size_t *lenp,
374 loff_t *ppos)
376 int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
378 if (ret || !write)
379 return ret;
381 sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
382 sysctl_sched_min_granularity);
384 return 0;
386 #endif
389 * delta *= P[w / rw]
391 static inline unsigned long
392 calc_delta_weight(unsigned long delta, struct sched_entity *se)
394 for_each_sched_entity(se) {
395 delta = calc_delta_mine(delta,
396 se->load.weight, &cfs_rq_of(se)->load);
399 return delta;
403 * delta /= w
405 static inline unsigned long
406 calc_delta_fair(unsigned long delta, struct sched_entity *se)
408 if (unlikely(se->load.weight != NICE_0_LOAD))
409 delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load);
411 return delta;
415 * The idea is to set a period in which each task runs once.
417 * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch
418 * this period because otherwise the slices get too small.
420 * p = (nr <= nl) ? l : l*nr/nl
422 static u64 __sched_period(unsigned long nr_running)
424 u64 period = sysctl_sched_latency;
425 unsigned long nr_latency = sched_nr_latency;
427 if (unlikely(nr_running > nr_latency)) {
428 period = sysctl_sched_min_granularity;
429 period *= nr_running;
432 return period;
436 * We calculate the wall-time slice from the period by taking a part
437 * proportional to the weight.
439 * s = p*P[w/rw]
441 static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
443 unsigned long nr_running = cfs_rq->nr_running;
445 if (unlikely(!se->on_rq))
446 nr_running++;
448 return calc_delta_weight(__sched_period(nr_running), se);
452 * We calculate the vruntime slice of a to be inserted task
454 * vs = s/w
456 static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
458 return calc_delta_fair(sched_slice(cfs_rq, se), se);
462 * Update the current task's runtime statistics. Skip current tasks that
463 * are not in our scheduling class.
465 static inline void
466 __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
467 unsigned long delta_exec)
469 unsigned long delta_exec_weighted;
471 schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max));
473 curr->sum_exec_runtime += delta_exec;
474 schedstat_add(cfs_rq, exec_clock, delta_exec);
475 delta_exec_weighted = calc_delta_fair(delta_exec, curr);
476 curr->vruntime += delta_exec_weighted;
477 update_min_vruntime(cfs_rq);
480 static void update_curr(struct cfs_rq *cfs_rq)
482 struct sched_entity *curr = cfs_rq->curr;
483 u64 now = rq_of(cfs_rq)->clock;
484 unsigned long delta_exec;
486 if (unlikely(!curr))
487 return;
490 * Get the amount of time the current task was running
491 * since the last time we changed load (this cannot
492 * overflow on 32 bits):
494 delta_exec = (unsigned long)(now - curr->exec_start);
496 __update_curr(cfs_rq, curr, delta_exec);
497 curr->exec_start = now;
499 if (entity_is_task(curr)) {
500 struct task_struct *curtask = task_of(curr);
502 cpuacct_charge(curtask, delta_exec);
503 account_group_exec_runtime(curtask, delta_exec);
507 static inline void
508 update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
510 schedstat_set(se->wait_start, rq_of(cfs_rq)->clock);
514 * Task is being enqueued - update stats:
516 static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
519 * Are we enqueueing a waiting task? (for current tasks
520 * a dequeue/enqueue event is a NOP)
522 if (se != cfs_rq->curr)
523 update_stats_wait_start(cfs_rq, se);
526 static void
527 update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
529 schedstat_set(se->wait_max, max(se->wait_max,
530 rq_of(cfs_rq)->clock - se->wait_start));
531 schedstat_set(se->wait_count, se->wait_count + 1);
532 schedstat_set(se->wait_sum, se->wait_sum +
533 rq_of(cfs_rq)->clock - se->wait_start);
534 schedstat_set(se->wait_start, 0);
537 static inline void
538 update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
541 * Mark the end of the wait period if dequeueing a
542 * waiting task:
544 if (se != cfs_rq->curr)
545 update_stats_wait_end(cfs_rq, se);
549 * We are picking a new current task - update its stats:
551 static inline void
552 update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
555 * We are starting a new run period:
557 se->exec_start = rq_of(cfs_rq)->clock;
560 /**************************************************
561 * Scheduling class queueing methods:
564 #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED
565 static void
566 add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
568 cfs_rq->task_weight += weight;
570 #else
571 static inline void
572 add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
575 #endif
577 static void
578 account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
580 update_load_add(&cfs_rq->load, se->load.weight);
581 if (!parent_entity(se))
582 inc_cpu_load(rq_of(cfs_rq), se->load.weight);
583 if (entity_is_task(se)) {
584 add_cfs_task_weight(cfs_rq, se->load.weight);
585 list_add(&se->group_node, &cfs_rq->tasks);
587 cfs_rq->nr_running++;
588 se->on_rq = 1;
591 static void
592 account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
594 update_load_sub(&cfs_rq->load, se->load.weight);
595 if (!parent_entity(se))
596 dec_cpu_load(rq_of(cfs_rq), se->load.weight);
597 if (entity_is_task(se)) {
598 add_cfs_task_weight(cfs_rq, -se->load.weight);
599 list_del_init(&se->group_node);
601 cfs_rq->nr_running--;
602 se->on_rq = 0;
605 static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
607 #ifdef CONFIG_SCHEDSTATS
608 if (se->sleep_start) {
609 u64 delta = rq_of(cfs_rq)->clock - se->sleep_start;
610 struct task_struct *tsk = task_of(se);
612 if ((s64)delta < 0)
613 delta = 0;
615 if (unlikely(delta > se->sleep_max))
616 se->sleep_max = delta;
618 se->sleep_start = 0;
619 se->sum_sleep_runtime += delta;
621 account_scheduler_latency(tsk, delta >> 10, 1);
623 if (se->block_start) {
624 u64 delta = rq_of(cfs_rq)->clock - se->block_start;
625 struct task_struct *tsk = task_of(se);
627 if ((s64)delta < 0)
628 delta = 0;
630 if (unlikely(delta > se->block_max))
631 se->block_max = delta;
633 se->block_start = 0;
634 se->sum_sleep_runtime += delta;
637 * Blocking time is in units of nanosecs, so shift by 20 to
638 * get a milliseconds-range estimation of the amount of
639 * time that the task spent sleeping:
641 if (unlikely(prof_on == SLEEP_PROFILING)) {
643 profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk),
644 delta >> 20);
646 account_scheduler_latency(tsk, delta >> 10, 0);
648 #endif
651 static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se)
653 #ifdef CONFIG_SCHED_DEBUG
654 s64 d = se->vruntime - cfs_rq->min_vruntime;
656 if (d < 0)
657 d = -d;
659 if (d > 3*sysctl_sched_latency)
660 schedstat_inc(cfs_rq, nr_spread_over);
661 #endif
664 static void
665 place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
667 u64 vruntime = cfs_rq->min_vruntime;
670 * The 'current' period is already promised to the current tasks,
671 * however the extra weight of the new task will slow them down a
672 * little, place the new task so that it fits in the slot that
673 * stays open at the end.
675 if (initial && sched_feat(START_DEBIT))
676 vruntime += sched_vslice(cfs_rq, se);
678 if (!initial) {
679 /* sleeps upto a single latency don't count. */
680 if (sched_feat(NEW_FAIR_SLEEPERS)) {
681 unsigned long thresh = sysctl_sched_latency;
684 * convert the sleeper threshold into virtual time
686 if (sched_feat(NORMALIZED_SLEEPER))
687 thresh = calc_delta_fair(thresh, se);
689 vruntime -= thresh;
692 /* ensure we never gain time by being placed backwards. */
693 vruntime = max_vruntime(se->vruntime, vruntime);
696 se->vruntime = vruntime;
699 static void
700 enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup)
703 * Update run-time statistics of the 'current'.
705 update_curr(cfs_rq);
706 account_entity_enqueue(cfs_rq, se);
708 if (wakeup) {
709 place_entity(cfs_rq, se, 0);
710 enqueue_sleeper(cfs_rq, se);
713 update_stats_enqueue(cfs_rq, se);
714 check_spread(cfs_rq, se);
715 if (se != cfs_rq->curr)
716 __enqueue_entity(cfs_rq, se);
719 static void
720 dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep)
723 * Update run-time statistics of the 'current'.
725 update_curr(cfs_rq);
727 update_stats_dequeue(cfs_rq, se);
728 if (sleep) {
729 #ifdef CONFIG_SCHEDSTATS
730 if (entity_is_task(se)) {
731 struct task_struct *tsk = task_of(se);
733 if (tsk->state & TASK_INTERRUPTIBLE)
734 se->sleep_start = rq_of(cfs_rq)->clock;
735 if (tsk->state & TASK_UNINTERRUPTIBLE)
736 se->block_start = rq_of(cfs_rq)->clock;
738 #endif
741 if (cfs_rq->last == se)
742 cfs_rq->last = NULL;
744 if (cfs_rq->next == se)
745 cfs_rq->next = NULL;
747 if (se != cfs_rq->curr)
748 __dequeue_entity(cfs_rq, se);
749 account_entity_dequeue(cfs_rq, se);
750 update_min_vruntime(cfs_rq);
754 * Preempt the current task with a newly woken task if needed:
756 static void
757 check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
759 unsigned long ideal_runtime, delta_exec;
761 ideal_runtime = sched_slice(cfs_rq, curr);
762 delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
763 if (delta_exec > ideal_runtime)
764 resched_task(rq_of(cfs_rq)->curr);
767 static void
768 set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
770 /* 'current' is not kept within the tree. */
771 if (se->on_rq) {
773 * Any task has to be enqueued before it get to execute on
774 * a CPU. So account for the time it spent waiting on the
775 * runqueue.
777 update_stats_wait_end(cfs_rq, se);
778 __dequeue_entity(cfs_rq, se);
781 update_stats_curr_start(cfs_rq, se);
782 cfs_rq->curr = se;
783 #ifdef CONFIG_SCHEDSTATS
785 * Track our maximum slice length, if the CPU's load is at
786 * least twice that of our own weight (i.e. dont track it
787 * when there are only lesser-weight tasks around):
789 if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
790 se->slice_max = max(se->slice_max,
791 se->sum_exec_runtime - se->prev_sum_exec_runtime);
793 #endif
794 se->prev_sum_exec_runtime = se->sum_exec_runtime;
797 static int
798 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);
800 static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
802 struct sched_entity *se = __pick_next_entity(cfs_rq);
804 if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, se) < 1)
805 return cfs_rq->next;
807 if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, se) < 1)
808 return cfs_rq->last;
810 return se;
813 static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
816 * If still on the runqueue then deactivate_task()
817 * was not called and update_curr() has to be done:
819 if (prev->on_rq)
820 update_curr(cfs_rq);
822 check_spread(cfs_rq, prev);
823 if (prev->on_rq) {
824 update_stats_wait_start(cfs_rq, prev);
825 /* Put 'current' back into the tree. */
826 __enqueue_entity(cfs_rq, prev);
828 cfs_rq->curr = NULL;
831 static void
832 entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
835 * Update run-time statistics of the 'current'.
837 update_curr(cfs_rq);
839 #ifdef CONFIG_SCHED_HRTICK
841 * queued ticks are scheduled to match the slice, so don't bother
842 * validating it and just reschedule.
844 if (queued) {
845 resched_task(rq_of(cfs_rq)->curr);
846 return;
849 * don't let the period tick interfere with the hrtick preemption
851 if (!sched_feat(DOUBLE_TICK) &&
852 hrtimer_active(&rq_of(cfs_rq)->hrtick_timer))
853 return;
854 #endif
856 if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT))
857 check_preempt_tick(cfs_rq, curr);
860 /**************************************************
861 * CFS operations on tasks:
864 #ifdef CONFIG_SCHED_HRTICK
865 static void hrtick_start_fair(struct rq *rq, struct task_struct *p)
867 struct sched_entity *se = &p->se;
868 struct cfs_rq *cfs_rq = cfs_rq_of(se);
870 WARN_ON(task_rq(p) != rq);
872 if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) {
873 u64 slice = sched_slice(cfs_rq, se);
874 u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime;
875 s64 delta = slice - ran;
877 if (delta < 0) {
878 if (rq->curr == p)
879 resched_task(p);
880 return;
884 * Don't schedule slices shorter than 10000ns, that just
885 * doesn't make sense. Rely on vruntime for fairness.
887 if (rq->curr != p)
888 delta = max_t(s64, 10000LL, delta);
890 hrtick_start(rq, delta);
895 * called from enqueue/dequeue and updates the hrtick when the
896 * current task is from our class and nr_running is low enough
897 * to matter.
899 static void hrtick_update(struct rq *rq)
901 struct task_struct *curr = rq->curr;
903 if (curr->sched_class != &fair_sched_class)
904 return;
906 if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
907 hrtick_start_fair(rq, curr);
909 #else /* !CONFIG_SCHED_HRTICK */
910 static inline void
911 hrtick_start_fair(struct rq *rq, struct task_struct *p)
915 static inline void hrtick_update(struct rq *rq)
918 #endif
921 * The enqueue_task method is called before nr_running is
922 * increased. Here we update the fair scheduling stats and
923 * then put the task into the rbtree:
925 static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup)
927 struct cfs_rq *cfs_rq;
928 struct sched_entity *se = &p->se;
930 for_each_sched_entity(se) {
931 if (se->on_rq)
932 break;
933 cfs_rq = cfs_rq_of(se);
934 enqueue_entity(cfs_rq, se, wakeup);
935 wakeup = 1;
938 hrtick_update(rq);
942 * The dequeue_task method is called before nr_running is
943 * decreased. We remove the task from the rbtree and
944 * update the fair scheduling stats:
946 static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep)
948 struct cfs_rq *cfs_rq;
949 struct sched_entity *se = &p->se;
951 for_each_sched_entity(se) {
952 cfs_rq = cfs_rq_of(se);
953 dequeue_entity(cfs_rq, se, sleep);
954 /* Don't dequeue parent if it has other entities besides us */
955 if (cfs_rq->load.weight)
956 break;
957 sleep = 1;
960 hrtick_update(rq);
964 * sched_yield() support is very simple - we dequeue and enqueue.
966 * If compat_yield is turned on then we requeue to the end of the tree.
968 static void yield_task_fair(struct rq *rq)
970 struct task_struct *curr = rq->curr;
971 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
972 struct sched_entity *rightmost, *se = &curr->se;
975 * Are we the only task in the tree?
977 if (unlikely(cfs_rq->nr_running == 1))
978 return;
980 if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) {
981 update_rq_clock(rq);
983 * Update run-time statistics of the 'current'.
985 update_curr(cfs_rq);
987 return;
990 * Find the rightmost entry in the rbtree:
992 rightmost = __pick_last_entity(cfs_rq);
994 * Already in the rightmost position?
996 if (unlikely(!rightmost || rightmost->vruntime < se->vruntime))
997 return;
1000 * Minimally necessary key value to be last in the tree:
1001 * Upon rescheduling, sched_class::put_prev_task() will place
1002 * 'current' within the tree based on its new key value.
1004 se->vruntime = rightmost->vruntime + 1;
1008 * wake_idle() will wake a task on an idle cpu if task->cpu is
1009 * not idle and an idle cpu is available. The span of cpus to
1010 * search starts with cpus closest then further out as needed,
1011 * so we always favor a closer, idle cpu.
1012 * Domains may include CPUs that are not usable for migration,
1013 * hence we need to mask them out (cpu_active_map)
1015 * Returns the CPU we should wake onto.
1017 #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
1018 static int wake_idle(int cpu, struct task_struct *p)
1020 cpumask_t tmp;
1021 struct sched_domain *sd;
1022 int i;
1025 * If it is idle, then it is the best cpu to run this task.
1027 * This cpu is also the best, if it has more than one task already.
1028 * Siblings must be also busy(in most cases) as they didn't already
1029 * pickup the extra load from this cpu and hence we need not check
1030 * sibling runqueue info. This will avoid the checks and cache miss
1031 * penalities associated with that.
1033 if (idle_cpu(cpu) || cpu_rq(cpu)->cfs.nr_running > 1)
1034 return cpu;
1036 for_each_domain(cpu, sd) {
1037 if ((sd->flags & SD_WAKE_IDLE)
1038 || ((sd->flags & SD_WAKE_IDLE_FAR)
1039 && !task_hot(p, task_rq(p)->clock, sd))) {
1040 cpus_and(tmp, sd->span, p->cpus_allowed);
1041 cpus_and(tmp, tmp, cpu_active_map);
1042 for_each_cpu_mask_nr(i, tmp) {
1043 if (idle_cpu(i)) {
1044 if (i != task_cpu(p)) {
1045 schedstat_inc(p,
1046 se.nr_wakeups_idle);
1048 return i;
1051 } else {
1052 break;
1055 return cpu;
1057 #else /* !ARCH_HAS_SCHED_WAKE_IDLE*/
1058 static inline int wake_idle(int cpu, struct task_struct *p)
1060 return cpu;
1062 #endif
1064 #ifdef CONFIG_SMP
1066 #ifdef CONFIG_FAIR_GROUP_SCHED
1068 * effective_load() calculates the load change as seen from the root_task_group
1070 * Adding load to a group doesn't make a group heavier, but can cause movement
1071 * of group shares between cpus. Assuming the shares were perfectly aligned one
1072 * can calculate the shift in shares.
1074 * The problem is that perfectly aligning the shares is rather expensive, hence
1075 * we try to avoid doing that too often - see update_shares(), which ratelimits
1076 * this change.
1078 * We compensate this by not only taking the current delta into account, but
1079 * also considering the delta between when the shares were last adjusted and
1080 * now.
1082 * We still saw a performance dip, some tracing learned us that between
1083 * cgroup:/ and cgroup:/foo balancing the number of affine wakeups increased
1084 * significantly. Therefore try to bias the error in direction of failing
1085 * the affine wakeup.
1088 static long effective_load(struct task_group *tg, int cpu,
1089 long wl, long wg)
1091 struct sched_entity *se = tg->se[cpu];
1093 if (!tg->parent)
1094 return wl;
1097 * By not taking the decrease of shares on the other cpu into
1098 * account our error leans towards reducing the affine wakeups.
1100 if (!wl && sched_feat(ASYM_EFF_LOAD))
1101 return wl;
1103 for_each_sched_entity(se) {
1104 long S, rw, s, a, b;
1105 long more_w;
1108 * Instead of using this increment, also add the difference
1109 * between when the shares were last updated and now.
1111 more_w = se->my_q->load.weight - se->my_q->rq_weight;
1112 wl += more_w;
1113 wg += more_w;
1115 S = se->my_q->tg->shares;
1116 s = se->my_q->shares;
1117 rw = se->my_q->rq_weight;
1119 a = S*(rw + wl);
1120 b = S*rw + s*wg;
1122 wl = s*(a-b);
1124 if (likely(b))
1125 wl /= b;
1128 * Assume the group is already running and will
1129 * thus already be accounted for in the weight.
1131 * That is, moving shares between CPUs, does not
1132 * alter the group weight.
1134 wg = 0;
1137 return wl;
1140 #else
1142 static inline unsigned long effective_load(struct task_group *tg, int cpu,
1143 unsigned long wl, unsigned long wg)
1145 return wl;
1148 #endif
1150 static int
1151 wake_affine(struct sched_domain *this_sd, struct rq *this_rq,
1152 struct task_struct *p, int prev_cpu, int this_cpu, int sync,
1153 int idx, unsigned long load, unsigned long this_load,
1154 unsigned int imbalance)
1156 struct task_struct *curr = this_rq->curr;
1157 struct task_group *tg;
1158 unsigned long tl = this_load;
1159 unsigned long tl_per_task;
1160 unsigned long weight;
1161 int balanced;
1163 if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS))
1164 return 0;
1166 if (sync && (curr->se.avg_overlap > sysctl_sched_migration_cost ||
1167 p->se.avg_overlap > sysctl_sched_migration_cost))
1168 sync = 0;
1171 * If sync wakeup then subtract the (maximum possible)
1172 * effect of the currently running task from the load
1173 * of the current CPU:
1175 if (sync) {
1176 tg = task_group(current);
1177 weight = current->se.load.weight;
1179 tl += effective_load(tg, this_cpu, -weight, -weight);
1180 load += effective_load(tg, prev_cpu, 0, -weight);
1183 tg = task_group(p);
1184 weight = p->se.load.weight;
1186 balanced = 100*(tl + effective_load(tg, this_cpu, weight, weight)) <=
1187 imbalance*(load + effective_load(tg, prev_cpu, 0, weight));
1190 * If the currently running task will sleep within
1191 * a reasonable amount of time then attract this newly
1192 * woken task:
1194 if (sync && balanced)
1195 return 1;
1197 schedstat_inc(p, se.nr_wakeups_affine_attempts);
1198 tl_per_task = cpu_avg_load_per_task(this_cpu);
1200 if (balanced || (tl <= load && tl + target_load(prev_cpu, idx) <=
1201 tl_per_task)) {
1203 * This domain has SD_WAKE_AFFINE and
1204 * p is cache cold in this domain, and
1205 * there is no bad imbalance.
1207 schedstat_inc(this_sd, ttwu_move_affine);
1208 schedstat_inc(p, se.nr_wakeups_affine);
1210 return 1;
1212 return 0;
1215 static int select_task_rq_fair(struct task_struct *p, int sync)
1217 struct sched_domain *sd, *this_sd = NULL;
1218 int prev_cpu, this_cpu, new_cpu;
1219 unsigned long load, this_load;
1220 struct rq *this_rq;
1221 unsigned int imbalance;
1222 int idx;
1224 prev_cpu = task_cpu(p);
1225 this_cpu = smp_processor_id();
1226 this_rq = cpu_rq(this_cpu);
1227 new_cpu = prev_cpu;
1229 if (prev_cpu == this_cpu)
1230 goto out;
1232 * 'this_sd' is the first domain that both
1233 * this_cpu and prev_cpu are present in:
1235 for_each_domain(this_cpu, sd) {
1236 if (cpu_isset(prev_cpu, sd->span)) {
1237 this_sd = sd;
1238 break;
1242 if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed)))
1243 goto out;
1246 * Check for affine wakeup and passive balancing possibilities.
1248 if (!this_sd)
1249 goto out;
1251 idx = this_sd->wake_idx;
1253 imbalance = 100 + (this_sd->imbalance_pct - 100) / 2;
1255 load = source_load(prev_cpu, idx);
1256 this_load = target_load(this_cpu, idx);
1258 if (wake_affine(this_sd, this_rq, p, prev_cpu, this_cpu, sync, idx,
1259 load, this_load, imbalance))
1260 return this_cpu;
1263 * Start passive balancing when half the imbalance_pct
1264 * limit is reached.
1266 if (this_sd->flags & SD_WAKE_BALANCE) {
1267 if (imbalance*this_load <= 100*load) {
1268 schedstat_inc(this_sd, ttwu_move_balance);
1269 schedstat_inc(p, se.nr_wakeups_passive);
1270 return this_cpu;
1274 out:
1275 return wake_idle(new_cpu, p);
1277 #endif /* CONFIG_SMP */
1279 static unsigned long wakeup_gran(struct sched_entity *se)
1281 unsigned long gran = sysctl_sched_wakeup_granularity;
1284 * More easily preempt - nice tasks, while not making it harder for
1285 * + nice tasks.
1287 if (!sched_feat(ASYM_GRAN) || se->load.weight > NICE_0_LOAD)
1288 gran = calc_delta_fair(sysctl_sched_wakeup_granularity, se);
1290 return gran;
1294 * Should 'se' preempt 'curr'.
1296 * |s1
1297 * |s2
1298 * |s3
1300 * |<--->|c
1302 * w(c, s1) = -1
1303 * w(c, s2) = 0
1304 * w(c, s3) = 1
1307 static int
1308 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
1310 s64 gran, vdiff = curr->vruntime - se->vruntime;
1312 if (vdiff <= 0)
1313 return -1;
1315 gran = wakeup_gran(curr);
1316 if (vdiff > gran)
1317 return 1;
1319 return 0;
1323 * Preempt the current task with a newly woken task if needed:
1325 static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync)
1327 struct task_struct *curr = rq->curr;
1328 struct sched_entity *se = &curr->se, *pse = &p->se;
1330 if (unlikely(rt_prio(p->prio))) {
1331 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
1333 update_rq_clock(rq);
1334 update_curr(cfs_rq);
1335 resched_task(curr);
1336 return;
1339 if (unlikely(p->sched_class != &fair_sched_class))
1340 return;
1342 if (unlikely(se == pse))
1343 return;
1346 * Only set the backward buddy when the current task is still on the
1347 * rq. This can happen when a wakeup gets interleaved with schedule on
1348 * the ->pre_schedule() or idle_balance() point, either of which can
1349 * drop the rq lock.
1351 * Also, during early boot the idle thread is in the fair class, for
1352 * obvious reasons its a bad idea to schedule back to the idle thread.
1354 if (sched_feat(LAST_BUDDY) && likely(se->on_rq && curr != rq->idle))
1355 cfs_rq_of(se)->last = se;
1356 cfs_rq_of(pse)->next = pse;
1359 * We can come here with TIF_NEED_RESCHED already set from new task
1360 * wake up path.
1362 if (test_tsk_need_resched(curr))
1363 return;
1366 * Batch tasks do not preempt (their preemption is driven by
1367 * the tick):
1369 if (unlikely(p->policy == SCHED_BATCH))
1370 return;
1372 if (!sched_feat(WAKEUP_PREEMPT))
1373 return;
1375 if (sched_feat(WAKEUP_OVERLAP) && (sync ||
1376 (se->avg_overlap < sysctl_sched_migration_cost &&
1377 pse->avg_overlap < sysctl_sched_migration_cost))) {
1378 resched_task(curr);
1379 return;
1382 find_matching_se(&se, &pse);
1384 while (se) {
1385 BUG_ON(!pse);
1387 if (wakeup_preempt_entity(se, pse) == 1) {
1388 resched_task(curr);
1389 break;
1392 se = parent_entity(se);
1393 pse = parent_entity(pse);
1397 static struct task_struct *pick_next_task_fair(struct rq *rq)
1399 struct task_struct *p;
1400 struct cfs_rq *cfs_rq = &rq->cfs;
1401 struct sched_entity *se;
1403 if (unlikely(!cfs_rq->nr_running))
1404 return NULL;
1406 do {
1407 se = pick_next_entity(cfs_rq);
1408 set_next_entity(cfs_rq, se);
1409 cfs_rq = group_cfs_rq(se);
1410 } while (cfs_rq);
1412 p = task_of(se);
1413 hrtick_start_fair(rq, p);
1415 return p;
1419 * Account for a descheduled task:
1421 static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
1423 struct sched_entity *se = &prev->se;
1424 struct cfs_rq *cfs_rq;
1426 for_each_sched_entity(se) {
1427 cfs_rq = cfs_rq_of(se);
1428 put_prev_entity(cfs_rq, se);
1432 #ifdef CONFIG_SMP
1433 /**************************************************
1434 * Fair scheduling class load-balancing methods:
1438 * Load-balancing iterator. Note: while the runqueue stays locked
1439 * during the whole iteration, the current task might be
1440 * dequeued so the iterator has to be dequeue-safe. Here we
1441 * achieve that by always pre-iterating before returning
1442 * the current task:
1444 static struct task_struct *
1445 __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next)
1447 struct task_struct *p = NULL;
1448 struct sched_entity *se;
1450 if (next == &cfs_rq->tasks)
1451 return NULL;
1453 se = list_entry(next, struct sched_entity, group_node);
1454 p = task_of(se);
1455 cfs_rq->balance_iterator = next->next;
1457 return p;
1460 static struct task_struct *load_balance_start_fair(void *arg)
1462 struct cfs_rq *cfs_rq = arg;
1464 return __load_balance_iterator(cfs_rq, cfs_rq->tasks.next);
1467 static struct task_struct *load_balance_next_fair(void *arg)
1469 struct cfs_rq *cfs_rq = arg;
1471 return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator);
1474 static unsigned long
1475 __load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1476 unsigned long max_load_move, struct sched_domain *sd,
1477 enum cpu_idle_type idle, int *all_pinned, int *this_best_prio,
1478 struct cfs_rq *cfs_rq)
1480 struct rq_iterator cfs_rq_iterator;
1482 cfs_rq_iterator.start = load_balance_start_fair;
1483 cfs_rq_iterator.next = load_balance_next_fair;
1484 cfs_rq_iterator.arg = cfs_rq;
1486 return balance_tasks(this_rq, this_cpu, busiest,
1487 max_load_move, sd, idle, all_pinned,
1488 this_best_prio, &cfs_rq_iterator);
1491 #ifdef CONFIG_FAIR_GROUP_SCHED
1492 static unsigned long
1493 load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1494 unsigned long max_load_move,
1495 struct sched_domain *sd, enum cpu_idle_type idle,
1496 int *all_pinned, int *this_best_prio)
1498 long rem_load_move = max_load_move;
1499 int busiest_cpu = cpu_of(busiest);
1500 struct task_group *tg;
1502 rcu_read_lock();
1503 update_h_load(busiest_cpu);
1505 list_for_each_entry_rcu(tg, &task_groups, list) {
1506 struct cfs_rq *busiest_cfs_rq = tg->cfs_rq[busiest_cpu];
1507 unsigned long busiest_h_load = busiest_cfs_rq->h_load;
1508 unsigned long busiest_weight = busiest_cfs_rq->load.weight;
1509 u64 rem_load, moved_load;
1512 * empty group
1514 if (!busiest_cfs_rq->task_weight)
1515 continue;
1517 rem_load = (u64)rem_load_move * busiest_weight;
1518 rem_load = div_u64(rem_load, busiest_h_load + 1);
1520 moved_load = __load_balance_fair(this_rq, this_cpu, busiest,
1521 rem_load, sd, idle, all_pinned, this_best_prio,
1522 tg->cfs_rq[busiest_cpu]);
1524 if (!moved_load)
1525 continue;
1527 moved_load *= busiest_h_load;
1528 moved_load = div_u64(moved_load, busiest_weight + 1);
1530 rem_load_move -= moved_load;
1531 if (rem_load_move < 0)
1532 break;
1534 rcu_read_unlock();
1536 return max_load_move - rem_load_move;
1538 #else
1539 static unsigned long
1540 load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1541 unsigned long max_load_move,
1542 struct sched_domain *sd, enum cpu_idle_type idle,
1543 int *all_pinned, int *this_best_prio)
1545 return __load_balance_fair(this_rq, this_cpu, busiest,
1546 max_load_move, sd, idle, all_pinned,
1547 this_best_prio, &busiest->cfs);
1549 #endif
1551 static int
1552 move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1553 struct sched_domain *sd, enum cpu_idle_type idle)
1555 struct cfs_rq *busy_cfs_rq;
1556 struct rq_iterator cfs_rq_iterator;
1558 cfs_rq_iterator.start = load_balance_start_fair;
1559 cfs_rq_iterator.next = load_balance_next_fair;
1561 for_each_leaf_cfs_rq(busiest, busy_cfs_rq) {
1563 * pass busy_cfs_rq argument into
1564 * load_balance_[start|next]_fair iterators
1566 cfs_rq_iterator.arg = busy_cfs_rq;
1567 if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle,
1568 &cfs_rq_iterator))
1569 return 1;
1572 return 0;
1574 #endif /* CONFIG_SMP */
1577 * scheduler tick hitting a task of our scheduling class:
1579 static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
1581 struct cfs_rq *cfs_rq;
1582 struct sched_entity *se = &curr->se;
1584 for_each_sched_entity(se) {
1585 cfs_rq = cfs_rq_of(se);
1586 entity_tick(cfs_rq, se, queued);
1590 #define swap(a, b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0)
1593 * Share the fairness runtime between parent and child, thus the
1594 * total amount of pressure for CPU stays equal - new tasks
1595 * get a chance to run but frequent forkers are not allowed to
1596 * monopolize the CPU. Note: the parent runqueue is locked,
1597 * the child is not running yet.
1599 static void task_new_fair(struct rq *rq, struct task_struct *p)
1601 struct cfs_rq *cfs_rq = task_cfs_rq(p);
1602 struct sched_entity *se = &p->se, *curr = cfs_rq->curr;
1603 int this_cpu = smp_processor_id();
1605 sched_info_queued(p);
1607 update_curr(cfs_rq);
1608 place_entity(cfs_rq, se, 1);
1610 /* 'curr' will be NULL if the child belongs to a different group */
1611 if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) &&
1612 curr && curr->vruntime < se->vruntime) {
1614 * Upon rescheduling, sched_class::put_prev_task() will place
1615 * 'current' within the tree based on its new key value.
1617 swap(curr->vruntime, se->vruntime);
1618 resched_task(rq->curr);
1621 enqueue_task_fair(rq, p, 0);
1625 * Priority of the task has changed. Check to see if we preempt
1626 * the current task.
1628 static void prio_changed_fair(struct rq *rq, struct task_struct *p,
1629 int oldprio, int running)
1632 * Reschedule if we are currently running on this runqueue and
1633 * our priority decreased, or if we are not currently running on
1634 * this runqueue and our priority is higher than the current's
1636 if (running) {
1637 if (p->prio > oldprio)
1638 resched_task(rq->curr);
1639 } else
1640 check_preempt_curr(rq, p, 0);
1644 * We switched to the sched_fair class.
1646 static void switched_to_fair(struct rq *rq, struct task_struct *p,
1647 int running)
1650 * We were most likely switched from sched_rt, so
1651 * kick off the schedule if running, otherwise just see
1652 * if we can still preempt the current task.
1654 if (running)
1655 resched_task(rq->curr);
1656 else
1657 check_preempt_curr(rq, p, 0);
1660 /* Account for a task changing its policy or group.
1662 * This routine is mostly called to set cfs_rq->curr field when a task
1663 * migrates between groups/classes.
1665 static void set_curr_task_fair(struct rq *rq)
1667 struct sched_entity *se = &rq->curr->se;
1669 for_each_sched_entity(se)
1670 set_next_entity(cfs_rq_of(se), se);
1673 #ifdef CONFIG_FAIR_GROUP_SCHED
1674 static void moved_group_fair(struct task_struct *p)
1676 struct cfs_rq *cfs_rq = task_cfs_rq(p);
1678 update_curr(cfs_rq);
1679 place_entity(cfs_rq, &p->se, 1);
1681 #endif
1684 * All the scheduling class methods:
1686 static const struct sched_class fair_sched_class = {
1687 .next = &idle_sched_class,
1688 .enqueue_task = enqueue_task_fair,
1689 .dequeue_task = dequeue_task_fair,
1690 .yield_task = yield_task_fair,
1692 .check_preempt_curr = check_preempt_wakeup,
1694 .pick_next_task = pick_next_task_fair,
1695 .put_prev_task = put_prev_task_fair,
1697 #ifdef CONFIG_SMP
1698 .select_task_rq = select_task_rq_fair,
1700 .load_balance = load_balance_fair,
1701 .move_one_task = move_one_task_fair,
1702 #endif
1704 .set_curr_task = set_curr_task_fair,
1705 .task_tick = task_tick_fair,
1706 .task_new = task_new_fair,
1708 .prio_changed = prio_changed_fair,
1709 .switched_to = switched_to_fair,
1711 #ifdef CONFIG_FAIR_GROUP_SCHED
1712 .moved_group = moved_group_fair,
1713 #endif
1716 #ifdef CONFIG_SCHED_DEBUG
1717 static void print_cfs_stats(struct seq_file *m, int cpu)
1719 struct cfs_rq *cfs_rq;
1721 rcu_read_lock();
1722 for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
1723 print_cfs_rq(m, cpu, cfs_rq);
1724 rcu_read_unlock();
1726 #endif