drivers/net/hamradio/6pack.c: move a dereference below a NULL test
[linux-2.6/mini2440.git] / kernel / sched_fair.c
blob8e1352c75557308bf1e892bb260b39e9feffd10f
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 /= w
391 static inline unsigned long
392 calc_delta_fair(unsigned long delta, struct sched_entity *se)
394 if (unlikely(se->load.weight != NICE_0_LOAD))
395 delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load);
397 return delta;
401 * The idea is to set a period in which each task runs once.
403 * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch
404 * this period because otherwise the slices get too small.
406 * p = (nr <= nl) ? l : l*nr/nl
408 static u64 __sched_period(unsigned long nr_running)
410 u64 period = sysctl_sched_latency;
411 unsigned long nr_latency = sched_nr_latency;
413 if (unlikely(nr_running > nr_latency)) {
414 period = sysctl_sched_min_granularity;
415 period *= nr_running;
418 return period;
422 * We calculate the wall-time slice from the period by taking a part
423 * proportional to the weight.
425 * s = p*P[w/rw]
427 static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
429 u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
431 for_each_sched_entity(se) {
432 struct load_weight *load = &cfs_rq->load;
434 if (unlikely(!se->on_rq)) {
435 struct load_weight lw = cfs_rq->load;
437 update_load_add(&lw, se->load.weight);
438 load = &lw;
440 slice = calc_delta_mine(slice, se->load.weight, load);
442 return slice;
446 * We calculate the vruntime slice of a to be inserted task
448 * vs = s/w
450 static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
452 return calc_delta_fair(sched_slice(cfs_rq, se), se);
456 * Update the current task's runtime statistics. Skip current tasks that
457 * are not in our scheduling class.
459 static inline void
460 __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
461 unsigned long delta_exec)
463 unsigned long delta_exec_weighted;
465 schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max));
467 curr->sum_exec_runtime += delta_exec;
468 schedstat_add(cfs_rq, exec_clock, delta_exec);
469 delta_exec_weighted = calc_delta_fair(delta_exec, curr);
470 curr->vruntime += delta_exec_weighted;
471 update_min_vruntime(cfs_rq);
474 static void update_curr(struct cfs_rq *cfs_rq)
476 struct sched_entity *curr = cfs_rq->curr;
477 u64 now = rq_of(cfs_rq)->clock;
478 unsigned long delta_exec;
480 if (unlikely(!curr))
481 return;
484 * Get the amount of time the current task was running
485 * since the last time we changed load (this cannot
486 * overflow on 32 bits):
488 delta_exec = (unsigned long)(now - curr->exec_start);
489 if (!delta_exec)
490 return;
492 __update_curr(cfs_rq, curr, delta_exec);
493 curr->exec_start = now;
495 if (entity_is_task(curr)) {
496 struct task_struct *curtask = task_of(curr);
498 cpuacct_charge(curtask, delta_exec);
499 account_group_exec_runtime(curtask, delta_exec);
503 static inline void
504 update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
506 schedstat_set(se->wait_start, rq_of(cfs_rq)->clock);
510 * Task is being enqueued - update stats:
512 static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
515 * Are we enqueueing a waiting task? (for current tasks
516 * a dequeue/enqueue event is a NOP)
518 if (se != cfs_rq->curr)
519 update_stats_wait_start(cfs_rq, se);
522 static void
523 update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
525 schedstat_set(se->wait_max, max(se->wait_max,
526 rq_of(cfs_rq)->clock - se->wait_start));
527 schedstat_set(se->wait_count, se->wait_count + 1);
528 schedstat_set(se->wait_sum, se->wait_sum +
529 rq_of(cfs_rq)->clock - se->wait_start);
530 schedstat_set(se->wait_start, 0);
533 static inline void
534 update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
537 * Mark the end of the wait period if dequeueing a
538 * waiting task:
540 if (se != cfs_rq->curr)
541 update_stats_wait_end(cfs_rq, se);
545 * We are picking a new current task - update its stats:
547 static inline void
548 update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
551 * We are starting a new run period:
553 se->exec_start = rq_of(cfs_rq)->clock;
556 /**************************************************
557 * Scheduling class queueing methods:
560 #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED
561 static void
562 add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
564 cfs_rq->task_weight += weight;
566 #else
567 static inline void
568 add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
571 #endif
573 static void
574 account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
576 update_load_add(&cfs_rq->load, se->load.weight);
577 if (!parent_entity(se))
578 inc_cpu_load(rq_of(cfs_rq), se->load.weight);
579 if (entity_is_task(se)) {
580 add_cfs_task_weight(cfs_rq, se->load.weight);
581 list_add(&se->group_node, &cfs_rq->tasks);
583 cfs_rq->nr_running++;
584 se->on_rq = 1;
587 static void
588 account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
590 update_load_sub(&cfs_rq->load, se->load.weight);
591 if (!parent_entity(se))
592 dec_cpu_load(rq_of(cfs_rq), se->load.weight);
593 if (entity_is_task(se)) {
594 add_cfs_task_weight(cfs_rq, -se->load.weight);
595 list_del_init(&se->group_node);
597 cfs_rq->nr_running--;
598 se->on_rq = 0;
601 static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
603 #ifdef CONFIG_SCHEDSTATS
604 if (se->sleep_start) {
605 u64 delta = rq_of(cfs_rq)->clock - se->sleep_start;
606 struct task_struct *tsk = task_of(se);
608 if ((s64)delta < 0)
609 delta = 0;
611 if (unlikely(delta > se->sleep_max))
612 se->sleep_max = delta;
614 se->sleep_start = 0;
615 se->sum_sleep_runtime += delta;
617 account_scheduler_latency(tsk, delta >> 10, 1);
619 if (se->block_start) {
620 u64 delta = rq_of(cfs_rq)->clock - se->block_start;
621 struct task_struct *tsk = task_of(se);
623 if ((s64)delta < 0)
624 delta = 0;
626 if (unlikely(delta > se->block_max))
627 se->block_max = delta;
629 se->block_start = 0;
630 se->sum_sleep_runtime += delta;
633 * Blocking time is in units of nanosecs, so shift by 20 to
634 * get a milliseconds-range estimation of the amount of
635 * time that the task spent sleeping:
637 if (unlikely(prof_on == SLEEP_PROFILING)) {
639 profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk),
640 delta >> 20);
642 account_scheduler_latency(tsk, delta >> 10, 0);
644 #endif
647 static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se)
649 #ifdef CONFIG_SCHED_DEBUG
650 s64 d = se->vruntime - cfs_rq->min_vruntime;
652 if (d < 0)
653 d = -d;
655 if (d > 3*sysctl_sched_latency)
656 schedstat_inc(cfs_rq, nr_spread_over);
657 #endif
660 static void
661 place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
663 u64 vruntime = cfs_rq->min_vruntime;
666 * The 'current' period is already promised to the current tasks,
667 * however the extra weight of the new task will slow them down a
668 * little, place the new task so that it fits in the slot that
669 * stays open at the end.
671 if (initial && sched_feat(START_DEBIT))
672 vruntime += sched_vslice(cfs_rq, se);
674 if (!initial) {
675 /* sleeps upto a single latency don't count. */
676 if (sched_feat(NEW_FAIR_SLEEPERS)) {
677 unsigned long thresh = sysctl_sched_latency;
680 * convert the sleeper threshold into virtual time
682 if (sched_feat(NORMALIZED_SLEEPER))
683 thresh = calc_delta_fair(thresh, se);
685 vruntime -= thresh;
688 /* ensure we never gain time by being placed backwards. */
689 vruntime = max_vruntime(se->vruntime, vruntime);
692 se->vruntime = vruntime;
695 static void
696 enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup)
699 * Update run-time statistics of the 'current'.
701 update_curr(cfs_rq);
702 account_entity_enqueue(cfs_rq, se);
704 if (wakeup) {
705 place_entity(cfs_rq, se, 0);
706 enqueue_sleeper(cfs_rq, se);
709 update_stats_enqueue(cfs_rq, se);
710 check_spread(cfs_rq, se);
711 if (se != cfs_rq->curr)
712 __enqueue_entity(cfs_rq, se);
715 static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
717 if (cfs_rq->last == se)
718 cfs_rq->last = NULL;
720 if (cfs_rq->next == se)
721 cfs_rq->next = NULL;
724 static void
725 dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep)
728 * Update run-time statistics of the 'current'.
730 update_curr(cfs_rq);
732 update_stats_dequeue(cfs_rq, se);
733 if (sleep) {
734 #ifdef CONFIG_SCHEDSTATS
735 if (entity_is_task(se)) {
736 struct task_struct *tsk = task_of(se);
738 if (tsk->state & TASK_INTERRUPTIBLE)
739 se->sleep_start = rq_of(cfs_rq)->clock;
740 if (tsk->state & TASK_UNINTERRUPTIBLE)
741 se->block_start = rq_of(cfs_rq)->clock;
743 #endif
746 clear_buddies(cfs_rq, se);
748 if (se != cfs_rq->curr)
749 __dequeue_entity(cfs_rq, se);
750 account_entity_dequeue(cfs_rq, se);
751 update_min_vruntime(cfs_rq);
755 * Preempt the current task with a newly woken task if needed:
757 static void
758 check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
760 unsigned long ideal_runtime, delta_exec;
762 ideal_runtime = sched_slice(cfs_rq, curr);
763 delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
764 if (delta_exec > ideal_runtime)
765 resched_task(rq_of(cfs_rq)->curr);
768 static void
769 set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
771 /* 'current' is not kept within the tree. */
772 if (se->on_rq) {
774 * Any task has to be enqueued before it get to execute on
775 * a CPU. So account for the time it spent waiting on the
776 * runqueue.
778 update_stats_wait_end(cfs_rq, se);
779 __dequeue_entity(cfs_rq, se);
782 update_stats_curr_start(cfs_rq, se);
783 cfs_rq->curr = se;
784 #ifdef CONFIG_SCHEDSTATS
786 * Track our maximum slice length, if the CPU's load is at
787 * least twice that of our own weight (i.e. dont track it
788 * when there are only lesser-weight tasks around):
790 if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
791 se->slice_max = max(se->slice_max,
792 se->sum_exec_runtime - se->prev_sum_exec_runtime);
794 #endif
795 se->prev_sum_exec_runtime = se->sum_exec_runtime;
798 static int
799 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);
801 static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
803 struct sched_entity *se = __pick_next_entity(cfs_rq);
805 if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, se) < 1)
806 return cfs_rq->next;
808 if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, se) < 1)
809 return cfs_rq->last;
811 return se;
814 static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
817 * If still on the runqueue then deactivate_task()
818 * was not called and update_curr() has to be done:
820 if (prev->on_rq)
821 update_curr(cfs_rq);
823 check_spread(cfs_rq, prev);
824 if (prev->on_rq) {
825 update_stats_wait_start(cfs_rq, prev);
826 /* Put 'current' back into the tree. */
827 __enqueue_entity(cfs_rq, prev);
829 cfs_rq->curr = NULL;
832 static void
833 entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
836 * Update run-time statistics of the 'current'.
838 update_curr(cfs_rq);
840 #ifdef CONFIG_SCHED_HRTICK
842 * queued ticks are scheduled to match the slice, so don't bother
843 * validating it and just reschedule.
845 if (queued) {
846 resched_task(rq_of(cfs_rq)->curr);
847 return;
850 * don't let the period tick interfere with the hrtick preemption
852 if (!sched_feat(DOUBLE_TICK) &&
853 hrtimer_active(&rq_of(cfs_rq)->hrtick_timer))
854 return;
855 #endif
857 if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT))
858 check_preempt_tick(cfs_rq, curr);
861 /**************************************************
862 * CFS operations on tasks:
865 #ifdef CONFIG_SCHED_HRTICK
866 static void hrtick_start_fair(struct rq *rq, struct task_struct *p)
868 struct sched_entity *se = &p->se;
869 struct cfs_rq *cfs_rq = cfs_rq_of(se);
871 WARN_ON(task_rq(p) != rq);
873 if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) {
874 u64 slice = sched_slice(cfs_rq, se);
875 u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime;
876 s64 delta = slice - ran;
878 if (delta < 0) {
879 if (rq->curr == p)
880 resched_task(p);
881 return;
885 * Don't schedule slices shorter than 10000ns, that just
886 * doesn't make sense. Rely on vruntime for fairness.
888 if (rq->curr != p)
889 delta = max_t(s64, 10000LL, delta);
891 hrtick_start(rq, delta);
896 * called from enqueue/dequeue and updates the hrtick when the
897 * current task is from our class and nr_running is low enough
898 * to matter.
900 static void hrtick_update(struct rq *rq)
902 struct task_struct *curr = rq->curr;
904 if (curr->sched_class != &fair_sched_class)
905 return;
907 if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
908 hrtick_start_fair(rq, curr);
910 #else /* !CONFIG_SCHED_HRTICK */
911 static inline void
912 hrtick_start_fair(struct rq *rq, struct task_struct *p)
916 static inline void hrtick_update(struct rq *rq)
919 #endif
922 * The enqueue_task method is called before nr_running is
923 * increased. Here we update the fair scheduling stats and
924 * then put the task into the rbtree:
926 static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup)
928 struct cfs_rq *cfs_rq;
929 struct sched_entity *se = &p->se;
931 for_each_sched_entity(se) {
932 if (se->on_rq)
933 break;
934 cfs_rq = cfs_rq_of(se);
935 enqueue_entity(cfs_rq, se, wakeup);
936 wakeup = 1;
939 hrtick_update(rq);
943 * The dequeue_task method is called before nr_running is
944 * decreased. We remove the task from the rbtree and
945 * update the fair scheduling stats:
947 static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep)
949 struct cfs_rq *cfs_rq;
950 struct sched_entity *se = &p->se;
952 for_each_sched_entity(se) {
953 cfs_rq = cfs_rq_of(se);
954 dequeue_entity(cfs_rq, se, sleep);
955 /* Don't dequeue parent if it has other entities besides us */
956 if (cfs_rq->load.weight)
957 break;
958 sleep = 1;
961 hrtick_update(rq);
965 * sched_yield() support is very simple - we dequeue and enqueue.
967 * If compat_yield is turned on then we requeue to the end of the tree.
969 static void yield_task_fair(struct rq *rq)
971 struct task_struct *curr = rq->curr;
972 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
973 struct sched_entity *rightmost, *se = &curr->se;
976 * Are we the only task in the tree?
978 if (unlikely(cfs_rq->nr_running == 1))
979 return;
981 clear_buddies(cfs_rq, se);
983 if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) {
984 update_rq_clock(rq);
986 * Update run-time statistics of the 'current'.
988 update_curr(cfs_rq);
990 return;
993 * Find the rightmost entry in the rbtree:
995 rightmost = __pick_last_entity(cfs_rq);
997 * Already in the rightmost position?
999 if (unlikely(!rightmost || rightmost->vruntime < se->vruntime))
1000 return;
1003 * Minimally necessary key value to be last in the tree:
1004 * Upon rescheduling, sched_class::put_prev_task() will place
1005 * 'current' within the tree based on its new key value.
1007 se->vruntime = rightmost->vruntime + 1;
1011 * wake_idle() will wake a task on an idle cpu if task->cpu is
1012 * not idle and an idle cpu is available. The span of cpus to
1013 * search starts with cpus closest then further out as needed,
1014 * so we always favor a closer, idle cpu.
1015 * Domains may include CPUs that are not usable for migration,
1016 * hence we need to mask them out (cpu_active_mask)
1018 * Returns the CPU we should wake onto.
1020 #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
1021 static int wake_idle(int cpu, struct task_struct *p)
1023 struct sched_domain *sd;
1024 int i;
1025 unsigned int chosen_wakeup_cpu;
1026 int this_cpu;
1029 * At POWERSAVINGS_BALANCE_WAKEUP level, if both this_cpu and prev_cpu
1030 * are idle and this is not a kernel thread and this task's affinity
1031 * allows it to be moved to preferred cpu, then just move!
1034 this_cpu = smp_processor_id();
1035 chosen_wakeup_cpu =
1036 cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu;
1038 if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP &&
1039 idle_cpu(cpu) && idle_cpu(this_cpu) &&
1040 p->mm && !(p->flags & PF_KTHREAD) &&
1041 cpu_isset(chosen_wakeup_cpu, p->cpus_allowed))
1042 return chosen_wakeup_cpu;
1045 * If it is idle, then it is the best cpu to run this task.
1047 * This cpu is also the best, if it has more than one task already.
1048 * Siblings must be also busy(in most cases) as they didn't already
1049 * pickup the extra load from this cpu and hence we need not check
1050 * sibling runqueue info. This will avoid the checks and cache miss
1051 * penalities associated with that.
1053 if (idle_cpu(cpu) || cpu_rq(cpu)->cfs.nr_running > 1)
1054 return cpu;
1056 for_each_domain(cpu, sd) {
1057 if ((sd->flags & SD_WAKE_IDLE)
1058 || ((sd->flags & SD_WAKE_IDLE_FAR)
1059 && !task_hot(p, task_rq(p)->clock, sd))) {
1060 for_each_cpu_and(i, sched_domain_span(sd),
1061 &p->cpus_allowed) {
1062 if (cpu_active(i) && idle_cpu(i)) {
1063 if (i != task_cpu(p)) {
1064 schedstat_inc(p,
1065 se.nr_wakeups_idle);
1067 return i;
1070 } else {
1071 break;
1074 return cpu;
1076 #else /* !ARCH_HAS_SCHED_WAKE_IDLE*/
1077 static inline int wake_idle(int cpu, struct task_struct *p)
1079 return cpu;
1081 #endif
1083 #ifdef CONFIG_SMP
1085 #ifdef CONFIG_FAIR_GROUP_SCHED
1087 * effective_load() calculates the load change as seen from the root_task_group
1089 * Adding load to a group doesn't make a group heavier, but can cause movement
1090 * of group shares between cpus. Assuming the shares were perfectly aligned one
1091 * can calculate the shift in shares.
1093 * The problem is that perfectly aligning the shares is rather expensive, hence
1094 * we try to avoid doing that too often - see update_shares(), which ratelimits
1095 * this change.
1097 * We compensate this by not only taking the current delta into account, but
1098 * also considering the delta between when the shares were last adjusted and
1099 * now.
1101 * We still saw a performance dip, some tracing learned us that between
1102 * cgroup:/ and cgroup:/foo balancing the number of affine wakeups increased
1103 * significantly. Therefore try to bias the error in direction of failing
1104 * the affine wakeup.
1107 static long effective_load(struct task_group *tg, int cpu,
1108 long wl, long wg)
1110 struct sched_entity *se = tg->se[cpu];
1112 if (!tg->parent)
1113 return wl;
1116 * By not taking the decrease of shares on the other cpu into
1117 * account our error leans towards reducing the affine wakeups.
1119 if (!wl && sched_feat(ASYM_EFF_LOAD))
1120 return wl;
1122 for_each_sched_entity(se) {
1123 long S, rw, s, a, b;
1124 long more_w;
1127 * Instead of using this increment, also add the difference
1128 * between when the shares were last updated and now.
1130 more_w = se->my_q->load.weight - se->my_q->rq_weight;
1131 wl += more_w;
1132 wg += more_w;
1134 S = se->my_q->tg->shares;
1135 s = se->my_q->shares;
1136 rw = se->my_q->rq_weight;
1138 a = S*(rw + wl);
1139 b = S*rw + s*wg;
1141 wl = s*(a-b);
1143 if (likely(b))
1144 wl /= b;
1147 * Assume the group is already running and will
1148 * thus already be accounted for in the weight.
1150 * That is, moving shares between CPUs, does not
1151 * alter the group weight.
1153 wg = 0;
1156 return wl;
1159 #else
1161 static inline unsigned long effective_load(struct task_group *tg, int cpu,
1162 unsigned long wl, unsigned long wg)
1164 return wl;
1167 #endif
1169 static int
1170 wake_affine(struct sched_domain *this_sd, struct rq *this_rq,
1171 struct task_struct *p, int prev_cpu, int this_cpu, int sync,
1172 int idx, unsigned long load, unsigned long this_load,
1173 unsigned int imbalance)
1175 struct task_struct *curr = this_rq->curr;
1176 struct task_group *tg;
1177 unsigned long tl = this_load;
1178 unsigned long tl_per_task;
1179 unsigned long weight;
1180 int balanced;
1182 if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS))
1183 return 0;
1185 if (sync && (curr->se.avg_overlap > sysctl_sched_migration_cost ||
1186 p->se.avg_overlap > sysctl_sched_migration_cost))
1187 sync = 0;
1190 * If sync wakeup then subtract the (maximum possible)
1191 * effect of the currently running task from the load
1192 * of the current CPU:
1194 if (sync) {
1195 tg = task_group(current);
1196 weight = current->se.load.weight;
1198 tl += effective_load(tg, this_cpu, -weight, -weight);
1199 load += effective_load(tg, prev_cpu, 0, -weight);
1202 tg = task_group(p);
1203 weight = p->se.load.weight;
1205 balanced = 100*(tl + effective_load(tg, this_cpu, weight, weight)) <=
1206 imbalance*(load + effective_load(tg, prev_cpu, 0, weight));
1209 * If the currently running task will sleep within
1210 * a reasonable amount of time then attract this newly
1211 * woken task:
1213 if (sync && balanced)
1214 return 1;
1216 schedstat_inc(p, se.nr_wakeups_affine_attempts);
1217 tl_per_task = cpu_avg_load_per_task(this_cpu);
1219 if (balanced || (tl <= load && tl + target_load(prev_cpu, idx) <=
1220 tl_per_task)) {
1222 * This domain has SD_WAKE_AFFINE and
1223 * p is cache cold in this domain, and
1224 * there is no bad imbalance.
1226 schedstat_inc(this_sd, ttwu_move_affine);
1227 schedstat_inc(p, se.nr_wakeups_affine);
1229 return 1;
1231 return 0;
1234 static int select_task_rq_fair(struct task_struct *p, int sync)
1236 struct sched_domain *sd, *this_sd = NULL;
1237 int prev_cpu, this_cpu, new_cpu;
1238 unsigned long load, this_load;
1239 struct rq *this_rq;
1240 unsigned int imbalance;
1241 int idx;
1243 prev_cpu = task_cpu(p);
1244 this_cpu = smp_processor_id();
1245 this_rq = cpu_rq(this_cpu);
1246 new_cpu = prev_cpu;
1248 if (prev_cpu == this_cpu)
1249 goto out;
1251 * 'this_sd' is the first domain that both
1252 * this_cpu and prev_cpu are present in:
1254 for_each_domain(this_cpu, sd) {
1255 if (cpumask_test_cpu(prev_cpu, sched_domain_span(sd))) {
1256 this_sd = sd;
1257 break;
1261 if (unlikely(!cpumask_test_cpu(this_cpu, &p->cpus_allowed)))
1262 goto out;
1265 * Check for affine wakeup and passive balancing possibilities.
1267 if (!this_sd)
1268 goto out;
1270 idx = this_sd->wake_idx;
1272 imbalance = 100 + (this_sd->imbalance_pct - 100) / 2;
1274 load = source_load(prev_cpu, idx);
1275 this_load = target_load(this_cpu, idx);
1277 if (wake_affine(this_sd, this_rq, p, prev_cpu, this_cpu, sync, idx,
1278 load, this_load, imbalance))
1279 return this_cpu;
1282 * Start passive balancing when half the imbalance_pct
1283 * limit is reached.
1285 if (this_sd->flags & SD_WAKE_BALANCE) {
1286 if (imbalance*this_load <= 100*load) {
1287 schedstat_inc(this_sd, ttwu_move_balance);
1288 schedstat_inc(p, se.nr_wakeups_passive);
1289 return this_cpu;
1293 out:
1294 return wake_idle(new_cpu, p);
1296 #endif /* CONFIG_SMP */
1298 static unsigned long wakeup_gran(struct sched_entity *se)
1300 unsigned long gran = sysctl_sched_wakeup_granularity;
1303 * More easily preempt - nice tasks, while not making it harder for
1304 * + nice tasks.
1306 if (!sched_feat(ASYM_GRAN) || se->load.weight > NICE_0_LOAD)
1307 gran = calc_delta_fair(sysctl_sched_wakeup_granularity, se);
1309 return gran;
1313 * Should 'se' preempt 'curr'.
1315 * |s1
1316 * |s2
1317 * |s3
1319 * |<--->|c
1321 * w(c, s1) = -1
1322 * w(c, s2) = 0
1323 * w(c, s3) = 1
1326 static int
1327 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
1329 s64 gran, vdiff = curr->vruntime - se->vruntime;
1331 if (vdiff <= 0)
1332 return -1;
1334 gran = wakeup_gran(curr);
1335 if (vdiff > gran)
1336 return 1;
1338 return 0;
1341 static void set_last_buddy(struct sched_entity *se)
1343 for_each_sched_entity(se)
1344 cfs_rq_of(se)->last = se;
1347 static void set_next_buddy(struct sched_entity *se)
1349 for_each_sched_entity(se)
1350 cfs_rq_of(se)->next = se;
1354 * Preempt the current task with a newly woken task if needed:
1356 static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync)
1358 struct task_struct *curr = rq->curr;
1359 struct sched_entity *se = &curr->se, *pse = &p->se;
1360 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
1362 update_curr(cfs_rq);
1364 if (unlikely(rt_prio(p->prio))) {
1365 resched_task(curr);
1366 return;
1369 if (unlikely(p->sched_class != &fair_sched_class))
1370 return;
1372 if (unlikely(se == pse))
1373 return;
1376 * Only set the backward buddy when the current task is still on the
1377 * rq. This can happen when a wakeup gets interleaved with schedule on
1378 * the ->pre_schedule() or idle_balance() point, either of which can
1379 * drop the rq lock.
1381 * Also, during early boot the idle thread is in the fair class, for
1382 * obvious reasons its a bad idea to schedule back to the idle thread.
1384 if (sched_feat(LAST_BUDDY) && likely(se->on_rq && curr != rq->idle))
1385 set_last_buddy(se);
1386 set_next_buddy(pse);
1389 * We can come here with TIF_NEED_RESCHED already set from new task
1390 * wake up path.
1392 if (test_tsk_need_resched(curr))
1393 return;
1396 * Batch tasks do not preempt (their preemption is driven by
1397 * the tick):
1399 if (unlikely(p->policy == SCHED_BATCH))
1400 return;
1402 if (!sched_feat(WAKEUP_PREEMPT))
1403 return;
1405 if (sched_feat(WAKEUP_OVERLAP) && (sync ||
1406 (se->avg_overlap < sysctl_sched_migration_cost &&
1407 pse->avg_overlap < sysctl_sched_migration_cost))) {
1408 resched_task(curr);
1409 return;
1412 find_matching_se(&se, &pse);
1414 while (se) {
1415 BUG_ON(!pse);
1417 if (wakeup_preempt_entity(se, pse) == 1) {
1418 resched_task(curr);
1419 break;
1422 se = parent_entity(se);
1423 pse = parent_entity(pse);
1427 static struct task_struct *pick_next_task_fair(struct rq *rq)
1429 struct task_struct *p;
1430 struct cfs_rq *cfs_rq = &rq->cfs;
1431 struct sched_entity *se;
1433 if (unlikely(!cfs_rq->nr_running))
1434 return NULL;
1436 do {
1437 se = pick_next_entity(cfs_rq);
1438 set_next_entity(cfs_rq, se);
1439 cfs_rq = group_cfs_rq(se);
1440 } while (cfs_rq);
1442 p = task_of(se);
1443 hrtick_start_fair(rq, p);
1445 return p;
1449 * Account for a descheduled task:
1451 static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
1453 struct sched_entity *se = &prev->se;
1454 struct cfs_rq *cfs_rq;
1456 for_each_sched_entity(se) {
1457 cfs_rq = cfs_rq_of(se);
1458 put_prev_entity(cfs_rq, se);
1462 #ifdef CONFIG_SMP
1463 /**************************************************
1464 * Fair scheduling class load-balancing methods:
1468 * Load-balancing iterator. Note: while the runqueue stays locked
1469 * during the whole iteration, the current task might be
1470 * dequeued so the iterator has to be dequeue-safe. Here we
1471 * achieve that by always pre-iterating before returning
1472 * the current task:
1474 static struct task_struct *
1475 __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next)
1477 struct task_struct *p = NULL;
1478 struct sched_entity *se;
1480 if (next == &cfs_rq->tasks)
1481 return NULL;
1483 se = list_entry(next, struct sched_entity, group_node);
1484 p = task_of(se);
1485 cfs_rq->balance_iterator = next->next;
1487 return p;
1490 static struct task_struct *load_balance_start_fair(void *arg)
1492 struct cfs_rq *cfs_rq = arg;
1494 return __load_balance_iterator(cfs_rq, cfs_rq->tasks.next);
1497 static struct task_struct *load_balance_next_fair(void *arg)
1499 struct cfs_rq *cfs_rq = arg;
1501 return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator);
1504 static unsigned long
1505 __load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1506 unsigned long max_load_move, struct sched_domain *sd,
1507 enum cpu_idle_type idle, int *all_pinned, int *this_best_prio,
1508 struct cfs_rq *cfs_rq)
1510 struct rq_iterator cfs_rq_iterator;
1512 cfs_rq_iterator.start = load_balance_start_fair;
1513 cfs_rq_iterator.next = load_balance_next_fair;
1514 cfs_rq_iterator.arg = cfs_rq;
1516 return balance_tasks(this_rq, this_cpu, busiest,
1517 max_load_move, sd, idle, all_pinned,
1518 this_best_prio, &cfs_rq_iterator);
1521 #ifdef CONFIG_FAIR_GROUP_SCHED
1522 static unsigned long
1523 load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1524 unsigned long max_load_move,
1525 struct sched_domain *sd, enum cpu_idle_type idle,
1526 int *all_pinned, int *this_best_prio)
1528 long rem_load_move = max_load_move;
1529 int busiest_cpu = cpu_of(busiest);
1530 struct task_group *tg;
1532 rcu_read_lock();
1533 update_h_load(busiest_cpu);
1535 list_for_each_entry_rcu(tg, &task_groups, list) {
1536 struct cfs_rq *busiest_cfs_rq = tg->cfs_rq[busiest_cpu];
1537 unsigned long busiest_h_load = busiest_cfs_rq->h_load;
1538 unsigned long busiest_weight = busiest_cfs_rq->load.weight;
1539 u64 rem_load, moved_load;
1542 * empty group
1544 if (!busiest_cfs_rq->task_weight)
1545 continue;
1547 rem_load = (u64)rem_load_move * busiest_weight;
1548 rem_load = div_u64(rem_load, busiest_h_load + 1);
1550 moved_load = __load_balance_fair(this_rq, this_cpu, busiest,
1551 rem_load, sd, idle, all_pinned, this_best_prio,
1552 tg->cfs_rq[busiest_cpu]);
1554 if (!moved_load)
1555 continue;
1557 moved_load *= busiest_h_load;
1558 moved_load = div_u64(moved_load, busiest_weight + 1);
1560 rem_load_move -= moved_load;
1561 if (rem_load_move < 0)
1562 break;
1564 rcu_read_unlock();
1566 return max_load_move - rem_load_move;
1568 #else
1569 static unsigned long
1570 load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1571 unsigned long max_load_move,
1572 struct sched_domain *sd, enum cpu_idle_type idle,
1573 int *all_pinned, int *this_best_prio)
1575 return __load_balance_fair(this_rq, this_cpu, busiest,
1576 max_load_move, sd, idle, all_pinned,
1577 this_best_prio, &busiest->cfs);
1579 #endif
1581 static int
1582 move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1583 struct sched_domain *sd, enum cpu_idle_type idle)
1585 struct cfs_rq *busy_cfs_rq;
1586 struct rq_iterator cfs_rq_iterator;
1588 cfs_rq_iterator.start = load_balance_start_fair;
1589 cfs_rq_iterator.next = load_balance_next_fair;
1591 for_each_leaf_cfs_rq(busiest, busy_cfs_rq) {
1593 * pass busy_cfs_rq argument into
1594 * load_balance_[start|next]_fair iterators
1596 cfs_rq_iterator.arg = busy_cfs_rq;
1597 if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle,
1598 &cfs_rq_iterator))
1599 return 1;
1602 return 0;
1604 #endif /* CONFIG_SMP */
1607 * scheduler tick hitting a task of our scheduling class:
1609 static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
1611 struct cfs_rq *cfs_rq;
1612 struct sched_entity *se = &curr->se;
1614 for_each_sched_entity(se) {
1615 cfs_rq = cfs_rq_of(se);
1616 entity_tick(cfs_rq, se, queued);
1621 * Share the fairness runtime between parent and child, thus the
1622 * total amount of pressure for CPU stays equal - new tasks
1623 * get a chance to run but frequent forkers are not allowed to
1624 * monopolize the CPU. Note: the parent runqueue is locked,
1625 * the child is not running yet.
1627 static void task_new_fair(struct rq *rq, struct task_struct *p)
1629 struct cfs_rq *cfs_rq = task_cfs_rq(p);
1630 struct sched_entity *se = &p->se, *curr = cfs_rq->curr;
1631 int this_cpu = smp_processor_id();
1633 sched_info_queued(p);
1635 update_curr(cfs_rq);
1636 place_entity(cfs_rq, se, 1);
1638 /* 'curr' will be NULL if the child belongs to a different group */
1639 if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) &&
1640 curr && curr->vruntime < se->vruntime) {
1642 * Upon rescheduling, sched_class::put_prev_task() will place
1643 * 'current' within the tree based on its new key value.
1645 swap(curr->vruntime, se->vruntime);
1646 resched_task(rq->curr);
1649 enqueue_task_fair(rq, p, 0);
1653 * Priority of the task has changed. Check to see if we preempt
1654 * the current task.
1656 static void prio_changed_fair(struct rq *rq, struct task_struct *p,
1657 int oldprio, int running)
1660 * Reschedule if we are currently running on this runqueue and
1661 * our priority decreased, or if we are not currently running on
1662 * this runqueue and our priority is higher than the current's
1664 if (running) {
1665 if (p->prio > oldprio)
1666 resched_task(rq->curr);
1667 } else
1668 check_preempt_curr(rq, p, 0);
1672 * We switched to the sched_fair class.
1674 static void switched_to_fair(struct rq *rq, struct task_struct *p,
1675 int running)
1678 * We were most likely switched from sched_rt, so
1679 * kick off the schedule if running, otherwise just see
1680 * if we can still preempt the current task.
1682 if (running)
1683 resched_task(rq->curr);
1684 else
1685 check_preempt_curr(rq, p, 0);
1688 /* Account for a task changing its policy or group.
1690 * This routine is mostly called to set cfs_rq->curr field when a task
1691 * migrates between groups/classes.
1693 static void set_curr_task_fair(struct rq *rq)
1695 struct sched_entity *se = &rq->curr->se;
1697 for_each_sched_entity(se)
1698 set_next_entity(cfs_rq_of(se), se);
1701 #ifdef CONFIG_FAIR_GROUP_SCHED
1702 static void moved_group_fair(struct task_struct *p)
1704 struct cfs_rq *cfs_rq = task_cfs_rq(p);
1706 update_curr(cfs_rq);
1707 place_entity(cfs_rq, &p->se, 1);
1709 #endif
1712 * All the scheduling class methods:
1714 static const struct sched_class fair_sched_class = {
1715 .next = &idle_sched_class,
1716 .enqueue_task = enqueue_task_fair,
1717 .dequeue_task = dequeue_task_fair,
1718 .yield_task = yield_task_fair,
1720 .check_preempt_curr = check_preempt_wakeup,
1722 .pick_next_task = pick_next_task_fair,
1723 .put_prev_task = put_prev_task_fair,
1725 #ifdef CONFIG_SMP
1726 .select_task_rq = select_task_rq_fair,
1728 .load_balance = load_balance_fair,
1729 .move_one_task = move_one_task_fair,
1730 #endif
1732 .set_curr_task = set_curr_task_fair,
1733 .task_tick = task_tick_fair,
1734 .task_new = task_new_fair,
1736 .prio_changed = prio_changed_fair,
1737 .switched_to = switched_to_fair,
1739 #ifdef CONFIG_FAIR_GROUP_SCHED
1740 .moved_group = moved_group_fair,
1741 #endif
1744 #ifdef CONFIG_SCHED_DEBUG
1745 static void print_cfs_stats(struct seq_file *m, int cpu)
1747 struct cfs_rq *cfs_rq;
1749 rcu_read_lock();
1750 for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
1751 print_cfs_rq(m, cpu, cfs_rq);
1752 rcu_read_unlock();
1754 #endif