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>
24 * Targeted preemption latency for CPU-bound tasks:
25 * (default: 20ms, units: nanoseconds)
27 * NOTE: this latency value is not the same as the concept of
28 * 'timeslice length' - timeslices in CFS are of variable length.
29 * (to see the precise effective timeslice length of your workload,
30 * run vmstat and monitor the context-switches field)
32 * On SMP systems the value of this is multiplied by the log2 of the
33 * number of CPUs. (i.e. factor 2x on 2-way systems, 3x on 4-way
34 * systems, 4x on 8-way systems, 5x on 16-way systems, etc.)
35 * Targeted preemption latency for CPU-bound tasks:
37 const_debug
unsigned int sysctl_sched_latency
= 20000000ULL;
40 * After fork, child runs first. (default) If set to 0 then
41 * parent will (try to) run first.
43 const_debug
unsigned int sysctl_sched_child_runs_first
= 1;
46 * Minimal preemption granularity for CPU-bound tasks:
47 * (default: 2 msec, units: nanoseconds)
49 unsigned int sysctl_sched_min_granularity __read_mostly
= 2000000ULL;
52 * sys_sched_yield() compat mode
54 * This option switches the agressive yield implementation of the
55 * old scheduler back on.
57 unsigned int __read_mostly sysctl_sched_compat_yield
;
60 * SCHED_BATCH wake-up granularity.
61 * (default: 25 msec, units: nanoseconds)
63 * This option delays the preemption effects of decoupled workloads
64 * and reduces their over-scheduling. Synchronous workloads will still
65 * have immediate wakeup/sleep latencies.
67 const_debug
unsigned int sysctl_sched_batch_wakeup_granularity
= 25000000UL;
70 * SCHED_OTHER wake-up granularity.
71 * (default: 1 msec, units: nanoseconds)
73 * This option delays the preemption effects of decoupled workloads
74 * and reduces their over-scheduling. Synchronous workloads will still
75 * have immediate wakeup/sleep latencies.
77 const_debug
unsigned int sysctl_sched_wakeup_granularity
= 2000000UL;
79 extern struct sched_class fair_sched_class
;
81 /**************************************************************
82 * CFS operations on generic schedulable entities:
85 #ifdef CONFIG_FAIR_GROUP_SCHED
87 /* cpu runqueue to which this cfs_rq is attached */
88 static inline struct rq
*rq_of(struct cfs_rq
*cfs_rq
)
93 /* An entity is a task if it doesn't "own" a runqueue */
94 #define entity_is_task(se) (!se->my_q)
96 #else /* CONFIG_FAIR_GROUP_SCHED */
98 static inline struct rq
*rq_of(struct cfs_rq
*cfs_rq
)
100 return container_of(cfs_rq
, struct rq
, cfs
);
103 #define entity_is_task(se) 1
105 #endif /* CONFIG_FAIR_GROUP_SCHED */
107 static inline struct task_struct
*task_of(struct sched_entity
*se
)
109 return container_of(se
, struct task_struct
, se
);
113 /**************************************************************
114 * Scheduling class tree data structure manipulation methods:
118 max_vruntime(u64 min_vruntime
, u64 vruntime
)
120 if ((vruntime
> min_vruntime
) ||
121 (min_vruntime
> (1ULL << 61) && vruntime
< (1ULL << 50)))
122 min_vruntime
= vruntime
;
128 set_leftmost(struct cfs_rq
*cfs_rq
, struct rb_node
*leftmost
)
130 struct sched_entity
*se
;
132 cfs_rq
->rb_leftmost
= leftmost
;
134 se
= rb_entry(leftmost
, struct sched_entity
, run_node
);
138 entity_key(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
140 return se
->vruntime
- cfs_rq
->min_vruntime
;
144 * Enqueue an entity into the rb-tree:
147 __enqueue_entity(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
149 struct rb_node
**link
= &cfs_rq
->tasks_timeline
.rb_node
;
150 struct rb_node
*parent
= NULL
;
151 struct sched_entity
*entry
;
152 s64 key
= entity_key(cfs_rq
, se
);
156 * Find the right place in the rbtree:
160 entry
= rb_entry(parent
, struct sched_entity
, run_node
);
162 * We dont care about collisions. Nodes with
163 * the same key stay together.
165 if (key
< entity_key(cfs_rq
, entry
)) {
166 link
= &parent
->rb_left
;
168 link
= &parent
->rb_right
;
174 * Maintain a cache of leftmost tree entries (it is frequently
178 set_leftmost(cfs_rq
, &se
->run_node
);
180 rb_link_node(&se
->run_node
, parent
, link
);
181 rb_insert_color(&se
->run_node
, &cfs_rq
->tasks_timeline
);
185 __dequeue_entity(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
187 if (cfs_rq
->rb_leftmost
== &se
->run_node
)
188 set_leftmost(cfs_rq
, rb_next(&se
->run_node
));
190 rb_erase(&se
->run_node
, &cfs_rq
->tasks_timeline
);
193 static inline struct rb_node
*first_fair(struct cfs_rq
*cfs_rq
)
195 return cfs_rq
->rb_leftmost
;
198 static struct sched_entity
*__pick_next_entity(struct cfs_rq
*cfs_rq
)
200 return rb_entry(first_fair(cfs_rq
), struct sched_entity
, run_node
);
203 static inline struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
)
205 struct rb_node
**link
= &cfs_rq
->tasks_timeline
.rb_node
;
206 struct sched_entity
*se
= NULL
;
207 struct rb_node
*parent
;
211 se
= rb_entry(parent
, struct sched_entity
, run_node
);
212 link
= &parent
->rb_right
;
218 /**************************************************************
219 * Scheduling class statistics methods:
222 static u64
__sched_period(unsigned long nr_running
)
224 u64 period
= sysctl_sched_latency
;
225 unsigned long nr_latency
=
226 sysctl_sched_latency
/ sysctl_sched_min_granularity
;
228 if (unlikely(nr_running
> nr_latency
)) {
229 period
*= nr_running
;
230 do_div(period
, nr_latency
);
236 static u64
sched_slice(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
238 u64 period
= __sched_period(cfs_rq
->nr_running
);
240 period
*= se
->load
.weight
;
241 do_div(period
, cfs_rq
->load
.weight
);
246 static u64
__sched_vslice(unsigned long nr_running
)
248 u64 period
= __sched_period(nr_running
);
250 do_div(period
, nr_running
);
256 * Update the current task's runtime statistics. Skip current tasks that
257 * are not in our scheduling class.
260 __update_curr(struct cfs_rq
*cfs_rq
, struct sched_entity
*curr
,
261 unsigned long delta_exec
)
263 unsigned long delta_exec_weighted
;
264 u64 next_vruntime
, min_vruntime
;
266 schedstat_set(curr
->exec_max
, max((u64
)delta_exec
, curr
->exec_max
));
268 curr
->sum_exec_runtime
+= delta_exec
;
269 schedstat_add(cfs_rq
, exec_clock
, delta_exec
);
270 delta_exec_weighted
= delta_exec
;
271 if (unlikely(curr
->load
.weight
!= NICE_0_LOAD
)) {
272 delta_exec_weighted
= calc_delta_fair(delta_exec_weighted
,
275 curr
->vruntime
+= delta_exec_weighted
;
278 * maintain cfs_rq->min_vruntime to be a monotonic increasing
279 * value tracking the leftmost vruntime in the tree.
281 if (first_fair(cfs_rq
)) {
282 next_vruntime
= __pick_next_entity(cfs_rq
)->vruntime
;
284 /* min_vruntime() := !max_vruntime() */
285 min_vruntime
= max_vruntime(curr
->vruntime
, next_vruntime
);
286 if (min_vruntime
== next_vruntime
)
287 min_vruntime
= curr
->vruntime
;
289 min_vruntime
= next_vruntime
;
291 min_vruntime
= curr
->vruntime
;
293 cfs_rq
->min_vruntime
=
294 max_vruntime(cfs_rq
->min_vruntime
, min_vruntime
);
297 static void update_curr(struct cfs_rq
*cfs_rq
)
299 struct sched_entity
*curr
= cfs_rq
->curr
;
300 u64 now
= rq_of(cfs_rq
)->clock
;
301 unsigned long delta_exec
;
307 * Get the amount of time the current task was running
308 * since the last time we changed load (this cannot
309 * overflow on 32 bits):
311 delta_exec
= (unsigned long)(now
- curr
->exec_start
);
313 __update_curr(cfs_rq
, curr
, delta_exec
);
314 curr
->exec_start
= now
;
318 update_stats_wait_start(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
320 schedstat_set(se
->wait_start
, rq_of(cfs_rq
)->clock
);
323 static inline unsigned long
324 calc_weighted(unsigned long delta
, struct sched_entity
*se
)
326 unsigned long weight
= se
->load
.weight
;
328 if (unlikely(weight
!= NICE_0_LOAD
))
329 return (u64
)delta
* se
->load
.weight
>> NICE_0_SHIFT
;
335 * Task is being enqueued - update stats:
337 static void update_stats_enqueue(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
340 * Are we enqueueing a waiting task? (for current tasks
341 * a dequeue/enqueue event is a NOP)
343 if (se
!= cfs_rq
->curr
)
344 update_stats_wait_start(cfs_rq
, se
);
348 update_stats_wait_end(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
350 schedstat_set(se
->wait_max
, max(se
->wait_max
,
351 rq_of(cfs_rq
)->clock
- se
->wait_start
));
352 schedstat_set(se
->wait_start
, 0);
356 update_stats_dequeue(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
360 * Mark the end of the wait period if dequeueing a
363 if (se
!= cfs_rq
->curr
)
364 update_stats_wait_end(cfs_rq
, se
);
368 * We are picking a new current task - update its stats:
371 update_stats_curr_start(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
374 * We are starting a new run period:
376 se
->exec_start
= rq_of(cfs_rq
)->clock
;
380 * We are descheduling a task - update its stats:
383 update_stats_curr_end(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
388 /**************************************************
389 * Scheduling class queueing methods:
393 account_entity_enqueue(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
395 update_load_add(&cfs_rq
->load
, se
->load
.weight
);
396 cfs_rq
->nr_running
++;
401 account_entity_dequeue(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
403 update_load_sub(&cfs_rq
->load
, se
->load
.weight
);
404 cfs_rq
->nr_running
--;
408 static void enqueue_sleeper(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
410 #ifdef CONFIG_SCHEDSTATS
411 if (se
->sleep_start
) {
412 u64 delta
= rq_of(cfs_rq
)->clock
- se
->sleep_start
;
417 if (unlikely(delta
> se
->sleep_max
))
418 se
->sleep_max
= delta
;
421 se
->sum_sleep_runtime
+= delta
;
423 if (se
->block_start
) {
424 u64 delta
= rq_of(cfs_rq
)->clock
- se
->block_start
;
429 if (unlikely(delta
> se
->block_max
))
430 se
->block_max
= delta
;
433 se
->sum_sleep_runtime
+= delta
;
436 * Blocking time is in units of nanosecs, so shift by 20 to
437 * get a milliseconds-range estimation of the amount of
438 * time that the task spent sleeping:
440 if (unlikely(prof_on
== SLEEP_PROFILING
)) {
441 struct task_struct
*tsk
= task_of(se
);
443 profile_hits(SLEEP_PROFILING
, (void *)get_wchan(tsk
),
450 static void check_spread(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
452 #ifdef CONFIG_SCHED_DEBUG
453 s64 d
= se
->vruntime
- cfs_rq
->min_vruntime
;
458 if (d
> 3*sysctl_sched_latency
)
459 schedstat_inc(cfs_rq
, nr_spread_over
);
464 place_entity(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
, int initial
)
468 vruntime
= cfs_rq
->min_vruntime
;
470 if (sched_feat(USE_TREE_AVG
)) {
471 struct sched_entity
*last
= __pick_last_entity(cfs_rq
);
473 vruntime
+= last
->vruntime
;
476 } else if (sched_feat(APPROX_AVG
) && cfs_rq
->nr_running
)
477 vruntime
+= __sched_vslice(cfs_rq
->nr_running
)/2;
479 if (initial
&& sched_feat(START_DEBIT
))
480 vruntime
+= __sched_vslice(cfs_rq
->nr_running
+ 1);
483 if (sched_feat(NEW_FAIR_SLEEPERS
))
484 vruntime
-= sysctl_sched_latency
;
486 vruntime
= max_t(s64
, vruntime
, se
->vruntime
);
489 se
->vruntime
= vruntime
;
494 enqueue_entity(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
, int wakeup
)
497 * Update the fair clock.
502 /* se->vruntime += cfs_rq->min_vruntime; */
503 place_entity(cfs_rq
, se
, 0);
504 enqueue_sleeper(cfs_rq
, se
);
507 update_stats_enqueue(cfs_rq
, se
);
508 check_spread(cfs_rq
, se
);
509 if (se
!= cfs_rq
->curr
)
510 __enqueue_entity(cfs_rq
, se
);
511 account_entity_enqueue(cfs_rq
, se
);
515 dequeue_entity(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
, int sleep
)
517 update_stats_dequeue(cfs_rq
, se
);
519 #ifdef CONFIG_SCHEDSTATS
520 if (entity_is_task(se
)) {
521 struct task_struct
*tsk
= task_of(se
);
523 if (tsk
->state
& TASK_INTERRUPTIBLE
)
524 se
->sleep_start
= rq_of(cfs_rq
)->clock
;
525 if (tsk
->state
& TASK_UNINTERRUPTIBLE
)
526 se
->block_start
= rq_of(cfs_rq
)->clock
;
531 if (se
!= cfs_rq
->curr
)
532 __dequeue_entity(cfs_rq
, se
);
533 account_entity_dequeue(cfs_rq
, se
);
537 * Preempt the current task with a newly woken task if needed:
540 check_preempt_tick(struct cfs_rq
*cfs_rq
, struct sched_entity
*curr
)
542 unsigned long ideal_runtime
, delta_exec
;
544 ideal_runtime
= sched_slice(cfs_rq
, curr
);
545 delta_exec
= curr
->sum_exec_runtime
- curr
->prev_sum_exec_runtime
;
546 if (delta_exec
> ideal_runtime
)
547 resched_task(rq_of(cfs_rq
)->curr
);
551 set_next_entity(struct cfs_rq
*cfs_rq
, struct sched_entity
*se
)
553 /* 'current' is not kept within the tree. */
556 * Any task has to be enqueued before it get to execute on
557 * a CPU. So account for the time it spent waiting on the
560 update_stats_wait_end(cfs_rq
, se
);
561 __dequeue_entity(cfs_rq
, se
);
564 update_stats_curr_start(cfs_rq
, se
);
566 #ifdef CONFIG_SCHEDSTATS
568 * Track our maximum slice length, if the CPU's load is at
569 * least twice that of our own weight (i.e. dont track it
570 * when there are only lesser-weight tasks around):
572 if (rq_of(cfs_rq
)->load
.weight
>= 2*se
->load
.weight
) {
573 se
->slice_max
= max(se
->slice_max
,
574 se
->sum_exec_runtime
- se
->prev_sum_exec_runtime
);
577 se
->prev_sum_exec_runtime
= se
->sum_exec_runtime
;
580 static struct sched_entity
*pick_next_entity(struct cfs_rq
*cfs_rq
)
582 struct sched_entity
*se
= __pick_next_entity(cfs_rq
);
584 set_next_entity(cfs_rq
, se
);
589 static void put_prev_entity(struct cfs_rq
*cfs_rq
, struct sched_entity
*prev
)
592 * If still on the runqueue then deactivate_task()
593 * was not called and update_curr() has to be done:
598 update_stats_curr_end(cfs_rq
, prev
);
600 check_spread(cfs_rq
, prev
);
602 update_stats_wait_start(cfs_rq
, prev
);
603 /* Put 'current' back into the tree. */
604 __enqueue_entity(cfs_rq
, prev
);
609 static void entity_tick(struct cfs_rq
*cfs_rq
, struct sched_entity
*curr
)
612 * Update run-time statistics of the 'current'.
616 if (cfs_rq
->nr_running
> 1)
617 check_preempt_tick(cfs_rq
, curr
);
620 /**************************************************
621 * CFS operations on tasks:
624 #ifdef CONFIG_FAIR_GROUP_SCHED
626 /* Walk up scheduling entities hierarchy */
627 #define for_each_sched_entity(se) \
628 for (; se; se = se->parent)
630 static inline struct cfs_rq
*task_cfs_rq(struct task_struct
*p
)
635 /* runqueue on which this entity is (to be) queued */
636 static inline struct cfs_rq
*cfs_rq_of(struct sched_entity
*se
)
641 /* runqueue "owned" by this group */
642 static inline struct cfs_rq
*group_cfs_rq(struct sched_entity
*grp
)
647 /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on
648 * another cpu ('this_cpu')
650 static inline struct cfs_rq
*cpu_cfs_rq(struct cfs_rq
*cfs_rq
, int this_cpu
)
652 return cfs_rq
->tg
->cfs_rq
[this_cpu
];
655 /* Iterate thr' all leaf cfs_rq's on a runqueue */
656 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
657 list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)
659 /* Do the two (enqueued) tasks belong to the same group ? */
660 static inline int is_same_group(struct task_struct
*curr
, struct task_struct
*p
)
662 if (curr
->se
.cfs_rq
== p
->se
.cfs_rq
)
668 #else /* CONFIG_FAIR_GROUP_SCHED */
670 #define for_each_sched_entity(se) \
671 for (; se; se = NULL)
673 static inline struct cfs_rq
*task_cfs_rq(struct task_struct
*p
)
675 return &task_rq(p
)->cfs
;
678 static inline struct cfs_rq
*cfs_rq_of(struct sched_entity
*se
)
680 struct task_struct
*p
= task_of(se
);
681 struct rq
*rq
= task_rq(p
);
686 /* runqueue "owned" by this group */
687 static inline struct cfs_rq
*group_cfs_rq(struct sched_entity
*grp
)
692 static inline struct cfs_rq
*cpu_cfs_rq(struct cfs_rq
*cfs_rq
, int this_cpu
)
694 return &cpu_rq(this_cpu
)->cfs
;
697 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
698 for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)
700 static inline int is_same_group(struct task_struct
*curr
, struct task_struct
*p
)
705 #endif /* CONFIG_FAIR_GROUP_SCHED */
708 * The enqueue_task method is called before nr_running is
709 * increased. Here we update the fair scheduling stats and
710 * then put the task into the rbtree:
712 static void enqueue_task_fair(struct rq
*rq
, struct task_struct
*p
, int wakeup
)
714 struct cfs_rq
*cfs_rq
;
715 struct sched_entity
*se
= &p
->se
;
717 for_each_sched_entity(se
) {
720 cfs_rq
= cfs_rq_of(se
);
721 enqueue_entity(cfs_rq
, se
, wakeup
);
726 * The dequeue_task method is called before nr_running is
727 * decreased. We remove the task from the rbtree and
728 * update the fair scheduling stats:
730 static void dequeue_task_fair(struct rq
*rq
, struct task_struct
*p
, int sleep
)
732 struct cfs_rq
*cfs_rq
;
733 struct sched_entity
*se
= &p
->se
;
735 for_each_sched_entity(se
) {
736 cfs_rq
= cfs_rq_of(se
);
737 dequeue_entity(cfs_rq
, se
, sleep
);
738 /* Don't dequeue parent if it has other entities besides us */
739 if (cfs_rq
->load
.weight
)
745 * sched_yield() support is very simple - we dequeue and enqueue.
747 * If compat_yield is turned on then we requeue to the end of the tree.
749 static void yield_task_fair(struct rq
*rq
)
751 struct cfs_rq
*cfs_rq
= task_cfs_rq(rq
->curr
);
752 struct rb_node
**link
= &cfs_rq
->tasks_timeline
.rb_node
;
753 struct sched_entity
*rightmost
, *se
= &rq
->curr
->se
;
754 struct rb_node
*parent
;
757 * Are we the only task in the tree?
759 if (unlikely(cfs_rq
->nr_running
== 1))
762 if (likely(!sysctl_sched_compat_yield
)) {
763 __update_rq_clock(rq
);
765 * Dequeue and enqueue the task to update its
766 * position within the tree:
768 dequeue_entity(cfs_rq
, se
, 0);
769 enqueue_entity(cfs_rq
, se
, 0);
774 * Find the rightmost entry in the rbtree:
778 link
= &parent
->rb_right
;
781 rightmost
= rb_entry(parent
, struct sched_entity
, run_node
);
783 * Already in the rightmost position?
785 if (unlikely(rightmost
== se
))
789 * Minimally necessary key value to be last in the tree:
791 se
->vruntime
= rightmost
->vruntime
+ 1;
793 if (cfs_rq
->rb_leftmost
== &se
->run_node
)
794 cfs_rq
->rb_leftmost
= rb_next(&se
->run_node
);
796 * Relink the task to the rightmost position:
798 rb_erase(&se
->run_node
, &cfs_rq
->tasks_timeline
);
799 rb_link_node(&se
->run_node
, parent
, link
);
800 rb_insert_color(&se
->run_node
, &cfs_rq
->tasks_timeline
);
804 * Preempt the current task with a newly woken task if needed:
806 static void check_preempt_wakeup(struct rq
*rq
, struct task_struct
*p
)
808 struct task_struct
*curr
= rq
->curr
;
809 struct cfs_rq
*cfs_rq
= task_cfs_rq(curr
);
811 if (unlikely(rt_prio(p
->prio
))) {
817 if (is_same_group(curr
, p
)) {
818 s64 delta
= curr
->se
.vruntime
- p
->se
.vruntime
;
820 if (delta
> (s64
)sysctl_sched_wakeup_granularity
)
825 static struct task_struct
*pick_next_task_fair(struct rq
*rq
)
827 struct cfs_rq
*cfs_rq
= &rq
->cfs
;
828 struct sched_entity
*se
;
830 if (unlikely(!cfs_rq
->nr_running
))
834 se
= pick_next_entity(cfs_rq
);
835 cfs_rq
= group_cfs_rq(se
);
842 * Account for a descheduled task:
844 static void put_prev_task_fair(struct rq
*rq
, struct task_struct
*prev
)
846 struct sched_entity
*se
= &prev
->se
;
847 struct cfs_rq
*cfs_rq
;
849 for_each_sched_entity(se
) {
850 cfs_rq
= cfs_rq_of(se
);
851 put_prev_entity(cfs_rq
, se
);
855 /**************************************************
856 * Fair scheduling class load-balancing methods:
860 * Load-balancing iterator. Note: while the runqueue stays locked
861 * during the whole iteration, the current task might be
862 * dequeued so the iterator has to be dequeue-safe. Here we
863 * achieve that by always pre-iterating before returning
866 static inline struct task_struct
*
867 __load_balance_iterator(struct cfs_rq
*cfs_rq
, struct rb_node
*curr
)
869 struct task_struct
*p
;
874 p
= rb_entry(curr
, struct task_struct
, se
.run_node
);
875 cfs_rq
->rb_load_balance_curr
= rb_next(curr
);
880 static struct task_struct
*load_balance_start_fair(void *arg
)
882 struct cfs_rq
*cfs_rq
= arg
;
884 return __load_balance_iterator(cfs_rq
, first_fair(cfs_rq
));
887 static struct task_struct
*load_balance_next_fair(void *arg
)
889 struct cfs_rq
*cfs_rq
= arg
;
891 return __load_balance_iterator(cfs_rq
, cfs_rq
->rb_load_balance_curr
);
894 #ifdef CONFIG_FAIR_GROUP_SCHED
895 static int cfs_rq_best_prio(struct cfs_rq
*cfs_rq
)
897 struct sched_entity
*curr
;
898 struct task_struct
*p
;
900 if (!cfs_rq
->nr_running
)
905 curr
= __pick_next_entity(cfs_rq
);
914 load_balance_fair(struct rq
*this_rq
, int this_cpu
, struct rq
*busiest
,
915 unsigned long max_nr_move
, unsigned long max_load_move
,
916 struct sched_domain
*sd
, enum cpu_idle_type idle
,
917 int *all_pinned
, int *this_best_prio
)
919 struct cfs_rq
*busy_cfs_rq
;
920 unsigned long load_moved
, total_nr_moved
= 0, nr_moved
;
921 long rem_load_move
= max_load_move
;
922 struct rq_iterator cfs_rq_iterator
;
924 cfs_rq_iterator
.start
= load_balance_start_fair
;
925 cfs_rq_iterator
.next
= load_balance_next_fair
;
927 for_each_leaf_cfs_rq(busiest
, busy_cfs_rq
) {
928 #ifdef CONFIG_FAIR_GROUP_SCHED
929 struct cfs_rq
*this_cfs_rq
;
931 unsigned long maxload
;
933 this_cfs_rq
= cpu_cfs_rq(busy_cfs_rq
, this_cpu
);
935 imbalance
= busy_cfs_rq
->load
.weight
- this_cfs_rq
->load
.weight
;
936 /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */
940 /* Don't pull more than imbalance/2 */
942 maxload
= min(rem_load_move
, imbalance
);
944 *this_best_prio
= cfs_rq_best_prio(this_cfs_rq
);
946 # define maxload rem_load_move
948 /* pass busy_cfs_rq argument into
949 * load_balance_[start|next]_fair iterators
951 cfs_rq_iterator
.arg
= busy_cfs_rq
;
952 nr_moved
= balance_tasks(this_rq
, this_cpu
, busiest
,
953 max_nr_move
, maxload
, sd
, idle
, all_pinned
,
954 &load_moved
, this_best_prio
, &cfs_rq_iterator
);
956 total_nr_moved
+= nr_moved
;
957 max_nr_move
-= nr_moved
;
958 rem_load_move
-= load_moved
;
960 if (max_nr_move
<= 0 || rem_load_move
<= 0)
964 return max_load_move
- rem_load_move
;
968 * scheduler tick hitting a task of our scheduling class:
970 static void task_tick_fair(struct rq
*rq
, struct task_struct
*curr
)
972 struct cfs_rq
*cfs_rq
;
973 struct sched_entity
*se
= &curr
->se
;
975 for_each_sched_entity(se
) {
976 cfs_rq
= cfs_rq_of(se
);
977 entity_tick(cfs_rq
, se
);
981 #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0)
984 * Share the fairness runtime between parent and child, thus the
985 * total amount of pressure for CPU stays equal - new tasks
986 * get a chance to run but frequent forkers are not allowed to
987 * monopolize the CPU. Note: the parent runqueue is locked,
988 * the child is not running yet.
990 static void task_new_fair(struct rq
*rq
, struct task_struct
*p
)
992 struct cfs_rq
*cfs_rq
= task_cfs_rq(p
);
993 struct sched_entity
*se
= &p
->se
, *curr
= cfs_rq
->curr
;
995 sched_info_queued(p
);
998 place_entity(cfs_rq
, se
, 1);
1000 if (sysctl_sched_child_runs_first
&&
1001 curr
->vruntime
< se
->vruntime
) {
1003 * Upon rescheduling, sched_class::put_prev_task() will place
1004 * 'current' within the tree based on its new key value.
1006 swap(curr
->vruntime
, se
->vruntime
);
1009 update_stats_enqueue(cfs_rq
, se
);
1010 check_spread(cfs_rq
, se
);
1011 check_spread(cfs_rq
, curr
);
1012 __enqueue_entity(cfs_rq
, se
);
1013 account_entity_enqueue(cfs_rq
, se
);
1014 resched_task(rq
->curr
);
1017 /* Account for a task changing its policy or group.
1019 * This routine is mostly called to set cfs_rq->curr field when a task
1020 * migrates between groups/classes.
1022 static void set_curr_task_fair(struct rq
*rq
)
1024 struct sched_entity
*se
= &rq
->curr
->se
;
1026 for_each_sched_entity(se
)
1027 set_next_entity(cfs_rq_of(se
), se
);
1031 * All the scheduling class methods:
1033 struct sched_class fair_sched_class __read_mostly
= {
1034 .enqueue_task
= enqueue_task_fair
,
1035 .dequeue_task
= dequeue_task_fair
,
1036 .yield_task
= yield_task_fair
,
1038 .check_preempt_curr
= check_preempt_wakeup
,
1040 .pick_next_task
= pick_next_task_fair
,
1041 .put_prev_task
= put_prev_task_fair
,
1043 .load_balance
= load_balance_fair
,
1045 .set_curr_task
= set_curr_task_fair
,
1046 .task_tick
= task_tick_fair
,
1047 .task_new
= task_new_fair
,
1050 #ifdef CONFIG_SCHED_DEBUG
1051 static void print_cfs_stats(struct seq_file
*m
, int cpu
)
1053 struct cfs_rq
*cfs_rq
;
1055 #ifdef CONFIG_FAIR_GROUP_SCHED
1056 print_cfs_rq(m
, cpu
, &cpu_rq(cpu
)->cfs
);
1058 for_each_leaf_cfs_rq(cpu_rq(cpu
), cfs_rq
)
1059 print_cfs_rq(m
, cpu
, cfs_rq
);