1 // SPDX-License-Identifier: GPL-2.0
3 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
8 int sched_rr_timeslice = RR_TIMESLICE;
9 int sysctl_sched_rr_timeslice = (MSEC_PER_SEC / HZ) * RR_TIMESLICE;
11 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
13 struct rt_bandwidth def_rt_bandwidth;
15 static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
17 struct rt_bandwidth *rt_b =
18 container_of(timer, struct rt_bandwidth, rt_period_timer);
22 raw_spin_lock(&rt_b->rt_runtime_lock);
24 overrun = hrtimer_forward_now(timer, rt_b->rt_period);
28 raw_spin_unlock(&rt_b->rt_runtime_lock);
29 idle = do_sched_rt_period_timer(rt_b, overrun);
30 raw_spin_lock(&rt_b->rt_runtime_lock);
33 rt_b->rt_period_active = 0;
34 raw_spin_unlock(&rt_b->rt_runtime_lock);
36 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
39 void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
41 rt_b->rt_period = ns_to_ktime(period);
42 rt_b->rt_runtime = runtime;
44 raw_spin_lock_init(&rt_b->rt_runtime_lock);
46 hrtimer_init(&rt_b->rt_period_timer,
47 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
48 rt_b->rt_period_timer.function = sched_rt_period_timer;
51 static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
53 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
56 raw_spin_lock(&rt_b->rt_runtime_lock);
57 if (!rt_b->rt_period_active) {
58 rt_b->rt_period_active = 1;
60 * SCHED_DEADLINE updates the bandwidth, as a run away
61 * RT task with a DL task could hog a CPU. But DL does
62 * not reset the period. If a deadline task was running
63 * without an RT task running, it can cause RT tasks to
64 * throttle when they start up. Kick the timer right away
65 * to update the period.
67 hrtimer_forward_now(&rt_b->rt_period_timer, ns_to_ktime(0));
68 hrtimer_start_expires(&rt_b->rt_period_timer, HRTIMER_MODE_ABS_PINNED);
70 raw_spin_unlock(&rt_b->rt_runtime_lock);
73 void init_rt_rq(struct rt_rq *rt_rq)
75 struct rt_prio_array *array;
78 array = &rt_rq->active;
79 for (i = 0; i < MAX_RT_PRIO; i++) {
80 INIT_LIST_HEAD(array->queue + i);
81 __clear_bit(i, array->bitmap);
83 /* delimiter for bitsearch: */
84 __set_bit(MAX_RT_PRIO, array->bitmap);
86 #if defined CONFIG_SMP
87 rt_rq->highest_prio.curr = MAX_RT_PRIO;
88 rt_rq->highest_prio.next = MAX_RT_PRIO;
89 rt_rq->rt_nr_migratory = 0;
90 rt_rq->overloaded = 0;
91 plist_head_init(&rt_rq->pushable_tasks);
92 #endif /* CONFIG_SMP */
93 /* We start is dequeued state, because no RT tasks are queued */
97 rt_rq->rt_throttled = 0;
98 rt_rq->rt_runtime = 0;
99 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
102 #ifdef CONFIG_RT_GROUP_SCHED
103 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
105 hrtimer_cancel(&rt_b->rt_period_timer);
108 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
110 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
112 #ifdef CONFIG_SCHED_DEBUG
113 WARN_ON_ONCE(!rt_entity_is_task(rt_se));
115 return container_of(rt_se, struct task_struct, rt);
118 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
123 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
128 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
130 struct rt_rq *rt_rq = rt_se->rt_rq;
135 void free_rt_sched_group(struct task_group *tg)
140 destroy_rt_bandwidth(&tg->rt_bandwidth);
142 for_each_possible_cpu(i) {
153 void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
154 struct sched_rt_entity *rt_se, int cpu,
155 struct sched_rt_entity *parent)
157 struct rq *rq = cpu_rq(cpu);
159 rt_rq->highest_prio.curr = MAX_RT_PRIO;
160 rt_rq->rt_nr_boosted = 0;
164 tg->rt_rq[cpu] = rt_rq;
165 tg->rt_se[cpu] = rt_se;
171 rt_se->rt_rq = &rq->rt;
173 rt_se->rt_rq = parent->my_q;
176 rt_se->parent = parent;
177 INIT_LIST_HEAD(&rt_se->run_list);
180 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
183 struct sched_rt_entity *rt_se;
186 tg->rt_rq = kcalloc(nr_cpu_ids, sizeof(rt_rq), GFP_KERNEL);
189 tg->rt_se = kcalloc(nr_cpu_ids, sizeof(rt_se), GFP_KERNEL);
193 init_rt_bandwidth(&tg->rt_bandwidth,
194 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
196 for_each_possible_cpu(i) {
197 rt_rq = kzalloc_node(sizeof(struct rt_rq),
198 GFP_KERNEL, cpu_to_node(i));
202 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
203 GFP_KERNEL, cpu_to_node(i));
208 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
209 init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
220 #else /* CONFIG_RT_GROUP_SCHED */
222 #define rt_entity_is_task(rt_se) (1)
224 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
226 return container_of(rt_se, struct task_struct, rt);
229 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
231 return container_of(rt_rq, struct rq, rt);
234 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
236 struct task_struct *p = rt_task_of(rt_se);
241 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
243 struct rq *rq = rq_of_rt_se(rt_se);
248 void free_rt_sched_group(struct task_group *tg) { }
250 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
254 #endif /* CONFIG_RT_GROUP_SCHED */
258 static void pull_rt_task(struct rq *this_rq);
260 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
262 /* Try to pull RT tasks here if we lower this rq's prio */
263 return rq->rt.highest_prio.curr > prev->prio;
266 static inline int rt_overloaded(struct rq *rq)
268 return atomic_read(&rq->rd->rto_count);
271 static inline void rt_set_overload(struct rq *rq)
276 cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
278 * Make sure the mask is visible before we set
279 * the overload count. That is checked to determine
280 * if we should look at the mask. It would be a shame
281 * if we looked at the mask, but the mask was not
284 * Matched by the barrier in pull_rt_task().
287 atomic_inc(&rq->rd->rto_count);
290 static inline void rt_clear_overload(struct rq *rq)
295 /* the order here really doesn't matter */
296 atomic_dec(&rq->rd->rto_count);
297 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
300 static void update_rt_migration(struct rt_rq *rt_rq)
302 if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
303 if (!rt_rq->overloaded) {
304 rt_set_overload(rq_of_rt_rq(rt_rq));
305 rt_rq->overloaded = 1;
307 } else if (rt_rq->overloaded) {
308 rt_clear_overload(rq_of_rt_rq(rt_rq));
309 rt_rq->overloaded = 0;
313 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
315 struct task_struct *p;
317 if (!rt_entity_is_task(rt_se))
320 p = rt_task_of(rt_se);
321 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
323 rt_rq->rt_nr_total++;
324 if (p->nr_cpus_allowed > 1)
325 rt_rq->rt_nr_migratory++;
327 update_rt_migration(rt_rq);
330 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
332 struct task_struct *p;
334 if (!rt_entity_is_task(rt_se))
337 p = rt_task_of(rt_se);
338 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
340 rt_rq->rt_nr_total--;
341 if (p->nr_cpus_allowed > 1)
342 rt_rq->rt_nr_migratory--;
344 update_rt_migration(rt_rq);
347 static inline int has_pushable_tasks(struct rq *rq)
349 return !plist_head_empty(&rq->rt.pushable_tasks);
352 static DEFINE_PER_CPU(struct callback_head, rt_push_head);
353 static DEFINE_PER_CPU(struct callback_head, rt_pull_head);
355 static void push_rt_tasks(struct rq *);
356 static void pull_rt_task(struct rq *);
358 static inline void rt_queue_push_tasks(struct rq *rq)
360 if (!has_pushable_tasks(rq))
363 queue_balance_callback(rq, &per_cpu(rt_push_head, rq->cpu), push_rt_tasks);
366 static inline void rt_queue_pull_task(struct rq *rq)
368 queue_balance_callback(rq, &per_cpu(rt_pull_head, rq->cpu), pull_rt_task);
371 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
373 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
374 plist_node_init(&p->pushable_tasks, p->prio);
375 plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
377 /* Update the highest prio pushable task */
378 if (p->prio < rq->rt.highest_prio.next)
379 rq->rt.highest_prio.next = p->prio;
382 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
384 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
386 /* Update the new highest prio pushable task */
387 if (has_pushable_tasks(rq)) {
388 p = plist_first_entry(&rq->rt.pushable_tasks,
389 struct task_struct, pushable_tasks);
390 rq->rt.highest_prio.next = p->prio;
392 rq->rt.highest_prio.next = MAX_RT_PRIO;
397 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
401 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
406 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
411 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
415 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
420 static inline void pull_rt_task(struct rq *this_rq)
424 static inline void rt_queue_push_tasks(struct rq *rq)
427 #endif /* CONFIG_SMP */
429 static void enqueue_top_rt_rq(struct rt_rq *rt_rq);
430 static void dequeue_top_rt_rq(struct rt_rq *rt_rq);
432 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
437 #ifdef CONFIG_RT_GROUP_SCHED
439 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
444 return rt_rq->rt_runtime;
447 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
449 return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
452 typedef struct task_group *rt_rq_iter_t;
454 static inline struct task_group *next_task_group(struct task_group *tg)
457 tg = list_entry_rcu(tg->list.next,
458 typeof(struct task_group), list);
459 } while (&tg->list != &task_groups && task_group_is_autogroup(tg));
461 if (&tg->list == &task_groups)
467 #define for_each_rt_rq(rt_rq, iter, rq) \
468 for (iter = container_of(&task_groups, typeof(*iter), list); \
469 (iter = next_task_group(iter)) && \
470 (rt_rq = iter->rt_rq[cpu_of(rq)]);)
472 #define for_each_sched_rt_entity(rt_se) \
473 for (; rt_se; rt_se = rt_se->parent)
475 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
480 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
481 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
483 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
485 struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
486 struct rq *rq = rq_of_rt_rq(rt_rq);
487 struct sched_rt_entity *rt_se;
489 int cpu = cpu_of(rq);
491 rt_se = rt_rq->tg->rt_se[cpu];
493 if (rt_rq->rt_nr_running) {
495 enqueue_top_rt_rq(rt_rq);
496 else if (!on_rt_rq(rt_se))
497 enqueue_rt_entity(rt_se, 0);
499 if (rt_rq->highest_prio.curr < curr->prio)
504 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
506 struct sched_rt_entity *rt_se;
507 int cpu = cpu_of(rq_of_rt_rq(rt_rq));
509 rt_se = rt_rq->tg->rt_se[cpu];
512 dequeue_top_rt_rq(rt_rq);
513 /* Kick cpufreq (see the comment in kernel/sched/sched.h). */
514 cpufreq_update_util(rq_of_rt_rq(rt_rq), 0);
516 else if (on_rt_rq(rt_se))
517 dequeue_rt_entity(rt_se, 0);
520 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
522 return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
525 static int rt_se_boosted(struct sched_rt_entity *rt_se)
527 struct rt_rq *rt_rq = group_rt_rq(rt_se);
528 struct task_struct *p;
531 return !!rt_rq->rt_nr_boosted;
533 p = rt_task_of(rt_se);
534 return p->prio != p->normal_prio;
538 static inline const struct cpumask *sched_rt_period_mask(void)
540 return this_rq()->rd->span;
543 static inline const struct cpumask *sched_rt_period_mask(void)
545 return cpu_online_mask;
550 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
552 return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
555 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
557 return &rt_rq->tg->rt_bandwidth;
560 #else /* !CONFIG_RT_GROUP_SCHED */
562 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
564 return rt_rq->rt_runtime;
567 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
569 return ktime_to_ns(def_rt_bandwidth.rt_period);
572 typedef struct rt_rq *rt_rq_iter_t;
574 #define for_each_rt_rq(rt_rq, iter, rq) \
575 for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
577 #define for_each_sched_rt_entity(rt_se) \
578 for (; rt_se; rt_se = NULL)
580 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
585 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
587 struct rq *rq = rq_of_rt_rq(rt_rq);
589 if (!rt_rq->rt_nr_running)
592 enqueue_top_rt_rq(rt_rq);
596 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
598 dequeue_top_rt_rq(rt_rq);
601 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
603 return rt_rq->rt_throttled;
606 static inline const struct cpumask *sched_rt_period_mask(void)
608 return cpu_online_mask;
612 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
614 return &cpu_rq(cpu)->rt;
617 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
619 return &def_rt_bandwidth;
622 #endif /* CONFIG_RT_GROUP_SCHED */
624 bool sched_rt_bandwidth_account(struct rt_rq *rt_rq)
626 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
628 return (hrtimer_active(&rt_b->rt_period_timer) ||
629 rt_rq->rt_time < rt_b->rt_runtime);
634 * We ran out of runtime, see if we can borrow some from our neighbours.
636 static void do_balance_runtime(struct rt_rq *rt_rq)
638 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
639 struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd;
643 weight = cpumask_weight(rd->span);
645 raw_spin_lock(&rt_b->rt_runtime_lock);
646 rt_period = ktime_to_ns(rt_b->rt_period);
647 for_each_cpu(i, rd->span) {
648 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
654 raw_spin_lock(&iter->rt_runtime_lock);
656 * Either all rqs have inf runtime and there's nothing to steal
657 * or __disable_runtime() below sets a specific rq to inf to
658 * indicate its been disabled and disalow stealing.
660 if (iter->rt_runtime == RUNTIME_INF)
664 * From runqueues with spare time, take 1/n part of their
665 * spare time, but no more than our period.
667 diff = iter->rt_runtime - iter->rt_time;
669 diff = div_u64((u64)diff, weight);
670 if (rt_rq->rt_runtime + diff > rt_period)
671 diff = rt_period - rt_rq->rt_runtime;
672 iter->rt_runtime -= diff;
673 rt_rq->rt_runtime += diff;
674 if (rt_rq->rt_runtime == rt_period) {
675 raw_spin_unlock(&iter->rt_runtime_lock);
680 raw_spin_unlock(&iter->rt_runtime_lock);
682 raw_spin_unlock(&rt_b->rt_runtime_lock);
686 * Ensure this RQ takes back all the runtime it lend to its neighbours.
688 static void __disable_runtime(struct rq *rq)
690 struct root_domain *rd = rq->rd;
694 if (unlikely(!scheduler_running))
697 for_each_rt_rq(rt_rq, iter, rq) {
698 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
702 raw_spin_lock(&rt_b->rt_runtime_lock);
703 raw_spin_lock(&rt_rq->rt_runtime_lock);
705 * Either we're all inf and nobody needs to borrow, or we're
706 * already disabled and thus have nothing to do, or we have
707 * exactly the right amount of runtime to take out.
709 if (rt_rq->rt_runtime == RUNTIME_INF ||
710 rt_rq->rt_runtime == rt_b->rt_runtime)
712 raw_spin_unlock(&rt_rq->rt_runtime_lock);
715 * Calculate the difference between what we started out with
716 * and what we current have, that's the amount of runtime
717 * we lend and now have to reclaim.
719 want = rt_b->rt_runtime - rt_rq->rt_runtime;
722 * Greedy reclaim, take back as much as we can.
724 for_each_cpu(i, rd->span) {
725 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
729 * Can't reclaim from ourselves or disabled runqueues.
731 if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
734 raw_spin_lock(&iter->rt_runtime_lock);
736 diff = min_t(s64, iter->rt_runtime, want);
737 iter->rt_runtime -= diff;
740 iter->rt_runtime -= want;
743 raw_spin_unlock(&iter->rt_runtime_lock);
749 raw_spin_lock(&rt_rq->rt_runtime_lock);
751 * We cannot be left wanting - that would mean some runtime
752 * leaked out of the system.
757 * Disable all the borrow logic by pretending we have inf
758 * runtime - in which case borrowing doesn't make sense.
760 rt_rq->rt_runtime = RUNTIME_INF;
761 rt_rq->rt_throttled = 0;
762 raw_spin_unlock(&rt_rq->rt_runtime_lock);
763 raw_spin_unlock(&rt_b->rt_runtime_lock);
765 /* Make rt_rq available for pick_next_task() */
766 sched_rt_rq_enqueue(rt_rq);
770 static void __enable_runtime(struct rq *rq)
775 if (unlikely(!scheduler_running))
779 * Reset each runqueue's bandwidth settings
781 for_each_rt_rq(rt_rq, iter, rq) {
782 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
784 raw_spin_lock(&rt_b->rt_runtime_lock);
785 raw_spin_lock(&rt_rq->rt_runtime_lock);
786 rt_rq->rt_runtime = rt_b->rt_runtime;
788 rt_rq->rt_throttled = 0;
789 raw_spin_unlock(&rt_rq->rt_runtime_lock);
790 raw_spin_unlock(&rt_b->rt_runtime_lock);
794 static void balance_runtime(struct rt_rq *rt_rq)
796 if (!sched_feat(RT_RUNTIME_SHARE))
799 if (rt_rq->rt_time > rt_rq->rt_runtime) {
800 raw_spin_unlock(&rt_rq->rt_runtime_lock);
801 do_balance_runtime(rt_rq);
802 raw_spin_lock(&rt_rq->rt_runtime_lock);
805 #else /* !CONFIG_SMP */
806 static inline void balance_runtime(struct rt_rq *rt_rq) {}
807 #endif /* CONFIG_SMP */
809 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
811 int i, idle = 1, throttled = 0;
812 const struct cpumask *span;
814 span = sched_rt_period_mask();
815 #ifdef CONFIG_RT_GROUP_SCHED
817 * FIXME: isolated CPUs should really leave the root task group,
818 * whether they are isolcpus or were isolated via cpusets, lest
819 * the timer run on a CPU which does not service all runqueues,
820 * potentially leaving other CPUs indefinitely throttled. If
821 * isolation is really required, the user will turn the throttle
822 * off to kill the perturbations it causes anyway. Meanwhile,
823 * this maintains functionality for boot and/or troubleshooting.
825 if (rt_b == &root_task_group.rt_bandwidth)
826 span = cpu_online_mask;
828 for_each_cpu(i, span) {
830 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
831 struct rq *rq = rq_of_rt_rq(rt_rq);
835 * When span == cpu_online_mask, taking each rq->lock
836 * can be time-consuming. Try to avoid it when possible.
838 raw_spin_lock(&rt_rq->rt_runtime_lock);
839 skip = !rt_rq->rt_time && !rt_rq->rt_nr_running;
840 raw_spin_unlock(&rt_rq->rt_runtime_lock);
844 raw_spin_lock(&rq->lock);
847 if (rt_rq->rt_time) {
850 raw_spin_lock(&rt_rq->rt_runtime_lock);
851 if (rt_rq->rt_throttled)
852 balance_runtime(rt_rq);
853 runtime = rt_rq->rt_runtime;
854 rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
855 if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
856 rt_rq->rt_throttled = 0;
860 * When we're idle and a woken (rt) task is
861 * throttled check_preempt_curr() will set
862 * skip_update and the time between the wakeup
863 * and this unthrottle will get accounted as
866 if (rt_rq->rt_nr_running && rq->curr == rq->idle)
867 rq_clock_cancel_skipupdate(rq);
869 if (rt_rq->rt_time || rt_rq->rt_nr_running)
871 raw_spin_unlock(&rt_rq->rt_runtime_lock);
872 } else if (rt_rq->rt_nr_running) {
874 if (!rt_rq_throttled(rt_rq))
877 if (rt_rq->rt_throttled)
881 sched_rt_rq_enqueue(rt_rq);
882 raw_spin_unlock(&rq->lock);
885 if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
891 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
893 #ifdef CONFIG_RT_GROUP_SCHED
894 struct rt_rq *rt_rq = group_rt_rq(rt_se);
897 return rt_rq->highest_prio.curr;
900 return rt_task_of(rt_se)->prio;
903 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
905 u64 runtime = sched_rt_runtime(rt_rq);
907 if (rt_rq->rt_throttled)
908 return rt_rq_throttled(rt_rq);
910 if (runtime >= sched_rt_period(rt_rq))
913 balance_runtime(rt_rq);
914 runtime = sched_rt_runtime(rt_rq);
915 if (runtime == RUNTIME_INF)
918 if (rt_rq->rt_time > runtime) {
919 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
922 * Don't actually throttle groups that have no runtime assigned
923 * but accrue some time due to boosting.
925 if (likely(rt_b->rt_runtime)) {
926 rt_rq->rt_throttled = 1;
927 printk_deferred_once("sched: RT throttling activated\n");
930 * In case we did anyway, make it go away,
931 * replenishment is a joke, since it will replenish us
937 if (rt_rq_throttled(rt_rq)) {
938 sched_rt_rq_dequeue(rt_rq);
947 * Update the current task's runtime statistics. Skip current tasks that
948 * are not in our scheduling class.
950 static void update_curr_rt(struct rq *rq)
952 struct task_struct *curr = rq->curr;
953 struct sched_rt_entity *rt_se = &curr->rt;
957 if (curr->sched_class != &rt_sched_class)
960 now = rq_clock_task(rq);
961 delta_exec = now - curr->se.exec_start;
962 if (unlikely((s64)delta_exec <= 0))
965 schedstat_set(curr->se.statistics.exec_max,
966 max(curr->se.statistics.exec_max, delta_exec));
968 curr->se.sum_exec_runtime += delta_exec;
969 account_group_exec_runtime(curr, delta_exec);
971 curr->se.exec_start = now;
972 cgroup_account_cputime(curr, delta_exec);
974 sched_rt_avg_update(rq, delta_exec);
976 if (!rt_bandwidth_enabled())
979 for_each_sched_rt_entity(rt_se) {
980 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
982 if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
983 raw_spin_lock(&rt_rq->rt_runtime_lock);
984 rt_rq->rt_time += delta_exec;
985 if (sched_rt_runtime_exceeded(rt_rq))
987 raw_spin_unlock(&rt_rq->rt_runtime_lock);
993 dequeue_top_rt_rq(struct rt_rq *rt_rq)
995 struct rq *rq = rq_of_rt_rq(rt_rq);
997 BUG_ON(&rq->rt != rt_rq);
999 if (!rt_rq->rt_queued)
1002 BUG_ON(!rq->nr_running);
1004 sub_nr_running(rq, rt_rq->rt_nr_running);
1005 rt_rq->rt_queued = 0;
1010 enqueue_top_rt_rq(struct rt_rq *rt_rq)
1012 struct rq *rq = rq_of_rt_rq(rt_rq);
1014 BUG_ON(&rq->rt != rt_rq);
1016 if (rt_rq->rt_queued)
1019 if (rt_rq_throttled(rt_rq))
1022 if (rt_rq->rt_nr_running) {
1023 add_nr_running(rq, rt_rq->rt_nr_running);
1024 rt_rq->rt_queued = 1;
1027 /* Kick cpufreq (see the comment in kernel/sched/sched.h). */
1028 cpufreq_update_util(rq, 0);
1031 #if defined CONFIG_SMP
1034 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1036 struct rq *rq = rq_of_rt_rq(rt_rq);
1038 #ifdef CONFIG_RT_GROUP_SCHED
1040 * Change rq's cpupri only if rt_rq is the top queue.
1042 if (&rq->rt != rt_rq)
1045 if (rq->online && prio < prev_prio)
1046 cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
1050 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1052 struct rq *rq = rq_of_rt_rq(rt_rq);
1054 #ifdef CONFIG_RT_GROUP_SCHED
1056 * Change rq's cpupri only if rt_rq is the top queue.
1058 if (&rq->rt != rt_rq)
1061 if (rq->online && rt_rq->highest_prio.curr != prev_prio)
1062 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
1065 #else /* CONFIG_SMP */
1068 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1070 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1072 #endif /* CONFIG_SMP */
1074 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
1076 inc_rt_prio(struct rt_rq *rt_rq, int prio)
1078 int prev_prio = rt_rq->highest_prio.curr;
1080 if (prio < prev_prio)
1081 rt_rq->highest_prio.curr = prio;
1083 inc_rt_prio_smp(rt_rq, prio, prev_prio);
1087 dec_rt_prio(struct rt_rq *rt_rq, int prio)
1089 int prev_prio = rt_rq->highest_prio.curr;
1091 if (rt_rq->rt_nr_running) {
1093 WARN_ON(prio < prev_prio);
1096 * This may have been our highest task, and therefore
1097 * we may have some recomputation to do
1099 if (prio == prev_prio) {
1100 struct rt_prio_array *array = &rt_rq->active;
1102 rt_rq->highest_prio.curr =
1103 sched_find_first_bit(array->bitmap);
1107 rt_rq->highest_prio.curr = MAX_RT_PRIO;
1109 dec_rt_prio_smp(rt_rq, prio, prev_prio);
1114 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
1115 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
1117 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1119 #ifdef CONFIG_RT_GROUP_SCHED
1122 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1124 if (rt_se_boosted(rt_se))
1125 rt_rq->rt_nr_boosted++;
1128 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
1132 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1134 if (rt_se_boosted(rt_se))
1135 rt_rq->rt_nr_boosted--;
1137 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
1140 #else /* CONFIG_RT_GROUP_SCHED */
1143 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1145 start_rt_bandwidth(&def_rt_bandwidth);
1149 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
1151 #endif /* CONFIG_RT_GROUP_SCHED */
1154 unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se)
1156 struct rt_rq *group_rq = group_rt_rq(rt_se);
1159 return group_rq->rt_nr_running;
1165 unsigned int rt_se_rr_nr_running(struct sched_rt_entity *rt_se)
1167 struct rt_rq *group_rq = group_rt_rq(rt_se);
1168 struct task_struct *tsk;
1171 return group_rq->rr_nr_running;
1173 tsk = rt_task_of(rt_se);
1175 return (tsk->policy == SCHED_RR) ? 1 : 0;
1179 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1181 int prio = rt_se_prio(rt_se);
1183 WARN_ON(!rt_prio(prio));
1184 rt_rq->rt_nr_running += rt_se_nr_running(rt_se);
1185 rt_rq->rr_nr_running += rt_se_rr_nr_running(rt_se);
1187 inc_rt_prio(rt_rq, prio);
1188 inc_rt_migration(rt_se, rt_rq);
1189 inc_rt_group(rt_se, rt_rq);
1193 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1195 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
1196 WARN_ON(!rt_rq->rt_nr_running);
1197 rt_rq->rt_nr_running -= rt_se_nr_running(rt_se);
1198 rt_rq->rr_nr_running -= rt_se_rr_nr_running(rt_se);
1200 dec_rt_prio(rt_rq, rt_se_prio(rt_se));
1201 dec_rt_migration(rt_se, rt_rq);
1202 dec_rt_group(rt_se, rt_rq);
1206 * Change rt_se->run_list location unless SAVE && !MOVE
1208 * assumes ENQUEUE/DEQUEUE flags match
1210 static inline bool move_entity(unsigned int flags)
1212 if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
1218 static void __delist_rt_entity(struct sched_rt_entity *rt_se, struct rt_prio_array *array)
1220 list_del_init(&rt_se->run_list);
1222 if (list_empty(array->queue + rt_se_prio(rt_se)))
1223 __clear_bit(rt_se_prio(rt_se), array->bitmap);
1228 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1230 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1231 struct rt_prio_array *array = &rt_rq->active;
1232 struct rt_rq *group_rq = group_rt_rq(rt_se);
1233 struct list_head *queue = array->queue + rt_se_prio(rt_se);
1236 * Don't enqueue the group if its throttled, or when empty.
1237 * The latter is a consequence of the former when a child group
1238 * get throttled and the current group doesn't have any other
1241 if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) {
1243 __delist_rt_entity(rt_se, array);
1247 if (move_entity(flags)) {
1248 WARN_ON_ONCE(rt_se->on_list);
1249 if (flags & ENQUEUE_HEAD)
1250 list_add(&rt_se->run_list, queue);
1252 list_add_tail(&rt_se->run_list, queue);
1254 __set_bit(rt_se_prio(rt_se), array->bitmap);
1259 inc_rt_tasks(rt_se, rt_rq);
1262 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1264 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1265 struct rt_prio_array *array = &rt_rq->active;
1267 if (move_entity(flags)) {
1268 WARN_ON_ONCE(!rt_se->on_list);
1269 __delist_rt_entity(rt_se, array);
1273 dec_rt_tasks(rt_se, rt_rq);
1277 * Because the prio of an upper entry depends on the lower
1278 * entries, we must remove entries top - down.
1280 static void dequeue_rt_stack(struct sched_rt_entity *rt_se, unsigned int flags)
1282 struct sched_rt_entity *back = NULL;
1284 for_each_sched_rt_entity(rt_se) {
1289 dequeue_top_rt_rq(rt_rq_of_se(back));
1291 for (rt_se = back; rt_se; rt_se = rt_se->back) {
1292 if (on_rt_rq(rt_se))
1293 __dequeue_rt_entity(rt_se, flags);
1297 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1299 struct rq *rq = rq_of_rt_se(rt_se);
1301 dequeue_rt_stack(rt_se, flags);
1302 for_each_sched_rt_entity(rt_se)
1303 __enqueue_rt_entity(rt_se, flags);
1304 enqueue_top_rt_rq(&rq->rt);
1307 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1309 struct rq *rq = rq_of_rt_se(rt_se);
1311 dequeue_rt_stack(rt_se, flags);
1313 for_each_sched_rt_entity(rt_se) {
1314 struct rt_rq *rt_rq = group_rt_rq(rt_se);
1316 if (rt_rq && rt_rq->rt_nr_running)
1317 __enqueue_rt_entity(rt_se, flags);
1319 enqueue_top_rt_rq(&rq->rt);
1323 * Adding/removing a task to/from a priority array:
1326 enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1328 struct sched_rt_entity *rt_se = &p->rt;
1330 if (flags & ENQUEUE_WAKEUP)
1333 enqueue_rt_entity(rt_se, flags);
1335 if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1336 enqueue_pushable_task(rq, p);
1339 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1341 struct sched_rt_entity *rt_se = &p->rt;
1344 dequeue_rt_entity(rt_se, flags);
1346 dequeue_pushable_task(rq, p);
1350 * Put task to the head or the end of the run list without the overhead of
1351 * dequeue followed by enqueue.
1354 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
1356 if (on_rt_rq(rt_se)) {
1357 struct rt_prio_array *array = &rt_rq->active;
1358 struct list_head *queue = array->queue + rt_se_prio(rt_se);
1361 list_move(&rt_se->run_list, queue);
1363 list_move_tail(&rt_se->run_list, queue);
1367 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
1369 struct sched_rt_entity *rt_se = &p->rt;
1370 struct rt_rq *rt_rq;
1372 for_each_sched_rt_entity(rt_se) {
1373 rt_rq = rt_rq_of_se(rt_se);
1374 requeue_rt_entity(rt_rq, rt_se, head);
1378 static void yield_task_rt(struct rq *rq)
1380 requeue_task_rt(rq, rq->curr, 0);
1384 static int find_lowest_rq(struct task_struct *task);
1387 select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags)
1389 struct task_struct *curr;
1392 /* For anything but wake ups, just return the task_cpu */
1393 if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK)
1399 curr = READ_ONCE(rq->curr); /* unlocked access */
1402 * If the current task on @p's runqueue is an RT task, then
1403 * try to see if we can wake this RT task up on another
1404 * runqueue. Otherwise simply start this RT task
1405 * on its current runqueue.
1407 * We want to avoid overloading runqueues. If the woken
1408 * task is a higher priority, then it will stay on this CPU
1409 * and the lower prio task should be moved to another CPU.
1410 * Even though this will probably make the lower prio task
1411 * lose its cache, we do not want to bounce a higher task
1412 * around just because it gave up its CPU, perhaps for a
1415 * For equal prio tasks, we just let the scheduler sort it out.
1417 * Otherwise, just let it ride on the affined RQ and the
1418 * post-schedule router will push the preempted task away
1420 * This test is optimistic, if we get it wrong the load-balancer
1421 * will have to sort it out.
1423 if (curr && unlikely(rt_task(curr)) &&
1424 (curr->nr_cpus_allowed < 2 ||
1425 curr->prio <= p->prio)) {
1426 int target = find_lowest_rq(p);
1429 * Don't bother moving it if the destination CPU is
1430 * not running a lower priority task.
1433 p->prio < cpu_rq(target)->rt.highest_prio.curr)
1442 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
1445 * Current can't be migrated, useless to reschedule,
1446 * let's hope p can move out.
1448 if (rq->curr->nr_cpus_allowed == 1 ||
1449 !cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
1453 * p is migratable, so let's not schedule it and
1454 * see if it is pushed or pulled somewhere else.
1456 if (p->nr_cpus_allowed != 1
1457 && cpupri_find(&rq->rd->cpupri, p, NULL))
1461 * There appear to be other CPUs that can accept
1462 * the current task but none can run 'p', so lets reschedule
1463 * to try and push the current task away:
1465 requeue_task_rt(rq, p, 1);
1469 #endif /* CONFIG_SMP */
1472 * Preempt the current task with a newly woken task if needed:
1474 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1476 if (p->prio < rq->curr->prio) {
1485 * - the newly woken task is of equal priority to the current task
1486 * - the newly woken task is non-migratable while current is migratable
1487 * - current will be preempted on the next reschedule
1489 * we should check to see if current can readily move to a different
1490 * cpu. If so, we will reschedule to allow the push logic to try
1491 * to move current somewhere else, making room for our non-migratable
1494 if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr))
1495 check_preempt_equal_prio(rq, p);
1499 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1500 struct rt_rq *rt_rq)
1502 struct rt_prio_array *array = &rt_rq->active;
1503 struct sched_rt_entity *next = NULL;
1504 struct list_head *queue;
1507 idx = sched_find_first_bit(array->bitmap);
1508 BUG_ON(idx >= MAX_RT_PRIO);
1510 queue = array->queue + idx;
1511 next = list_entry(queue->next, struct sched_rt_entity, run_list);
1516 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1518 struct sched_rt_entity *rt_se;
1519 struct task_struct *p;
1520 struct rt_rq *rt_rq = &rq->rt;
1523 rt_se = pick_next_rt_entity(rq, rt_rq);
1525 rt_rq = group_rt_rq(rt_se);
1528 p = rt_task_of(rt_se);
1529 p->se.exec_start = rq_clock_task(rq);
1534 static struct task_struct *
1535 pick_next_task_rt(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
1537 struct task_struct *p;
1538 struct rt_rq *rt_rq = &rq->rt;
1540 if (need_pull_rt_task(rq, prev)) {
1542 * This is OK, because current is on_cpu, which avoids it being
1543 * picked for load-balance and preemption/IRQs are still
1544 * disabled avoiding further scheduler activity on it and we're
1545 * being very careful to re-start the picking loop.
1547 rq_unpin_lock(rq, rf);
1549 rq_repin_lock(rq, rf);
1551 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1552 * means a dl or stop task can slip in, in which case we need
1553 * to re-start task selection.
1555 if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) ||
1556 rq->dl.dl_nr_running))
1561 * We may dequeue prev's rt_rq in put_prev_task().
1562 * So, we update time before rt_nr_running check.
1564 if (prev->sched_class == &rt_sched_class)
1567 if (!rt_rq->rt_queued)
1570 put_prev_task(rq, prev);
1572 p = _pick_next_task_rt(rq);
1574 /* The running task is never eligible for pushing */
1575 dequeue_pushable_task(rq, p);
1577 rt_queue_push_tasks(rq);
1582 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1587 * The previous task needs to be made eligible for pushing
1588 * if it is still active
1590 if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
1591 enqueue_pushable_task(rq, p);
1596 /* Only try algorithms three times */
1597 #define RT_MAX_TRIES 3
1599 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1601 if (!task_running(rq, p) &&
1602 cpumask_test_cpu(cpu, &p->cpus_allowed))
1609 * Return the highest pushable rq's task, which is suitable to be executed
1610 * on the CPU, NULL otherwise
1612 static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu)
1614 struct plist_head *head = &rq->rt.pushable_tasks;
1615 struct task_struct *p;
1617 if (!has_pushable_tasks(rq))
1620 plist_for_each_entry(p, head, pushable_tasks) {
1621 if (pick_rt_task(rq, p, cpu))
1628 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1630 static int find_lowest_rq(struct task_struct *task)
1632 struct sched_domain *sd;
1633 struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask);
1634 int this_cpu = smp_processor_id();
1635 int cpu = task_cpu(task);
1637 /* Make sure the mask is initialized first */
1638 if (unlikely(!lowest_mask))
1641 if (task->nr_cpus_allowed == 1)
1642 return -1; /* No other targets possible */
1644 if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1645 return -1; /* No targets found */
1648 * At this point we have built a mask of CPUs representing the
1649 * lowest priority tasks in the system. Now we want to elect
1650 * the best one based on our affinity and topology.
1652 * We prioritize the last CPU that the task executed on since
1653 * it is most likely cache-hot in that location.
1655 if (cpumask_test_cpu(cpu, lowest_mask))
1659 * Otherwise, we consult the sched_domains span maps to figure
1660 * out which CPU is logically closest to our hot cache data.
1662 if (!cpumask_test_cpu(this_cpu, lowest_mask))
1663 this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1666 for_each_domain(cpu, sd) {
1667 if (sd->flags & SD_WAKE_AFFINE) {
1671 * "this_cpu" is cheaper to preempt than a
1674 if (this_cpu != -1 &&
1675 cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1680 best_cpu = cpumask_first_and(lowest_mask,
1681 sched_domain_span(sd));
1682 if (best_cpu < nr_cpu_ids) {
1691 * And finally, if there were no matches within the domains
1692 * just give the caller *something* to work with from the compatible
1698 cpu = cpumask_any(lowest_mask);
1699 if (cpu < nr_cpu_ids)
1705 /* Will lock the rq it finds */
1706 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1708 struct rq *lowest_rq = NULL;
1712 for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1713 cpu = find_lowest_rq(task);
1715 if ((cpu == -1) || (cpu == rq->cpu))
1718 lowest_rq = cpu_rq(cpu);
1720 if (lowest_rq->rt.highest_prio.curr <= task->prio) {
1722 * Target rq has tasks of equal or higher priority,
1723 * retrying does not release any lock and is unlikely
1724 * to yield a different result.
1730 /* if the prio of this runqueue changed, try again */
1731 if (double_lock_balance(rq, lowest_rq)) {
1733 * We had to unlock the run queue. In
1734 * the mean time, task could have
1735 * migrated already or had its affinity changed.
1736 * Also make sure that it wasn't scheduled on its rq.
1738 if (unlikely(task_rq(task) != rq ||
1739 !cpumask_test_cpu(lowest_rq->cpu, &task->cpus_allowed) ||
1740 task_running(rq, task) ||
1742 !task_on_rq_queued(task))) {
1744 double_unlock_balance(rq, lowest_rq);
1750 /* If this rq is still suitable use it. */
1751 if (lowest_rq->rt.highest_prio.curr > task->prio)
1755 double_unlock_balance(rq, lowest_rq);
1762 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1764 struct task_struct *p;
1766 if (!has_pushable_tasks(rq))
1769 p = plist_first_entry(&rq->rt.pushable_tasks,
1770 struct task_struct, pushable_tasks);
1772 BUG_ON(rq->cpu != task_cpu(p));
1773 BUG_ON(task_current(rq, p));
1774 BUG_ON(p->nr_cpus_allowed <= 1);
1776 BUG_ON(!task_on_rq_queued(p));
1777 BUG_ON(!rt_task(p));
1783 * If the current CPU has more than one RT task, see if the non
1784 * running task can migrate over to a CPU that is running a task
1785 * of lesser priority.
1787 static int push_rt_task(struct rq *rq)
1789 struct task_struct *next_task;
1790 struct rq *lowest_rq;
1793 if (!rq->rt.overloaded)
1796 next_task = pick_next_pushable_task(rq);
1801 if (unlikely(next_task == rq->curr)) {
1807 * It's possible that the next_task slipped in of
1808 * higher priority than current. If that's the case
1809 * just reschedule current.
1811 if (unlikely(next_task->prio < rq->curr->prio)) {
1816 /* We might release rq lock */
1817 get_task_struct(next_task);
1819 /* find_lock_lowest_rq locks the rq if found */
1820 lowest_rq = find_lock_lowest_rq(next_task, rq);
1822 struct task_struct *task;
1824 * find_lock_lowest_rq releases rq->lock
1825 * so it is possible that next_task has migrated.
1827 * We need to make sure that the task is still on the same
1828 * run-queue and is also still the next task eligible for
1831 task = pick_next_pushable_task(rq);
1832 if (task == next_task) {
1834 * The task hasn't migrated, and is still the next
1835 * eligible task, but we failed to find a run-queue
1836 * to push it to. Do not retry in this case, since
1837 * other CPUs will pull from us when ready.
1843 /* No more tasks, just exit */
1847 * Something has shifted, try again.
1849 put_task_struct(next_task);
1854 deactivate_task(rq, next_task, 0);
1855 set_task_cpu(next_task, lowest_rq->cpu);
1856 activate_task(lowest_rq, next_task, 0);
1859 resched_curr(lowest_rq);
1861 double_unlock_balance(rq, lowest_rq);
1864 put_task_struct(next_task);
1869 static void push_rt_tasks(struct rq *rq)
1871 /* push_rt_task will return true if it moved an RT */
1872 while (push_rt_task(rq))
1876 #ifdef HAVE_RT_PUSH_IPI
1879 * When a high priority task schedules out from a CPU and a lower priority
1880 * task is scheduled in, a check is made to see if there's any RT tasks
1881 * on other CPUs that are waiting to run because a higher priority RT task
1882 * is currently running on its CPU. In this case, the CPU with multiple RT
1883 * tasks queued on it (overloaded) needs to be notified that a CPU has opened
1884 * up that may be able to run one of its non-running queued RT tasks.
1886 * All CPUs with overloaded RT tasks need to be notified as there is currently
1887 * no way to know which of these CPUs have the highest priority task waiting
1888 * to run. Instead of trying to take a spinlock on each of these CPUs,
1889 * which has shown to cause large latency when done on machines with many
1890 * CPUs, sending an IPI to the CPUs to have them push off the overloaded
1891 * RT tasks waiting to run.
1893 * Just sending an IPI to each of the CPUs is also an issue, as on large
1894 * count CPU machines, this can cause an IPI storm on a CPU, especially
1895 * if its the only CPU with multiple RT tasks queued, and a large number
1896 * of CPUs scheduling a lower priority task at the same time.
1898 * Each root domain has its own irq work function that can iterate over
1899 * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT
1900 * tassk must be checked if there's one or many CPUs that are lowering
1901 * their priority, there's a single irq work iterator that will try to
1902 * push off RT tasks that are waiting to run.
1904 * When a CPU schedules a lower priority task, it will kick off the
1905 * irq work iterator that will jump to each CPU with overloaded RT tasks.
1906 * As it only takes the first CPU that schedules a lower priority task
1907 * to start the process, the rto_start variable is incremented and if
1908 * the atomic result is one, then that CPU will try to take the rto_lock.
1909 * This prevents high contention on the lock as the process handles all
1910 * CPUs scheduling lower priority tasks.
1912 * All CPUs that are scheduling a lower priority task will increment the
1913 * rt_loop_next variable. This will make sure that the irq work iterator
1914 * checks all RT overloaded CPUs whenever a CPU schedules a new lower
1915 * priority task, even if the iterator is in the middle of a scan. Incrementing
1916 * the rt_loop_next will cause the iterator to perform another scan.
1919 static int rto_next_cpu(struct root_domain *rd)
1925 * When starting the IPI RT pushing, the rto_cpu is set to -1,
1926 * rt_next_cpu() will simply return the first CPU found in
1929 * If rto_next_cpu() is called with rto_cpu is a valid CPU, it
1930 * will return the next CPU found in the rto_mask.
1932 * If there are no more CPUs left in the rto_mask, then a check is made
1933 * against rto_loop and rto_loop_next. rto_loop is only updated with
1934 * the rto_lock held, but any CPU may increment the rto_loop_next
1935 * without any locking.
1939 /* When rto_cpu is -1 this acts like cpumask_first() */
1940 cpu = cpumask_next(rd->rto_cpu, rd->rto_mask);
1944 if (cpu < nr_cpu_ids)
1950 * ACQUIRE ensures we see the @rto_mask changes
1951 * made prior to the @next value observed.
1953 * Matches WMB in rt_set_overload().
1955 next = atomic_read_acquire(&rd->rto_loop_next);
1957 if (rd->rto_loop == next)
1960 rd->rto_loop = next;
1966 static inline bool rto_start_trylock(atomic_t *v)
1968 return !atomic_cmpxchg_acquire(v, 0, 1);
1971 static inline void rto_start_unlock(atomic_t *v)
1973 atomic_set_release(v, 0);
1976 static void tell_cpu_to_push(struct rq *rq)
1980 /* Keep the loop going if the IPI is currently active */
1981 atomic_inc(&rq->rd->rto_loop_next);
1983 /* Only one CPU can initiate a loop at a time */
1984 if (!rto_start_trylock(&rq->rd->rto_loop_start))
1987 raw_spin_lock(&rq->rd->rto_lock);
1990 * The rto_cpu is updated under the lock, if it has a valid CPU
1991 * then the IPI is still running and will continue due to the
1992 * update to loop_next, and nothing needs to be done here.
1993 * Otherwise it is finishing up and an ipi needs to be sent.
1995 if (rq->rd->rto_cpu < 0)
1996 cpu = rto_next_cpu(rq->rd);
1998 raw_spin_unlock(&rq->rd->rto_lock);
2000 rto_start_unlock(&rq->rd->rto_loop_start);
2003 /* Make sure the rd does not get freed while pushing */
2004 sched_get_rd(rq->rd);
2005 irq_work_queue_on(&rq->rd->rto_push_work, cpu);
2009 /* Called from hardirq context */
2010 void rto_push_irq_work_func(struct irq_work *work)
2012 struct root_domain *rd =
2013 container_of(work, struct root_domain, rto_push_work);
2020 * We do not need to grab the lock to check for has_pushable_tasks.
2021 * When it gets updated, a check is made if a push is possible.
2023 if (has_pushable_tasks(rq)) {
2024 raw_spin_lock(&rq->lock);
2026 raw_spin_unlock(&rq->lock);
2029 raw_spin_lock(&rd->rto_lock);
2031 /* Pass the IPI to the next rt overloaded queue */
2032 cpu = rto_next_cpu(rd);
2034 raw_spin_unlock(&rd->rto_lock);
2041 /* Try the next RT overloaded CPU */
2042 irq_work_queue_on(&rd->rto_push_work, cpu);
2044 #endif /* HAVE_RT_PUSH_IPI */
2046 static void pull_rt_task(struct rq *this_rq)
2048 int this_cpu = this_rq->cpu, cpu;
2049 bool resched = false;
2050 struct task_struct *p;
2052 int rt_overload_count = rt_overloaded(this_rq);
2054 if (likely(!rt_overload_count))
2058 * Match the barrier from rt_set_overloaded; this guarantees that if we
2059 * see overloaded we must also see the rto_mask bit.
2063 /* If we are the only overloaded CPU do nothing */
2064 if (rt_overload_count == 1 &&
2065 cpumask_test_cpu(this_rq->cpu, this_rq->rd->rto_mask))
2068 #ifdef HAVE_RT_PUSH_IPI
2069 if (sched_feat(RT_PUSH_IPI)) {
2070 tell_cpu_to_push(this_rq);
2075 for_each_cpu(cpu, this_rq->rd->rto_mask) {
2076 if (this_cpu == cpu)
2079 src_rq = cpu_rq(cpu);
2082 * Don't bother taking the src_rq->lock if the next highest
2083 * task is known to be lower-priority than our current task.
2084 * This may look racy, but if this value is about to go
2085 * logically higher, the src_rq will push this task away.
2086 * And if its going logically lower, we do not care
2088 if (src_rq->rt.highest_prio.next >=
2089 this_rq->rt.highest_prio.curr)
2093 * We can potentially drop this_rq's lock in
2094 * double_lock_balance, and another CPU could
2097 double_lock_balance(this_rq, src_rq);
2100 * We can pull only a task, which is pushable
2101 * on its rq, and no others.
2103 p = pick_highest_pushable_task(src_rq, this_cpu);
2106 * Do we have an RT task that preempts
2107 * the to-be-scheduled task?
2109 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
2110 WARN_ON(p == src_rq->curr);
2111 WARN_ON(!task_on_rq_queued(p));
2114 * There's a chance that p is higher in priority
2115 * than what's currently running on its CPU.
2116 * This is just that p is wakeing up and hasn't
2117 * had a chance to schedule. We only pull
2118 * p if it is lower in priority than the
2119 * current task on the run queue
2121 if (p->prio < src_rq->curr->prio)
2126 deactivate_task(src_rq, p, 0);
2127 set_task_cpu(p, this_cpu);
2128 activate_task(this_rq, p, 0);
2130 * We continue with the search, just in
2131 * case there's an even higher prio task
2132 * in another runqueue. (low likelihood
2137 double_unlock_balance(this_rq, src_rq);
2141 resched_curr(this_rq);
2145 * If we are not running and we are not going to reschedule soon, we should
2146 * try to push tasks away now
2148 static void task_woken_rt(struct rq *rq, struct task_struct *p)
2150 if (!task_running(rq, p) &&
2151 !test_tsk_need_resched(rq->curr) &&
2152 p->nr_cpus_allowed > 1 &&
2153 (dl_task(rq->curr) || rt_task(rq->curr)) &&
2154 (rq->curr->nr_cpus_allowed < 2 ||
2155 rq->curr->prio <= p->prio))
2159 /* Assumes rq->lock is held */
2160 static void rq_online_rt(struct rq *rq)
2162 if (rq->rt.overloaded)
2163 rt_set_overload(rq);
2165 __enable_runtime(rq);
2167 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
2170 /* Assumes rq->lock is held */
2171 static void rq_offline_rt(struct rq *rq)
2173 if (rq->rt.overloaded)
2174 rt_clear_overload(rq);
2176 __disable_runtime(rq);
2178 cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
2182 * When switch from the rt queue, we bring ourselves to a position
2183 * that we might want to pull RT tasks from other runqueues.
2185 static void switched_from_rt(struct rq *rq, struct task_struct *p)
2188 * If there are other RT tasks then we will reschedule
2189 * and the scheduling of the other RT tasks will handle
2190 * the balancing. But if we are the last RT task
2191 * we may need to handle the pulling of RT tasks
2194 if (!task_on_rq_queued(p) || rq->rt.rt_nr_running)
2197 rt_queue_pull_task(rq);
2200 void __init init_sched_rt_class(void)
2204 for_each_possible_cpu(i) {
2205 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
2206 GFP_KERNEL, cpu_to_node(i));
2209 #endif /* CONFIG_SMP */
2212 * When switching a task to RT, we may overload the runqueue
2213 * with RT tasks. In this case we try to push them off to
2216 static void switched_to_rt(struct rq *rq, struct task_struct *p)
2219 * If we are already running, then there's nothing
2220 * that needs to be done. But if we are not running
2221 * we may need to preempt the current running task.
2222 * If that current running task is also an RT task
2223 * then see if we can move to another run queue.
2225 if (task_on_rq_queued(p) && rq->curr != p) {
2227 if (p->nr_cpus_allowed > 1 && rq->rt.overloaded)
2228 rt_queue_push_tasks(rq);
2229 #endif /* CONFIG_SMP */
2230 if (p->prio < rq->curr->prio && cpu_online(cpu_of(rq)))
2236 * Priority of the task has changed. This may cause
2237 * us to initiate a push or pull.
2240 prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio)
2242 if (!task_on_rq_queued(p))
2245 if (rq->curr == p) {
2248 * If our priority decreases while running, we
2249 * may need to pull tasks to this runqueue.
2251 if (oldprio < p->prio)
2252 rt_queue_pull_task(rq);
2255 * If there's a higher priority task waiting to run
2258 if (p->prio > rq->rt.highest_prio.curr)
2261 /* For UP simply resched on drop of prio */
2262 if (oldprio < p->prio)
2264 #endif /* CONFIG_SMP */
2267 * This task is not running, but if it is
2268 * greater than the current running task
2271 if (p->prio < rq->curr->prio)
2276 #ifdef CONFIG_POSIX_TIMERS
2277 static void watchdog(struct rq *rq, struct task_struct *p)
2279 unsigned long soft, hard;
2281 /* max may change after cur was read, this will be fixed next tick */
2282 soft = task_rlimit(p, RLIMIT_RTTIME);
2283 hard = task_rlimit_max(p, RLIMIT_RTTIME);
2285 if (soft != RLIM_INFINITY) {
2288 if (p->rt.watchdog_stamp != jiffies) {
2290 p->rt.watchdog_stamp = jiffies;
2293 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
2294 if (p->rt.timeout > next)
2295 p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
2299 static inline void watchdog(struct rq *rq, struct task_struct *p) { }
2303 * scheduler tick hitting a task of our scheduling class.
2305 * NOTE: This function can be called remotely by the tick offload that
2306 * goes along full dynticks. Therefore no local assumption can be made
2307 * and everything must be accessed through the @rq and @curr passed in
2310 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
2312 struct sched_rt_entity *rt_se = &p->rt;
2319 * RR tasks need a special form of timeslice management.
2320 * FIFO tasks have no timeslices.
2322 if (p->policy != SCHED_RR)
2325 if (--p->rt.time_slice)
2328 p->rt.time_slice = sched_rr_timeslice;
2331 * Requeue to the end of queue if we (and all of our ancestors) are not
2332 * the only element on the queue
2334 for_each_sched_rt_entity(rt_se) {
2335 if (rt_se->run_list.prev != rt_se->run_list.next) {
2336 requeue_task_rt(rq, p, 0);
2343 static void set_curr_task_rt(struct rq *rq)
2345 struct task_struct *p = rq->curr;
2347 p->se.exec_start = rq_clock_task(rq);
2349 /* The running task is never eligible for pushing */
2350 dequeue_pushable_task(rq, p);
2353 static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
2356 * Time slice is 0 for SCHED_FIFO tasks
2358 if (task->policy == SCHED_RR)
2359 return sched_rr_timeslice;
2364 const struct sched_class rt_sched_class = {
2365 .next = &fair_sched_class,
2366 .enqueue_task = enqueue_task_rt,
2367 .dequeue_task = dequeue_task_rt,
2368 .yield_task = yield_task_rt,
2370 .check_preempt_curr = check_preempt_curr_rt,
2372 .pick_next_task = pick_next_task_rt,
2373 .put_prev_task = put_prev_task_rt,
2376 .select_task_rq = select_task_rq_rt,
2378 .set_cpus_allowed = set_cpus_allowed_common,
2379 .rq_online = rq_online_rt,
2380 .rq_offline = rq_offline_rt,
2381 .task_woken = task_woken_rt,
2382 .switched_from = switched_from_rt,
2385 .set_curr_task = set_curr_task_rt,
2386 .task_tick = task_tick_rt,
2388 .get_rr_interval = get_rr_interval_rt,
2390 .prio_changed = prio_changed_rt,
2391 .switched_to = switched_to_rt,
2393 .update_curr = update_curr_rt,
2396 #ifdef CONFIG_RT_GROUP_SCHED
2398 * Ensure that the real time constraints are schedulable.
2400 static DEFINE_MUTEX(rt_constraints_mutex);
2402 /* Must be called with tasklist_lock held */
2403 static inline int tg_has_rt_tasks(struct task_group *tg)
2405 struct task_struct *g, *p;
2408 * Autogroups do not have RT tasks; see autogroup_create().
2410 if (task_group_is_autogroup(tg))
2413 for_each_process_thread(g, p) {
2414 if (rt_task(p) && task_group(p) == tg)
2421 struct rt_schedulable_data {
2422 struct task_group *tg;
2427 static int tg_rt_schedulable(struct task_group *tg, void *data)
2429 struct rt_schedulable_data *d = data;
2430 struct task_group *child;
2431 unsigned long total, sum = 0;
2432 u64 period, runtime;
2434 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
2435 runtime = tg->rt_bandwidth.rt_runtime;
2438 period = d->rt_period;
2439 runtime = d->rt_runtime;
2443 * Cannot have more runtime than the period.
2445 if (runtime > period && runtime != RUNTIME_INF)
2449 * Ensure we don't starve existing RT tasks.
2451 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
2454 total = to_ratio(period, runtime);
2457 * Nobody can have more than the global setting allows.
2459 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
2463 * The sum of our children's runtime should not exceed our own.
2465 list_for_each_entry_rcu(child, &tg->children, siblings) {
2466 period = ktime_to_ns(child->rt_bandwidth.rt_period);
2467 runtime = child->rt_bandwidth.rt_runtime;
2469 if (child == d->tg) {
2470 period = d->rt_period;
2471 runtime = d->rt_runtime;
2474 sum += to_ratio(period, runtime);
2483 static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
2487 struct rt_schedulable_data data = {
2489 .rt_period = period,
2490 .rt_runtime = runtime,
2494 ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
2500 static int tg_set_rt_bandwidth(struct task_group *tg,
2501 u64 rt_period, u64 rt_runtime)
2506 * Disallowing the root group RT runtime is BAD, it would disallow the
2507 * kernel creating (and or operating) RT threads.
2509 if (tg == &root_task_group && rt_runtime == 0)
2512 /* No period doesn't make any sense. */
2516 mutex_lock(&rt_constraints_mutex);
2517 read_lock(&tasklist_lock);
2518 err = __rt_schedulable(tg, rt_period, rt_runtime);
2522 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
2523 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
2524 tg->rt_bandwidth.rt_runtime = rt_runtime;
2526 for_each_possible_cpu(i) {
2527 struct rt_rq *rt_rq = tg->rt_rq[i];
2529 raw_spin_lock(&rt_rq->rt_runtime_lock);
2530 rt_rq->rt_runtime = rt_runtime;
2531 raw_spin_unlock(&rt_rq->rt_runtime_lock);
2533 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
2535 read_unlock(&tasklist_lock);
2536 mutex_unlock(&rt_constraints_mutex);
2541 int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
2543 u64 rt_runtime, rt_period;
2545 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
2546 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
2547 if (rt_runtime_us < 0)
2548 rt_runtime = RUNTIME_INF;
2550 return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2553 long sched_group_rt_runtime(struct task_group *tg)
2557 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
2560 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
2561 do_div(rt_runtime_us, NSEC_PER_USEC);
2562 return rt_runtime_us;
2565 int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
2567 u64 rt_runtime, rt_period;
2569 rt_period = rt_period_us * NSEC_PER_USEC;
2570 rt_runtime = tg->rt_bandwidth.rt_runtime;
2572 return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2575 long sched_group_rt_period(struct task_group *tg)
2579 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
2580 do_div(rt_period_us, NSEC_PER_USEC);
2581 return rt_period_us;
2584 static int sched_rt_global_constraints(void)
2588 mutex_lock(&rt_constraints_mutex);
2589 read_lock(&tasklist_lock);
2590 ret = __rt_schedulable(NULL, 0, 0);
2591 read_unlock(&tasklist_lock);
2592 mutex_unlock(&rt_constraints_mutex);
2597 int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
2599 /* Don't accept realtime tasks when there is no way for them to run */
2600 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
2606 #else /* !CONFIG_RT_GROUP_SCHED */
2607 static int sched_rt_global_constraints(void)
2609 unsigned long flags;
2612 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
2613 for_each_possible_cpu(i) {
2614 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
2616 raw_spin_lock(&rt_rq->rt_runtime_lock);
2617 rt_rq->rt_runtime = global_rt_runtime();
2618 raw_spin_unlock(&rt_rq->rt_runtime_lock);
2620 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
2624 #endif /* CONFIG_RT_GROUP_SCHED */
2626 static int sched_rt_global_validate(void)
2628 if (sysctl_sched_rt_period <= 0)
2631 if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
2632 (sysctl_sched_rt_runtime > sysctl_sched_rt_period))
2638 static void sched_rt_do_global(void)
2640 def_rt_bandwidth.rt_runtime = global_rt_runtime();
2641 def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period());
2644 int sched_rt_handler(struct ctl_table *table, int write,
2645 void __user *buffer, size_t *lenp,
2648 int old_period, old_runtime;
2649 static DEFINE_MUTEX(mutex);
2653 old_period = sysctl_sched_rt_period;
2654 old_runtime = sysctl_sched_rt_runtime;
2656 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2658 if (!ret && write) {
2659 ret = sched_rt_global_validate();
2663 ret = sched_dl_global_validate();
2667 ret = sched_rt_global_constraints();
2671 sched_rt_do_global();
2672 sched_dl_do_global();
2676 sysctl_sched_rt_period = old_period;
2677 sysctl_sched_rt_runtime = old_runtime;
2679 mutex_unlock(&mutex);
2684 int sched_rr_handler(struct ctl_table *table, int write,
2685 void __user *buffer, size_t *lenp,
2689 static DEFINE_MUTEX(mutex);
2692 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2694 * Make sure that internally we keep jiffies.
2695 * Also, writing zero resets the timeslice to default:
2697 if (!ret && write) {
2698 sched_rr_timeslice =
2699 sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE :
2700 msecs_to_jiffies(sysctl_sched_rr_timeslice);
2702 mutex_unlock(&mutex);
2707 #ifdef CONFIG_SCHED_DEBUG
2708 void print_rt_stats(struct seq_file *m, int cpu)
2711 struct rt_rq *rt_rq;
2714 for_each_rt_rq(rt_rq, iter, cpu_rq(cpu))
2715 print_rt_rq(m, cpu, rt_rq);
2718 #endif /* CONFIG_SCHED_DEBUG */