1 // SPDX-License-Identifier: GPL-2.0
3 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
9 #include <linux/slab.h>
10 #include <linux/irq_work.h>
12 int sched_rr_timeslice = RR_TIMESLICE;
13 int sysctl_sched_rr_timeslice = (MSEC_PER_SEC / HZ) * RR_TIMESLICE;
15 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
17 struct rt_bandwidth def_rt_bandwidth;
19 static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
21 struct rt_bandwidth *rt_b =
22 container_of(timer, struct rt_bandwidth, rt_period_timer);
26 raw_spin_lock(&rt_b->rt_runtime_lock);
28 overrun = hrtimer_forward_now(timer, rt_b->rt_period);
32 raw_spin_unlock(&rt_b->rt_runtime_lock);
33 idle = do_sched_rt_period_timer(rt_b, overrun);
34 raw_spin_lock(&rt_b->rt_runtime_lock);
37 rt_b->rt_period_active = 0;
38 raw_spin_unlock(&rt_b->rt_runtime_lock);
40 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
43 void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
45 rt_b->rt_period = ns_to_ktime(period);
46 rt_b->rt_runtime = runtime;
48 raw_spin_lock_init(&rt_b->rt_runtime_lock);
50 hrtimer_init(&rt_b->rt_period_timer,
51 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
52 rt_b->rt_period_timer.function = sched_rt_period_timer;
55 static inline void do_start_rt_bandwidth(struct rt_bandwidth *rt_b)
57 raw_spin_lock(&rt_b->rt_runtime_lock);
58 if (!rt_b->rt_period_active) {
59 rt_b->rt_period_active = 1;
61 * SCHED_DEADLINE updates the bandwidth, as a run away
62 * RT task with a DL task could hog a CPU. But DL does
63 * not reset the period. If a deadline task was running
64 * without an RT task running, it can cause RT tasks to
65 * throttle when they start up. Kick the timer right away
66 * to update the period.
68 hrtimer_forward_now(&rt_b->rt_period_timer, ns_to_ktime(0));
69 hrtimer_start_expires(&rt_b->rt_period_timer, HRTIMER_MODE_ABS_PINNED);
71 raw_spin_unlock(&rt_b->rt_runtime_lock);
74 static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
76 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
79 do_start_rt_bandwidth(rt_b);
82 void init_rt_rq(struct rt_rq *rt_rq)
84 struct rt_prio_array *array;
87 array = &rt_rq->active;
88 for (i = 0; i < MAX_RT_PRIO; i++) {
89 INIT_LIST_HEAD(array->queue + i);
90 __clear_bit(i, array->bitmap);
92 /* delimiter for bitsearch: */
93 __set_bit(MAX_RT_PRIO, array->bitmap);
95 #if defined CONFIG_SMP
96 rt_rq->highest_prio.curr = MAX_RT_PRIO;
97 rt_rq->highest_prio.next = MAX_RT_PRIO;
98 rt_rq->rt_nr_migratory = 0;
99 rt_rq->overloaded = 0;
100 plist_head_init(&rt_rq->pushable_tasks);
101 #endif /* CONFIG_SMP */
102 /* We start is dequeued state, because no RT tasks are queued */
103 rt_rq->rt_queued = 0;
106 rt_rq->rt_throttled = 0;
107 rt_rq->rt_runtime = 0;
108 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
111 #ifdef CONFIG_RT_GROUP_SCHED
112 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
114 hrtimer_cancel(&rt_b->rt_period_timer);
117 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
119 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
121 #ifdef CONFIG_SCHED_DEBUG
122 WARN_ON_ONCE(!rt_entity_is_task(rt_se));
124 return container_of(rt_se, struct task_struct, rt);
127 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
132 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
137 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
139 struct rt_rq *rt_rq = rt_se->rt_rq;
144 void free_rt_sched_group(struct task_group *tg)
149 destroy_rt_bandwidth(&tg->rt_bandwidth);
151 for_each_possible_cpu(i) {
162 void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
163 struct sched_rt_entity *rt_se, int cpu,
164 struct sched_rt_entity *parent)
166 struct rq *rq = cpu_rq(cpu);
168 rt_rq->highest_prio.curr = MAX_RT_PRIO;
169 rt_rq->rt_nr_boosted = 0;
173 tg->rt_rq[cpu] = rt_rq;
174 tg->rt_se[cpu] = rt_se;
180 rt_se->rt_rq = &rq->rt;
182 rt_se->rt_rq = parent->my_q;
185 rt_se->parent = parent;
186 INIT_LIST_HEAD(&rt_se->run_list);
189 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
192 struct sched_rt_entity *rt_se;
195 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
198 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
202 init_rt_bandwidth(&tg->rt_bandwidth,
203 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
205 for_each_possible_cpu(i) {
206 rt_rq = kzalloc_node(sizeof(struct rt_rq),
207 GFP_KERNEL, cpu_to_node(i));
211 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
212 GFP_KERNEL, cpu_to_node(i));
217 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
218 init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
229 #else /* CONFIG_RT_GROUP_SCHED */
231 #define rt_entity_is_task(rt_se) (1)
233 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
235 return container_of(rt_se, struct task_struct, rt);
238 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
240 return container_of(rt_rq, struct rq, rt);
243 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
245 struct task_struct *p = rt_task_of(rt_se);
250 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
252 struct rq *rq = rq_of_rt_se(rt_se);
257 void free_rt_sched_group(struct task_group *tg) { }
259 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
263 #endif /* CONFIG_RT_GROUP_SCHED */
267 static void pull_rt_task(struct rq *this_rq);
269 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
271 /* Try to pull RT tasks here if we lower this rq's prio */
272 return rq->rt.highest_prio.curr > prev->prio;
275 static inline int rt_overloaded(struct rq *rq)
277 return atomic_read(&rq->rd->rto_count);
280 static inline void rt_set_overload(struct rq *rq)
285 cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
287 * Make sure the mask is visible before we set
288 * the overload count. That is checked to determine
289 * if we should look at the mask. It would be a shame
290 * if we looked at the mask, but the mask was not
293 * Matched by the barrier in pull_rt_task().
296 atomic_inc(&rq->rd->rto_count);
299 static inline void rt_clear_overload(struct rq *rq)
304 /* the order here really doesn't matter */
305 atomic_dec(&rq->rd->rto_count);
306 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
309 static void update_rt_migration(struct rt_rq *rt_rq)
311 if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
312 if (!rt_rq->overloaded) {
313 rt_set_overload(rq_of_rt_rq(rt_rq));
314 rt_rq->overloaded = 1;
316 } else if (rt_rq->overloaded) {
317 rt_clear_overload(rq_of_rt_rq(rt_rq));
318 rt_rq->overloaded = 0;
322 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
324 struct task_struct *p;
326 if (!rt_entity_is_task(rt_se))
329 p = rt_task_of(rt_se);
330 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
332 rt_rq->rt_nr_total++;
333 if (p->nr_cpus_allowed > 1)
334 rt_rq->rt_nr_migratory++;
336 update_rt_migration(rt_rq);
339 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
341 struct task_struct *p;
343 if (!rt_entity_is_task(rt_se))
346 p = rt_task_of(rt_se);
347 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
349 rt_rq->rt_nr_total--;
350 if (p->nr_cpus_allowed > 1)
351 rt_rq->rt_nr_migratory--;
353 update_rt_migration(rt_rq);
356 static inline int has_pushable_tasks(struct rq *rq)
358 return !plist_head_empty(&rq->rt.pushable_tasks);
361 static DEFINE_PER_CPU(struct callback_head, rt_push_head);
362 static DEFINE_PER_CPU(struct callback_head, rt_pull_head);
364 static void push_rt_tasks(struct rq *);
365 static void pull_rt_task(struct rq *);
367 static inline void queue_push_tasks(struct rq *rq)
369 if (!has_pushable_tasks(rq))
372 queue_balance_callback(rq, &per_cpu(rt_push_head, rq->cpu), push_rt_tasks);
375 static inline void queue_pull_task(struct rq *rq)
377 queue_balance_callback(rq, &per_cpu(rt_pull_head, rq->cpu), pull_rt_task);
380 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
382 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
383 plist_node_init(&p->pushable_tasks, p->prio);
384 plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
386 /* Update the highest prio pushable task */
387 if (p->prio < rq->rt.highest_prio.next)
388 rq->rt.highest_prio.next = p->prio;
391 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
393 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
395 /* Update the new highest prio pushable task */
396 if (has_pushable_tasks(rq)) {
397 p = plist_first_entry(&rq->rt.pushable_tasks,
398 struct task_struct, pushable_tasks);
399 rq->rt.highest_prio.next = p->prio;
401 rq->rt.highest_prio.next = MAX_RT_PRIO;
406 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
410 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
415 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
420 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
424 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
429 static inline void pull_rt_task(struct rq *this_rq)
433 static inline void queue_push_tasks(struct rq *rq)
436 #endif /* CONFIG_SMP */
438 static void enqueue_top_rt_rq(struct rt_rq *rt_rq);
439 static void dequeue_top_rt_rq(struct rt_rq *rt_rq);
441 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
446 #ifdef CONFIG_RT_GROUP_SCHED
448 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
453 return rt_rq->rt_runtime;
456 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
458 return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
461 typedef struct task_group *rt_rq_iter_t;
463 static inline struct task_group *next_task_group(struct task_group *tg)
466 tg = list_entry_rcu(tg->list.next,
467 typeof(struct task_group), list);
468 } while (&tg->list != &task_groups && task_group_is_autogroup(tg));
470 if (&tg->list == &task_groups)
476 #define for_each_rt_rq(rt_rq, iter, rq) \
477 for (iter = container_of(&task_groups, typeof(*iter), list); \
478 (iter = next_task_group(iter)) && \
479 (rt_rq = iter->rt_rq[cpu_of(rq)]);)
481 #define for_each_sched_rt_entity(rt_se) \
482 for (; rt_se; rt_se = rt_se->parent)
484 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
489 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
490 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
492 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
494 struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
495 struct rq *rq = rq_of_rt_rq(rt_rq);
496 struct sched_rt_entity *rt_se;
498 int cpu = cpu_of(rq);
500 rt_se = rt_rq->tg->rt_se[cpu];
502 if (rt_rq->rt_nr_running) {
504 enqueue_top_rt_rq(rt_rq);
505 else if (!on_rt_rq(rt_se))
506 enqueue_rt_entity(rt_se, 0);
508 if (rt_rq->highest_prio.curr < curr->prio)
513 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
515 struct sched_rt_entity *rt_se;
516 int cpu = cpu_of(rq_of_rt_rq(rt_rq));
518 rt_se = rt_rq->tg->rt_se[cpu];
521 dequeue_top_rt_rq(rt_rq);
522 else if (on_rt_rq(rt_se))
523 dequeue_rt_entity(rt_se, 0);
526 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
528 return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
531 static int rt_se_boosted(struct sched_rt_entity *rt_se)
533 struct rt_rq *rt_rq = group_rt_rq(rt_se);
534 struct task_struct *p;
537 return !!rt_rq->rt_nr_boosted;
539 p = rt_task_of(rt_se);
540 return p->prio != p->normal_prio;
544 static inline const struct cpumask *sched_rt_period_mask(void)
546 return this_rq()->rd->span;
549 static inline const struct cpumask *sched_rt_period_mask(void)
551 return cpu_online_mask;
556 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
558 return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
561 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
563 return &rt_rq->tg->rt_bandwidth;
566 #else /* !CONFIG_RT_GROUP_SCHED */
568 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
570 return rt_rq->rt_runtime;
573 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
575 return ktime_to_ns(def_rt_bandwidth.rt_period);
578 typedef struct rt_rq *rt_rq_iter_t;
580 #define for_each_rt_rq(rt_rq, iter, rq) \
581 for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
583 #define for_each_sched_rt_entity(rt_se) \
584 for (; rt_se; rt_se = NULL)
586 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
591 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
593 struct rq *rq = rq_of_rt_rq(rt_rq);
595 if (!rt_rq->rt_nr_running)
598 enqueue_top_rt_rq(rt_rq);
602 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
604 dequeue_top_rt_rq(rt_rq);
607 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
609 return rt_rq->rt_throttled;
612 static inline const struct cpumask *sched_rt_period_mask(void)
614 return cpu_online_mask;
618 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
620 return &cpu_rq(cpu)->rt;
623 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
625 return &def_rt_bandwidth;
628 #endif /* CONFIG_RT_GROUP_SCHED */
630 bool sched_rt_bandwidth_account(struct rt_rq *rt_rq)
632 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
634 return (hrtimer_active(&rt_b->rt_period_timer) ||
635 rt_rq->rt_time < rt_b->rt_runtime);
640 * We ran out of runtime, see if we can borrow some from our neighbours.
642 static void do_balance_runtime(struct rt_rq *rt_rq)
644 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
645 struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd;
649 weight = cpumask_weight(rd->span);
651 raw_spin_lock(&rt_b->rt_runtime_lock);
652 rt_period = ktime_to_ns(rt_b->rt_period);
653 for_each_cpu(i, rd->span) {
654 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
660 raw_spin_lock(&iter->rt_runtime_lock);
662 * Either all rqs have inf runtime and there's nothing to steal
663 * or __disable_runtime() below sets a specific rq to inf to
664 * indicate its been disabled and disalow stealing.
666 if (iter->rt_runtime == RUNTIME_INF)
670 * From runqueues with spare time, take 1/n part of their
671 * spare time, but no more than our period.
673 diff = iter->rt_runtime - iter->rt_time;
675 diff = div_u64((u64)diff, weight);
676 if (rt_rq->rt_runtime + diff > rt_period)
677 diff = rt_period - rt_rq->rt_runtime;
678 iter->rt_runtime -= diff;
679 rt_rq->rt_runtime += diff;
680 if (rt_rq->rt_runtime == rt_period) {
681 raw_spin_unlock(&iter->rt_runtime_lock);
686 raw_spin_unlock(&iter->rt_runtime_lock);
688 raw_spin_unlock(&rt_b->rt_runtime_lock);
692 * Ensure this RQ takes back all the runtime it lend to its neighbours.
694 static void __disable_runtime(struct rq *rq)
696 struct root_domain *rd = rq->rd;
700 if (unlikely(!scheduler_running))
703 for_each_rt_rq(rt_rq, iter, rq) {
704 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
708 raw_spin_lock(&rt_b->rt_runtime_lock);
709 raw_spin_lock(&rt_rq->rt_runtime_lock);
711 * Either we're all inf and nobody needs to borrow, or we're
712 * already disabled and thus have nothing to do, or we have
713 * exactly the right amount of runtime to take out.
715 if (rt_rq->rt_runtime == RUNTIME_INF ||
716 rt_rq->rt_runtime == rt_b->rt_runtime)
718 raw_spin_unlock(&rt_rq->rt_runtime_lock);
721 * Calculate the difference between what we started out with
722 * and what we current have, that's the amount of runtime
723 * we lend and now have to reclaim.
725 want = rt_b->rt_runtime - rt_rq->rt_runtime;
728 * Greedy reclaim, take back as much as we can.
730 for_each_cpu(i, rd->span) {
731 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
735 * Can't reclaim from ourselves or disabled runqueues.
737 if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
740 raw_spin_lock(&iter->rt_runtime_lock);
742 diff = min_t(s64, iter->rt_runtime, want);
743 iter->rt_runtime -= diff;
746 iter->rt_runtime -= want;
749 raw_spin_unlock(&iter->rt_runtime_lock);
755 raw_spin_lock(&rt_rq->rt_runtime_lock);
757 * We cannot be left wanting - that would mean some runtime
758 * leaked out of the system.
763 * Disable all the borrow logic by pretending we have inf
764 * runtime - in which case borrowing doesn't make sense.
766 rt_rq->rt_runtime = RUNTIME_INF;
767 rt_rq->rt_throttled = 0;
768 raw_spin_unlock(&rt_rq->rt_runtime_lock);
769 raw_spin_unlock(&rt_b->rt_runtime_lock);
771 /* Make rt_rq available for pick_next_task() */
772 sched_rt_rq_enqueue(rt_rq);
776 static void __enable_runtime(struct rq *rq)
781 if (unlikely(!scheduler_running))
785 * Reset each runqueue's bandwidth settings
787 for_each_rt_rq(rt_rq, iter, rq) {
788 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
790 raw_spin_lock(&rt_b->rt_runtime_lock);
791 raw_spin_lock(&rt_rq->rt_runtime_lock);
792 rt_rq->rt_runtime = rt_b->rt_runtime;
794 rt_rq->rt_throttled = 0;
795 raw_spin_unlock(&rt_rq->rt_runtime_lock);
796 raw_spin_unlock(&rt_b->rt_runtime_lock);
800 static void balance_runtime(struct rt_rq *rt_rq)
802 if (!sched_feat(RT_RUNTIME_SHARE))
805 if (rt_rq->rt_time > rt_rq->rt_runtime) {
806 raw_spin_unlock(&rt_rq->rt_runtime_lock);
807 do_balance_runtime(rt_rq);
808 raw_spin_lock(&rt_rq->rt_runtime_lock);
811 #else /* !CONFIG_SMP */
812 static inline void balance_runtime(struct rt_rq *rt_rq) {}
813 #endif /* CONFIG_SMP */
815 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
817 int i, idle = 1, throttled = 0;
818 const struct cpumask *span;
820 span = sched_rt_period_mask();
821 #ifdef CONFIG_RT_GROUP_SCHED
823 * FIXME: isolated CPUs should really leave the root task group,
824 * whether they are isolcpus or were isolated via cpusets, lest
825 * the timer run on a CPU which does not service all runqueues,
826 * potentially leaving other CPUs indefinitely throttled. If
827 * isolation is really required, the user will turn the throttle
828 * off to kill the perturbations it causes anyway. Meanwhile,
829 * this maintains functionality for boot and/or troubleshooting.
831 if (rt_b == &root_task_group.rt_bandwidth)
832 span = cpu_online_mask;
834 for_each_cpu(i, span) {
836 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
837 struct rq *rq = rq_of_rt_rq(rt_rq);
841 * When span == cpu_online_mask, taking each rq->lock
842 * can be time-consuming. Try to avoid it when possible.
844 raw_spin_lock(&rt_rq->rt_runtime_lock);
845 if (!sched_feat(RT_RUNTIME_SHARE) && rt_rq->rt_runtime != RUNTIME_INF)
846 rt_rq->rt_runtime = rt_b->rt_runtime;
847 skip = !rt_rq->rt_time && !rt_rq->rt_nr_running;
848 raw_spin_unlock(&rt_rq->rt_runtime_lock);
852 raw_spin_lock(&rq->lock);
855 if (rt_rq->rt_time) {
858 raw_spin_lock(&rt_rq->rt_runtime_lock);
859 if (rt_rq->rt_throttled)
860 balance_runtime(rt_rq);
861 runtime = rt_rq->rt_runtime;
862 rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
863 if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
864 rt_rq->rt_throttled = 0;
868 * When we're idle and a woken (rt) task is
869 * throttled check_preempt_curr() will set
870 * skip_update and the time between the wakeup
871 * and this unthrottle will get accounted as
874 if (rt_rq->rt_nr_running && rq->curr == rq->idle)
875 rq_clock_skip_update(rq, false);
877 if (rt_rq->rt_time || rt_rq->rt_nr_running)
879 raw_spin_unlock(&rt_rq->rt_runtime_lock);
880 } else if (rt_rq->rt_nr_running) {
882 if (!rt_rq_throttled(rt_rq))
885 if (rt_rq->rt_throttled)
889 sched_rt_rq_enqueue(rt_rq);
890 raw_spin_unlock(&rq->lock);
893 if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
899 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
901 #ifdef CONFIG_RT_GROUP_SCHED
902 struct rt_rq *rt_rq = group_rt_rq(rt_se);
905 return rt_rq->highest_prio.curr;
908 return rt_task_of(rt_se)->prio;
911 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
913 u64 runtime = sched_rt_runtime(rt_rq);
915 if (rt_rq->rt_throttled)
916 return rt_rq_throttled(rt_rq);
918 if (runtime >= sched_rt_period(rt_rq))
921 balance_runtime(rt_rq);
922 runtime = sched_rt_runtime(rt_rq);
923 if (runtime == RUNTIME_INF)
926 if (rt_rq->rt_time > runtime) {
927 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
930 * Don't actually throttle groups that have no runtime assigned
931 * but accrue some time due to boosting.
933 if (likely(rt_b->rt_runtime)) {
934 rt_rq->rt_throttled = 1;
935 printk_deferred_once("sched: RT throttling activated\n");
938 * In case we did anyway, make it go away,
939 * replenishment is a joke, since it will replenish us
945 if (rt_rq_throttled(rt_rq)) {
946 sched_rt_rq_dequeue(rt_rq);
955 * Update the current task's runtime statistics. Skip current tasks that
956 * are not in our scheduling class.
958 static void update_curr_rt(struct rq *rq)
960 struct task_struct *curr = rq->curr;
961 struct sched_rt_entity *rt_se = &curr->rt;
964 if (curr->sched_class != &rt_sched_class)
967 delta_exec = rq_clock_task(rq) - curr->se.exec_start;
968 if (unlikely((s64)delta_exec <= 0))
971 /* Kick cpufreq (see the comment in kernel/sched/sched.h). */
972 cpufreq_update_util(rq, SCHED_CPUFREQ_RT);
974 schedstat_set(curr->se.statistics.exec_max,
975 max(curr->se.statistics.exec_max, delta_exec));
977 curr->se.sum_exec_runtime += delta_exec;
978 account_group_exec_runtime(curr, delta_exec);
980 curr->se.exec_start = rq_clock_task(rq);
981 cpuacct_charge(curr, delta_exec);
983 sched_rt_avg_update(rq, delta_exec);
985 if (!rt_bandwidth_enabled())
988 for_each_sched_rt_entity(rt_se) {
989 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
992 if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
993 raw_spin_lock(&rt_rq->rt_runtime_lock);
994 rt_rq->rt_time += delta_exec;
995 exceeded = sched_rt_runtime_exceeded(rt_rq);
998 raw_spin_unlock(&rt_rq->rt_runtime_lock);
1000 do_start_rt_bandwidth(sched_rt_bandwidth(rt_rq));
1006 dequeue_top_rt_rq(struct rt_rq *rt_rq)
1008 struct rq *rq = rq_of_rt_rq(rt_rq);
1010 BUG_ON(&rq->rt != rt_rq);
1012 if (!rt_rq->rt_queued)
1015 BUG_ON(!rq->nr_running);
1017 sub_nr_running(rq, rt_rq->rt_nr_running);
1018 rt_rq->rt_queued = 0;
1022 enqueue_top_rt_rq(struct rt_rq *rt_rq)
1024 struct rq *rq = rq_of_rt_rq(rt_rq);
1026 BUG_ON(&rq->rt != rt_rq);
1028 if (rt_rq->rt_queued)
1030 if (rt_rq_throttled(rt_rq) || !rt_rq->rt_nr_running)
1033 add_nr_running(rq, rt_rq->rt_nr_running);
1034 rt_rq->rt_queued = 1;
1037 #if defined CONFIG_SMP
1040 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1042 struct rq *rq = rq_of_rt_rq(rt_rq);
1044 #ifdef CONFIG_RT_GROUP_SCHED
1046 * Change rq's cpupri only if rt_rq is the top queue.
1048 if (&rq->rt != rt_rq)
1051 if (rq->online && prio < prev_prio)
1052 cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
1056 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1058 struct rq *rq = rq_of_rt_rq(rt_rq);
1060 #ifdef CONFIG_RT_GROUP_SCHED
1062 * Change rq's cpupri only if rt_rq is the top queue.
1064 if (&rq->rt != rt_rq)
1067 if (rq->online && rt_rq->highest_prio.curr != prev_prio)
1068 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
1071 #else /* CONFIG_SMP */
1074 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1076 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1078 #endif /* CONFIG_SMP */
1080 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
1082 inc_rt_prio(struct rt_rq *rt_rq, int prio)
1084 int prev_prio = rt_rq->highest_prio.curr;
1086 if (prio < prev_prio)
1087 rt_rq->highest_prio.curr = prio;
1089 inc_rt_prio_smp(rt_rq, prio, prev_prio);
1093 dec_rt_prio(struct rt_rq *rt_rq, int prio)
1095 int prev_prio = rt_rq->highest_prio.curr;
1097 if (rt_rq->rt_nr_running) {
1099 WARN_ON(prio < prev_prio);
1102 * This may have been our highest task, and therefore
1103 * we may have some recomputation to do
1105 if (prio == prev_prio) {
1106 struct rt_prio_array *array = &rt_rq->active;
1108 rt_rq->highest_prio.curr =
1109 sched_find_first_bit(array->bitmap);
1113 rt_rq->highest_prio.curr = MAX_RT_PRIO;
1115 dec_rt_prio_smp(rt_rq, prio, prev_prio);
1120 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
1121 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
1123 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1125 #ifdef CONFIG_RT_GROUP_SCHED
1128 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1130 if (rt_se_boosted(rt_se))
1131 rt_rq->rt_nr_boosted++;
1134 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
1138 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1140 if (rt_se_boosted(rt_se))
1141 rt_rq->rt_nr_boosted--;
1143 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
1146 #else /* CONFIG_RT_GROUP_SCHED */
1149 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1151 start_rt_bandwidth(&def_rt_bandwidth);
1155 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
1157 #endif /* CONFIG_RT_GROUP_SCHED */
1160 unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se)
1162 struct rt_rq *group_rq = group_rt_rq(rt_se);
1165 return group_rq->rt_nr_running;
1171 unsigned int rt_se_rr_nr_running(struct sched_rt_entity *rt_se)
1173 struct rt_rq *group_rq = group_rt_rq(rt_se);
1174 struct task_struct *tsk;
1177 return group_rq->rr_nr_running;
1179 tsk = rt_task_of(rt_se);
1181 return (tsk->policy == SCHED_RR) ? 1 : 0;
1185 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1187 int prio = rt_se_prio(rt_se);
1189 WARN_ON(!rt_prio(prio));
1190 rt_rq->rt_nr_running += rt_se_nr_running(rt_se);
1191 rt_rq->rr_nr_running += rt_se_rr_nr_running(rt_se);
1193 inc_rt_prio(rt_rq, prio);
1194 inc_rt_migration(rt_se, rt_rq);
1195 inc_rt_group(rt_se, rt_rq);
1199 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1201 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
1202 WARN_ON(!rt_rq->rt_nr_running);
1203 rt_rq->rt_nr_running -= rt_se_nr_running(rt_se);
1204 rt_rq->rr_nr_running -= rt_se_rr_nr_running(rt_se);
1206 dec_rt_prio(rt_rq, rt_se_prio(rt_se));
1207 dec_rt_migration(rt_se, rt_rq);
1208 dec_rt_group(rt_se, rt_rq);
1212 * Change rt_se->run_list location unless SAVE && !MOVE
1214 * assumes ENQUEUE/DEQUEUE flags match
1216 static inline bool move_entity(unsigned int flags)
1218 if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
1224 static void __delist_rt_entity(struct sched_rt_entity *rt_se, struct rt_prio_array *array)
1226 list_del_init(&rt_se->run_list);
1228 if (list_empty(array->queue + rt_se_prio(rt_se)))
1229 __clear_bit(rt_se_prio(rt_se), array->bitmap);
1234 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1236 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1237 struct rt_prio_array *array = &rt_rq->active;
1238 struct rt_rq *group_rq = group_rt_rq(rt_se);
1239 struct list_head *queue = array->queue + rt_se_prio(rt_se);
1242 * Don't enqueue the group if its throttled, or when empty.
1243 * The latter is a consequence of the former when a child group
1244 * get throttled and the current group doesn't have any other
1247 if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) {
1249 __delist_rt_entity(rt_se, array);
1253 if (move_entity(flags)) {
1254 WARN_ON_ONCE(rt_se->on_list);
1255 if (flags & ENQUEUE_HEAD)
1256 list_add(&rt_se->run_list, queue);
1258 list_add_tail(&rt_se->run_list, queue);
1260 __set_bit(rt_se_prio(rt_se), array->bitmap);
1265 inc_rt_tasks(rt_se, rt_rq);
1268 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1270 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1271 struct rt_prio_array *array = &rt_rq->active;
1273 if (move_entity(flags)) {
1274 WARN_ON_ONCE(!rt_se->on_list);
1275 __delist_rt_entity(rt_se, array);
1279 dec_rt_tasks(rt_se, rt_rq);
1283 * Because the prio of an upper entry depends on the lower
1284 * entries, we must remove entries top - down.
1286 static void dequeue_rt_stack(struct sched_rt_entity *rt_se, unsigned int flags)
1288 struct sched_rt_entity *back = NULL;
1290 for_each_sched_rt_entity(rt_se) {
1295 dequeue_top_rt_rq(rt_rq_of_se(back));
1297 for (rt_se = back; rt_se; rt_se = rt_se->back) {
1298 if (on_rt_rq(rt_se))
1299 __dequeue_rt_entity(rt_se, flags);
1303 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1305 struct rq *rq = rq_of_rt_se(rt_se);
1307 dequeue_rt_stack(rt_se, flags);
1308 for_each_sched_rt_entity(rt_se)
1309 __enqueue_rt_entity(rt_se, flags);
1310 enqueue_top_rt_rq(&rq->rt);
1313 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1315 struct rq *rq = rq_of_rt_se(rt_se);
1317 dequeue_rt_stack(rt_se, flags);
1319 for_each_sched_rt_entity(rt_se) {
1320 struct rt_rq *rt_rq = group_rt_rq(rt_se);
1322 if (rt_rq && rt_rq->rt_nr_running)
1323 __enqueue_rt_entity(rt_se, flags);
1325 enqueue_top_rt_rq(&rq->rt);
1329 * Adding/removing a task to/from a priority array:
1332 enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1334 struct sched_rt_entity *rt_se = &p->rt;
1336 if (flags & ENQUEUE_WAKEUP)
1339 enqueue_rt_entity(rt_se, flags);
1341 if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1342 enqueue_pushable_task(rq, p);
1345 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1347 struct sched_rt_entity *rt_se = &p->rt;
1350 dequeue_rt_entity(rt_se, flags);
1352 dequeue_pushable_task(rq, p);
1356 * Put task to the head or the end of the run list without the overhead of
1357 * dequeue followed by enqueue.
1360 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
1362 if (on_rt_rq(rt_se)) {
1363 struct rt_prio_array *array = &rt_rq->active;
1364 struct list_head *queue = array->queue + rt_se_prio(rt_se);
1367 list_move(&rt_se->run_list, queue);
1369 list_move_tail(&rt_se->run_list, queue);
1373 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
1375 struct sched_rt_entity *rt_se = &p->rt;
1376 struct rt_rq *rt_rq;
1378 for_each_sched_rt_entity(rt_se) {
1379 rt_rq = rt_rq_of_se(rt_se);
1380 requeue_rt_entity(rt_rq, rt_se, head);
1384 static void yield_task_rt(struct rq *rq)
1386 requeue_task_rt(rq, rq->curr, 0);
1390 static int find_lowest_rq(struct task_struct *task);
1393 select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags)
1395 struct task_struct *curr;
1398 /* For anything but wake ups, just return the task_cpu */
1399 if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK)
1405 curr = READ_ONCE(rq->curr); /* unlocked access */
1408 * If the current task on @p's runqueue is an RT task, then
1409 * try to see if we can wake this RT task up on another
1410 * runqueue. Otherwise simply start this RT task
1411 * on its current runqueue.
1413 * We want to avoid overloading runqueues. If the woken
1414 * task is a higher priority, then it will stay on this CPU
1415 * and the lower prio task should be moved to another CPU.
1416 * Even though this will probably make the lower prio task
1417 * lose its cache, we do not want to bounce a higher task
1418 * around just because it gave up its CPU, perhaps for a
1421 * For equal prio tasks, we just let the scheduler sort it out.
1423 * Otherwise, just let it ride on the affined RQ and the
1424 * post-schedule router will push the preempted task away
1426 * This test is optimistic, if we get it wrong the load-balancer
1427 * will have to sort it out.
1429 if (curr && unlikely(rt_task(curr)) &&
1430 (curr->nr_cpus_allowed < 2 ||
1431 curr->prio <= p->prio)) {
1432 int target = find_lowest_rq(p);
1435 * Don't bother moving it if the destination CPU is
1436 * not running a lower priority task.
1439 p->prio < cpu_rq(target)->rt.highest_prio.curr)
1448 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
1451 * Current can't be migrated, useless to reschedule,
1452 * let's hope p can move out.
1454 if (rq->curr->nr_cpus_allowed == 1 ||
1455 !cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
1459 * p is migratable, so let's not schedule it and
1460 * see if it is pushed or pulled somewhere else.
1462 if (p->nr_cpus_allowed != 1
1463 && cpupri_find(&rq->rd->cpupri, p, NULL))
1467 * There appears to be other cpus that can accept
1468 * current and none to run 'p', so lets reschedule
1469 * to try and push current away:
1471 requeue_task_rt(rq, p, 1);
1475 #endif /* CONFIG_SMP */
1478 * Preempt the current task with a newly woken task if needed:
1480 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1482 if (p->prio < rq->curr->prio) {
1491 * - the newly woken task is of equal priority to the current task
1492 * - the newly woken task is non-migratable while current is migratable
1493 * - current will be preempted on the next reschedule
1495 * we should check to see if current can readily move to a different
1496 * cpu. If so, we will reschedule to allow the push logic to try
1497 * to move current somewhere else, making room for our non-migratable
1500 if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr))
1501 check_preempt_equal_prio(rq, p);
1505 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1506 struct rt_rq *rt_rq)
1508 struct rt_prio_array *array = &rt_rq->active;
1509 struct sched_rt_entity *next = NULL;
1510 struct list_head *queue;
1513 idx = sched_find_first_bit(array->bitmap);
1514 BUG_ON(idx >= MAX_RT_PRIO);
1516 queue = array->queue + idx;
1517 next = list_entry(queue->next, struct sched_rt_entity, run_list);
1522 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1524 struct sched_rt_entity *rt_se;
1525 struct task_struct *p;
1526 struct rt_rq *rt_rq = &rq->rt;
1529 rt_se = pick_next_rt_entity(rq, rt_rq);
1531 rt_rq = group_rt_rq(rt_se);
1534 p = rt_task_of(rt_se);
1535 p->se.exec_start = rq_clock_task(rq);
1540 static struct task_struct *
1541 pick_next_task_rt(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
1543 struct task_struct *p;
1544 struct rt_rq *rt_rq = &rq->rt;
1546 if (need_pull_rt_task(rq, prev)) {
1548 * This is OK, because current is on_cpu, which avoids it being
1549 * picked for load-balance and preemption/IRQs are still
1550 * disabled avoiding further scheduler activity on it and we're
1551 * being very careful to re-start the picking loop.
1553 rq_unpin_lock(rq, rf);
1555 rq_repin_lock(rq, rf);
1557 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1558 * means a dl or stop task can slip in, in which case we need
1559 * to re-start task selection.
1561 if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) ||
1562 rq->dl.dl_nr_running))
1567 * We may dequeue prev's rt_rq in put_prev_task().
1568 * So, we update time before rt_nr_running check.
1570 if (prev->sched_class == &rt_sched_class)
1573 if (!rt_rq->rt_queued)
1576 put_prev_task(rq, prev);
1578 p = _pick_next_task_rt(rq);
1580 /* The running task is never eligible for pushing */
1581 dequeue_pushable_task(rq, p);
1583 queue_push_tasks(rq);
1588 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1593 * The previous task needs to be made eligible for pushing
1594 * if it is still active
1596 if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
1597 enqueue_pushable_task(rq, p);
1602 /* Only try algorithms three times */
1603 #define RT_MAX_TRIES 3
1605 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1607 if (!task_running(rq, p) &&
1608 cpumask_test_cpu(cpu, &p->cpus_allowed))
1614 * Return the highest pushable rq's task, which is suitable to be executed
1615 * on the cpu, NULL otherwise
1617 static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu)
1619 struct plist_head *head = &rq->rt.pushable_tasks;
1620 struct task_struct *p;
1622 if (!has_pushable_tasks(rq))
1625 plist_for_each_entry(p, head, pushable_tasks) {
1626 if (pick_rt_task(rq, p, cpu))
1633 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1635 static int find_lowest_rq(struct task_struct *task)
1637 struct sched_domain *sd;
1638 struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask);
1639 int this_cpu = smp_processor_id();
1640 int cpu = task_cpu(task);
1642 /* Make sure the mask is initialized first */
1643 if (unlikely(!lowest_mask))
1646 if (task->nr_cpus_allowed == 1)
1647 return -1; /* No other targets possible */
1649 if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1650 return -1; /* No targets found */
1653 * At this point we have built a mask of cpus representing the
1654 * lowest priority tasks in the system. Now we want to elect
1655 * the best one based on our affinity and topology.
1657 * We prioritize the last cpu that the task executed on since
1658 * it is most likely cache-hot in that location.
1660 if (cpumask_test_cpu(cpu, lowest_mask))
1664 * Otherwise, we consult the sched_domains span maps to figure
1665 * out which cpu is logically closest to our hot cache data.
1667 if (!cpumask_test_cpu(this_cpu, lowest_mask))
1668 this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1671 for_each_domain(cpu, sd) {
1672 if (sd->flags & SD_WAKE_AFFINE) {
1676 * "this_cpu" is cheaper to preempt than a
1679 if (this_cpu != -1 &&
1680 cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1685 best_cpu = cpumask_first_and(lowest_mask,
1686 sched_domain_span(sd));
1687 if (best_cpu < nr_cpu_ids) {
1696 * And finally, if there were no matches within the domains
1697 * just give the caller *something* to work with from the compatible
1703 cpu = cpumask_any(lowest_mask);
1704 if (cpu < nr_cpu_ids)
1709 /* Will lock the rq it finds */
1710 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1712 struct rq *lowest_rq = NULL;
1716 for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1717 cpu = find_lowest_rq(task);
1719 if ((cpu == -1) || (cpu == rq->cpu))
1722 lowest_rq = cpu_rq(cpu);
1724 if (lowest_rq->rt.highest_prio.curr <= task->prio) {
1726 * Target rq has tasks of equal or higher priority,
1727 * retrying does not release any lock and is unlikely
1728 * to yield a different result.
1734 /* if the prio of this runqueue changed, try again */
1735 if (double_lock_balance(rq, lowest_rq)) {
1737 * We had to unlock the run queue. In
1738 * the mean time, task could have
1739 * migrated already or had its affinity changed.
1740 * Also make sure that it wasn't scheduled on its rq.
1742 if (unlikely(task_rq(task) != rq ||
1743 !cpumask_test_cpu(lowest_rq->cpu, &task->cpus_allowed) ||
1744 task_running(rq, task) ||
1746 !task_on_rq_queued(task))) {
1748 double_unlock_balance(rq, lowest_rq);
1754 /* If this rq is still suitable use it. */
1755 if (lowest_rq->rt.highest_prio.curr > task->prio)
1759 double_unlock_balance(rq, lowest_rq);
1766 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1768 struct task_struct *p;
1770 if (!has_pushable_tasks(rq))
1773 p = plist_first_entry(&rq->rt.pushable_tasks,
1774 struct task_struct, pushable_tasks);
1776 BUG_ON(rq->cpu != task_cpu(p));
1777 BUG_ON(task_current(rq, p));
1778 BUG_ON(p->nr_cpus_allowed <= 1);
1780 BUG_ON(!task_on_rq_queued(p));
1781 BUG_ON(!rt_task(p));
1787 * If the current CPU has more than one RT task, see if the non
1788 * running task can migrate over to a CPU that is running a task
1789 * of lesser priority.
1791 static int push_rt_task(struct rq *rq)
1793 struct task_struct *next_task;
1794 struct rq *lowest_rq;
1797 if (!rq->rt.overloaded)
1800 next_task = pick_next_pushable_task(rq);
1805 if (unlikely(next_task == rq->curr)) {
1811 * It's possible that the next_task slipped in of
1812 * higher priority than current. If that's the case
1813 * just reschedule current.
1815 if (unlikely(next_task->prio < rq->curr->prio)) {
1820 /* We might release rq lock */
1821 get_task_struct(next_task);
1823 /* find_lock_lowest_rq locks the rq if found */
1824 lowest_rq = find_lock_lowest_rq(next_task, rq);
1826 struct task_struct *task;
1828 * find_lock_lowest_rq releases rq->lock
1829 * so it is possible that next_task has migrated.
1831 * We need to make sure that the task is still on the same
1832 * run-queue and is also still the next task eligible for
1835 task = pick_next_pushable_task(rq);
1836 if (task == next_task) {
1838 * The task hasn't migrated, and is still the next
1839 * eligible task, but we failed to find a run-queue
1840 * to push it to. Do not retry in this case, since
1841 * other cpus will pull from us when ready.
1847 /* No more tasks, just exit */
1851 * Something has shifted, try again.
1853 put_task_struct(next_task);
1858 deactivate_task(rq, next_task, 0);
1859 set_task_cpu(next_task, lowest_rq->cpu);
1860 activate_task(lowest_rq, next_task, 0);
1863 resched_curr(lowest_rq);
1865 double_unlock_balance(rq, lowest_rq);
1868 put_task_struct(next_task);
1873 static void push_rt_tasks(struct rq *rq)
1875 /* push_rt_task will return true if it moved an RT */
1876 while (push_rt_task(rq))
1880 #ifdef HAVE_RT_PUSH_IPI
1883 * When a high priority task schedules out from a CPU and a lower priority
1884 * task is scheduled in, a check is made to see if there's any RT tasks
1885 * on other CPUs that are waiting to run because a higher priority RT task
1886 * is currently running on its CPU. In this case, the CPU with multiple RT
1887 * tasks queued on it (overloaded) needs to be notified that a CPU has opened
1888 * up that may be able to run one of its non-running queued RT tasks.
1890 * All CPUs with overloaded RT tasks need to be notified as there is currently
1891 * no way to know which of these CPUs have the highest priority task waiting
1892 * to run. Instead of trying to take a spinlock on each of these CPUs,
1893 * which has shown to cause large latency when done on machines with many
1894 * CPUs, sending an IPI to the CPUs to have them push off the overloaded
1895 * RT tasks waiting to run.
1897 * Just sending an IPI to each of the CPUs is also an issue, as on large
1898 * count CPU machines, this can cause an IPI storm on a CPU, especially
1899 * if its the only CPU with multiple RT tasks queued, and a large number
1900 * of CPUs scheduling a lower priority task at the same time.
1902 * Each root domain has its own irq work function that can iterate over
1903 * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT
1904 * tassk must be checked if there's one or many CPUs that are lowering
1905 * their priority, there's a single irq work iterator that will try to
1906 * push off RT tasks that are waiting to run.
1908 * When a CPU schedules a lower priority task, it will kick off the
1909 * irq work iterator that will jump to each CPU with overloaded RT tasks.
1910 * As it only takes the first CPU that schedules a lower priority task
1911 * to start the process, the rto_start variable is incremented and if
1912 * the atomic result is one, then that CPU will try to take the rto_lock.
1913 * This prevents high contention on the lock as the process handles all
1914 * CPUs scheduling lower priority tasks.
1916 * All CPUs that are scheduling a lower priority task will increment the
1917 * rt_loop_next variable. This will make sure that the irq work iterator
1918 * checks all RT overloaded CPUs whenever a CPU schedules a new lower
1919 * priority task, even if the iterator is in the middle of a scan. Incrementing
1920 * the rt_loop_next will cause the iterator to perform another scan.
1923 static int rto_next_cpu(struct root_domain *rd)
1929 * When starting the IPI RT pushing, the rto_cpu is set to -1,
1930 * rt_next_cpu() will simply return the first CPU found in
1933 * If rto_next_cpu() is called with rto_cpu is a valid cpu, it
1934 * will return the next CPU found in the rto_mask.
1936 * If there are no more CPUs left in the rto_mask, then a check is made
1937 * against rto_loop and rto_loop_next. rto_loop is only updated with
1938 * the rto_lock held, but any CPU may increment the rto_loop_next
1939 * without any locking.
1943 /* When rto_cpu is -1 this acts like cpumask_first() */
1944 cpu = cpumask_next(rd->rto_cpu, rd->rto_mask);
1948 if (cpu < nr_cpu_ids)
1954 * ACQUIRE ensures we see the @rto_mask changes
1955 * made prior to the @next value observed.
1957 * Matches WMB in rt_set_overload().
1959 next = atomic_read_acquire(&rd->rto_loop_next);
1961 if (rd->rto_loop == next)
1964 rd->rto_loop = next;
1970 static inline bool rto_start_trylock(atomic_t *v)
1972 return !atomic_cmpxchg_acquire(v, 0, 1);
1975 static inline void rto_start_unlock(atomic_t *v)
1977 atomic_set_release(v, 0);
1980 static void tell_cpu_to_push(struct rq *rq)
1984 /* Keep the loop going if the IPI is currently active */
1985 atomic_inc(&rq->rd->rto_loop_next);
1987 /* Only one CPU can initiate a loop at a time */
1988 if (!rto_start_trylock(&rq->rd->rto_loop_start))
1991 raw_spin_lock(&rq->rd->rto_lock);
1994 * The rto_cpu is updated under the lock, if it has a valid cpu
1995 * then the IPI is still running and will continue due to the
1996 * update to loop_next, and nothing needs to be done here.
1997 * Otherwise it is finishing up and an ipi needs to be sent.
1999 if (rq->rd->rto_cpu < 0)
2000 cpu = rto_next_cpu(rq->rd);
2002 raw_spin_unlock(&rq->rd->rto_lock);
2004 rto_start_unlock(&rq->rd->rto_loop_start);
2007 /* Make sure the rd does not get freed while pushing */
2008 sched_get_rd(rq->rd);
2009 irq_work_queue_on(&rq->rd->rto_push_work, cpu);
2013 /* Called from hardirq context */
2014 void rto_push_irq_work_func(struct irq_work *work)
2016 struct root_domain *rd =
2017 container_of(work, struct root_domain, rto_push_work);
2024 * We do not need to grab the lock to check for has_pushable_tasks.
2025 * When it gets updated, a check is made if a push is possible.
2027 if (has_pushable_tasks(rq)) {
2028 raw_spin_lock(&rq->lock);
2030 raw_spin_unlock(&rq->lock);
2033 raw_spin_lock(&rd->rto_lock);
2035 /* Pass the IPI to the next rt overloaded queue */
2036 cpu = rto_next_cpu(rd);
2038 raw_spin_unlock(&rd->rto_lock);
2045 /* Try the next RT overloaded CPU */
2046 irq_work_queue_on(&rd->rto_push_work, cpu);
2048 #endif /* HAVE_RT_PUSH_IPI */
2050 static void pull_rt_task(struct rq *this_rq)
2052 int this_cpu = this_rq->cpu, cpu;
2053 bool resched = false;
2054 struct task_struct *p;
2056 int rt_overload_count = rt_overloaded(this_rq);
2058 if (likely(!rt_overload_count))
2062 * Match the barrier from rt_set_overloaded; this guarantees that if we
2063 * see overloaded we must also see the rto_mask bit.
2067 /* If we are the only overloaded CPU do nothing */
2068 if (rt_overload_count == 1 &&
2069 cpumask_test_cpu(this_rq->cpu, this_rq->rd->rto_mask))
2072 #ifdef HAVE_RT_PUSH_IPI
2073 if (sched_feat(RT_PUSH_IPI)) {
2074 tell_cpu_to_push(this_rq);
2079 for_each_cpu(cpu, this_rq->rd->rto_mask) {
2080 if (this_cpu == cpu)
2083 src_rq = cpu_rq(cpu);
2086 * Don't bother taking the src_rq->lock if the next highest
2087 * task is known to be lower-priority than our current task.
2088 * This may look racy, but if this value is about to go
2089 * logically higher, the src_rq will push this task away.
2090 * And if its going logically lower, we do not care
2092 if (src_rq->rt.highest_prio.next >=
2093 this_rq->rt.highest_prio.curr)
2097 * We can potentially drop this_rq's lock in
2098 * double_lock_balance, and another CPU could
2101 double_lock_balance(this_rq, src_rq);
2104 * We can pull only a task, which is pushable
2105 * on its rq, and no others.
2107 p = pick_highest_pushable_task(src_rq, this_cpu);
2110 * Do we have an RT task that preempts
2111 * the to-be-scheduled task?
2113 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
2114 WARN_ON(p == src_rq->curr);
2115 WARN_ON(!task_on_rq_queued(p));
2118 * There's a chance that p is higher in priority
2119 * than what's currently running on its cpu.
2120 * This is just that p is wakeing up and hasn't
2121 * had a chance to schedule. We only pull
2122 * p if it is lower in priority than the
2123 * current task on the run queue
2125 if (p->prio < src_rq->curr->prio)
2130 deactivate_task(src_rq, p, 0);
2131 set_task_cpu(p, this_cpu);
2132 activate_task(this_rq, p, 0);
2134 * We continue with the search, just in
2135 * case there's an even higher prio task
2136 * in another runqueue. (low likelihood
2141 double_unlock_balance(this_rq, src_rq);
2145 resched_curr(this_rq);
2149 * If we are not running and we are not going to reschedule soon, we should
2150 * try to push tasks away now
2152 static void task_woken_rt(struct rq *rq, struct task_struct *p)
2154 if (!task_running(rq, p) &&
2155 !test_tsk_need_resched(rq->curr) &&
2156 p->nr_cpus_allowed > 1 &&
2157 (dl_task(rq->curr) || rt_task(rq->curr)) &&
2158 (rq->curr->nr_cpus_allowed < 2 ||
2159 rq->curr->prio <= p->prio))
2163 /* Assumes rq->lock is held */
2164 static void rq_online_rt(struct rq *rq)
2166 if (rq->rt.overloaded)
2167 rt_set_overload(rq);
2169 __enable_runtime(rq);
2171 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
2174 /* Assumes rq->lock is held */
2175 static void rq_offline_rt(struct rq *rq)
2177 if (rq->rt.overloaded)
2178 rt_clear_overload(rq);
2180 __disable_runtime(rq);
2182 cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
2186 * When switch from the rt queue, we bring ourselves to a position
2187 * that we might want to pull RT tasks from other runqueues.
2189 static void switched_from_rt(struct rq *rq, struct task_struct *p)
2192 * If there are other RT tasks then we will reschedule
2193 * and the scheduling of the other RT tasks will handle
2194 * the balancing. But if we are the last RT task
2195 * we may need to handle the pulling of RT tasks
2198 if (!task_on_rq_queued(p) || rq->rt.rt_nr_running)
2201 queue_pull_task(rq);
2204 void __init init_sched_rt_class(void)
2208 for_each_possible_cpu(i) {
2209 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
2210 GFP_KERNEL, cpu_to_node(i));
2213 #endif /* CONFIG_SMP */
2216 * When switching a task to RT, we may overload the runqueue
2217 * with RT tasks. In this case we try to push them off to
2220 static void switched_to_rt(struct rq *rq, struct task_struct *p)
2223 * If we are already running, then there's nothing
2224 * that needs to be done. But if we are not running
2225 * we may need to preempt the current running task.
2226 * If that current running task is also an RT task
2227 * then see if we can move to another run queue.
2229 if (task_on_rq_queued(p) && rq->curr != p) {
2231 if (p->nr_cpus_allowed > 1 && rq->rt.overloaded)
2232 queue_push_tasks(rq);
2233 #endif /* CONFIG_SMP */
2234 if (p->prio < rq->curr->prio && cpu_online(cpu_of(rq)))
2240 * Priority of the task has changed. This may cause
2241 * us to initiate a push or pull.
2244 prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio)
2246 if (!task_on_rq_queued(p))
2249 if (rq->curr == p) {
2252 * If our priority decreases while running, we
2253 * may need to pull tasks to this runqueue.
2255 if (oldprio < p->prio)
2256 queue_pull_task(rq);
2259 * If there's a higher priority task waiting to run
2262 if (p->prio > rq->rt.highest_prio.curr)
2265 /* For UP simply resched on drop of prio */
2266 if (oldprio < p->prio)
2268 #endif /* CONFIG_SMP */
2271 * This task is not running, but if it is
2272 * greater than the current running task
2275 if (p->prio < rq->curr->prio)
2280 #ifdef CONFIG_POSIX_TIMERS
2281 static void watchdog(struct rq *rq, struct task_struct *p)
2283 unsigned long soft, hard;
2285 /* max may change after cur was read, this will be fixed next tick */
2286 soft = task_rlimit(p, RLIMIT_RTTIME);
2287 hard = task_rlimit_max(p, RLIMIT_RTTIME);
2289 if (soft != RLIM_INFINITY) {
2292 if (p->rt.watchdog_stamp != jiffies) {
2294 p->rt.watchdog_stamp = jiffies;
2297 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
2298 if (p->rt.timeout > next)
2299 p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
2303 static inline void watchdog(struct rq *rq, struct task_struct *p) { }
2306 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
2308 struct sched_rt_entity *rt_se = &p->rt;
2315 * RR tasks need a special form of timeslice management.
2316 * FIFO tasks have no timeslices.
2318 if (p->policy != SCHED_RR)
2321 if (--p->rt.time_slice)
2324 p->rt.time_slice = sched_rr_timeslice;
2327 * Requeue to the end of queue if we (and all of our ancestors) are not
2328 * the only element on the queue
2330 for_each_sched_rt_entity(rt_se) {
2331 if (rt_se->run_list.prev != rt_se->run_list.next) {
2332 requeue_task_rt(rq, p, 0);
2339 static void set_curr_task_rt(struct rq *rq)
2341 struct task_struct *p = rq->curr;
2343 p->se.exec_start = rq_clock_task(rq);
2345 /* The running task is never eligible for pushing */
2346 dequeue_pushable_task(rq, p);
2349 static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
2352 * Time slice is 0 for SCHED_FIFO tasks
2354 if (task->policy == SCHED_RR)
2355 return sched_rr_timeslice;
2360 const struct sched_class rt_sched_class = {
2361 .next = &fair_sched_class,
2362 .enqueue_task = enqueue_task_rt,
2363 .dequeue_task = dequeue_task_rt,
2364 .yield_task = yield_task_rt,
2366 .check_preempt_curr = check_preempt_curr_rt,
2368 .pick_next_task = pick_next_task_rt,
2369 .put_prev_task = put_prev_task_rt,
2372 .select_task_rq = select_task_rq_rt,
2374 .set_cpus_allowed = set_cpus_allowed_common,
2375 .rq_online = rq_online_rt,
2376 .rq_offline = rq_offline_rt,
2377 .task_woken = task_woken_rt,
2378 .switched_from = switched_from_rt,
2381 .set_curr_task = set_curr_task_rt,
2382 .task_tick = task_tick_rt,
2384 .get_rr_interval = get_rr_interval_rt,
2386 .prio_changed = prio_changed_rt,
2387 .switched_to = switched_to_rt,
2389 .update_curr = update_curr_rt,
2392 #ifdef CONFIG_RT_GROUP_SCHED
2394 * Ensure that the real time constraints are schedulable.
2396 static DEFINE_MUTEX(rt_constraints_mutex);
2398 /* Must be called with tasklist_lock held */
2399 static inline int tg_has_rt_tasks(struct task_group *tg)
2401 struct task_struct *g, *p;
2404 * Autogroups do not have RT tasks; see autogroup_create().
2406 if (task_group_is_autogroup(tg))
2409 for_each_process_thread(g, p) {
2410 if (rt_task(p) && task_group(p) == tg)
2417 struct rt_schedulable_data {
2418 struct task_group *tg;
2423 static int tg_rt_schedulable(struct task_group *tg, void *data)
2425 struct rt_schedulable_data *d = data;
2426 struct task_group *child;
2427 unsigned long total, sum = 0;
2428 u64 period, runtime;
2430 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
2431 runtime = tg->rt_bandwidth.rt_runtime;
2434 period = d->rt_period;
2435 runtime = d->rt_runtime;
2439 * Cannot have more runtime than the period.
2441 if (runtime > period && runtime != RUNTIME_INF)
2445 * Ensure we don't starve existing RT tasks.
2447 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
2450 total = to_ratio(period, runtime);
2453 * Nobody can have more than the global setting allows.
2455 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
2459 * The sum of our children's runtime should not exceed our own.
2461 list_for_each_entry_rcu(child, &tg->children, siblings) {
2462 period = ktime_to_ns(child->rt_bandwidth.rt_period);
2463 runtime = child->rt_bandwidth.rt_runtime;
2465 if (child == d->tg) {
2466 period = d->rt_period;
2467 runtime = d->rt_runtime;
2470 sum += to_ratio(period, runtime);
2479 static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
2483 struct rt_schedulable_data data = {
2485 .rt_period = period,
2486 .rt_runtime = runtime,
2490 ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
2496 static int tg_set_rt_bandwidth(struct task_group *tg,
2497 u64 rt_period, u64 rt_runtime)
2502 * Disallowing the root group RT runtime is BAD, it would disallow the
2503 * kernel creating (and or operating) RT threads.
2505 if (tg == &root_task_group && rt_runtime == 0)
2508 /* No period doesn't make any sense. */
2512 mutex_lock(&rt_constraints_mutex);
2513 read_lock(&tasklist_lock);
2514 err = __rt_schedulable(tg, rt_period, rt_runtime);
2518 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
2519 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
2520 tg->rt_bandwidth.rt_runtime = rt_runtime;
2522 for_each_possible_cpu(i) {
2523 struct rt_rq *rt_rq = tg->rt_rq[i];
2525 raw_spin_lock(&rt_rq->rt_runtime_lock);
2526 rt_rq->rt_runtime = rt_runtime;
2527 raw_spin_unlock(&rt_rq->rt_runtime_lock);
2529 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
2531 read_unlock(&tasklist_lock);
2532 mutex_unlock(&rt_constraints_mutex);
2537 int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
2539 u64 rt_runtime, rt_period;
2541 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
2542 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
2543 if (rt_runtime_us < 0)
2544 rt_runtime = RUNTIME_INF;
2545 else if ((u64)rt_runtime_us > U64_MAX / NSEC_PER_USEC)
2548 return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2551 long sched_group_rt_runtime(struct task_group *tg)
2555 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
2558 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
2559 do_div(rt_runtime_us, NSEC_PER_USEC);
2560 return rt_runtime_us;
2563 int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
2565 u64 rt_runtime, rt_period;
2567 if (rt_period_us > U64_MAX / NSEC_PER_USEC)
2570 rt_period = rt_period_us * NSEC_PER_USEC;
2571 rt_runtime = tg->rt_bandwidth.rt_runtime;
2573 return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2576 long sched_group_rt_period(struct task_group *tg)
2580 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
2581 do_div(rt_period_us, NSEC_PER_USEC);
2582 return rt_period_us;
2585 static int sched_rt_global_constraints(void)
2589 mutex_lock(&rt_constraints_mutex);
2590 read_lock(&tasklist_lock);
2591 ret = __rt_schedulable(NULL, 0, 0);
2592 read_unlock(&tasklist_lock);
2593 mutex_unlock(&rt_constraints_mutex);
2598 int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
2600 /* Don't accept realtime tasks when there is no way for them to run */
2601 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
2607 #else /* !CONFIG_RT_GROUP_SCHED */
2608 static int sched_rt_global_constraints(void)
2610 unsigned long flags;
2613 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
2614 for_each_possible_cpu(i) {
2615 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
2617 raw_spin_lock(&rt_rq->rt_runtime_lock);
2618 rt_rq->rt_runtime = global_rt_runtime();
2619 raw_spin_unlock(&rt_rq->rt_runtime_lock);
2621 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
2625 #endif /* CONFIG_RT_GROUP_SCHED */
2627 static int sched_rt_global_validate(void)
2629 if (sysctl_sched_rt_period <= 0)
2632 if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
2633 (sysctl_sched_rt_runtime > sysctl_sched_rt_period))
2639 static void sched_rt_do_global(void)
2641 unsigned long flags;
2643 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
2644 def_rt_bandwidth.rt_runtime = global_rt_runtime();
2645 def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period());
2646 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
2649 int sched_rt_handler(struct ctl_table *table, int write,
2650 void __user *buffer, size_t *lenp,
2653 int old_period, old_runtime;
2654 static DEFINE_MUTEX(mutex);
2658 old_period = sysctl_sched_rt_period;
2659 old_runtime = sysctl_sched_rt_runtime;
2661 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2663 if (!ret && write) {
2664 ret = sched_rt_global_validate();
2668 ret = sched_dl_global_validate();
2672 ret = sched_rt_global_constraints();
2676 sched_rt_do_global();
2677 sched_dl_do_global();
2681 sysctl_sched_rt_period = old_period;
2682 sysctl_sched_rt_runtime = old_runtime;
2684 mutex_unlock(&mutex);
2689 int sched_rr_handler(struct ctl_table *table, int write,
2690 void __user *buffer, size_t *lenp,
2694 static DEFINE_MUTEX(mutex);
2697 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2699 * Make sure that internally we keep jiffies.
2700 * Also, writing zero resets the timeslice to default:
2702 if (!ret && write) {
2703 sched_rr_timeslice =
2704 sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE :
2705 msecs_to_jiffies(sysctl_sched_rr_timeslice);
2707 mutex_unlock(&mutex);
2711 #ifdef CONFIG_SCHED_DEBUG
2712 void print_rt_stats(struct seq_file *m, int cpu)
2715 struct rt_rq *rt_rq;
2718 for_each_rt_rq(rt_rq, iter, cpu_rq(cpu))
2719 print_rt_rq(m, cpu, rt_rq);
2722 #endif /* CONFIG_SCHED_DEBUG */