2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
8 #include <linux/slab.h>
10 int sched_rr_timeslice = RR_TIMESLICE;
12 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
14 struct rt_bandwidth def_rt_bandwidth;
16 static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
18 struct rt_bandwidth *rt_b =
19 container_of(timer, struct rt_bandwidth, rt_period_timer);
25 now = hrtimer_cb_get_time(timer);
26 overrun = hrtimer_forward(timer, now, rt_b->rt_period);
31 idle = do_sched_rt_period_timer(rt_b, overrun);
34 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
37 void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
39 rt_b->rt_period = ns_to_ktime(period);
40 rt_b->rt_runtime = runtime;
42 raw_spin_lock_init(&rt_b->rt_runtime_lock);
44 hrtimer_init(&rt_b->rt_period_timer,
45 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
46 rt_b->rt_period_timer.function = sched_rt_period_timer;
49 static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
51 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
54 if (hrtimer_active(&rt_b->rt_period_timer))
57 raw_spin_lock(&rt_b->rt_runtime_lock);
58 start_bandwidth_timer(&rt_b->rt_period_timer, rt_b->rt_period);
59 raw_spin_unlock(&rt_b->rt_runtime_lock);
62 void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
64 struct rt_prio_array *array;
67 array = &rt_rq->active;
68 for (i = 0; i < MAX_RT_PRIO; i++) {
69 INIT_LIST_HEAD(array->queue + i);
70 __clear_bit(i, array->bitmap);
72 /* delimiter for bitsearch: */
73 __set_bit(MAX_RT_PRIO, array->bitmap);
75 #if defined CONFIG_SMP
76 rt_rq->highest_prio.curr = MAX_RT_PRIO;
77 rt_rq->highest_prio.next = MAX_RT_PRIO;
78 rt_rq->rt_nr_migratory = 0;
79 rt_rq->overloaded = 0;
80 plist_head_init(&rt_rq->pushable_tasks);
84 rt_rq->rt_throttled = 0;
85 rt_rq->rt_runtime = 0;
86 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
89 #ifdef CONFIG_RT_GROUP_SCHED
90 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
92 hrtimer_cancel(&rt_b->rt_period_timer);
95 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
97 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
99 #ifdef CONFIG_SCHED_DEBUG
100 WARN_ON_ONCE(!rt_entity_is_task(rt_se));
102 return container_of(rt_se, struct task_struct, rt);
105 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
110 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
115 void free_rt_sched_group(struct task_group *tg)
120 destroy_rt_bandwidth(&tg->rt_bandwidth);
122 for_each_possible_cpu(i) {
133 void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
134 struct sched_rt_entity *rt_se, int cpu,
135 struct sched_rt_entity *parent)
137 struct rq *rq = cpu_rq(cpu);
139 rt_rq->highest_prio.curr = MAX_RT_PRIO;
140 rt_rq->rt_nr_boosted = 0;
144 tg->rt_rq[cpu] = rt_rq;
145 tg->rt_se[cpu] = rt_se;
151 rt_se->rt_rq = &rq->rt;
153 rt_se->rt_rq = parent->my_q;
156 rt_se->parent = parent;
157 INIT_LIST_HEAD(&rt_se->run_list);
160 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
163 struct sched_rt_entity *rt_se;
166 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
169 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
173 init_rt_bandwidth(&tg->rt_bandwidth,
174 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
176 for_each_possible_cpu(i) {
177 rt_rq = kzalloc_node(sizeof(struct rt_rq),
178 GFP_KERNEL, cpu_to_node(i));
182 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
183 GFP_KERNEL, cpu_to_node(i));
187 init_rt_rq(rt_rq, cpu_rq(i));
188 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
189 init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
200 #else /* CONFIG_RT_GROUP_SCHED */
202 #define rt_entity_is_task(rt_se) (1)
204 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
206 return container_of(rt_se, struct task_struct, rt);
209 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
211 return container_of(rt_rq, struct rq, rt);
214 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
216 struct task_struct *p = rt_task_of(rt_se);
217 struct rq *rq = task_rq(p);
222 void free_rt_sched_group(struct task_group *tg) { }
224 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
228 #endif /* CONFIG_RT_GROUP_SCHED */
232 static inline int rt_overloaded(struct rq *rq)
234 return atomic_read(&rq->rd->rto_count);
237 static inline void rt_set_overload(struct rq *rq)
242 cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
244 * Make sure the mask is visible before we set
245 * the overload count. That is checked to determine
246 * if we should look at the mask. It would be a shame
247 * if we looked at the mask, but the mask was not
251 atomic_inc(&rq->rd->rto_count);
254 static inline void rt_clear_overload(struct rq *rq)
259 /* the order here really doesn't matter */
260 atomic_dec(&rq->rd->rto_count);
261 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
264 static void update_rt_migration(struct rt_rq *rt_rq)
266 if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
267 if (!rt_rq->overloaded) {
268 rt_set_overload(rq_of_rt_rq(rt_rq));
269 rt_rq->overloaded = 1;
271 } else if (rt_rq->overloaded) {
272 rt_clear_overload(rq_of_rt_rq(rt_rq));
273 rt_rq->overloaded = 0;
277 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
279 struct task_struct *p;
281 if (!rt_entity_is_task(rt_se))
284 p = rt_task_of(rt_se);
285 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
287 rt_rq->rt_nr_total++;
288 if (p->nr_cpus_allowed > 1)
289 rt_rq->rt_nr_migratory++;
291 update_rt_migration(rt_rq);
294 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
296 struct task_struct *p;
298 if (!rt_entity_is_task(rt_se))
301 p = rt_task_of(rt_se);
302 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
304 rt_rq->rt_nr_total--;
305 if (p->nr_cpus_allowed > 1)
306 rt_rq->rt_nr_migratory--;
308 update_rt_migration(rt_rq);
311 static inline int has_pushable_tasks(struct rq *rq)
313 return !plist_head_empty(&rq->rt.pushable_tasks);
316 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
318 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
319 plist_node_init(&p->pushable_tasks, p->prio);
320 plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
322 /* Update the highest prio pushable task */
323 if (p->prio < rq->rt.highest_prio.next)
324 rq->rt.highest_prio.next = p->prio;
327 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
329 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
331 /* Update the new highest prio pushable task */
332 if (has_pushable_tasks(rq)) {
333 p = plist_first_entry(&rq->rt.pushable_tasks,
334 struct task_struct, pushable_tasks);
335 rq->rt.highest_prio.next = p->prio;
337 rq->rt.highest_prio.next = MAX_RT_PRIO;
342 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
346 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
351 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
356 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
360 #endif /* CONFIG_SMP */
362 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
364 return !list_empty(&rt_se->run_list);
367 #ifdef CONFIG_RT_GROUP_SCHED
369 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
374 return rt_rq->rt_runtime;
377 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
379 return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
382 typedef struct task_group *rt_rq_iter_t;
384 static inline struct task_group *next_task_group(struct task_group *tg)
387 tg = list_entry_rcu(tg->list.next,
388 typeof(struct task_group), list);
389 } while (&tg->list != &task_groups && task_group_is_autogroup(tg));
391 if (&tg->list == &task_groups)
397 #define for_each_rt_rq(rt_rq, iter, rq) \
398 for (iter = container_of(&task_groups, typeof(*iter), list); \
399 (iter = next_task_group(iter)) && \
400 (rt_rq = iter->rt_rq[cpu_of(rq)]);)
402 static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq)
404 list_add_rcu(&rt_rq->leaf_rt_rq_list,
405 &rq_of_rt_rq(rt_rq)->leaf_rt_rq_list);
408 static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq)
410 list_del_rcu(&rt_rq->leaf_rt_rq_list);
413 #define for_each_leaf_rt_rq(rt_rq, rq) \
414 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
416 #define for_each_sched_rt_entity(rt_se) \
417 for (; rt_se; rt_se = rt_se->parent)
419 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
424 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head);
425 static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
427 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
429 struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
430 struct sched_rt_entity *rt_se;
432 int cpu = cpu_of(rq_of_rt_rq(rt_rq));
434 rt_se = rt_rq->tg->rt_se[cpu];
436 if (rt_rq->rt_nr_running) {
437 if (rt_se && !on_rt_rq(rt_se))
438 enqueue_rt_entity(rt_se, false);
439 if (rt_rq->highest_prio.curr < curr->prio)
444 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
446 struct sched_rt_entity *rt_se;
447 int cpu = cpu_of(rq_of_rt_rq(rt_rq));
449 rt_se = rt_rq->tg->rt_se[cpu];
451 if (rt_se && on_rt_rq(rt_se))
452 dequeue_rt_entity(rt_se);
455 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
457 return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
460 static int rt_se_boosted(struct sched_rt_entity *rt_se)
462 struct rt_rq *rt_rq = group_rt_rq(rt_se);
463 struct task_struct *p;
466 return !!rt_rq->rt_nr_boosted;
468 p = rt_task_of(rt_se);
469 return p->prio != p->normal_prio;
473 static inline const struct cpumask *sched_rt_period_mask(void)
475 return this_rq()->rd->span;
478 static inline const struct cpumask *sched_rt_period_mask(void)
480 return cpu_online_mask;
485 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
487 return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
490 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
492 return &rt_rq->tg->rt_bandwidth;
495 #else /* !CONFIG_RT_GROUP_SCHED */
497 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
499 return rt_rq->rt_runtime;
502 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
504 return ktime_to_ns(def_rt_bandwidth.rt_period);
507 typedef struct rt_rq *rt_rq_iter_t;
509 #define for_each_rt_rq(rt_rq, iter, rq) \
510 for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
512 static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq)
516 static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq)
520 #define for_each_leaf_rt_rq(rt_rq, rq) \
521 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
523 #define for_each_sched_rt_entity(rt_se) \
524 for (; rt_se; rt_se = NULL)
526 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
531 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
533 if (rt_rq->rt_nr_running)
534 resched_task(rq_of_rt_rq(rt_rq)->curr);
537 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
541 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
543 return rt_rq->rt_throttled;
546 static inline const struct cpumask *sched_rt_period_mask(void)
548 return cpu_online_mask;
552 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
554 return &cpu_rq(cpu)->rt;
557 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
559 return &def_rt_bandwidth;
562 #endif /* CONFIG_RT_GROUP_SCHED */
566 * We ran out of runtime, see if we can borrow some from our neighbours.
568 static int do_balance_runtime(struct rt_rq *rt_rq)
570 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
571 struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd;
572 int i, weight, more = 0;
575 weight = cpumask_weight(rd->span);
577 raw_spin_lock(&rt_b->rt_runtime_lock);
578 rt_period = ktime_to_ns(rt_b->rt_period);
579 for_each_cpu(i, rd->span) {
580 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
586 raw_spin_lock(&iter->rt_runtime_lock);
588 * Either all rqs have inf runtime and there's nothing to steal
589 * or __disable_runtime() below sets a specific rq to inf to
590 * indicate its been disabled and disalow stealing.
592 if (iter->rt_runtime == RUNTIME_INF)
596 * From runqueues with spare time, take 1/n part of their
597 * spare time, but no more than our period.
599 diff = iter->rt_runtime - iter->rt_time;
601 diff = div_u64((u64)diff, weight);
602 if (rt_rq->rt_runtime + diff > rt_period)
603 diff = rt_period - rt_rq->rt_runtime;
604 iter->rt_runtime -= diff;
605 rt_rq->rt_runtime += diff;
607 if (rt_rq->rt_runtime == rt_period) {
608 raw_spin_unlock(&iter->rt_runtime_lock);
613 raw_spin_unlock(&iter->rt_runtime_lock);
615 raw_spin_unlock(&rt_b->rt_runtime_lock);
621 * Ensure this RQ takes back all the runtime it lend to its neighbours.
623 static void __disable_runtime(struct rq *rq)
625 struct root_domain *rd = rq->rd;
629 if (unlikely(!scheduler_running))
632 for_each_rt_rq(rt_rq, iter, rq) {
633 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
637 raw_spin_lock(&rt_b->rt_runtime_lock);
638 raw_spin_lock(&rt_rq->rt_runtime_lock);
640 * Either we're all inf and nobody needs to borrow, or we're
641 * already disabled and thus have nothing to do, or we have
642 * exactly the right amount of runtime to take out.
644 if (rt_rq->rt_runtime == RUNTIME_INF ||
645 rt_rq->rt_runtime == rt_b->rt_runtime)
647 raw_spin_unlock(&rt_rq->rt_runtime_lock);
650 * Calculate the difference between what we started out with
651 * and what we current have, that's the amount of runtime
652 * we lend and now have to reclaim.
654 want = rt_b->rt_runtime - rt_rq->rt_runtime;
657 * Greedy reclaim, take back as much as we can.
659 for_each_cpu(i, rd->span) {
660 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
664 * Can't reclaim from ourselves or disabled runqueues.
666 if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
669 raw_spin_lock(&iter->rt_runtime_lock);
671 diff = min_t(s64, iter->rt_runtime, want);
672 iter->rt_runtime -= diff;
675 iter->rt_runtime -= want;
678 raw_spin_unlock(&iter->rt_runtime_lock);
684 raw_spin_lock(&rt_rq->rt_runtime_lock);
686 * We cannot be left wanting - that would mean some runtime
687 * leaked out of the system.
692 * Disable all the borrow logic by pretending we have inf
693 * runtime - in which case borrowing doesn't make sense.
695 rt_rq->rt_runtime = RUNTIME_INF;
696 rt_rq->rt_throttled = 0;
697 raw_spin_unlock(&rt_rq->rt_runtime_lock);
698 raw_spin_unlock(&rt_b->rt_runtime_lock);
702 static void __enable_runtime(struct rq *rq)
707 if (unlikely(!scheduler_running))
711 * Reset each runqueue's bandwidth settings
713 for_each_rt_rq(rt_rq, iter, rq) {
714 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
716 raw_spin_lock(&rt_b->rt_runtime_lock);
717 raw_spin_lock(&rt_rq->rt_runtime_lock);
718 rt_rq->rt_runtime = rt_b->rt_runtime;
720 rt_rq->rt_throttled = 0;
721 raw_spin_unlock(&rt_rq->rt_runtime_lock);
722 raw_spin_unlock(&rt_b->rt_runtime_lock);
726 static int balance_runtime(struct rt_rq *rt_rq)
730 if (!sched_feat(RT_RUNTIME_SHARE))
733 if (rt_rq->rt_time > rt_rq->rt_runtime) {
734 raw_spin_unlock(&rt_rq->rt_runtime_lock);
735 more = do_balance_runtime(rt_rq);
736 raw_spin_lock(&rt_rq->rt_runtime_lock);
741 #else /* !CONFIG_SMP */
742 static inline int balance_runtime(struct rt_rq *rt_rq)
746 #endif /* CONFIG_SMP */
748 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
750 int i, idle = 1, throttled = 0;
751 const struct cpumask *span;
753 span = sched_rt_period_mask();
754 #ifdef CONFIG_RT_GROUP_SCHED
756 * FIXME: isolated CPUs should really leave the root task group,
757 * whether they are isolcpus or were isolated via cpusets, lest
758 * the timer run on a CPU which does not service all runqueues,
759 * potentially leaving other CPUs indefinitely throttled. If
760 * isolation is really required, the user will turn the throttle
761 * off to kill the perturbations it causes anyway. Meanwhile,
762 * this maintains functionality for boot and/or troubleshooting.
764 if (rt_b == &root_task_group.rt_bandwidth)
765 span = cpu_online_mask;
767 for_each_cpu(i, span) {
769 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
770 struct rq *rq = rq_of_rt_rq(rt_rq);
772 raw_spin_lock(&rq->lock);
773 if (rt_rq->rt_time) {
776 raw_spin_lock(&rt_rq->rt_runtime_lock);
777 if (rt_rq->rt_throttled)
778 balance_runtime(rt_rq);
779 runtime = rt_rq->rt_runtime;
780 rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
781 if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
782 rt_rq->rt_throttled = 0;
786 * Force a clock update if the CPU was idle,
787 * lest wakeup -> unthrottle time accumulate.
789 if (rt_rq->rt_nr_running && rq->curr == rq->idle)
790 rq->skip_clock_update = -1;
792 if (rt_rq->rt_time || rt_rq->rt_nr_running)
794 raw_spin_unlock(&rt_rq->rt_runtime_lock);
795 } else if (rt_rq->rt_nr_running) {
797 if (!rt_rq_throttled(rt_rq))
800 if (rt_rq->rt_throttled)
804 sched_rt_rq_enqueue(rt_rq);
805 raw_spin_unlock(&rq->lock);
808 if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
814 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
816 #ifdef CONFIG_RT_GROUP_SCHED
817 struct rt_rq *rt_rq = group_rt_rq(rt_se);
820 return rt_rq->highest_prio.curr;
823 return rt_task_of(rt_se)->prio;
826 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
828 u64 runtime = sched_rt_runtime(rt_rq);
830 if (rt_rq->rt_throttled)
831 return rt_rq_throttled(rt_rq);
833 if (runtime >= sched_rt_period(rt_rq))
836 balance_runtime(rt_rq);
837 runtime = sched_rt_runtime(rt_rq);
838 if (runtime == RUNTIME_INF)
841 if (rt_rq->rt_time > runtime) {
842 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
845 * Don't actually throttle groups that have no runtime assigned
846 * but accrue some time due to boosting.
848 if (likely(rt_b->rt_runtime)) {
849 static bool once = false;
851 rt_rq->rt_throttled = 1;
855 printk_sched("sched: RT throttling activated\n");
859 * In case we did anyway, make it go away,
860 * replenishment is a joke, since it will replenish us
866 if (rt_rq_throttled(rt_rq)) {
867 sched_rt_rq_dequeue(rt_rq);
876 * Update the current task's runtime statistics. Skip current tasks that
877 * are not in our scheduling class.
879 static void update_curr_rt(struct rq *rq)
881 struct task_struct *curr = rq->curr;
882 struct sched_rt_entity *rt_se = &curr->rt;
883 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
886 if (curr->sched_class != &rt_sched_class)
889 delta_exec = rq_clock_task(rq) - curr->se.exec_start;
890 if (unlikely((s64)delta_exec <= 0))
893 schedstat_set(curr->se.statistics.exec_max,
894 max(curr->se.statistics.exec_max, delta_exec));
896 curr->se.sum_exec_runtime += delta_exec;
897 account_group_exec_runtime(curr, delta_exec);
899 curr->se.exec_start = rq_clock_task(rq);
900 cpuacct_charge(curr, delta_exec);
902 sched_rt_avg_update(rq, delta_exec);
904 if (!rt_bandwidth_enabled())
907 for_each_sched_rt_entity(rt_se) {
908 rt_rq = rt_rq_of_se(rt_se);
910 if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
911 raw_spin_lock(&rt_rq->rt_runtime_lock);
912 rt_rq->rt_time += delta_exec;
913 if (sched_rt_runtime_exceeded(rt_rq))
915 raw_spin_unlock(&rt_rq->rt_runtime_lock);
920 #if defined CONFIG_SMP
923 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
925 struct rq *rq = rq_of_rt_rq(rt_rq);
927 if (rq->online && prio < prev_prio)
928 cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
932 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
934 struct rq *rq = rq_of_rt_rq(rt_rq);
936 if (rq->online && rt_rq->highest_prio.curr != prev_prio)
937 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
940 #else /* CONFIG_SMP */
943 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
945 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
947 #endif /* CONFIG_SMP */
949 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
951 inc_rt_prio(struct rt_rq *rt_rq, int prio)
953 int prev_prio = rt_rq->highest_prio.curr;
955 if (prio < prev_prio)
956 rt_rq->highest_prio.curr = prio;
958 inc_rt_prio_smp(rt_rq, prio, prev_prio);
962 dec_rt_prio(struct rt_rq *rt_rq, int prio)
964 int prev_prio = rt_rq->highest_prio.curr;
966 if (rt_rq->rt_nr_running) {
968 WARN_ON(prio < prev_prio);
971 * This may have been our highest task, and therefore
972 * we may have some recomputation to do
974 if (prio == prev_prio) {
975 struct rt_prio_array *array = &rt_rq->active;
977 rt_rq->highest_prio.curr =
978 sched_find_first_bit(array->bitmap);
982 rt_rq->highest_prio.curr = MAX_RT_PRIO;
984 dec_rt_prio_smp(rt_rq, prio, prev_prio);
989 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
990 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
992 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
994 #ifdef CONFIG_RT_GROUP_SCHED
997 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
999 if (rt_se_boosted(rt_se))
1000 rt_rq->rt_nr_boosted++;
1003 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
1007 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1009 if (rt_se_boosted(rt_se))
1010 rt_rq->rt_nr_boosted--;
1012 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
1015 #else /* CONFIG_RT_GROUP_SCHED */
1018 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1020 start_rt_bandwidth(&def_rt_bandwidth);
1024 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
1026 #endif /* CONFIG_RT_GROUP_SCHED */
1029 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1031 int prio = rt_se_prio(rt_se);
1033 WARN_ON(!rt_prio(prio));
1034 rt_rq->rt_nr_running++;
1036 inc_rt_prio(rt_rq, prio);
1037 inc_rt_migration(rt_se, rt_rq);
1038 inc_rt_group(rt_se, rt_rq);
1042 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1044 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
1045 WARN_ON(!rt_rq->rt_nr_running);
1046 rt_rq->rt_nr_running--;
1048 dec_rt_prio(rt_rq, rt_se_prio(rt_se));
1049 dec_rt_migration(rt_se, rt_rq);
1050 dec_rt_group(rt_se, rt_rq);
1053 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
1055 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1056 struct rt_prio_array *array = &rt_rq->active;
1057 struct rt_rq *group_rq = group_rt_rq(rt_se);
1058 struct list_head *queue = array->queue + rt_se_prio(rt_se);
1061 * Don't enqueue the group if its throttled, or when empty.
1062 * The latter is a consequence of the former when a child group
1063 * get throttled and the current group doesn't have any other
1066 if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
1069 if (!rt_rq->rt_nr_running)
1070 list_add_leaf_rt_rq(rt_rq);
1073 list_add(&rt_se->run_list, queue);
1075 list_add_tail(&rt_se->run_list, queue);
1076 __set_bit(rt_se_prio(rt_se), array->bitmap);
1078 inc_rt_tasks(rt_se, rt_rq);
1081 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
1083 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1084 struct rt_prio_array *array = &rt_rq->active;
1086 list_del_init(&rt_se->run_list);
1087 if (list_empty(array->queue + rt_se_prio(rt_se)))
1088 __clear_bit(rt_se_prio(rt_se), array->bitmap);
1090 dec_rt_tasks(rt_se, rt_rq);
1091 if (!rt_rq->rt_nr_running)
1092 list_del_leaf_rt_rq(rt_rq);
1096 * Because the prio of an upper entry depends on the lower
1097 * entries, we must remove entries top - down.
1099 static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
1101 struct sched_rt_entity *back = NULL;
1103 for_each_sched_rt_entity(rt_se) {
1108 for (rt_se = back; rt_se; rt_se = rt_se->back) {
1109 if (on_rt_rq(rt_se))
1110 __dequeue_rt_entity(rt_se);
1114 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
1116 dequeue_rt_stack(rt_se);
1117 for_each_sched_rt_entity(rt_se)
1118 __enqueue_rt_entity(rt_se, head);
1121 static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
1123 dequeue_rt_stack(rt_se);
1125 for_each_sched_rt_entity(rt_se) {
1126 struct rt_rq *rt_rq = group_rt_rq(rt_se);
1128 if (rt_rq && rt_rq->rt_nr_running)
1129 __enqueue_rt_entity(rt_se, false);
1134 * Adding/removing a task to/from a priority array:
1137 enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1139 struct sched_rt_entity *rt_se = &p->rt;
1141 if (flags & ENQUEUE_WAKEUP)
1144 enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD);
1146 if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1147 enqueue_pushable_task(rq, p);
1152 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1154 struct sched_rt_entity *rt_se = &p->rt;
1157 dequeue_rt_entity(rt_se);
1159 dequeue_pushable_task(rq, p);
1165 * Put task to the head or the end of the run list without the overhead of
1166 * dequeue followed by enqueue.
1169 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
1171 if (on_rt_rq(rt_se)) {
1172 struct rt_prio_array *array = &rt_rq->active;
1173 struct list_head *queue = array->queue + rt_se_prio(rt_se);
1176 list_move(&rt_se->run_list, queue);
1178 list_move_tail(&rt_se->run_list, queue);
1182 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
1184 struct sched_rt_entity *rt_se = &p->rt;
1185 struct rt_rq *rt_rq;
1187 for_each_sched_rt_entity(rt_se) {
1188 rt_rq = rt_rq_of_se(rt_se);
1189 requeue_rt_entity(rt_rq, rt_se, head);
1193 static void yield_task_rt(struct rq *rq)
1195 requeue_task_rt(rq, rq->curr, 0);
1199 static int find_lowest_rq(struct task_struct *task);
1202 select_task_rq_rt(struct task_struct *p, int sd_flag, int flags)
1204 struct task_struct *curr;
1210 if (p->nr_cpus_allowed == 1)
1213 /* For anything but wake ups, just return the task_cpu */
1214 if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK)
1220 curr = ACCESS_ONCE(rq->curr); /* unlocked access */
1223 * If the current task on @p's runqueue is an RT task, then
1224 * try to see if we can wake this RT task up on another
1225 * runqueue. Otherwise simply start this RT task
1226 * on its current runqueue.
1228 * We want to avoid overloading runqueues. If the woken
1229 * task is a higher priority, then it will stay on this CPU
1230 * and the lower prio task should be moved to another CPU.
1231 * Even though this will probably make the lower prio task
1232 * lose its cache, we do not want to bounce a higher task
1233 * around just because it gave up its CPU, perhaps for a
1236 * For equal prio tasks, we just let the scheduler sort it out.
1238 * Otherwise, just let it ride on the affined RQ and the
1239 * post-schedule router will push the preempted task away
1241 * This test is optimistic, if we get it wrong the load-balancer
1242 * will have to sort it out.
1244 if (curr && unlikely(rt_task(curr)) &&
1245 (curr->nr_cpus_allowed < 2 ||
1246 curr->prio <= p->prio) &&
1247 (p->nr_cpus_allowed > 1)) {
1248 int target = find_lowest_rq(p);
1259 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
1261 if (rq->curr->nr_cpus_allowed == 1)
1264 if (p->nr_cpus_allowed != 1
1265 && cpupri_find(&rq->rd->cpupri, p, NULL))
1268 if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
1272 * There appears to be other cpus that can accept
1273 * current and none to run 'p', so lets reschedule
1274 * to try and push current away:
1276 requeue_task_rt(rq, p, 1);
1277 resched_task(rq->curr);
1280 #endif /* CONFIG_SMP */
1283 * Preempt the current task with a newly woken task if needed:
1285 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1287 if (p->prio < rq->curr->prio) {
1288 resched_task(rq->curr);
1296 * - the newly woken task is of equal priority to the current task
1297 * - the newly woken task is non-migratable while current is migratable
1298 * - current will be preempted on the next reschedule
1300 * we should check to see if current can readily move to a different
1301 * cpu. If so, we will reschedule to allow the push logic to try
1302 * to move current somewhere else, making room for our non-migratable
1305 if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr))
1306 check_preempt_equal_prio(rq, p);
1310 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1311 struct rt_rq *rt_rq)
1313 struct rt_prio_array *array = &rt_rq->active;
1314 struct sched_rt_entity *next = NULL;
1315 struct list_head *queue;
1318 idx = sched_find_first_bit(array->bitmap);
1319 BUG_ON(idx >= MAX_RT_PRIO);
1321 queue = array->queue + idx;
1322 next = list_entry(queue->next, struct sched_rt_entity, run_list);
1327 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1329 struct sched_rt_entity *rt_se;
1330 struct task_struct *p;
1331 struct rt_rq *rt_rq;
1335 if (!rt_rq->rt_nr_running)
1338 if (rt_rq_throttled(rt_rq))
1342 rt_se = pick_next_rt_entity(rq, rt_rq);
1344 rt_rq = group_rt_rq(rt_se);
1347 p = rt_task_of(rt_se);
1348 p->se.exec_start = rq_clock_task(rq);
1353 static struct task_struct *pick_next_task_rt(struct rq *rq)
1355 struct task_struct *p = _pick_next_task_rt(rq);
1357 /* The running task is never eligible for pushing */
1359 dequeue_pushable_task(rq, p);
1363 * We detect this state here so that we can avoid taking the RQ
1364 * lock again later if there is no need to push
1366 rq->post_schedule = has_pushable_tasks(rq);
1372 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1377 * The previous task needs to be made eligible for pushing
1378 * if it is still active
1380 if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
1381 enqueue_pushable_task(rq, p);
1386 /* Only try algorithms three times */
1387 #define RT_MAX_TRIES 3
1389 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1391 if (!task_running(rq, p) &&
1392 cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
1397 /* Return the second highest RT task, NULL otherwise */
1398 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
1400 struct task_struct *next = NULL;
1401 struct sched_rt_entity *rt_se;
1402 struct rt_prio_array *array;
1403 struct rt_rq *rt_rq;
1406 for_each_leaf_rt_rq(rt_rq, rq) {
1407 array = &rt_rq->active;
1408 idx = sched_find_first_bit(array->bitmap);
1410 if (idx >= MAX_RT_PRIO)
1412 if (next && next->prio <= idx)
1414 list_for_each_entry(rt_se, array->queue + idx, run_list) {
1415 struct task_struct *p;
1417 if (!rt_entity_is_task(rt_se))
1420 p = rt_task_of(rt_se);
1421 if (pick_rt_task(rq, p, cpu)) {
1427 idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
1435 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1437 static int find_lowest_rq(struct task_struct *task)
1439 struct sched_domain *sd;
1440 struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask);
1441 int this_cpu = smp_processor_id();
1442 int cpu = task_cpu(task);
1444 /* Make sure the mask is initialized first */
1445 if (unlikely(!lowest_mask))
1448 if (task->nr_cpus_allowed == 1)
1449 return -1; /* No other targets possible */
1451 if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1452 return -1; /* No targets found */
1455 * At this point we have built a mask of cpus representing the
1456 * lowest priority tasks in the system. Now we want to elect
1457 * the best one based on our affinity and topology.
1459 * We prioritize the last cpu that the task executed on since
1460 * it is most likely cache-hot in that location.
1462 if (cpumask_test_cpu(cpu, lowest_mask))
1466 * Otherwise, we consult the sched_domains span maps to figure
1467 * out which cpu is logically closest to our hot cache data.
1469 if (!cpumask_test_cpu(this_cpu, lowest_mask))
1470 this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1473 for_each_domain(cpu, sd) {
1474 if (sd->flags & SD_WAKE_AFFINE) {
1478 * "this_cpu" is cheaper to preempt than a
1481 if (this_cpu != -1 &&
1482 cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1487 best_cpu = cpumask_first_and(lowest_mask,
1488 sched_domain_span(sd));
1489 if (best_cpu < nr_cpu_ids) {
1498 * And finally, if there were no matches within the domains
1499 * just give the caller *something* to work with from the compatible
1505 cpu = cpumask_any(lowest_mask);
1506 if (cpu < nr_cpu_ids)
1511 /* Will lock the rq it finds */
1512 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1514 struct rq *lowest_rq = NULL;
1518 for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1519 cpu = find_lowest_rq(task);
1521 if ((cpu == -1) || (cpu == rq->cpu))
1524 lowest_rq = cpu_rq(cpu);
1526 /* if the prio of this runqueue changed, try again */
1527 if (double_lock_balance(rq, lowest_rq)) {
1529 * We had to unlock the run queue. In
1530 * the mean time, task could have
1531 * migrated already or had its affinity changed.
1532 * Also make sure that it wasn't scheduled on its rq.
1534 if (unlikely(task_rq(task) != rq ||
1535 !cpumask_test_cpu(lowest_rq->cpu,
1536 tsk_cpus_allowed(task)) ||
1537 task_running(rq, task) ||
1540 double_unlock_balance(rq, lowest_rq);
1546 /* If this rq is still suitable use it. */
1547 if (lowest_rq->rt.highest_prio.curr > task->prio)
1551 double_unlock_balance(rq, lowest_rq);
1558 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1560 struct task_struct *p;
1562 if (!has_pushable_tasks(rq))
1565 p = plist_first_entry(&rq->rt.pushable_tasks,
1566 struct task_struct, pushable_tasks);
1568 BUG_ON(rq->cpu != task_cpu(p));
1569 BUG_ON(task_current(rq, p));
1570 BUG_ON(p->nr_cpus_allowed <= 1);
1573 BUG_ON(!rt_task(p));
1579 * If the current CPU has more than one RT task, see if the non
1580 * running task can migrate over to a CPU that is running a task
1581 * of lesser priority.
1583 static int push_rt_task(struct rq *rq)
1585 struct task_struct *next_task;
1586 struct rq *lowest_rq;
1589 if (!rq->rt.overloaded)
1592 next_task = pick_next_pushable_task(rq);
1597 if (unlikely(next_task == rq->curr)) {
1603 * It's possible that the next_task slipped in of
1604 * higher priority than current. If that's the case
1605 * just reschedule current.
1607 if (unlikely(next_task->prio < rq->curr->prio)) {
1608 resched_task(rq->curr);
1612 /* We might release rq lock */
1613 get_task_struct(next_task);
1615 /* find_lock_lowest_rq locks the rq if found */
1616 lowest_rq = find_lock_lowest_rq(next_task, rq);
1618 struct task_struct *task;
1620 * find_lock_lowest_rq releases rq->lock
1621 * so it is possible that next_task has migrated.
1623 * We need to make sure that the task is still on the same
1624 * run-queue and is also still the next task eligible for
1627 task = pick_next_pushable_task(rq);
1628 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1630 * The task hasn't migrated, and is still the next
1631 * eligible task, but we failed to find a run-queue
1632 * to push it to. Do not retry in this case, since
1633 * other cpus will pull from us when ready.
1639 /* No more tasks, just exit */
1643 * Something has shifted, try again.
1645 put_task_struct(next_task);
1650 deactivate_task(rq, next_task, 0);
1651 set_task_cpu(next_task, lowest_rq->cpu);
1652 activate_task(lowest_rq, next_task, 0);
1655 resched_task(lowest_rq->curr);
1657 double_unlock_balance(rq, lowest_rq);
1660 put_task_struct(next_task);
1665 static void push_rt_tasks(struct rq *rq)
1667 /* push_rt_task will return true if it moved an RT */
1668 while (push_rt_task(rq))
1672 static int pull_rt_task(struct rq *this_rq)
1674 int this_cpu = this_rq->cpu, ret = 0, cpu;
1675 struct task_struct *p;
1678 if (likely(!rt_overloaded(this_rq)))
1681 for_each_cpu(cpu, this_rq->rd->rto_mask) {
1682 if (this_cpu == cpu)
1685 src_rq = cpu_rq(cpu);
1688 * Don't bother taking the src_rq->lock if the next highest
1689 * task is known to be lower-priority than our current task.
1690 * This may look racy, but if this value is about to go
1691 * logically higher, the src_rq will push this task away.
1692 * And if its going logically lower, we do not care
1694 if (src_rq->rt.highest_prio.next >=
1695 this_rq->rt.highest_prio.curr)
1699 * We can potentially drop this_rq's lock in
1700 * double_lock_balance, and another CPU could
1703 double_lock_balance(this_rq, src_rq);
1706 * Are there still pullable RT tasks?
1708 if (src_rq->rt.rt_nr_running <= 1)
1711 p = pick_next_highest_task_rt(src_rq, this_cpu);
1714 * Do we have an RT task that preempts
1715 * the to-be-scheduled task?
1717 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
1718 WARN_ON(p == src_rq->curr);
1722 * There's a chance that p is higher in priority
1723 * than what's currently running on its cpu.
1724 * This is just that p is wakeing up and hasn't
1725 * had a chance to schedule. We only pull
1726 * p if it is lower in priority than the
1727 * current task on the run queue
1729 if (p->prio < src_rq->curr->prio)
1734 deactivate_task(src_rq, p, 0);
1735 set_task_cpu(p, this_cpu);
1736 activate_task(this_rq, p, 0);
1738 * We continue with the search, just in
1739 * case there's an even higher prio task
1740 * in another runqueue. (low likelihood
1745 double_unlock_balance(this_rq, src_rq);
1751 static void pre_schedule_rt(struct rq *rq, struct task_struct *prev)
1753 /* Try to pull RT tasks here if we lower this rq's prio */
1754 if (rq->rt.highest_prio.curr > prev->prio)
1758 static void post_schedule_rt(struct rq *rq)
1764 * If we are not running and we are not going to reschedule soon, we should
1765 * try to push tasks away now
1767 static void task_woken_rt(struct rq *rq, struct task_struct *p)
1769 if (!task_running(rq, p) &&
1770 !test_tsk_need_resched(rq->curr) &&
1771 has_pushable_tasks(rq) &&
1772 p->nr_cpus_allowed > 1 &&
1773 rt_task(rq->curr) &&
1774 (rq->curr->nr_cpus_allowed < 2 ||
1775 rq->curr->prio <= p->prio))
1779 static void set_cpus_allowed_rt(struct task_struct *p,
1780 const struct cpumask *new_mask)
1785 BUG_ON(!rt_task(p));
1790 weight = cpumask_weight(new_mask);
1793 * Only update if the process changes its state from whether it
1794 * can migrate or not.
1796 if ((p->nr_cpus_allowed > 1) == (weight > 1))
1802 * The process used to be able to migrate OR it can now migrate
1805 if (!task_current(rq, p))
1806 dequeue_pushable_task(rq, p);
1807 BUG_ON(!rq->rt.rt_nr_migratory);
1808 rq->rt.rt_nr_migratory--;
1810 if (!task_current(rq, p))
1811 enqueue_pushable_task(rq, p);
1812 rq->rt.rt_nr_migratory++;
1815 update_rt_migration(&rq->rt);
1818 /* Assumes rq->lock is held */
1819 static void rq_online_rt(struct rq *rq)
1821 if (rq->rt.overloaded)
1822 rt_set_overload(rq);
1824 __enable_runtime(rq);
1826 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1829 /* Assumes rq->lock is held */
1830 static void rq_offline_rt(struct rq *rq)
1832 if (rq->rt.overloaded)
1833 rt_clear_overload(rq);
1835 __disable_runtime(rq);
1837 cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
1841 * When switch from the rt queue, we bring ourselves to a position
1842 * that we might want to pull RT tasks from other runqueues.
1844 static void switched_from_rt(struct rq *rq, struct task_struct *p)
1847 * If there are other RT tasks then we will reschedule
1848 * and the scheduling of the other RT tasks will handle
1849 * the balancing. But if we are the last RT task
1850 * we may need to handle the pulling of RT tasks
1853 if (!p->on_rq || rq->rt.rt_nr_running)
1856 if (pull_rt_task(rq))
1857 resched_task(rq->curr);
1860 void init_sched_rt_class(void)
1864 for_each_possible_cpu(i) {
1865 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
1866 GFP_KERNEL, cpu_to_node(i));
1869 #endif /* CONFIG_SMP */
1872 * When switching a task to RT, we may overload the runqueue
1873 * with RT tasks. In this case we try to push them off to
1876 static void switched_to_rt(struct rq *rq, struct task_struct *p)
1878 int check_resched = 1;
1881 * If we are already running, then there's nothing
1882 * that needs to be done. But if we are not running
1883 * we may need to preempt the current running task.
1884 * If that current running task is also an RT task
1885 * then see if we can move to another run queue.
1887 if (p->on_rq && rq->curr != p) {
1889 if (rq->rt.overloaded && push_rt_task(rq) &&
1890 /* Don't resched if we changed runqueues */
1893 #endif /* CONFIG_SMP */
1894 if (check_resched && p->prio < rq->curr->prio)
1895 resched_task(rq->curr);
1900 * Priority of the task has changed. This may cause
1901 * us to initiate a push or pull.
1904 prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio)
1909 if (rq->curr == p) {
1912 * If our priority decreases while running, we
1913 * may need to pull tasks to this runqueue.
1915 if (oldprio < p->prio)
1918 * If there's a higher priority task waiting to run
1919 * then reschedule. Note, the above pull_rt_task
1920 * can release the rq lock and p could migrate.
1921 * Only reschedule if p is still on the same runqueue.
1923 if (p->prio > rq->rt.highest_prio.curr && rq->curr == p)
1926 /* For UP simply resched on drop of prio */
1927 if (oldprio < p->prio)
1929 #endif /* CONFIG_SMP */
1932 * This task is not running, but if it is
1933 * greater than the current running task
1936 if (p->prio < rq->curr->prio)
1937 resched_task(rq->curr);
1941 static void watchdog(struct rq *rq, struct task_struct *p)
1943 unsigned long soft, hard;
1945 /* max may change after cur was read, this will be fixed next tick */
1946 soft = task_rlimit(p, RLIMIT_RTTIME);
1947 hard = task_rlimit_max(p, RLIMIT_RTTIME);
1949 if (soft != RLIM_INFINITY) {
1952 if (p->rt.watchdog_stamp != jiffies) {
1954 p->rt.watchdog_stamp = jiffies;
1957 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
1958 if (p->rt.timeout > next)
1959 p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
1963 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
1965 struct sched_rt_entity *rt_se = &p->rt;
1972 * RR tasks need a special form of timeslice management.
1973 * FIFO tasks have no timeslices.
1975 if (p->policy != SCHED_RR)
1978 if (--p->rt.time_slice)
1981 p->rt.time_slice = sched_rr_timeslice;
1984 * Requeue to the end of queue if we (and all of our ancestors) are the
1985 * only element on the queue
1987 for_each_sched_rt_entity(rt_se) {
1988 if (rt_se->run_list.prev != rt_se->run_list.next) {
1989 requeue_task_rt(rq, p, 0);
1990 set_tsk_need_resched(p);
1996 static void set_curr_task_rt(struct rq *rq)
1998 struct task_struct *p = rq->curr;
2000 p->se.exec_start = rq_clock_task(rq);
2002 /* The running task is never eligible for pushing */
2003 dequeue_pushable_task(rq, p);
2006 static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
2009 * Time slice is 0 for SCHED_FIFO tasks
2011 if (task->policy == SCHED_RR)
2012 return sched_rr_timeslice;
2017 const struct sched_class rt_sched_class = {
2018 .next = &fair_sched_class,
2019 .enqueue_task = enqueue_task_rt,
2020 .dequeue_task = dequeue_task_rt,
2021 .yield_task = yield_task_rt,
2023 .check_preempt_curr = check_preempt_curr_rt,
2025 .pick_next_task = pick_next_task_rt,
2026 .put_prev_task = put_prev_task_rt,
2029 .select_task_rq = select_task_rq_rt,
2031 .set_cpus_allowed = set_cpus_allowed_rt,
2032 .rq_online = rq_online_rt,
2033 .rq_offline = rq_offline_rt,
2034 .pre_schedule = pre_schedule_rt,
2035 .post_schedule = post_schedule_rt,
2036 .task_woken = task_woken_rt,
2037 .switched_from = switched_from_rt,
2040 .set_curr_task = set_curr_task_rt,
2041 .task_tick = task_tick_rt,
2043 .get_rr_interval = get_rr_interval_rt,
2045 .prio_changed = prio_changed_rt,
2046 .switched_to = switched_to_rt,
2049 #ifdef CONFIG_SCHED_DEBUG
2050 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2052 void print_rt_stats(struct seq_file *m, int cpu)
2055 struct rt_rq *rt_rq;
2058 for_each_rt_rq(rt_rq, iter, cpu_rq(cpu))
2059 print_rt_rq(m, cpu, rt_rq);
2062 #endif /* CONFIG_SCHED_DEBUG */