iwmc3200wifi: Fix test of unsigned in iwm_ntf_stop_resume_tx()
[linux-2.6/x86.git] / kernel / sched_rt.c
blob5c5fef3784152396980cd81e5c6c9ebe87c0d93c
1 /*
2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
3 * policies)
4 */
6 #ifdef CONFIG_RT_GROUP_SCHED
8 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
10 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
12 #ifdef CONFIG_SCHED_DEBUG
13 WARN_ON_ONCE(!rt_entity_is_task(rt_se));
14 #endif
15 return container_of(rt_se, struct task_struct, rt);
18 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
20 return rt_rq->rq;
23 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
25 return rt_se->rt_rq;
28 #else /* CONFIG_RT_GROUP_SCHED */
30 #define rt_entity_is_task(rt_se) (1)
32 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
34 return container_of(rt_se, struct task_struct, rt);
37 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
39 return container_of(rt_rq, struct rq, rt);
42 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
44 struct task_struct *p = rt_task_of(rt_se);
45 struct rq *rq = task_rq(p);
47 return &rq->rt;
50 #endif /* CONFIG_RT_GROUP_SCHED */
52 #ifdef CONFIG_SMP
54 static inline int rt_overloaded(struct rq *rq)
56 return atomic_read(&rq->rd->rto_count);
59 static inline void rt_set_overload(struct rq *rq)
61 if (!rq->online)
62 return;
64 cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
66 * Make sure the mask is visible before we set
67 * the overload count. That is checked to determine
68 * if we should look at the mask. It would be a shame
69 * if we looked at the mask, but the mask was not
70 * updated yet.
72 wmb();
73 atomic_inc(&rq->rd->rto_count);
76 static inline void rt_clear_overload(struct rq *rq)
78 if (!rq->online)
79 return;
81 /* the order here really doesn't matter */
82 atomic_dec(&rq->rd->rto_count);
83 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
86 static void update_rt_migration(struct rt_rq *rt_rq)
88 if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
89 if (!rt_rq->overloaded) {
90 rt_set_overload(rq_of_rt_rq(rt_rq));
91 rt_rq->overloaded = 1;
93 } else if (rt_rq->overloaded) {
94 rt_clear_overload(rq_of_rt_rq(rt_rq));
95 rt_rq->overloaded = 0;
99 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
101 if (!rt_entity_is_task(rt_se))
102 return;
104 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
106 rt_rq->rt_nr_total++;
107 if (rt_se->nr_cpus_allowed > 1)
108 rt_rq->rt_nr_migratory++;
110 update_rt_migration(rt_rq);
113 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
115 if (!rt_entity_is_task(rt_se))
116 return;
118 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
120 rt_rq->rt_nr_total--;
121 if (rt_se->nr_cpus_allowed > 1)
122 rt_rq->rt_nr_migratory--;
124 update_rt_migration(rt_rq);
127 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
129 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
130 plist_node_init(&p->pushable_tasks, p->prio);
131 plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
134 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
136 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
139 static inline int has_pushable_tasks(struct rq *rq)
141 return !plist_head_empty(&rq->rt.pushable_tasks);
144 #else
146 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
150 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
154 static inline
155 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
159 static inline
160 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
164 #endif /* CONFIG_SMP */
166 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
168 return !list_empty(&rt_se->run_list);
171 #ifdef CONFIG_RT_GROUP_SCHED
173 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
175 if (!rt_rq->tg)
176 return RUNTIME_INF;
178 return rt_rq->rt_runtime;
181 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
183 return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
186 #define for_each_leaf_rt_rq(rt_rq, rq) \
187 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
189 #define for_each_sched_rt_entity(rt_se) \
190 for (; rt_se; rt_se = rt_se->parent)
192 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
194 return rt_se->my_q;
197 static void enqueue_rt_entity(struct sched_rt_entity *rt_se);
198 static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
200 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
202 struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
203 struct sched_rt_entity *rt_se = rt_rq->rt_se;
205 if (rt_rq->rt_nr_running) {
206 if (rt_se && !on_rt_rq(rt_se))
207 enqueue_rt_entity(rt_se);
208 if (rt_rq->highest_prio.curr < curr->prio)
209 resched_task(curr);
213 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
215 struct sched_rt_entity *rt_se = rt_rq->rt_se;
217 if (rt_se && on_rt_rq(rt_se))
218 dequeue_rt_entity(rt_se);
221 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
223 return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
226 static int rt_se_boosted(struct sched_rt_entity *rt_se)
228 struct rt_rq *rt_rq = group_rt_rq(rt_se);
229 struct task_struct *p;
231 if (rt_rq)
232 return !!rt_rq->rt_nr_boosted;
234 p = rt_task_of(rt_se);
235 return p->prio != p->normal_prio;
238 #ifdef CONFIG_SMP
239 static inline const struct cpumask *sched_rt_period_mask(void)
241 return cpu_rq(smp_processor_id())->rd->span;
243 #else
244 static inline const struct cpumask *sched_rt_period_mask(void)
246 return cpu_online_mask;
248 #endif
250 static inline
251 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
253 return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
256 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
258 return &rt_rq->tg->rt_bandwidth;
261 #else /* !CONFIG_RT_GROUP_SCHED */
263 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
265 return rt_rq->rt_runtime;
268 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
270 return ktime_to_ns(def_rt_bandwidth.rt_period);
273 #define for_each_leaf_rt_rq(rt_rq, rq) \
274 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
276 #define for_each_sched_rt_entity(rt_se) \
277 for (; rt_se; rt_se = NULL)
279 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
281 return NULL;
284 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
286 if (rt_rq->rt_nr_running)
287 resched_task(rq_of_rt_rq(rt_rq)->curr);
290 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
294 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
296 return rt_rq->rt_throttled;
299 static inline const struct cpumask *sched_rt_period_mask(void)
301 return cpu_online_mask;
304 static inline
305 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
307 return &cpu_rq(cpu)->rt;
310 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
312 return &def_rt_bandwidth;
315 #endif /* CONFIG_RT_GROUP_SCHED */
317 #ifdef CONFIG_SMP
319 * We ran out of runtime, see if we can borrow some from our neighbours.
321 static int do_balance_runtime(struct rt_rq *rt_rq)
323 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
324 struct root_domain *rd = cpu_rq(smp_processor_id())->rd;
325 int i, weight, more = 0;
326 u64 rt_period;
328 weight = cpumask_weight(rd->span);
330 spin_lock(&rt_b->rt_runtime_lock);
331 rt_period = ktime_to_ns(rt_b->rt_period);
332 for_each_cpu(i, rd->span) {
333 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
334 s64 diff;
336 if (iter == rt_rq)
337 continue;
339 spin_lock(&iter->rt_runtime_lock);
341 * Either all rqs have inf runtime and there's nothing to steal
342 * or __disable_runtime() below sets a specific rq to inf to
343 * indicate its been disabled and disalow stealing.
345 if (iter->rt_runtime == RUNTIME_INF)
346 goto next;
349 * From runqueues with spare time, take 1/n part of their
350 * spare time, but no more than our period.
352 diff = iter->rt_runtime - iter->rt_time;
353 if (diff > 0) {
354 diff = div_u64((u64)diff, weight);
355 if (rt_rq->rt_runtime + diff > rt_period)
356 diff = rt_period - rt_rq->rt_runtime;
357 iter->rt_runtime -= diff;
358 rt_rq->rt_runtime += diff;
359 more = 1;
360 if (rt_rq->rt_runtime == rt_period) {
361 spin_unlock(&iter->rt_runtime_lock);
362 break;
365 next:
366 spin_unlock(&iter->rt_runtime_lock);
368 spin_unlock(&rt_b->rt_runtime_lock);
370 return more;
374 * Ensure this RQ takes back all the runtime it lend to its neighbours.
376 static void __disable_runtime(struct rq *rq)
378 struct root_domain *rd = rq->rd;
379 struct rt_rq *rt_rq;
381 if (unlikely(!scheduler_running))
382 return;
384 for_each_leaf_rt_rq(rt_rq, rq) {
385 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
386 s64 want;
387 int i;
389 spin_lock(&rt_b->rt_runtime_lock);
390 spin_lock(&rt_rq->rt_runtime_lock);
392 * Either we're all inf and nobody needs to borrow, or we're
393 * already disabled and thus have nothing to do, or we have
394 * exactly the right amount of runtime to take out.
396 if (rt_rq->rt_runtime == RUNTIME_INF ||
397 rt_rq->rt_runtime == rt_b->rt_runtime)
398 goto balanced;
399 spin_unlock(&rt_rq->rt_runtime_lock);
402 * Calculate the difference between what we started out with
403 * and what we current have, that's the amount of runtime
404 * we lend and now have to reclaim.
406 want = rt_b->rt_runtime - rt_rq->rt_runtime;
409 * Greedy reclaim, take back as much as we can.
411 for_each_cpu(i, rd->span) {
412 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
413 s64 diff;
416 * Can't reclaim from ourselves or disabled runqueues.
418 if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
419 continue;
421 spin_lock(&iter->rt_runtime_lock);
422 if (want > 0) {
423 diff = min_t(s64, iter->rt_runtime, want);
424 iter->rt_runtime -= diff;
425 want -= diff;
426 } else {
427 iter->rt_runtime -= want;
428 want -= want;
430 spin_unlock(&iter->rt_runtime_lock);
432 if (!want)
433 break;
436 spin_lock(&rt_rq->rt_runtime_lock);
438 * We cannot be left wanting - that would mean some runtime
439 * leaked out of the system.
441 BUG_ON(want);
442 balanced:
444 * Disable all the borrow logic by pretending we have inf
445 * runtime - in which case borrowing doesn't make sense.
447 rt_rq->rt_runtime = RUNTIME_INF;
448 spin_unlock(&rt_rq->rt_runtime_lock);
449 spin_unlock(&rt_b->rt_runtime_lock);
453 static void disable_runtime(struct rq *rq)
455 unsigned long flags;
457 spin_lock_irqsave(&rq->lock, flags);
458 __disable_runtime(rq);
459 spin_unlock_irqrestore(&rq->lock, flags);
462 static void __enable_runtime(struct rq *rq)
464 struct rt_rq *rt_rq;
466 if (unlikely(!scheduler_running))
467 return;
470 * Reset each runqueue's bandwidth settings
472 for_each_leaf_rt_rq(rt_rq, rq) {
473 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
475 spin_lock(&rt_b->rt_runtime_lock);
476 spin_lock(&rt_rq->rt_runtime_lock);
477 rt_rq->rt_runtime = rt_b->rt_runtime;
478 rt_rq->rt_time = 0;
479 rt_rq->rt_throttled = 0;
480 spin_unlock(&rt_rq->rt_runtime_lock);
481 spin_unlock(&rt_b->rt_runtime_lock);
485 static void enable_runtime(struct rq *rq)
487 unsigned long flags;
489 spin_lock_irqsave(&rq->lock, flags);
490 __enable_runtime(rq);
491 spin_unlock_irqrestore(&rq->lock, flags);
494 static int balance_runtime(struct rt_rq *rt_rq)
496 int more = 0;
498 if (rt_rq->rt_time > rt_rq->rt_runtime) {
499 spin_unlock(&rt_rq->rt_runtime_lock);
500 more = do_balance_runtime(rt_rq);
501 spin_lock(&rt_rq->rt_runtime_lock);
504 return more;
506 #else /* !CONFIG_SMP */
507 static inline int balance_runtime(struct rt_rq *rt_rq)
509 return 0;
511 #endif /* CONFIG_SMP */
513 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
515 int i, idle = 1;
516 const struct cpumask *span;
518 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
519 return 1;
521 span = sched_rt_period_mask();
522 for_each_cpu(i, span) {
523 int enqueue = 0;
524 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
525 struct rq *rq = rq_of_rt_rq(rt_rq);
527 spin_lock(&rq->lock);
528 if (rt_rq->rt_time) {
529 u64 runtime;
531 spin_lock(&rt_rq->rt_runtime_lock);
532 if (rt_rq->rt_throttled)
533 balance_runtime(rt_rq);
534 runtime = rt_rq->rt_runtime;
535 rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
536 if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
537 rt_rq->rt_throttled = 0;
538 enqueue = 1;
540 if (rt_rq->rt_time || rt_rq->rt_nr_running)
541 idle = 0;
542 spin_unlock(&rt_rq->rt_runtime_lock);
543 } else if (rt_rq->rt_nr_running)
544 idle = 0;
546 if (enqueue)
547 sched_rt_rq_enqueue(rt_rq);
548 spin_unlock(&rq->lock);
551 return idle;
554 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
556 #ifdef CONFIG_RT_GROUP_SCHED
557 struct rt_rq *rt_rq = group_rt_rq(rt_se);
559 if (rt_rq)
560 return rt_rq->highest_prio.curr;
561 #endif
563 return rt_task_of(rt_se)->prio;
566 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
568 u64 runtime = sched_rt_runtime(rt_rq);
570 if (rt_rq->rt_throttled)
571 return rt_rq_throttled(rt_rq);
573 if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq))
574 return 0;
576 balance_runtime(rt_rq);
577 runtime = sched_rt_runtime(rt_rq);
578 if (runtime == RUNTIME_INF)
579 return 0;
581 if (rt_rq->rt_time > runtime) {
582 rt_rq->rt_throttled = 1;
583 if (rt_rq_throttled(rt_rq)) {
584 sched_rt_rq_dequeue(rt_rq);
585 return 1;
589 return 0;
593 * Update the current task's runtime statistics. Skip current tasks that
594 * are not in our scheduling class.
596 static void update_curr_rt(struct rq *rq)
598 struct task_struct *curr = rq->curr;
599 struct sched_rt_entity *rt_se = &curr->rt;
600 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
601 u64 delta_exec;
603 if (!task_has_rt_policy(curr))
604 return;
606 delta_exec = rq->clock - curr->se.exec_start;
607 if (unlikely((s64)delta_exec < 0))
608 delta_exec = 0;
610 schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec));
612 curr->se.sum_exec_runtime += delta_exec;
613 account_group_exec_runtime(curr, delta_exec);
615 curr->se.exec_start = rq->clock;
616 cpuacct_charge(curr, delta_exec);
618 sched_rt_avg_update(rq, delta_exec);
620 if (!rt_bandwidth_enabled())
621 return;
623 for_each_sched_rt_entity(rt_se) {
624 rt_rq = rt_rq_of_se(rt_se);
626 if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
627 spin_lock(&rt_rq->rt_runtime_lock);
628 rt_rq->rt_time += delta_exec;
629 if (sched_rt_runtime_exceeded(rt_rq))
630 resched_task(curr);
631 spin_unlock(&rt_rq->rt_runtime_lock);
636 #if defined CONFIG_SMP
638 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu);
640 static inline int next_prio(struct rq *rq)
642 struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu);
644 if (next && rt_prio(next->prio))
645 return next->prio;
646 else
647 return MAX_RT_PRIO;
650 static void
651 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
653 struct rq *rq = rq_of_rt_rq(rt_rq);
655 if (prio < prev_prio) {
658 * If the new task is higher in priority than anything on the
659 * run-queue, we know that the previous high becomes our
660 * next-highest.
662 rt_rq->highest_prio.next = prev_prio;
664 if (rq->online)
665 cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
667 } else if (prio == rt_rq->highest_prio.curr)
669 * If the next task is equal in priority to the highest on
670 * the run-queue, then we implicitly know that the next highest
671 * task cannot be any lower than current
673 rt_rq->highest_prio.next = prio;
674 else if (prio < rt_rq->highest_prio.next)
676 * Otherwise, we need to recompute next-highest
678 rt_rq->highest_prio.next = next_prio(rq);
681 static void
682 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
684 struct rq *rq = rq_of_rt_rq(rt_rq);
686 if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next))
687 rt_rq->highest_prio.next = next_prio(rq);
689 if (rq->online && rt_rq->highest_prio.curr != prev_prio)
690 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
693 #else /* CONFIG_SMP */
695 static inline
696 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
697 static inline
698 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
700 #endif /* CONFIG_SMP */
702 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
703 static void
704 inc_rt_prio(struct rt_rq *rt_rq, int prio)
706 int prev_prio = rt_rq->highest_prio.curr;
708 if (prio < prev_prio)
709 rt_rq->highest_prio.curr = prio;
711 inc_rt_prio_smp(rt_rq, prio, prev_prio);
714 static void
715 dec_rt_prio(struct rt_rq *rt_rq, int prio)
717 int prev_prio = rt_rq->highest_prio.curr;
719 if (rt_rq->rt_nr_running) {
721 WARN_ON(prio < prev_prio);
724 * This may have been our highest task, and therefore
725 * we may have some recomputation to do
727 if (prio == prev_prio) {
728 struct rt_prio_array *array = &rt_rq->active;
730 rt_rq->highest_prio.curr =
731 sched_find_first_bit(array->bitmap);
734 } else
735 rt_rq->highest_prio.curr = MAX_RT_PRIO;
737 dec_rt_prio_smp(rt_rq, prio, prev_prio);
740 #else
742 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
743 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
745 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
747 #ifdef CONFIG_RT_GROUP_SCHED
749 static void
750 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
752 if (rt_se_boosted(rt_se))
753 rt_rq->rt_nr_boosted++;
755 if (rt_rq->tg)
756 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
759 static void
760 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
762 if (rt_se_boosted(rt_se))
763 rt_rq->rt_nr_boosted--;
765 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
768 #else /* CONFIG_RT_GROUP_SCHED */
770 static void
771 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
773 start_rt_bandwidth(&def_rt_bandwidth);
776 static inline
777 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
779 #endif /* CONFIG_RT_GROUP_SCHED */
781 static inline
782 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
784 int prio = rt_se_prio(rt_se);
786 WARN_ON(!rt_prio(prio));
787 rt_rq->rt_nr_running++;
789 inc_rt_prio(rt_rq, prio);
790 inc_rt_migration(rt_se, rt_rq);
791 inc_rt_group(rt_se, rt_rq);
794 static inline
795 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
797 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
798 WARN_ON(!rt_rq->rt_nr_running);
799 rt_rq->rt_nr_running--;
801 dec_rt_prio(rt_rq, rt_se_prio(rt_se));
802 dec_rt_migration(rt_se, rt_rq);
803 dec_rt_group(rt_se, rt_rq);
806 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se)
808 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
809 struct rt_prio_array *array = &rt_rq->active;
810 struct rt_rq *group_rq = group_rt_rq(rt_se);
811 struct list_head *queue = array->queue + rt_se_prio(rt_se);
814 * Don't enqueue the group if its throttled, or when empty.
815 * The latter is a consequence of the former when a child group
816 * get throttled and the current group doesn't have any other
817 * active members.
819 if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
820 return;
822 list_add_tail(&rt_se->run_list, queue);
823 __set_bit(rt_se_prio(rt_se), array->bitmap);
825 inc_rt_tasks(rt_se, rt_rq);
828 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
830 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
831 struct rt_prio_array *array = &rt_rq->active;
833 list_del_init(&rt_se->run_list);
834 if (list_empty(array->queue + rt_se_prio(rt_se)))
835 __clear_bit(rt_se_prio(rt_se), array->bitmap);
837 dec_rt_tasks(rt_se, rt_rq);
841 * Because the prio of an upper entry depends on the lower
842 * entries, we must remove entries top - down.
844 static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
846 struct sched_rt_entity *back = NULL;
848 for_each_sched_rt_entity(rt_se) {
849 rt_se->back = back;
850 back = rt_se;
853 for (rt_se = back; rt_se; rt_se = rt_se->back) {
854 if (on_rt_rq(rt_se))
855 __dequeue_rt_entity(rt_se);
859 static void enqueue_rt_entity(struct sched_rt_entity *rt_se)
861 dequeue_rt_stack(rt_se);
862 for_each_sched_rt_entity(rt_se)
863 __enqueue_rt_entity(rt_se);
866 static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
868 dequeue_rt_stack(rt_se);
870 for_each_sched_rt_entity(rt_se) {
871 struct rt_rq *rt_rq = group_rt_rq(rt_se);
873 if (rt_rq && rt_rq->rt_nr_running)
874 __enqueue_rt_entity(rt_se);
879 * Adding/removing a task to/from a priority array:
881 static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
883 struct sched_rt_entity *rt_se = &p->rt;
885 if (wakeup)
886 rt_se->timeout = 0;
888 enqueue_rt_entity(rt_se);
890 if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1)
891 enqueue_pushable_task(rq, p);
894 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
896 struct sched_rt_entity *rt_se = &p->rt;
898 update_curr_rt(rq);
899 dequeue_rt_entity(rt_se);
901 dequeue_pushable_task(rq, p);
905 * Put task to the end of the run list without the overhead of dequeue
906 * followed by enqueue.
908 static void
909 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
911 if (on_rt_rq(rt_se)) {
912 struct rt_prio_array *array = &rt_rq->active;
913 struct list_head *queue = array->queue + rt_se_prio(rt_se);
915 if (head)
916 list_move(&rt_se->run_list, queue);
917 else
918 list_move_tail(&rt_se->run_list, queue);
922 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
924 struct sched_rt_entity *rt_se = &p->rt;
925 struct rt_rq *rt_rq;
927 for_each_sched_rt_entity(rt_se) {
928 rt_rq = rt_rq_of_se(rt_se);
929 requeue_rt_entity(rt_rq, rt_se, head);
933 static void yield_task_rt(struct rq *rq)
935 requeue_task_rt(rq, rq->curr, 0);
938 #ifdef CONFIG_SMP
939 static int find_lowest_rq(struct task_struct *task);
941 static int select_task_rq_rt(struct task_struct *p, int sd_flag, int flags)
943 struct rq *rq = task_rq(p);
945 if (sd_flag != SD_BALANCE_WAKE)
946 return smp_processor_id();
949 * If the current task is an RT task, then
950 * try to see if we can wake this RT task up on another
951 * runqueue. Otherwise simply start this RT task
952 * on its current runqueue.
954 * We want to avoid overloading runqueues. Even if
955 * the RT task is of higher priority than the current RT task.
956 * RT tasks behave differently than other tasks. If
957 * one gets preempted, we try to push it off to another queue.
958 * So trying to keep a preempting RT task on the same
959 * cache hot CPU will force the running RT task to
960 * a cold CPU. So we waste all the cache for the lower
961 * RT task in hopes of saving some of a RT task
962 * that is just being woken and probably will have
963 * cold cache anyway.
965 if (unlikely(rt_task(rq->curr)) &&
966 (p->rt.nr_cpus_allowed > 1)) {
967 int cpu = find_lowest_rq(p);
969 return (cpu == -1) ? task_cpu(p) : cpu;
973 * Otherwise, just let it ride on the affined RQ and the
974 * post-schedule router will push the preempted task away
976 return task_cpu(p);
979 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
981 if (rq->curr->rt.nr_cpus_allowed == 1)
982 return;
984 if (p->rt.nr_cpus_allowed != 1
985 && cpupri_find(&rq->rd->cpupri, p, NULL))
986 return;
988 if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
989 return;
992 * There appears to be other cpus that can accept
993 * current and none to run 'p', so lets reschedule
994 * to try and push current away:
996 requeue_task_rt(rq, p, 1);
997 resched_task(rq->curr);
1000 #endif /* CONFIG_SMP */
1003 * Preempt the current task with a newly woken task if needed:
1005 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1007 if (p->prio < rq->curr->prio) {
1008 resched_task(rq->curr);
1009 return;
1012 #ifdef CONFIG_SMP
1014 * If:
1016 * - the newly woken task is of equal priority to the current task
1017 * - the newly woken task is non-migratable while current is migratable
1018 * - current will be preempted on the next reschedule
1020 * we should check to see if current can readily move to a different
1021 * cpu. If so, we will reschedule to allow the push logic to try
1022 * to move current somewhere else, making room for our non-migratable
1023 * task.
1025 if (p->prio == rq->curr->prio && !need_resched())
1026 check_preempt_equal_prio(rq, p);
1027 #endif
1030 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1031 struct rt_rq *rt_rq)
1033 struct rt_prio_array *array = &rt_rq->active;
1034 struct sched_rt_entity *next = NULL;
1035 struct list_head *queue;
1036 int idx;
1038 idx = sched_find_first_bit(array->bitmap);
1039 BUG_ON(idx >= MAX_RT_PRIO);
1041 queue = array->queue + idx;
1042 next = list_entry(queue->next, struct sched_rt_entity, run_list);
1044 return next;
1047 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1049 struct sched_rt_entity *rt_se;
1050 struct task_struct *p;
1051 struct rt_rq *rt_rq;
1053 rt_rq = &rq->rt;
1055 if (unlikely(!rt_rq->rt_nr_running))
1056 return NULL;
1058 if (rt_rq_throttled(rt_rq))
1059 return NULL;
1061 do {
1062 rt_se = pick_next_rt_entity(rq, rt_rq);
1063 BUG_ON(!rt_se);
1064 rt_rq = group_rt_rq(rt_se);
1065 } while (rt_rq);
1067 p = rt_task_of(rt_se);
1068 p->se.exec_start = rq->clock;
1070 return p;
1073 static struct task_struct *pick_next_task_rt(struct rq *rq)
1075 struct task_struct *p = _pick_next_task_rt(rq);
1077 /* The running task is never eligible for pushing */
1078 if (p)
1079 dequeue_pushable_task(rq, p);
1081 #ifdef CONFIG_SMP
1083 * We detect this state here so that we can avoid taking the RQ
1084 * lock again later if there is no need to push
1086 rq->post_schedule = has_pushable_tasks(rq);
1087 #endif
1089 return p;
1092 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1094 update_curr_rt(rq);
1095 p->se.exec_start = 0;
1098 * The previous task needs to be made eligible for pushing
1099 * if it is still active
1101 if (p->se.on_rq && p->rt.nr_cpus_allowed > 1)
1102 enqueue_pushable_task(rq, p);
1105 #ifdef CONFIG_SMP
1107 /* Only try algorithms three times */
1108 #define RT_MAX_TRIES 3
1110 static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep);
1112 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1114 if (!task_running(rq, p) &&
1115 (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) &&
1116 (p->rt.nr_cpus_allowed > 1))
1117 return 1;
1118 return 0;
1121 /* Return the second highest RT task, NULL otherwise */
1122 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
1124 struct task_struct *next = NULL;
1125 struct sched_rt_entity *rt_se;
1126 struct rt_prio_array *array;
1127 struct rt_rq *rt_rq;
1128 int idx;
1130 for_each_leaf_rt_rq(rt_rq, rq) {
1131 array = &rt_rq->active;
1132 idx = sched_find_first_bit(array->bitmap);
1133 next_idx:
1134 if (idx >= MAX_RT_PRIO)
1135 continue;
1136 if (next && next->prio < idx)
1137 continue;
1138 list_for_each_entry(rt_se, array->queue + idx, run_list) {
1139 struct task_struct *p = rt_task_of(rt_se);
1140 if (pick_rt_task(rq, p, cpu)) {
1141 next = p;
1142 break;
1145 if (!next) {
1146 idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
1147 goto next_idx;
1151 return next;
1154 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1156 static int find_lowest_rq(struct task_struct *task)
1158 struct sched_domain *sd;
1159 struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask);
1160 int this_cpu = smp_processor_id();
1161 int cpu = task_cpu(task);
1163 if (task->rt.nr_cpus_allowed == 1)
1164 return -1; /* No other targets possible */
1166 if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1167 return -1; /* No targets found */
1170 * At this point we have built a mask of cpus representing the
1171 * lowest priority tasks in the system. Now we want to elect
1172 * the best one based on our affinity and topology.
1174 * We prioritize the last cpu that the task executed on since
1175 * it is most likely cache-hot in that location.
1177 if (cpumask_test_cpu(cpu, lowest_mask))
1178 return cpu;
1181 * Otherwise, we consult the sched_domains span maps to figure
1182 * out which cpu is logically closest to our hot cache data.
1184 if (!cpumask_test_cpu(this_cpu, lowest_mask))
1185 this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1187 for_each_domain(cpu, sd) {
1188 if (sd->flags & SD_WAKE_AFFINE) {
1189 int best_cpu;
1192 * "this_cpu" is cheaper to preempt than a
1193 * remote processor.
1195 if (this_cpu != -1 &&
1196 cpumask_test_cpu(this_cpu, sched_domain_span(sd)))
1197 return this_cpu;
1199 best_cpu = cpumask_first_and(lowest_mask,
1200 sched_domain_span(sd));
1201 if (best_cpu < nr_cpu_ids)
1202 return best_cpu;
1207 * And finally, if there were no matches within the domains
1208 * just give the caller *something* to work with from the compatible
1209 * locations.
1211 if (this_cpu != -1)
1212 return this_cpu;
1214 cpu = cpumask_any(lowest_mask);
1215 if (cpu < nr_cpu_ids)
1216 return cpu;
1217 return -1;
1220 /* Will lock the rq it finds */
1221 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1223 struct rq *lowest_rq = NULL;
1224 int tries;
1225 int cpu;
1227 for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1228 cpu = find_lowest_rq(task);
1230 if ((cpu == -1) || (cpu == rq->cpu))
1231 break;
1233 lowest_rq = cpu_rq(cpu);
1235 /* if the prio of this runqueue changed, try again */
1236 if (double_lock_balance(rq, lowest_rq)) {
1238 * We had to unlock the run queue. In
1239 * the mean time, task could have
1240 * migrated already or had its affinity changed.
1241 * Also make sure that it wasn't scheduled on its rq.
1243 if (unlikely(task_rq(task) != rq ||
1244 !cpumask_test_cpu(lowest_rq->cpu,
1245 &task->cpus_allowed) ||
1246 task_running(rq, task) ||
1247 !task->se.on_rq)) {
1249 spin_unlock(&lowest_rq->lock);
1250 lowest_rq = NULL;
1251 break;
1255 /* If this rq is still suitable use it. */
1256 if (lowest_rq->rt.highest_prio.curr > task->prio)
1257 break;
1259 /* try again */
1260 double_unlock_balance(rq, lowest_rq);
1261 lowest_rq = NULL;
1264 return lowest_rq;
1267 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1269 struct task_struct *p;
1271 if (!has_pushable_tasks(rq))
1272 return NULL;
1274 p = plist_first_entry(&rq->rt.pushable_tasks,
1275 struct task_struct, pushable_tasks);
1277 BUG_ON(rq->cpu != task_cpu(p));
1278 BUG_ON(task_current(rq, p));
1279 BUG_ON(p->rt.nr_cpus_allowed <= 1);
1281 BUG_ON(!p->se.on_rq);
1282 BUG_ON(!rt_task(p));
1284 return p;
1288 * If the current CPU has more than one RT task, see if the non
1289 * running task can migrate over to a CPU that is running a task
1290 * of lesser priority.
1292 static int push_rt_task(struct rq *rq)
1294 struct task_struct *next_task;
1295 struct rq *lowest_rq;
1297 if (!rq->rt.overloaded)
1298 return 0;
1300 next_task = pick_next_pushable_task(rq);
1301 if (!next_task)
1302 return 0;
1304 retry:
1305 if (unlikely(next_task == rq->curr)) {
1306 WARN_ON(1);
1307 return 0;
1311 * It's possible that the next_task slipped in of
1312 * higher priority than current. If that's the case
1313 * just reschedule current.
1315 if (unlikely(next_task->prio < rq->curr->prio)) {
1316 resched_task(rq->curr);
1317 return 0;
1320 /* We might release rq lock */
1321 get_task_struct(next_task);
1323 /* find_lock_lowest_rq locks the rq if found */
1324 lowest_rq = find_lock_lowest_rq(next_task, rq);
1325 if (!lowest_rq) {
1326 struct task_struct *task;
1328 * find lock_lowest_rq releases rq->lock
1329 * so it is possible that next_task has migrated.
1331 * We need to make sure that the task is still on the same
1332 * run-queue and is also still the next task eligible for
1333 * pushing.
1335 task = pick_next_pushable_task(rq);
1336 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1338 * If we get here, the task hasnt moved at all, but
1339 * it has failed to push. We will not try again,
1340 * since the other cpus will pull from us when they
1341 * are ready.
1343 dequeue_pushable_task(rq, next_task);
1344 goto out;
1347 if (!task)
1348 /* No more tasks, just exit */
1349 goto out;
1352 * Something has shifted, try again.
1354 put_task_struct(next_task);
1355 next_task = task;
1356 goto retry;
1359 deactivate_task(rq, next_task, 0);
1360 set_task_cpu(next_task, lowest_rq->cpu);
1361 activate_task(lowest_rq, next_task, 0);
1363 resched_task(lowest_rq->curr);
1365 double_unlock_balance(rq, lowest_rq);
1367 out:
1368 put_task_struct(next_task);
1370 return 1;
1373 static void push_rt_tasks(struct rq *rq)
1375 /* push_rt_task will return true if it moved an RT */
1376 while (push_rt_task(rq))
1380 static int pull_rt_task(struct rq *this_rq)
1382 int this_cpu = this_rq->cpu, ret = 0, cpu;
1383 struct task_struct *p;
1384 struct rq *src_rq;
1386 if (likely(!rt_overloaded(this_rq)))
1387 return 0;
1389 for_each_cpu(cpu, this_rq->rd->rto_mask) {
1390 if (this_cpu == cpu)
1391 continue;
1393 src_rq = cpu_rq(cpu);
1396 * Don't bother taking the src_rq->lock if the next highest
1397 * task is known to be lower-priority than our current task.
1398 * This may look racy, but if this value is about to go
1399 * logically higher, the src_rq will push this task away.
1400 * And if its going logically lower, we do not care
1402 if (src_rq->rt.highest_prio.next >=
1403 this_rq->rt.highest_prio.curr)
1404 continue;
1407 * We can potentially drop this_rq's lock in
1408 * double_lock_balance, and another CPU could
1409 * alter this_rq
1411 double_lock_balance(this_rq, src_rq);
1414 * Are there still pullable RT tasks?
1416 if (src_rq->rt.rt_nr_running <= 1)
1417 goto skip;
1419 p = pick_next_highest_task_rt(src_rq, this_cpu);
1422 * Do we have an RT task that preempts
1423 * the to-be-scheduled task?
1425 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
1426 WARN_ON(p == src_rq->curr);
1427 WARN_ON(!p->se.on_rq);
1430 * There's a chance that p is higher in priority
1431 * than what's currently running on its cpu.
1432 * This is just that p is wakeing up and hasn't
1433 * had a chance to schedule. We only pull
1434 * p if it is lower in priority than the
1435 * current task on the run queue
1437 if (p->prio < src_rq->curr->prio)
1438 goto skip;
1440 ret = 1;
1442 deactivate_task(src_rq, p, 0);
1443 set_task_cpu(p, this_cpu);
1444 activate_task(this_rq, p, 0);
1446 * We continue with the search, just in
1447 * case there's an even higher prio task
1448 * in another runqueue. (low likelyhood
1449 * but possible)
1452 skip:
1453 double_unlock_balance(this_rq, src_rq);
1456 return ret;
1459 static void pre_schedule_rt(struct rq *rq, struct task_struct *prev)
1461 /* Try to pull RT tasks here if we lower this rq's prio */
1462 if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio)
1463 pull_rt_task(rq);
1466 static void post_schedule_rt(struct rq *rq)
1468 push_rt_tasks(rq);
1472 * If we are not running and we are not going to reschedule soon, we should
1473 * try to push tasks away now
1475 static void task_wake_up_rt(struct rq *rq, struct task_struct *p)
1477 if (!task_running(rq, p) &&
1478 !test_tsk_need_resched(rq->curr) &&
1479 has_pushable_tasks(rq) &&
1480 p->rt.nr_cpus_allowed > 1)
1481 push_rt_tasks(rq);
1484 static unsigned long
1485 load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
1486 unsigned long max_load_move,
1487 struct sched_domain *sd, enum cpu_idle_type idle,
1488 int *all_pinned, int *this_best_prio)
1490 /* don't touch RT tasks */
1491 return 0;
1494 static int
1495 move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
1496 struct sched_domain *sd, enum cpu_idle_type idle)
1498 /* don't touch RT tasks */
1499 return 0;
1502 static void set_cpus_allowed_rt(struct task_struct *p,
1503 const struct cpumask *new_mask)
1505 int weight = cpumask_weight(new_mask);
1507 BUG_ON(!rt_task(p));
1510 * Update the migration status of the RQ if we have an RT task
1511 * which is running AND changing its weight value.
1513 if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) {
1514 struct rq *rq = task_rq(p);
1516 if (!task_current(rq, p)) {
1518 * Make sure we dequeue this task from the pushable list
1519 * before going further. It will either remain off of
1520 * the list because we are no longer pushable, or it
1521 * will be requeued.
1523 if (p->rt.nr_cpus_allowed > 1)
1524 dequeue_pushable_task(rq, p);
1527 * Requeue if our weight is changing and still > 1
1529 if (weight > 1)
1530 enqueue_pushable_task(rq, p);
1534 if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) {
1535 rq->rt.rt_nr_migratory++;
1536 } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) {
1537 BUG_ON(!rq->rt.rt_nr_migratory);
1538 rq->rt.rt_nr_migratory--;
1541 update_rt_migration(&rq->rt);
1544 cpumask_copy(&p->cpus_allowed, new_mask);
1545 p->rt.nr_cpus_allowed = weight;
1548 /* Assumes rq->lock is held */
1549 static void rq_online_rt(struct rq *rq)
1551 if (rq->rt.overloaded)
1552 rt_set_overload(rq);
1554 __enable_runtime(rq);
1556 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1559 /* Assumes rq->lock is held */
1560 static void rq_offline_rt(struct rq *rq)
1562 if (rq->rt.overloaded)
1563 rt_clear_overload(rq);
1565 __disable_runtime(rq);
1567 cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
1571 * When switch from the rt queue, we bring ourselves to a position
1572 * that we might want to pull RT tasks from other runqueues.
1574 static void switched_from_rt(struct rq *rq, struct task_struct *p,
1575 int running)
1578 * If there are other RT tasks then we will reschedule
1579 * and the scheduling of the other RT tasks will handle
1580 * the balancing. But if we are the last RT task
1581 * we may need to handle the pulling of RT tasks
1582 * now.
1584 if (!rq->rt.rt_nr_running)
1585 pull_rt_task(rq);
1588 static inline void init_sched_rt_class(void)
1590 unsigned int i;
1592 for_each_possible_cpu(i)
1593 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
1594 GFP_KERNEL, cpu_to_node(i));
1596 #endif /* CONFIG_SMP */
1599 * When switching a task to RT, we may overload the runqueue
1600 * with RT tasks. In this case we try to push them off to
1601 * other runqueues.
1603 static void switched_to_rt(struct rq *rq, struct task_struct *p,
1604 int running)
1606 int check_resched = 1;
1609 * If we are already running, then there's nothing
1610 * that needs to be done. But if we are not running
1611 * we may need to preempt the current running task.
1612 * If that current running task is also an RT task
1613 * then see if we can move to another run queue.
1615 if (!running) {
1616 #ifdef CONFIG_SMP
1617 if (rq->rt.overloaded && push_rt_task(rq) &&
1618 /* Don't resched if we changed runqueues */
1619 rq != task_rq(p))
1620 check_resched = 0;
1621 #endif /* CONFIG_SMP */
1622 if (check_resched && p->prio < rq->curr->prio)
1623 resched_task(rq->curr);
1628 * Priority of the task has changed. This may cause
1629 * us to initiate a push or pull.
1631 static void prio_changed_rt(struct rq *rq, struct task_struct *p,
1632 int oldprio, int running)
1634 if (running) {
1635 #ifdef CONFIG_SMP
1637 * If our priority decreases while running, we
1638 * may need to pull tasks to this runqueue.
1640 if (oldprio < p->prio)
1641 pull_rt_task(rq);
1643 * If there's a higher priority task waiting to run
1644 * then reschedule. Note, the above pull_rt_task
1645 * can release the rq lock and p could migrate.
1646 * Only reschedule if p is still on the same runqueue.
1648 if (p->prio > rq->rt.highest_prio.curr && rq->curr == p)
1649 resched_task(p);
1650 #else
1651 /* For UP simply resched on drop of prio */
1652 if (oldprio < p->prio)
1653 resched_task(p);
1654 #endif /* CONFIG_SMP */
1655 } else {
1657 * This task is not running, but if it is
1658 * greater than the current running task
1659 * then reschedule.
1661 if (p->prio < rq->curr->prio)
1662 resched_task(rq->curr);
1666 static void watchdog(struct rq *rq, struct task_struct *p)
1668 unsigned long soft, hard;
1670 if (!p->signal)
1671 return;
1673 soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur;
1674 hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max;
1676 if (soft != RLIM_INFINITY) {
1677 unsigned long next;
1679 p->rt.timeout++;
1680 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
1681 if (p->rt.timeout > next)
1682 p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
1686 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
1688 update_curr_rt(rq);
1690 watchdog(rq, p);
1693 * RR tasks need a special form of timeslice management.
1694 * FIFO tasks have no timeslices.
1696 if (p->policy != SCHED_RR)
1697 return;
1699 if (--p->rt.time_slice)
1700 return;
1702 p->rt.time_slice = DEF_TIMESLICE;
1705 * Requeue to the end of queue if we are not the only element
1706 * on the queue:
1708 if (p->rt.run_list.prev != p->rt.run_list.next) {
1709 requeue_task_rt(rq, p, 0);
1710 set_tsk_need_resched(p);
1714 static void set_curr_task_rt(struct rq *rq)
1716 struct task_struct *p = rq->curr;
1718 p->se.exec_start = rq->clock;
1720 /* The running task is never eligible for pushing */
1721 dequeue_pushable_task(rq, p);
1724 unsigned int get_rr_interval_rt(struct task_struct *task)
1727 * Time slice is 0 for SCHED_FIFO tasks
1729 if (task->policy == SCHED_RR)
1730 return DEF_TIMESLICE;
1731 else
1732 return 0;
1735 static const struct sched_class rt_sched_class = {
1736 .next = &fair_sched_class,
1737 .enqueue_task = enqueue_task_rt,
1738 .dequeue_task = dequeue_task_rt,
1739 .yield_task = yield_task_rt,
1741 .check_preempt_curr = check_preempt_curr_rt,
1743 .pick_next_task = pick_next_task_rt,
1744 .put_prev_task = put_prev_task_rt,
1746 #ifdef CONFIG_SMP
1747 .select_task_rq = select_task_rq_rt,
1749 .load_balance = load_balance_rt,
1750 .move_one_task = move_one_task_rt,
1751 .set_cpus_allowed = set_cpus_allowed_rt,
1752 .rq_online = rq_online_rt,
1753 .rq_offline = rq_offline_rt,
1754 .pre_schedule = pre_schedule_rt,
1755 .post_schedule = post_schedule_rt,
1756 .task_wake_up = task_wake_up_rt,
1757 .switched_from = switched_from_rt,
1758 #endif
1760 .set_curr_task = set_curr_task_rt,
1761 .task_tick = task_tick_rt,
1763 .get_rr_interval = get_rr_interval_rt,
1765 .prio_changed = prio_changed_rt,
1766 .switched_to = switched_to_rt,
1769 #ifdef CONFIG_SCHED_DEBUG
1770 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
1772 static void print_rt_stats(struct seq_file *m, int cpu)
1774 struct rt_rq *rt_rq;
1776 rcu_read_lock();
1777 for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu))
1778 print_rt_rq(m, cpu, rt_rq);
1779 rcu_read_unlock();
1781 #endif /* CONFIG_SCHED_DEBUG */