xref: /openbmc/linux/kernel/sched/rt.c (revision e0f6d1a5)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
4  * policies)
5  */
6 #include "sched.h"
7 
8 int sched_rr_timeslice = RR_TIMESLICE;
9 int sysctl_sched_rr_timeslice = (MSEC_PER_SEC / HZ) * RR_TIMESLICE;
10 
11 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
12 
13 struct rt_bandwidth def_rt_bandwidth;
14 
15 static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
16 {
17 	struct rt_bandwidth *rt_b =
18 		container_of(timer, struct rt_bandwidth, rt_period_timer);
19 	int idle = 0;
20 	int overrun;
21 
22 	raw_spin_lock(&rt_b->rt_runtime_lock);
23 	for (;;) {
24 		overrun = hrtimer_forward_now(timer, rt_b->rt_period);
25 		if (!overrun)
26 			break;
27 
28 		raw_spin_unlock(&rt_b->rt_runtime_lock);
29 		idle = do_sched_rt_period_timer(rt_b, overrun);
30 		raw_spin_lock(&rt_b->rt_runtime_lock);
31 	}
32 	if (idle)
33 		rt_b->rt_period_active = 0;
34 	raw_spin_unlock(&rt_b->rt_runtime_lock);
35 
36 	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
37 }
38 
39 void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
40 {
41 	rt_b->rt_period = ns_to_ktime(period);
42 	rt_b->rt_runtime = runtime;
43 
44 	raw_spin_lock_init(&rt_b->rt_runtime_lock);
45 
46 	hrtimer_init(&rt_b->rt_period_timer,
47 			CLOCK_MONOTONIC, HRTIMER_MODE_REL);
48 	rt_b->rt_period_timer.function = sched_rt_period_timer;
49 }
50 
51 static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
52 {
53 	if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
54 		return;
55 
56 	raw_spin_lock(&rt_b->rt_runtime_lock);
57 	if (!rt_b->rt_period_active) {
58 		rt_b->rt_period_active = 1;
59 		/*
60 		 * SCHED_DEADLINE updates the bandwidth, as a run away
61 		 * RT task with a DL task could hog a CPU. But DL does
62 		 * not reset the period. If a deadline task was running
63 		 * without an RT task running, it can cause RT tasks to
64 		 * throttle when they start up. Kick the timer right away
65 		 * to update the period.
66 		 */
67 		hrtimer_forward_now(&rt_b->rt_period_timer, ns_to_ktime(0));
68 		hrtimer_start_expires(&rt_b->rt_period_timer, HRTIMER_MODE_ABS_PINNED);
69 	}
70 	raw_spin_unlock(&rt_b->rt_runtime_lock);
71 }
72 
73 void init_rt_rq(struct rt_rq *rt_rq)
74 {
75 	struct rt_prio_array *array;
76 	int i;
77 
78 	array = &rt_rq->active;
79 	for (i = 0; i < MAX_RT_PRIO; i++) {
80 		INIT_LIST_HEAD(array->queue + i);
81 		__clear_bit(i, array->bitmap);
82 	}
83 	/* delimiter for bitsearch: */
84 	__set_bit(MAX_RT_PRIO, array->bitmap);
85 
86 #if defined CONFIG_SMP
87 	rt_rq->highest_prio.curr = MAX_RT_PRIO;
88 	rt_rq->highest_prio.next = MAX_RT_PRIO;
89 	rt_rq->rt_nr_migratory = 0;
90 	rt_rq->overloaded = 0;
91 	plist_head_init(&rt_rq->pushable_tasks);
92 #endif /* CONFIG_SMP */
93 	/* We start is dequeued state, because no RT tasks are queued */
94 	rt_rq->rt_queued = 0;
95 
96 	rt_rq->rt_time = 0;
97 	rt_rq->rt_throttled = 0;
98 	rt_rq->rt_runtime = 0;
99 	raw_spin_lock_init(&rt_rq->rt_runtime_lock);
100 }
101 
102 #ifdef CONFIG_RT_GROUP_SCHED
103 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
104 {
105 	hrtimer_cancel(&rt_b->rt_period_timer);
106 }
107 
108 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
109 
110 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
111 {
112 #ifdef CONFIG_SCHED_DEBUG
113 	WARN_ON_ONCE(!rt_entity_is_task(rt_se));
114 #endif
115 	return container_of(rt_se, struct task_struct, rt);
116 }
117 
118 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
119 {
120 	return rt_rq->rq;
121 }
122 
123 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
124 {
125 	return rt_se->rt_rq;
126 }
127 
128 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
129 {
130 	struct rt_rq *rt_rq = rt_se->rt_rq;
131 
132 	return rt_rq->rq;
133 }
134 
135 void free_rt_sched_group(struct task_group *tg)
136 {
137 	int i;
138 
139 	if (tg->rt_se)
140 		destroy_rt_bandwidth(&tg->rt_bandwidth);
141 
142 	for_each_possible_cpu(i) {
143 		if (tg->rt_rq)
144 			kfree(tg->rt_rq[i]);
145 		if (tg->rt_se)
146 			kfree(tg->rt_se[i]);
147 	}
148 
149 	kfree(tg->rt_rq);
150 	kfree(tg->rt_se);
151 }
152 
153 void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
154 		struct sched_rt_entity *rt_se, int cpu,
155 		struct sched_rt_entity *parent)
156 {
157 	struct rq *rq = cpu_rq(cpu);
158 
159 	rt_rq->highest_prio.curr = MAX_RT_PRIO;
160 	rt_rq->rt_nr_boosted = 0;
161 	rt_rq->rq = rq;
162 	rt_rq->tg = tg;
163 
164 	tg->rt_rq[cpu] = rt_rq;
165 	tg->rt_se[cpu] = rt_se;
166 
167 	if (!rt_se)
168 		return;
169 
170 	if (!parent)
171 		rt_se->rt_rq = &rq->rt;
172 	else
173 		rt_se->rt_rq = parent->my_q;
174 
175 	rt_se->my_q = rt_rq;
176 	rt_se->parent = parent;
177 	INIT_LIST_HEAD(&rt_se->run_list);
178 }
179 
180 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
181 {
182 	struct rt_rq *rt_rq;
183 	struct sched_rt_entity *rt_se;
184 	int i;
185 
186 	tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
187 	if (!tg->rt_rq)
188 		goto err;
189 	tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
190 	if (!tg->rt_se)
191 		goto err;
192 
193 	init_rt_bandwidth(&tg->rt_bandwidth,
194 			ktime_to_ns(def_rt_bandwidth.rt_period), 0);
195 
196 	for_each_possible_cpu(i) {
197 		rt_rq = kzalloc_node(sizeof(struct rt_rq),
198 				     GFP_KERNEL, cpu_to_node(i));
199 		if (!rt_rq)
200 			goto err;
201 
202 		rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
203 				     GFP_KERNEL, cpu_to_node(i));
204 		if (!rt_se)
205 			goto err_free_rq;
206 
207 		init_rt_rq(rt_rq);
208 		rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
209 		init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
210 	}
211 
212 	return 1;
213 
214 err_free_rq:
215 	kfree(rt_rq);
216 err:
217 	return 0;
218 }
219 
220 #else /* CONFIG_RT_GROUP_SCHED */
221 
222 #define rt_entity_is_task(rt_se) (1)
223 
224 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
225 {
226 	return container_of(rt_se, struct task_struct, rt);
227 }
228 
229 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
230 {
231 	return container_of(rt_rq, struct rq, rt);
232 }
233 
234 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
235 {
236 	struct task_struct *p = rt_task_of(rt_se);
237 
238 	return task_rq(p);
239 }
240 
241 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
242 {
243 	struct rq *rq = rq_of_rt_se(rt_se);
244 
245 	return &rq->rt;
246 }
247 
248 void free_rt_sched_group(struct task_group *tg) { }
249 
250 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
251 {
252 	return 1;
253 }
254 #endif /* CONFIG_RT_GROUP_SCHED */
255 
256 #ifdef CONFIG_SMP
257 
258 static void pull_rt_task(struct rq *this_rq);
259 
260 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
261 {
262 	/* Try to pull RT tasks here if we lower this rq's prio */
263 	return rq->rt.highest_prio.curr > prev->prio;
264 }
265 
266 static inline int rt_overloaded(struct rq *rq)
267 {
268 	return atomic_read(&rq->rd->rto_count);
269 }
270 
271 static inline void rt_set_overload(struct rq *rq)
272 {
273 	if (!rq->online)
274 		return;
275 
276 	cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
277 	/*
278 	 * Make sure the mask is visible before we set
279 	 * the overload count. That is checked to determine
280 	 * if we should look at the mask. It would be a shame
281 	 * if we looked at the mask, but the mask was not
282 	 * updated yet.
283 	 *
284 	 * Matched by the barrier in pull_rt_task().
285 	 */
286 	smp_wmb();
287 	atomic_inc(&rq->rd->rto_count);
288 }
289 
290 static inline void rt_clear_overload(struct rq *rq)
291 {
292 	if (!rq->online)
293 		return;
294 
295 	/* the order here really doesn't matter */
296 	atomic_dec(&rq->rd->rto_count);
297 	cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
298 }
299 
300 static void update_rt_migration(struct rt_rq *rt_rq)
301 {
302 	if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
303 		if (!rt_rq->overloaded) {
304 			rt_set_overload(rq_of_rt_rq(rt_rq));
305 			rt_rq->overloaded = 1;
306 		}
307 	} else if (rt_rq->overloaded) {
308 		rt_clear_overload(rq_of_rt_rq(rt_rq));
309 		rt_rq->overloaded = 0;
310 	}
311 }
312 
313 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
314 {
315 	struct task_struct *p;
316 
317 	if (!rt_entity_is_task(rt_se))
318 		return;
319 
320 	p = rt_task_of(rt_se);
321 	rt_rq = &rq_of_rt_rq(rt_rq)->rt;
322 
323 	rt_rq->rt_nr_total++;
324 	if (p->nr_cpus_allowed > 1)
325 		rt_rq->rt_nr_migratory++;
326 
327 	update_rt_migration(rt_rq);
328 }
329 
330 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
331 {
332 	struct task_struct *p;
333 
334 	if (!rt_entity_is_task(rt_se))
335 		return;
336 
337 	p = rt_task_of(rt_se);
338 	rt_rq = &rq_of_rt_rq(rt_rq)->rt;
339 
340 	rt_rq->rt_nr_total--;
341 	if (p->nr_cpus_allowed > 1)
342 		rt_rq->rt_nr_migratory--;
343 
344 	update_rt_migration(rt_rq);
345 }
346 
347 static inline int has_pushable_tasks(struct rq *rq)
348 {
349 	return !plist_head_empty(&rq->rt.pushable_tasks);
350 }
351 
352 static DEFINE_PER_CPU(struct callback_head, rt_push_head);
353 static DEFINE_PER_CPU(struct callback_head, rt_pull_head);
354 
355 static void push_rt_tasks(struct rq *);
356 static void pull_rt_task(struct rq *);
357 
358 static inline void rt_queue_push_tasks(struct rq *rq)
359 {
360 	if (!has_pushable_tasks(rq))
361 		return;
362 
363 	queue_balance_callback(rq, &per_cpu(rt_push_head, rq->cpu), push_rt_tasks);
364 }
365 
366 static inline void rt_queue_pull_task(struct rq *rq)
367 {
368 	queue_balance_callback(rq, &per_cpu(rt_pull_head, rq->cpu), pull_rt_task);
369 }
370 
371 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
372 {
373 	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
374 	plist_node_init(&p->pushable_tasks, p->prio);
375 	plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
376 
377 	/* Update the highest prio pushable task */
378 	if (p->prio < rq->rt.highest_prio.next)
379 		rq->rt.highest_prio.next = p->prio;
380 }
381 
382 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
383 {
384 	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
385 
386 	/* Update the new highest prio pushable task */
387 	if (has_pushable_tasks(rq)) {
388 		p = plist_first_entry(&rq->rt.pushable_tasks,
389 				      struct task_struct, pushable_tasks);
390 		rq->rt.highest_prio.next = p->prio;
391 	} else
392 		rq->rt.highest_prio.next = MAX_RT_PRIO;
393 }
394 
395 #else
396 
397 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
398 {
399 }
400 
401 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
402 {
403 }
404 
405 static inline
406 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
407 {
408 }
409 
410 static inline
411 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
412 {
413 }
414 
415 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
416 {
417 	return false;
418 }
419 
420 static inline void pull_rt_task(struct rq *this_rq)
421 {
422 }
423 
424 static inline void rt_queue_push_tasks(struct rq *rq)
425 {
426 }
427 #endif /* CONFIG_SMP */
428 
429 static void enqueue_top_rt_rq(struct rt_rq *rt_rq);
430 static void dequeue_top_rt_rq(struct rt_rq *rt_rq);
431 
432 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
433 {
434 	return rt_se->on_rq;
435 }
436 
437 #ifdef CONFIG_RT_GROUP_SCHED
438 
439 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
440 {
441 	if (!rt_rq->tg)
442 		return RUNTIME_INF;
443 
444 	return rt_rq->rt_runtime;
445 }
446 
447 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
448 {
449 	return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
450 }
451 
452 typedef struct task_group *rt_rq_iter_t;
453 
454 static inline struct task_group *next_task_group(struct task_group *tg)
455 {
456 	do {
457 		tg = list_entry_rcu(tg->list.next,
458 			typeof(struct task_group), list);
459 	} while (&tg->list != &task_groups && task_group_is_autogroup(tg));
460 
461 	if (&tg->list == &task_groups)
462 		tg = NULL;
463 
464 	return tg;
465 }
466 
467 #define for_each_rt_rq(rt_rq, iter, rq)					\
468 	for (iter = container_of(&task_groups, typeof(*iter), list);	\
469 		(iter = next_task_group(iter)) &&			\
470 		(rt_rq = iter->rt_rq[cpu_of(rq)]);)
471 
472 #define for_each_sched_rt_entity(rt_se) \
473 	for (; rt_se; rt_se = rt_se->parent)
474 
475 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
476 {
477 	return rt_se->my_q;
478 }
479 
480 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
481 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
482 
483 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
484 {
485 	struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
486 	struct rq *rq = rq_of_rt_rq(rt_rq);
487 	struct sched_rt_entity *rt_se;
488 
489 	int cpu = cpu_of(rq);
490 
491 	rt_se = rt_rq->tg->rt_se[cpu];
492 
493 	if (rt_rq->rt_nr_running) {
494 		if (!rt_se)
495 			enqueue_top_rt_rq(rt_rq);
496 		else if (!on_rt_rq(rt_se))
497 			enqueue_rt_entity(rt_se, 0);
498 
499 		if (rt_rq->highest_prio.curr < curr->prio)
500 			resched_curr(rq);
501 	}
502 }
503 
504 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
505 {
506 	struct sched_rt_entity *rt_se;
507 	int cpu = cpu_of(rq_of_rt_rq(rt_rq));
508 
509 	rt_se = rt_rq->tg->rt_se[cpu];
510 
511 	if (!rt_se)
512 		dequeue_top_rt_rq(rt_rq);
513 	else if (on_rt_rq(rt_se))
514 		dequeue_rt_entity(rt_se, 0);
515 }
516 
517 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
518 {
519 	return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
520 }
521 
522 static int rt_se_boosted(struct sched_rt_entity *rt_se)
523 {
524 	struct rt_rq *rt_rq = group_rt_rq(rt_se);
525 	struct task_struct *p;
526 
527 	if (rt_rq)
528 		return !!rt_rq->rt_nr_boosted;
529 
530 	p = rt_task_of(rt_se);
531 	return p->prio != p->normal_prio;
532 }
533 
534 #ifdef CONFIG_SMP
535 static inline const struct cpumask *sched_rt_period_mask(void)
536 {
537 	return this_rq()->rd->span;
538 }
539 #else
540 static inline const struct cpumask *sched_rt_period_mask(void)
541 {
542 	return cpu_online_mask;
543 }
544 #endif
545 
546 static inline
547 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
548 {
549 	return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
550 }
551 
552 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
553 {
554 	return &rt_rq->tg->rt_bandwidth;
555 }
556 
557 #else /* !CONFIG_RT_GROUP_SCHED */
558 
559 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
560 {
561 	return rt_rq->rt_runtime;
562 }
563 
564 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
565 {
566 	return ktime_to_ns(def_rt_bandwidth.rt_period);
567 }
568 
569 typedef struct rt_rq *rt_rq_iter_t;
570 
571 #define for_each_rt_rq(rt_rq, iter, rq) \
572 	for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
573 
574 #define for_each_sched_rt_entity(rt_se) \
575 	for (; rt_se; rt_se = NULL)
576 
577 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
578 {
579 	return NULL;
580 }
581 
582 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
583 {
584 	struct rq *rq = rq_of_rt_rq(rt_rq);
585 
586 	if (!rt_rq->rt_nr_running)
587 		return;
588 
589 	enqueue_top_rt_rq(rt_rq);
590 	resched_curr(rq);
591 }
592 
593 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
594 {
595 	dequeue_top_rt_rq(rt_rq);
596 }
597 
598 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
599 {
600 	return rt_rq->rt_throttled;
601 }
602 
603 static inline const struct cpumask *sched_rt_period_mask(void)
604 {
605 	return cpu_online_mask;
606 }
607 
608 static inline
609 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
610 {
611 	return &cpu_rq(cpu)->rt;
612 }
613 
614 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
615 {
616 	return &def_rt_bandwidth;
617 }
618 
619 #endif /* CONFIG_RT_GROUP_SCHED */
620 
621 bool sched_rt_bandwidth_account(struct rt_rq *rt_rq)
622 {
623 	struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
624 
625 	return (hrtimer_active(&rt_b->rt_period_timer) ||
626 		rt_rq->rt_time < rt_b->rt_runtime);
627 }
628 
629 #ifdef CONFIG_SMP
630 /*
631  * We ran out of runtime, see if we can borrow some from our neighbours.
632  */
633 static void do_balance_runtime(struct rt_rq *rt_rq)
634 {
635 	struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
636 	struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd;
637 	int i, weight;
638 	u64 rt_period;
639 
640 	weight = cpumask_weight(rd->span);
641 
642 	raw_spin_lock(&rt_b->rt_runtime_lock);
643 	rt_period = ktime_to_ns(rt_b->rt_period);
644 	for_each_cpu(i, rd->span) {
645 		struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
646 		s64 diff;
647 
648 		if (iter == rt_rq)
649 			continue;
650 
651 		raw_spin_lock(&iter->rt_runtime_lock);
652 		/*
653 		 * Either all rqs have inf runtime and there's nothing to steal
654 		 * or __disable_runtime() below sets a specific rq to inf to
655 		 * indicate its been disabled and disalow stealing.
656 		 */
657 		if (iter->rt_runtime == RUNTIME_INF)
658 			goto next;
659 
660 		/*
661 		 * From runqueues with spare time, take 1/n part of their
662 		 * spare time, but no more than our period.
663 		 */
664 		diff = iter->rt_runtime - iter->rt_time;
665 		if (diff > 0) {
666 			diff = div_u64((u64)diff, weight);
667 			if (rt_rq->rt_runtime + diff > rt_period)
668 				diff = rt_period - rt_rq->rt_runtime;
669 			iter->rt_runtime -= diff;
670 			rt_rq->rt_runtime += diff;
671 			if (rt_rq->rt_runtime == rt_period) {
672 				raw_spin_unlock(&iter->rt_runtime_lock);
673 				break;
674 			}
675 		}
676 next:
677 		raw_spin_unlock(&iter->rt_runtime_lock);
678 	}
679 	raw_spin_unlock(&rt_b->rt_runtime_lock);
680 }
681 
682 /*
683  * Ensure this RQ takes back all the runtime it lend to its neighbours.
684  */
685 static void __disable_runtime(struct rq *rq)
686 {
687 	struct root_domain *rd = rq->rd;
688 	rt_rq_iter_t iter;
689 	struct rt_rq *rt_rq;
690 
691 	if (unlikely(!scheduler_running))
692 		return;
693 
694 	for_each_rt_rq(rt_rq, iter, rq) {
695 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
696 		s64 want;
697 		int i;
698 
699 		raw_spin_lock(&rt_b->rt_runtime_lock);
700 		raw_spin_lock(&rt_rq->rt_runtime_lock);
701 		/*
702 		 * Either we're all inf and nobody needs to borrow, or we're
703 		 * already disabled and thus have nothing to do, or we have
704 		 * exactly the right amount of runtime to take out.
705 		 */
706 		if (rt_rq->rt_runtime == RUNTIME_INF ||
707 				rt_rq->rt_runtime == rt_b->rt_runtime)
708 			goto balanced;
709 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
710 
711 		/*
712 		 * Calculate the difference between what we started out with
713 		 * and what we current have, that's the amount of runtime
714 		 * we lend and now have to reclaim.
715 		 */
716 		want = rt_b->rt_runtime - rt_rq->rt_runtime;
717 
718 		/*
719 		 * Greedy reclaim, take back as much as we can.
720 		 */
721 		for_each_cpu(i, rd->span) {
722 			struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
723 			s64 diff;
724 
725 			/*
726 			 * Can't reclaim from ourselves or disabled runqueues.
727 			 */
728 			if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
729 				continue;
730 
731 			raw_spin_lock(&iter->rt_runtime_lock);
732 			if (want > 0) {
733 				diff = min_t(s64, iter->rt_runtime, want);
734 				iter->rt_runtime -= diff;
735 				want -= diff;
736 			} else {
737 				iter->rt_runtime -= want;
738 				want -= want;
739 			}
740 			raw_spin_unlock(&iter->rt_runtime_lock);
741 
742 			if (!want)
743 				break;
744 		}
745 
746 		raw_spin_lock(&rt_rq->rt_runtime_lock);
747 		/*
748 		 * We cannot be left wanting - that would mean some runtime
749 		 * leaked out of the system.
750 		 */
751 		BUG_ON(want);
752 balanced:
753 		/*
754 		 * Disable all the borrow logic by pretending we have inf
755 		 * runtime - in which case borrowing doesn't make sense.
756 		 */
757 		rt_rq->rt_runtime = RUNTIME_INF;
758 		rt_rq->rt_throttled = 0;
759 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
760 		raw_spin_unlock(&rt_b->rt_runtime_lock);
761 
762 		/* Make rt_rq available for pick_next_task() */
763 		sched_rt_rq_enqueue(rt_rq);
764 	}
765 }
766 
767 static void __enable_runtime(struct rq *rq)
768 {
769 	rt_rq_iter_t iter;
770 	struct rt_rq *rt_rq;
771 
772 	if (unlikely(!scheduler_running))
773 		return;
774 
775 	/*
776 	 * Reset each runqueue's bandwidth settings
777 	 */
778 	for_each_rt_rq(rt_rq, iter, rq) {
779 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
780 
781 		raw_spin_lock(&rt_b->rt_runtime_lock);
782 		raw_spin_lock(&rt_rq->rt_runtime_lock);
783 		rt_rq->rt_runtime = rt_b->rt_runtime;
784 		rt_rq->rt_time = 0;
785 		rt_rq->rt_throttled = 0;
786 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
787 		raw_spin_unlock(&rt_b->rt_runtime_lock);
788 	}
789 }
790 
791 static void balance_runtime(struct rt_rq *rt_rq)
792 {
793 	if (!sched_feat(RT_RUNTIME_SHARE))
794 		return;
795 
796 	if (rt_rq->rt_time > rt_rq->rt_runtime) {
797 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
798 		do_balance_runtime(rt_rq);
799 		raw_spin_lock(&rt_rq->rt_runtime_lock);
800 	}
801 }
802 #else /* !CONFIG_SMP */
803 static inline void balance_runtime(struct rt_rq *rt_rq) {}
804 #endif /* CONFIG_SMP */
805 
806 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
807 {
808 	int i, idle = 1, throttled = 0;
809 	const struct cpumask *span;
810 
811 	span = sched_rt_period_mask();
812 #ifdef CONFIG_RT_GROUP_SCHED
813 	/*
814 	 * FIXME: isolated CPUs should really leave the root task group,
815 	 * whether they are isolcpus or were isolated via cpusets, lest
816 	 * the timer run on a CPU which does not service all runqueues,
817 	 * potentially leaving other CPUs indefinitely throttled.  If
818 	 * isolation is really required, the user will turn the throttle
819 	 * off to kill the perturbations it causes anyway.  Meanwhile,
820 	 * this maintains functionality for boot and/or troubleshooting.
821 	 */
822 	if (rt_b == &root_task_group.rt_bandwidth)
823 		span = cpu_online_mask;
824 #endif
825 	for_each_cpu(i, span) {
826 		int enqueue = 0;
827 		struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
828 		struct rq *rq = rq_of_rt_rq(rt_rq);
829 		int skip;
830 
831 		/*
832 		 * When span == cpu_online_mask, taking each rq->lock
833 		 * can be time-consuming. Try to avoid it when possible.
834 		 */
835 		raw_spin_lock(&rt_rq->rt_runtime_lock);
836 		skip = !rt_rq->rt_time && !rt_rq->rt_nr_running;
837 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
838 		if (skip)
839 			continue;
840 
841 		raw_spin_lock(&rq->lock);
842 		update_rq_clock(rq);
843 
844 		if (rt_rq->rt_time) {
845 			u64 runtime;
846 
847 			raw_spin_lock(&rt_rq->rt_runtime_lock);
848 			if (rt_rq->rt_throttled)
849 				balance_runtime(rt_rq);
850 			runtime = rt_rq->rt_runtime;
851 			rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
852 			if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
853 				rt_rq->rt_throttled = 0;
854 				enqueue = 1;
855 
856 				/*
857 				 * When we're idle and a woken (rt) task is
858 				 * throttled check_preempt_curr() will set
859 				 * skip_update and the time between the wakeup
860 				 * and this unthrottle will get accounted as
861 				 * 'runtime'.
862 				 */
863 				if (rt_rq->rt_nr_running && rq->curr == rq->idle)
864 					rq_clock_cancel_skipupdate(rq);
865 			}
866 			if (rt_rq->rt_time || rt_rq->rt_nr_running)
867 				idle = 0;
868 			raw_spin_unlock(&rt_rq->rt_runtime_lock);
869 		} else if (rt_rq->rt_nr_running) {
870 			idle = 0;
871 			if (!rt_rq_throttled(rt_rq))
872 				enqueue = 1;
873 		}
874 		if (rt_rq->rt_throttled)
875 			throttled = 1;
876 
877 		if (enqueue)
878 			sched_rt_rq_enqueue(rt_rq);
879 		raw_spin_unlock(&rq->lock);
880 	}
881 
882 	if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
883 		return 1;
884 
885 	return idle;
886 }
887 
888 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
889 {
890 #ifdef CONFIG_RT_GROUP_SCHED
891 	struct rt_rq *rt_rq = group_rt_rq(rt_se);
892 
893 	if (rt_rq)
894 		return rt_rq->highest_prio.curr;
895 #endif
896 
897 	return rt_task_of(rt_se)->prio;
898 }
899 
900 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
901 {
902 	u64 runtime = sched_rt_runtime(rt_rq);
903 
904 	if (rt_rq->rt_throttled)
905 		return rt_rq_throttled(rt_rq);
906 
907 	if (runtime >= sched_rt_period(rt_rq))
908 		return 0;
909 
910 	balance_runtime(rt_rq);
911 	runtime = sched_rt_runtime(rt_rq);
912 	if (runtime == RUNTIME_INF)
913 		return 0;
914 
915 	if (rt_rq->rt_time > runtime) {
916 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
917 
918 		/*
919 		 * Don't actually throttle groups that have no runtime assigned
920 		 * but accrue some time due to boosting.
921 		 */
922 		if (likely(rt_b->rt_runtime)) {
923 			rt_rq->rt_throttled = 1;
924 			printk_deferred_once("sched: RT throttling activated\n");
925 		} else {
926 			/*
927 			 * In case we did anyway, make it go away,
928 			 * replenishment is a joke, since it will replenish us
929 			 * with exactly 0 ns.
930 			 */
931 			rt_rq->rt_time = 0;
932 		}
933 
934 		if (rt_rq_throttled(rt_rq)) {
935 			sched_rt_rq_dequeue(rt_rq);
936 			return 1;
937 		}
938 	}
939 
940 	return 0;
941 }
942 
943 /*
944  * Update the current task's runtime statistics. Skip current tasks that
945  * are not in our scheduling class.
946  */
947 static void update_curr_rt(struct rq *rq)
948 {
949 	struct task_struct *curr = rq->curr;
950 	struct sched_rt_entity *rt_se = &curr->rt;
951 	u64 delta_exec;
952 	u64 now;
953 
954 	if (curr->sched_class != &rt_sched_class)
955 		return;
956 
957 	now = rq_clock_task(rq);
958 	delta_exec = now - curr->se.exec_start;
959 	if (unlikely((s64)delta_exec <= 0))
960 		return;
961 
962 	schedstat_set(curr->se.statistics.exec_max,
963 		      max(curr->se.statistics.exec_max, delta_exec));
964 
965 	curr->se.sum_exec_runtime += delta_exec;
966 	account_group_exec_runtime(curr, delta_exec);
967 
968 	curr->se.exec_start = now;
969 	cgroup_account_cputime(curr, delta_exec);
970 
971 	sched_rt_avg_update(rq, delta_exec);
972 
973 	if (!rt_bandwidth_enabled())
974 		return;
975 
976 	for_each_sched_rt_entity(rt_se) {
977 		struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
978 
979 		if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
980 			raw_spin_lock(&rt_rq->rt_runtime_lock);
981 			rt_rq->rt_time += delta_exec;
982 			if (sched_rt_runtime_exceeded(rt_rq))
983 				resched_curr(rq);
984 			raw_spin_unlock(&rt_rq->rt_runtime_lock);
985 		}
986 	}
987 }
988 
989 static void
990 dequeue_top_rt_rq(struct rt_rq *rt_rq)
991 {
992 	struct rq *rq = rq_of_rt_rq(rt_rq);
993 
994 	BUG_ON(&rq->rt != rt_rq);
995 
996 	if (!rt_rq->rt_queued)
997 		return;
998 
999 	BUG_ON(!rq->nr_running);
1000 
1001 	sub_nr_running(rq, rt_rq->rt_nr_running);
1002 	rt_rq->rt_queued = 0;
1003 
1004 	/* Kick cpufreq (see the comment in kernel/sched/sched.h). */
1005 	cpufreq_update_util(rq, 0);
1006 }
1007 
1008 static void
1009 enqueue_top_rt_rq(struct rt_rq *rt_rq)
1010 {
1011 	struct rq *rq = rq_of_rt_rq(rt_rq);
1012 
1013 	BUG_ON(&rq->rt != rt_rq);
1014 
1015 	if (rt_rq->rt_queued)
1016 		return;
1017 	if (rt_rq_throttled(rt_rq) || !rt_rq->rt_nr_running)
1018 		return;
1019 
1020 	add_nr_running(rq, rt_rq->rt_nr_running);
1021 	rt_rq->rt_queued = 1;
1022 
1023 	/* Kick cpufreq (see the comment in kernel/sched/sched.h). */
1024 	cpufreq_update_util(rq, 0);
1025 }
1026 
1027 #if defined CONFIG_SMP
1028 
1029 static void
1030 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1031 {
1032 	struct rq *rq = rq_of_rt_rq(rt_rq);
1033 
1034 #ifdef CONFIG_RT_GROUP_SCHED
1035 	/*
1036 	 * Change rq's cpupri only if rt_rq is the top queue.
1037 	 */
1038 	if (&rq->rt != rt_rq)
1039 		return;
1040 #endif
1041 	if (rq->online && prio < prev_prio)
1042 		cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
1043 }
1044 
1045 static void
1046 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1047 {
1048 	struct rq *rq = rq_of_rt_rq(rt_rq);
1049 
1050 #ifdef CONFIG_RT_GROUP_SCHED
1051 	/*
1052 	 * Change rq's cpupri only if rt_rq is the top queue.
1053 	 */
1054 	if (&rq->rt != rt_rq)
1055 		return;
1056 #endif
1057 	if (rq->online && rt_rq->highest_prio.curr != prev_prio)
1058 		cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
1059 }
1060 
1061 #else /* CONFIG_SMP */
1062 
1063 static inline
1064 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1065 static inline
1066 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1067 
1068 #endif /* CONFIG_SMP */
1069 
1070 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
1071 static void
1072 inc_rt_prio(struct rt_rq *rt_rq, int prio)
1073 {
1074 	int prev_prio = rt_rq->highest_prio.curr;
1075 
1076 	if (prio < prev_prio)
1077 		rt_rq->highest_prio.curr = prio;
1078 
1079 	inc_rt_prio_smp(rt_rq, prio, prev_prio);
1080 }
1081 
1082 static void
1083 dec_rt_prio(struct rt_rq *rt_rq, int prio)
1084 {
1085 	int prev_prio = rt_rq->highest_prio.curr;
1086 
1087 	if (rt_rq->rt_nr_running) {
1088 
1089 		WARN_ON(prio < prev_prio);
1090 
1091 		/*
1092 		 * This may have been our highest task, and therefore
1093 		 * we may have some recomputation to do
1094 		 */
1095 		if (prio == prev_prio) {
1096 			struct rt_prio_array *array = &rt_rq->active;
1097 
1098 			rt_rq->highest_prio.curr =
1099 				sched_find_first_bit(array->bitmap);
1100 		}
1101 
1102 	} else
1103 		rt_rq->highest_prio.curr = MAX_RT_PRIO;
1104 
1105 	dec_rt_prio_smp(rt_rq, prio, prev_prio);
1106 }
1107 
1108 #else
1109 
1110 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
1111 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
1112 
1113 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1114 
1115 #ifdef CONFIG_RT_GROUP_SCHED
1116 
1117 static void
1118 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1119 {
1120 	if (rt_se_boosted(rt_se))
1121 		rt_rq->rt_nr_boosted++;
1122 
1123 	if (rt_rq->tg)
1124 		start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
1125 }
1126 
1127 static void
1128 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1129 {
1130 	if (rt_se_boosted(rt_se))
1131 		rt_rq->rt_nr_boosted--;
1132 
1133 	WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
1134 }
1135 
1136 #else /* CONFIG_RT_GROUP_SCHED */
1137 
1138 static void
1139 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1140 {
1141 	start_rt_bandwidth(&def_rt_bandwidth);
1142 }
1143 
1144 static inline
1145 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
1146 
1147 #endif /* CONFIG_RT_GROUP_SCHED */
1148 
1149 static inline
1150 unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se)
1151 {
1152 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1153 
1154 	if (group_rq)
1155 		return group_rq->rt_nr_running;
1156 	else
1157 		return 1;
1158 }
1159 
1160 static inline
1161 unsigned int rt_se_rr_nr_running(struct sched_rt_entity *rt_se)
1162 {
1163 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1164 	struct task_struct *tsk;
1165 
1166 	if (group_rq)
1167 		return group_rq->rr_nr_running;
1168 
1169 	tsk = rt_task_of(rt_se);
1170 
1171 	return (tsk->policy == SCHED_RR) ? 1 : 0;
1172 }
1173 
1174 static inline
1175 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1176 {
1177 	int prio = rt_se_prio(rt_se);
1178 
1179 	WARN_ON(!rt_prio(prio));
1180 	rt_rq->rt_nr_running += rt_se_nr_running(rt_se);
1181 	rt_rq->rr_nr_running += rt_se_rr_nr_running(rt_se);
1182 
1183 	inc_rt_prio(rt_rq, prio);
1184 	inc_rt_migration(rt_se, rt_rq);
1185 	inc_rt_group(rt_se, rt_rq);
1186 }
1187 
1188 static inline
1189 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1190 {
1191 	WARN_ON(!rt_prio(rt_se_prio(rt_se)));
1192 	WARN_ON(!rt_rq->rt_nr_running);
1193 	rt_rq->rt_nr_running -= rt_se_nr_running(rt_se);
1194 	rt_rq->rr_nr_running -= rt_se_rr_nr_running(rt_se);
1195 
1196 	dec_rt_prio(rt_rq, rt_se_prio(rt_se));
1197 	dec_rt_migration(rt_se, rt_rq);
1198 	dec_rt_group(rt_se, rt_rq);
1199 }
1200 
1201 /*
1202  * Change rt_se->run_list location unless SAVE && !MOVE
1203  *
1204  * assumes ENQUEUE/DEQUEUE flags match
1205  */
1206 static inline bool move_entity(unsigned int flags)
1207 {
1208 	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
1209 		return false;
1210 
1211 	return true;
1212 }
1213 
1214 static void __delist_rt_entity(struct sched_rt_entity *rt_se, struct rt_prio_array *array)
1215 {
1216 	list_del_init(&rt_se->run_list);
1217 
1218 	if (list_empty(array->queue + rt_se_prio(rt_se)))
1219 		__clear_bit(rt_se_prio(rt_se), array->bitmap);
1220 
1221 	rt_se->on_list = 0;
1222 }
1223 
1224 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1225 {
1226 	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1227 	struct rt_prio_array *array = &rt_rq->active;
1228 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1229 	struct list_head *queue = array->queue + rt_se_prio(rt_se);
1230 
1231 	/*
1232 	 * Don't enqueue the group if its throttled, or when empty.
1233 	 * The latter is a consequence of the former when a child group
1234 	 * get throttled and the current group doesn't have any other
1235 	 * active members.
1236 	 */
1237 	if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) {
1238 		if (rt_se->on_list)
1239 			__delist_rt_entity(rt_se, array);
1240 		return;
1241 	}
1242 
1243 	if (move_entity(flags)) {
1244 		WARN_ON_ONCE(rt_se->on_list);
1245 		if (flags & ENQUEUE_HEAD)
1246 			list_add(&rt_se->run_list, queue);
1247 		else
1248 			list_add_tail(&rt_se->run_list, queue);
1249 
1250 		__set_bit(rt_se_prio(rt_se), array->bitmap);
1251 		rt_se->on_list = 1;
1252 	}
1253 	rt_se->on_rq = 1;
1254 
1255 	inc_rt_tasks(rt_se, rt_rq);
1256 }
1257 
1258 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1259 {
1260 	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1261 	struct rt_prio_array *array = &rt_rq->active;
1262 
1263 	if (move_entity(flags)) {
1264 		WARN_ON_ONCE(!rt_se->on_list);
1265 		__delist_rt_entity(rt_se, array);
1266 	}
1267 	rt_se->on_rq = 0;
1268 
1269 	dec_rt_tasks(rt_se, rt_rq);
1270 }
1271 
1272 /*
1273  * Because the prio of an upper entry depends on the lower
1274  * entries, we must remove entries top - down.
1275  */
1276 static void dequeue_rt_stack(struct sched_rt_entity *rt_se, unsigned int flags)
1277 {
1278 	struct sched_rt_entity *back = NULL;
1279 
1280 	for_each_sched_rt_entity(rt_se) {
1281 		rt_se->back = back;
1282 		back = rt_se;
1283 	}
1284 
1285 	dequeue_top_rt_rq(rt_rq_of_se(back));
1286 
1287 	for (rt_se = back; rt_se; rt_se = rt_se->back) {
1288 		if (on_rt_rq(rt_se))
1289 			__dequeue_rt_entity(rt_se, flags);
1290 	}
1291 }
1292 
1293 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1294 {
1295 	struct rq *rq = rq_of_rt_se(rt_se);
1296 
1297 	dequeue_rt_stack(rt_se, flags);
1298 	for_each_sched_rt_entity(rt_se)
1299 		__enqueue_rt_entity(rt_se, flags);
1300 	enqueue_top_rt_rq(&rq->rt);
1301 }
1302 
1303 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1304 {
1305 	struct rq *rq = rq_of_rt_se(rt_se);
1306 
1307 	dequeue_rt_stack(rt_se, flags);
1308 
1309 	for_each_sched_rt_entity(rt_se) {
1310 		struct rt_rq *rt_rq = group_rt_rq(rt_se);
1311 
1312 		if (rt_rq && rt_rq->rt_nr_running)
1313 			__enqueue_rt_entity(rt_se, flags);
1314 	}
1315 	enqueue_top_rt_rq(&rq->rt);
1316 }
1317 
1318 /*
1319  * Adding/removing a task to/from a priority array:
1320  */
1321 static void
1322 enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1323 {
1324 	struct sched_rt_entity *rt_se = &p->rt;
1325 
1326 	if (flags & ENQUEUE_WAKEUP)
1327 		rt_se->timeout = 0;
1328 
1329 	enqueue_rt_entity(rt_se, flags);
1330 
1331 	if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1332 		enqueue_pushable_task(rq, p);
1333 }
1334 
1335 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1336 {
1337 	struct sched_rt_entity *rt_se = &p->rt;
1338 
1339 	update_curr_rt(rq);
1340 	dequeue_rt_entity(rt_se, flags);
1341 
1342 	dequeue_pushable_task(rq, p);
1343 }
1344 
1345 /*
1346  * Put task to the head or the end of the run list without the overhead of
1347  * dequeue followed by enqueue.
1348  */
1349 static void
1350 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
1351 {
1352 	if (on_rt_rq(rt_se)) {
1353 		struct rt_prio_array *array = &rt_rq->active;
1354 		struct list_head *queue = array->queue + rt_se_prio(rt_se);
1355 
1356 		if (head)
1357 			list_move(&rt_se->run_list, queue);
1358 		else
1359 			list_move_tail(&rt_se->run_list, queue);
1360 	}
1361 }
1362 
1363 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
1364 {
1365 	struct sched_rt_entity *rt_se = &p->rt;
1366 	struct rt_rq *rt_rq;
1367 
1368 	for_each_sched_rt_entity(rt_se) {
1369 		rt_rq = rt_rq_of_se(rt_se);
1370 		requeue_rt_entity(rt_rq, rt_se, head);
1371 	}
1372 }
1373 
1374 static void yield_task_rt(struct rq *rq)
1375 {
1376 	requeue_task_rt(rq, rq->curr, 0);
1377 }
1378 
1379 #ifdef CONFIG_SMP
1380 static int find_lowest_rq(struct task_struct *task);
1381 
1382 static int
1383 select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags)
1384 {
1385 	struct task_struct *curr;
1386 	struct rq *rq;
1387 
1388 	/* For anything but wake ups, just return the task_cpu */
1389 	if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK)
1390 		goto out;
1391 
1392 	rq = cpu_rq(cpu);
1393 
1394 	rcu_read_lock();
1395 	curr = READ_ONCE(rq->curr); /* unlocked access */
1396 
1397 	/*
1398 	 * If the current task on @p's runqueue is an RT task, then
1399 	 * try to see if we can wake this RT task up on another
1400 	 * runqueue. Otherwise simply start this RT task
1401 	 * on its current runqueue.
1402 	 *
1403 	 * We want to avoid overloading runqueues. If the woken
1404 	 * task is a higher priority, then it will stay on this CPU
1405 	 * and the lower prio task should be moved to another CPU.
1406 	 * Even though this will probably make the lower prio task
1407 	 * lose its cache, we do not want to bounce a higher task
1408 	 * around just because it gave up its CPU, perhaps for a
1409 	 * lock?
1410 	 *
1411 	 * For equal prio tasks, we just let the scheduler sort it out.
1412 	 *
1413 	 * Otherwise, just let it ride on the affined RQ and the
1414 	 * post-schedule router will push the preempted task away
1415 	 *
1416 	 * This test is optimistic, if we get it wrong the load-balancer
1417 	 * will have to sort it out.
1418 	 */
1419 	if (curr && unlikely(rt_task(curr)) &&
1420 	    (curr->nr_cpus_allowed < 2 ||
1421 	     curr->prio <= p->prio)) {
1422 		int target = find_lowest_rq(p);
1423 
1424 		/*
1425 		 * Don't bother moving it if the destination CPU is
1426 		 * not running a lower priority task.
1427 		 */
1428 		if (target != -1 &&
1429 		    p->prio < cpu_rq(target)->rt.highest_prio.curr)
1430 			cpu = target;
1431 	}
1432 	rcu_read_unlock();
1433 
1434 out:
1435 	return cpu;
1436 }
1437 
1438 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
1439 {
1440 	/*
1441 	 * Current can't be migrated, useless to reschedule,
1442 	 * let's hope p can move out.
1443 	 */
1444 	if (rq->curr->nr_cpus_allowed == 1 ||
1445 	    !cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
1446 		return;
1447 
1448 	/*
1449 	 * p is migratable, so let's not schedule it and
1450 	 * see if it is pushed or pulled somewhere else.
1451 	 */
1452 	if (p->nr_cpus_allowed != 1
1453 	    && cpupri_find(&rq->rd->cpupri, p, NULL))
1454 		return;
1455 
1456 	/*
1457 	 * There appear to be other CPUs that can accept
1458 	 * the current task but none can run 'p', so lets reschedule
1459 	 * to try and push the current task away:
1460 	 */
1461 	requeue_task_rt(rq, p, 1);
1462 	resched_curr(rq);
1463 }
1464 
1465 #endif /* CONFIG_SMP */
1466 
1467 /*
1468  * Preempt the current task with a newly woken task if needed:
1469  */
1470 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1471 {
1472 	if (p->prio < rq->curr->prio) {
1473 		resched_curr(rq);
1474 		return;
1475 	}
1476 
1477 #ifdef CONFIG_SMP
1478 	/*
1479 	 * If:
1480 	 *
1481 	 * - the newly woken task is of equal priority to the current task
1482 	 * - the newly woken task is non-migratable while current is migratable
1483 	 * - current will be preempted on the next reschedule
1484 	 *
1485 	 * we should check to see if current can readily move to a different
1486 	 * cpu.  If so, we will reschedule to allow the push logic to try
1487 	 * to move current somewhere else, making room for our non-migratable
1488 	 * task.
1489 	 */
1490 	if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr))
1491 		check_preempt_equal_prio(rq, p);
1492 #endif
1493 }
1494 
1495 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1496 						   struct rt_rq *rt_rq)
1497 {
1498 	struct rt_prio_array *array = &rt_rq->active;
1499 	struct sched_rt_entity *next = NULL;
1500 	struct list_head *queue;
1501 	int idx;
1502 
1503 	idx = sched_find_first_bit(array->bitmap);
1504 	BUG_ON(idx >= MAX_RT_PRIO);
1505 
1506 	queue = array->queue + idx;
1507 	next = list_entry(queue->next, struct sched_rt_entity, run_list);
1508 
1509 	return next;
1510 }
1511 
1512 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1513 {
1514 	struct sched_rt_entity *rt_se;
1515 	struct task_struct *p;
1516 	struct rt_rq *rt_rq  = &rq->rt;
1517 
1518 	do {
1519 		rt_se = pick_next_rt_entity(rq, rt_rq);
1520 		BUG_ON(!rt_se);
1521 		rt_rq = group_rt_rq(rt_se);
1522 	} while (rt_rq);
1523 
1524 	p = rt_task_of(rt_se);
1525 	p->se.exec_start = rq_clock_task(rq);
1526 
1527 	return p;
1528 }
1529 
1530 static struct task_struct *
1531 pick_next_task_rt(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
1532 {
1533 	struct task_struct *p;
1534 	struct rt_rq *rt_rq = &rq->rt;
1535 
1536 	if (need_pull_rt_task(rq, prev)) {
1537 		/*
1538 		 * This is OK, because current is on_cpu, which avoids it being
1539 		 * picked for load-balance and preemption/IRQs are still
1540 		 * disabled avoiding further scheduler activity on it and we're
1541 		 * being very careful to re-start the picking loop.
1542 		 */
1543 		rq_unpin_lock(rq, rf);
1544 		pull_rt_task(rq);
1545 		rq_repin_lock(rq, rf);
1546 		/*
1547 		 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1548 		 * means a dl or stop task can slip in, in which case we need
1549 		 * to re-start task selection.
1550 		 */
1551 		if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) ||
1552 			     rq->dl.dl_nr_running))
1553 			return RETRY_TASK;
1554 	}
1555 
1556 	/*
1557 	 * We may dequeue prev's rt_rq in put_prev_task().
1558 	 * So, we update time before rt_nr_running check.
1559 	 */
1560 	if (prev->sched_class == &rt_sched_class)
1561 		update_curr_rt(rq);
1562 
1563 	if (!rt_rq->rt_queued)
1564 		return NULL;
1565 
1566 	put_prev_task(rq, prev);
1567 
1568 	p = _pick_next_task_rt(rq);
1569 
1570 	/* The running task is never eligible for pushing */
1571 	dequeue_pushable_task(rq, p);
1572 
1573 	rt_queue_push_tasks(rq);
1574 
1575 	return p;
1576 }
1577 
1578 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1579 {
1580 	update_curr_rt(rq);
1581 
1582 	/*
1583 	 * The previous task needs to be made eligible for pushing
1584 	 * if it is still active
1585 	 */
1586 	if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
1587 		enqueue_pushable_task(rq, p);
1588 }
1589 
1590 #ifdef CONFIG_SMP
1591 
1592 /* Only try algorithms three times */
1593 #define RT_MAX_TRIES 3
1594 
1595 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1596 {
1597 	if (!task_running(rq, p) &&
1598 	    cpumask_test_cpu(cpu, &p->cpus_allowed))
1599 		return 1;
1600 
1601 	return 0;
1602 }
1603 
1604 /*
1605  * Return the highest pushable rq's task, which is suitable to be executed
1606  * on the CPU, NULL otherwise
1607  */
1608 static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu)
1609 {
1610 	struct plist_head *head = &rq->rt.pushable_tasks;
1611 	struct task_struct *p;
1612 
1613 	if (!has_pushable_tasks(rq))
1614 		return NULL;
1615 
1616 	plist_for_each_entry(p, head, pushable_tasks) {
1617 		if (pick_rt_task(rq, p, cpu))
1618 			return p;
1619 	}
1620 
1621 	return NULL;
1622 }
1623 
1624 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1625 
1626 static int find_lowest_rq(struct task_struct *task)
1627 {
1628 	struct sched_domain *sd;
1629 	struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask);
1630 	int this_cpu = smp_processor_id();
1631 	int cpu      = task_cpu(task);
1632 
1633 	/* Make sure the mask is initialized first */
1634 	if (unlikely(!lowest_mask))
1635 		return -1;
1636 
1637 	if (task->nr_cpus_allowed == 1)
1638 		return -1; /* No other targets possible */
1639 
1640 	if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1641 		return -1; /* No targets found */
1642 
1643 	/*
1644 	 * At this point we have built a mask of CPUs representing the
1645 	 * lowest priority tasks in the system.  Now we want to elect
1646 	 * the best one based on our affinity and topology.
1647 	 *
1648 	 * We prioritize the last CPU that the task executed on since
1649 	 * it is most likely cache-hot in that location.
1650 	 */
1651 	if (cpumask_test_cpu(cpu, lowest_mask))
1652 		return cpu;
1653 
1654 	/*
1655 	 * Otherwise, we consult the sched_domains span maps to figure
1656 	 * out which CPU is logically closest to our hot cache data.
1657 	 */
1658 	if (!cpumask_test_cpu(this_cpu, lowest_mask))
1659 		this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1660 
1661 	rcu_read_lock();
1662 	for_each_domain(cpu, sd) {
1663 		if (sd->flags & SD_WAKE_AFFINE) {
1664 			int best_cpu;
1665 
1666 			/*
1667 			 * "this_cpu" is cheaper to preempt than a
1668 			 * remote processor.
1669 			 */
1670 			if (this_cpu != -1 &&
1671 			    cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1672 				rcu_read_unlock();
1673 				return this_cpu;
1674 			}
1675 
1676 			best_cpu = cpumask_first_and(lowest_mask,
1677 						     sched_domain_span(sd));
1678 			if (best_cpu < nr_cpu_ids) {
1679 				rcu_read_unlock();
1680 				return best_cpu;
1681 			}
1682 		}
1683 	}
1684 	rcu_read_unlock();
1685 
1686 	/*
1687 	 * And finally, if there were no matches within the domains
1688 	 * just give the caller *something* to work with from the compatible
1689 	 * locations.
1690 	 */
1691 	if (this_cpu != -1)
1692 		return this_cpu;
1693 
1694 	cpu = cpumask_any(lowest_mask);
1695 	if (cpu < nr_cpu_ids)
1696 		return cpu;
1697 
1698 	return -1;
1699 }
1700 
1701 /* Will lock the rq it finds */
1702 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1703 {
1704 	struct rq *lowest_rq = NULL;
1705 	int tries;
1706 	int cpu;
1707 
1708 	for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1709 		cpu = find_lowest_rq(task);
1710 
1711 		if ((cpu == -1) || (cpu == rq->cpu))
1712 			break;
1713 
1714 		lowest_rq = cpu_rq(cpu);
1715 
1716 		if (lowest_rq->rt.highest_prio.curr <= task->prio) {
1717 			/*
1718 			 * Target rq has tasks of equal or higher priority,
1719 			 * retrying does not release any lock and is unlikely
1720 			 * to yield a different result.
1721 			 */
1722 			lowest_rq = NULL;
1723 			break;
1724 		}
1725 
1726 		/* if the prio of this runqueue changed, try again */
1727 		if (double_lock_balance(rq, lowest_rq)) {
1728 			/*
1729 			 * We had to unlock the run queue. In
1730 			 * the mean time, task could have
1731 			 * migrated already or had its affinity changed.
1732 			 * Also make sure that it wasn't scheduled on its rq.
1733 			 */
1734 			if (unlikely(task_rq(task) != rq ||
1735 				     !cpumask_test_cpu(lowest_rq->cpu, &task->cpus_allowed) ||
1736 				     task_running(rq, task) ||
1737 				     !rt_task(task) ||
1738 				     !task_on_rq_queued(task))) {
1739 
1740 				double_unlock_balance(rq, lowest_rq);
1741 				lowest_rq = NULL;
1742 				break;
1743 			}
1744 		}
1745 
1746 		/* If this rq is still suitable use it. */
1747 		if (lowest_rq->rt.highest_prio.curr > task->prio)
1748 			break;
1749 
1750 		/* try again */
1751 		double_unlock_balance(rq, lowest_rq);
1752 		lowest_rq = NULL;
1753 	}
1754 
1755 	return lowest_rq;
1756 }
1757 
1758 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1759 {
1760 	struct task_struct *p;
1761 
1762 	if (!has_pushable_tasks(rq))
1763 		return NULL;
1764 
1765 	p = plist_first_entry(&rq->rt.pushable_tasks,
1766 			      struct task_struct, pushable_tasks);
1767 
1768 	BUG_ON(rq->cpu != task_cpu(p));
1769 	BUG_ON(task_current(rq, p));
1770 	BUG_ON(p->nr_cpus_allowed <= 1);
1771 
1772 	BUG_ON(!task_on_rq_queued(p));
1773 	BUG_ON(!rt_task(p));
1774 
1775 	return p;
1776 }
1777 
1778 /*
1779  * If the current CPU has more than one RT task, see if the non
1780  * running task can migrate over to a CPU that is running a task
1781  * of lesser priority.
1782  */
1783 static int push_rt_task(struct rq *rq)
1784 {
1785 	struct task_struct *next_task;
1786 	struct rq *lowest_rq;
1787 	int ret = 0;
1788 
1789 	if (!rq->rt.overloaded)
1790 		return 0;
1791 
1792 	next_task = pick_next_pushable_task(rq);
1793 	if (!next_task)
1794 		return 0;
1795 
1796 retry:
1797 	if (unlikely(next_task == rq->curr)) {
1798 		WARN_ON(1);
1799 		return 0;
1800 	}
1801 
1802 	/*
1803 	 * It's possible that the next_task slipped in of
1804 	 * higher priority than current. If that's the case
1805 	 * just reschedule current.
1806 	 */
1807 	if (unlikely(next_task->prio < rq->curr->prio)) {
1808 		resched_curr(rq);
1809 		return 0;
1810 	}
1811 
1812 	/* We might release rq lock */
1813 	get_task_struct(next_task);
1814 
1815 	/* find_lock_lowest_rq locks the rq if found */
1816 	lowest_rq = find_lock_lowest_rq(next_task, rq);
1817 	if (!lowest_rq) {
1818 		struct task_struct *task;
1819 		/*
1820 		 * find_lock_lowest_rq releases rq->lock
1821 		 * so it is possible that next_task has migrated.
1822 		 *
1823 		 * We need to make sure that the task is still on the same
1824 		 * run-queue and is also still the next task eligible for
1825 		 * pushing.
1826 		 */
1827 		task = pick_next_pushable_task(rq);
1828 		if (task == next_task) {
1829 			/*
1830 			 * The task hasn't migrated, and is still the next
1831 			 * eligible task, but we failed to find a run-queue
1832 			 * to push it to.  Do not retry in this case, since
1833 			 * other CPUs will pull from us when ready.
1834 			 */
1835 			goto out;
1836 		}
1837 
1838 		if (!task)
1839 			/* No more tasks, just exit */
1840 			goto out;
1841 
1842 		/*
1843 		 * Something has shifted, try again.
1844 		 */
1845 		put_task_struct(next_task);
1846 		next_task = task;
1847 		goto retry;
1848 	}
1849 
1850 	deactivate_task(rq, next_task, 0);
1851 	set_task_cpu(next_task, lowest_rq->cpu);
1852 	activate_task(lowest_rq, next_task, 0);
1853 	ret = 1;
1854 
1855 	resched_curr(lowest_rq);
1856 
1857 	double_unlock_balance(rq, lowest_rq);
1858 
1859 out:
1860 	put_task_struct(next_task);
1861 
1862 	return ret;
1863 }
1864 
1865 static void push_rt_tasks(struct rq *rq)
1866 {
1867 	/* push_rt_task will return true if it moved an RT */
1868 	while (push_rt_task(rq))
1869 		;
1870 }
1871 
1872 #ifdef HAVE_RT_PUSH_IPI
1873 
1874 /*
1875  * When a high priority task schedules out from a CPU and a lower priority
1876  * task is scheduled in, a check is made to see if there's any RT tasks
1877  * on other CPUs that are waiting to run because a higher priority RT task
1878  * is currently running on its CPU. In this case, the CPU with multiple RT
1879  * tasks queued on it (overloaded) needs to be notified that a CPU has opened
1880  * up that may be able to run one of its non-running queued RT tasks.
1881  *
1882  * All CPUs with overloaded RT tasks need to be notified as there is currently
1883  * no way to know which of these CPUs have the highest priority task waiting
1884  * to run. Instead of trying to take a spinlock on each of these CPUs,
1885  * which has shown to cause large latency when done on machines with many
1886  * CPUs, sending an IPI to the CPUs to have them push off the overloaded
1887  * RT tasks waiting to run.
1888  *
1889  * Just sending an IPI to each of the CPUs is also an issue, as on large
1890  * count CPU machines, this can cause an IPI storm on a CPU, especially
1891  * if its the only CPU with multiple RT tasks queued, and a large number
1892  * of CPUs scheduling a lower priority task at the same time.
1893  *
1894  * Each root domain has its own irq work function that can iterate over
1895  * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT
1896  * tassk must be checked if there's one or many CPUs that are lowering
1897  * their priority, there's a single irq work iterator that will try to
1898  * push off RT tasks that are waiting to run.
1899  *
1900  * When a CPU schedules a lower priority task, it will kick off the
1901  * irq work iterator that will jump to each CPU with overloaded RT tasks.
1902  * As it only takes the first CPU that schedules a lower priority task
1903  * to start the process, the rto_start variable is incremented and if
1904  * the atomic result is one, then that CPU will try to take the rto_lock.
1905  * This prevents high contention on the lock as the process handles all
1906  * CPUs scheduling lower priority tasks.
1907  *
1908  * All CPUs that are scheduling a lower priority task will increment the
1909  * rt_loop_next variable. This will make sure that the irq work iterator
1910  * checks all RT overloaded CPUs whenever a CPU schedules a new lower
1911  * priority task, even if the iterator is in the middle of a scan. Incrementing
1912  * the rt_loop_next will cause the iterator to perform another scan.
1913  *
1914  */
1915 static int rto_next_cpu(struct root_domain *rd)
1916 {
1917 	int next;
1918 	int cpu;
1919 
1920 	/*
1921 	 * When starting the IPI RT pushing, the rto_cpu is set to -1,
1922 	 * rt_next_cpu() will simply return the first CPU found in
1923 	 * the rto_mask.
1924 	 *
1925 	 * If rto_next_cpu() is called with rto_cpu is a valid CPU, it
1926 	 * will return the next CPU found in the rto_mask.
1927 	 *
1928 	 * If there are no more CPUs left in the rto_mask, then a check is made
1929 	 * against rto_loop and rto_loop_next. rto_loop is only updated with
1930 	 * the rto_lock held, but any CPU may increment the rto_loop_next
1931 	 * without any locking.
1932 	 */
1933 	for (;;) {
1934 
1935 		/* When rto_cpu is -1 this acts like cpumask_first() */
1936 		cpu = cpumask_next(rd->rto_cpu, rd->rto_mask);
1937 
1938 		rd->rto_cpu = cpu;
1939 
1940 		if (cpu < nr_cpu_ids)
1941 			return cpu;
1942 
1943 		rd->rto_cpu = -1;
1944 
1945 		/*
1946 		 * ACQUIRE ensures we see the @rto_mask changes
1947 		 * made prior to the @next value observed.
1948 		 *
1949 		 * Matches WMB in rt_set_overload().
1950 		 */
1951 		next = atomic_read_acquire(&rd->rto_loop_next);
1952 
1953 		if (rd->rto_loop == next)
1954 			break;
1955 
1956 		rd->rto_loop = next;
1957 	}
1958 
1959 	return -1;
1960 }
1961 
1962 static inline bool rto_start_trylock(atomic_t *v)
1963 {
1964 	return !atomic_cmpxchg_acquire(v, 0, 1);
1965 }
1966 
1967 static inline void rto_start_unlock(atomic_t *v)
1968 {
1969 	atomic_set_release(v, 0);
1970 }
1971 
1972 static void tell_cpu_to_push(struct rq *rq)
1973 {
1974 	int cpu = -1;
1975 
1976 	/* Keep the loop going if the IPI is currently active */
1977 	atomic_inc(&rq->rd->rto_loop_next);
1978 
1979 	/* Only one CPU can initiate a loop at a time */
1980 	if (!rto_start_trylock(&rq->rd->rto_loop_start))
1981 		return;
1982 
1983 	raw_spin_lock(&rq->rd->rto_lock);
1984 
1985 	/*
1986 	 * The rto_cpu is updated under the lock, if it has a valid CPU
1987 	 * then the IPI is still running and will continue due to the
1988 	 * update to loop_next, and nothing needs to be done here.
1989 	 * Otherwise it is finishing up and an ipi needs to be sent.
1990 	 */
1991 	if (rq->rd->rto_cpu < 0)
1992 		cpu = rto_next_cpu(rq->rd);
1993 
1994 	raw_spin_unlock(&rq->rd->rto_lock);
1995 
1996 	rto_start_unlock(&rq->rd->rto_loop_start);
1997 
1998 	if (cpu >= 0) {
1999 		/* Make sure the rd does not get freed while pushing */
2000 		sched_get_rd(rq->rd);
2001 		irq_work_queue_on(&rq->rd->rto_push_work, cpu);
2002 	}
2003 }
2004 
2005 /* Called from hardirq context */
2006 void rto_push_irq_work_func(struct irq_work *work)
2007 {
2008 	struct root_domain *rd =
2009 		container_of(work, struct root_domain, rto_push_work);
2010 	struct rq *rq;
2011 	int cpu;
2012 
2013 	rq = this_rq();
2014 
2015 	/*
2016 	 * We do not need to grab the lock to check for has_pushable_tasks.
2017 	 * When it gets updated, a check is made if a push is possible.
2018 	 */
2019 	if (has_pushable_tasks(rq)) {
2020 		raw_spin_lock(&rq->lock);
2021 		push_rt_tasks(rq);
2022 		raw_spin_unlock(&rq->lock);
2023 	}
2024 
2025 	raw_spin_lock(&rd->rto_lock);
2026 
2027 	/* Pass the IPI to the next rt overloaded queue */
2028 	cpu = rto_next_cpu(rd);
2029 
2030 	raw_spin_unlock(&rd->rto_lock);
2031 
2032 	if (cpu < 0) {
2033 		sched_put_rd(rd);
2034 		return;
2035 	}
2036 
2037 	/* Try the next RT overloaded CPU */
2038 	irq_work_queue_on(&rd->rto_push_work, cpu);
2039 }
2040 #endif /* HAVE_RT_PUSH_IPI */
2041 
2042 static void pull_rt_task(struct rq *this_rq)
2043 {
2044 	int this_cpu = this_rq->cpu, cpu;
2045 	bool resched = false;
2046 	struct task_struct *p;
2047 	struct rq *src_rq;
2048 	int rt_overload_count = rt_overloaded(this_rq);
2049 
2050 	if (likely(!rt_overload_count))
2051 		return;
2052 
2053 	/*
2054 	 * Match the barrier from rt_set_overloaded; this guarantees that if we
2055 	 * see overloaded we must also see the rto_mask bit.
2056 	 */
2057 	smp_rmb();
2058 
2059 	/* If we are the only overloaded CPU do nothing */
2060 	if (rt_overload_count == 1 &&
2061 	    cpumask_test_cpu(this_rq->cpu, this_rq->rd->rto_mask))
2062 		return;
2063 
2064 #ifdef HAVE_RT_PUSH_IPI
2065 	if (sched_feat(RT_PUSH_IPI)) {
2066 		tell_cpu_to_push(this_rq);
2067 		return;
2068 	}
2069 #endif
2070 
2071 	for_each_cpu(cpu, this_rq->rd->rto_mask) {
2072 		if (this_cpu == cpu)
2073 			continue;
2074 
2075 		src_rq = cpu_rq(cpu);
2076 
2077 		/*
2078 		 * Don't bother taking the src_rq->lock if the next highest
2079 		 * task is known to be lower-priority than our current task.
2080 		 * This may look racy, but if this value is about to go
2081 		 * logically higher, the src_rq will push this task away.
2082 		 * And if its going logically lower, we do not care
2083 		 */
2084 		if (src_rq->rt.highest_prio.next >=
2085 		    this_rq->rt.highest_prio.curr)
2086 			continue;
2087 
2088 		/*
2089 		 * We can potentially drop this_rq's lock in
2090 		 * double_lock_balance, and another CPU could
2091 		 * alter this_rq
2092 		 */
2093 		double_lock_balance(this_rq, src_rq);
2094 
2095 		/*
2096 		 * We can pull only a task, which is pushable
2097 		 * on its rq, and no others.
2098 		 */
2099 		p = pick_highest_pushable_task(src_rq, this_cpu);
2100 
2101 		/*
2102 		 * Do we have an RT task that preempts
2103 		 * the to-be-scheduled task?
2104 		 */
2105 		if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
2106 			WARN_ON(p == src_rq->curr);
2107 			WARN_ON(!task_on_rq_queued(p));
2108 
2109 			/*
2110 			 * There's a chance that p is higher in priority
2111 			 * than what's currently running on its CPU.
2112 			 * This is just that p is wakeing up and hasn't
2113 			 * had a chance to schedule. We only pull
2114 			 * p if it is lower in priority than the
2115 			 * current task on the run queue
2116 			 */
2117 			if (p->prio < src_rq->curr->prio)
2118 				goto skip;
2119 
2120 			resched = true;
2121 
2122 			deactivate_task(src_rq, p, 0);
2123 			set_task_cpu(p, this_cpu);
2124 			activate_task(this_rq, p, 0);
2125 			/*
2126 			 * We continue with the search, just in
2127 			 * case there's an even higher prio task
2128 			 * in another runqueue. (low likelihood
2129 			 * but possible)
2130 			 */
2131 		}
2132 skip:
2133 		double_unlock_balance(this_rq, src_rq);
2134 	}
2135 
2136 	if (resched)
2137 		resched_curr(this_rq);
2138 }
2139 
2140 /*
2141  * If we are not running and we are not going to reschedule soon, we should
2142  * try to push tasks away now
2143  */
2144 static void task_woken_rt(struct rq *rq, struct task_struct *p)
2145 {
2146 	if (!task_running(rq, p) &&
2147 	    !test_tsk_need_resched(rq->curr) &&
2148 	    p->nr_cpus_allowed > 1 &&
2149 	    (dl_task(rq->curr) || rt_task(rq->curr)) &&
2150 	    (rq->curr->nr_cpus_allowed < 2 ||
2151 	     rq->curr->prio <= p->prio))
2152 		push_rt_tasks(rq);
2153 }
2154 
2155 /* Assumes rq->lock is held */
2156 static void rq_online_rt(struct rq *rq)
2157 {
2158 	if (rq->rt.overloaded)
2159 		rt_set_overload(rq);
2160 
2161 	__enable_runtime(rq);
2162 
2163 	cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
2164 }
2165 
2166 /* Assumes rq->lock is held */
2167 static void rq_offline_rt(struct rq *rq)
2168 {
2169 	if (rq->rt.overloaded)
2170 		rt_clear_overload(rq);
2171 
2172 	__disable_runtime(rq);
2173 
2174 	cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
2175 }
2176 
2177 /*
2178  * When switch from the rt queue, we bring ourselves to a position
2179  * that we might want to pull RT tasks from other runqueues.
2180  */
2181 static void switched_from_rt(struct rq *rq, struct task_struct *p)
2182 {
2183 	/*
2184 	 * If there are other RT tasks then we will reschedule
2185 	 * and the scheduling of the other RT tasks will handle
2186 	 * the balancing. But if we are the last RT task
2187 	 * we may need to handle the pulling of RT tasks
2188 	 * now.
2189 	 */
2190 	if (!task_on_rq_queued(p) || rq->rt.rt_nr_running)
2191 		return;
2192 
2193 	rt_queue_pull_task(rq);
2194 }
2195 
2196 void __init init_sched_rt_class(void)
2197 {
2198 	unsigned int i;
2199 
2200 	for_each_possible_cpu(i) {
2201 		zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
2202 					GFP_KERNEL, cpu_to_node(i));
2203 	}
2204 }
2205 #endif /* CONFIG_SMP */
2206 
2207 /*
2208  * When switching a task to RT, we may overload the runqueue
2209  * with RT tasks. In this case we try to push them off to
2210  * other runqueues.
2211  */
2212 static void switched_to_rt(struct rq *rq, struct task_struct *p)
2213 {
2214 	/*
2215 	 * If we are already running, then there's nothing
2216 	 * that needs to be done. But if we are not running
2217 	 * we may need to preempt the current running task.
2218 	 * If that current running task is also an RT task
2219 	 * then see if we can move to another run queue.
2220 	 */
2221 	if (task_on_rq_queued(p) && rq->curr != p) {
2222 #ifdef CONFIG_SMP
2223 		if (p->nr_cpus_allowed > 1 && rq->rt.overloaded)
2224 			rt_queue_push_tasks(rq);
2225 #endif /* CONFIG_SMP */
2226 		if (p->prio < rq->curr->prio && cpu_online(cpu_of(rq)))
2227 			resched_curr(rq);
2228 	}
2229 }
2230 
2231 /*
2232  * Priority of the task has changed. This may cause
2233  * us to initiate a push or pull.
2234  */
2235 static void
2236 prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio)
2237 {
2238 	if (!task_on_rq_queued(p))
2239 		return;
2240 
2241 	if (rq->curr == p) {
2242 #ifdef CONFIG_SMP
2243 		/*
2244 		 * If our priority decreases while running, we
2245 		 * may need to pull tasks to this runqueue.
2246 		 */
2247 		if (oldprio < p->prio)
2248 			rt_queue_pull_task(rq);
2249 
2250 		/*
2251 		 * If there's a higher priority task waiting to run
2252 		 * then reschedule.
2253 		 */
2254 		if (p->prio > rq->rt.highest_prio.curr)
2255 			resched_curr(rq);
2256 #else
2257 		/* For UP simply resched on drop of prio */
2258 		if (oldprio < p->prio)
2259 			resched_curr(rq);
2260 #endif /* CONFIG_SMP */
2261 	} else {
2262 		/*
2263 		 * This task is not running, but if it is
2264 		 * greater than the current running task
2265 		 * then reschedule.
2266 		 */
2267 		if (p->prio < rq->curr->prio)
2268 			resched_curr(rq);
2269 	}
2270 }
2271 
2272 #ifdef CONFIG_POSIX_TIMERS
2273 static void watchdog(struct rq *rq, struct task_struct *p)
2274 {
2275 	unsigned long soft, hard;
2276 
2277 	/* max may change after cur was read, this will be fixed next tick */
2278 	soft = task_rlimit(p, RLIMIT_RTTIME);
2279 	hard = task_rlimit_max(p, RLIMIT_RTTIME);
2280 
2281 	if (soft != RLIM_INFINITY) {
2282 		unsigned long next;
2283 
2284 		if (p->rt.watchdog_stamp != jiffies) {
2285 			p->rt.timeout++;
2286 			p->rt.watchdog_stamp = jiffies;
2287 		}
2288 
2289 		next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
2290 		if (p->rt.timeout > next)
2291 			p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
2292 	}
2293 }
2294 #else
2295 static inline void watchdog(struct rq *rq, struct task_struct *p) { }
2296 #endif
2297 
2298 /*
2299  * scheduler tick hitting a task of our scheduling class.
2300  *
2301  * NOTE: This function can be called remotely by the tick offload that
2302  * goes along full dynticks. Therefore no local assumption can be made
2303  * and everything must be accessed through the @rq and @curr passed in
2304  * parameters.
2305  */
2306 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
2307 {
2308 	struct sched_rt_entity *rt_se = &p->rt;
2309 
2310 	update_curr_rt(rq);
2311 
2312 	watchdog(rq, p);
2313 
2314 	/*
2315 	 * RR tasks need a special form of timeslice management.
2316 	 * FIFO tasks have no timeslices.
2317 	 */
2318 	if (p->policy != SCHED_RR)
2319 		return;
2320 
2321 	if (--p->rt.time_slice)
2322 		return;
2323 
2324 	p->rt.time_slice = sched_rr_timeslice;
2325 
2326 	/*
2327 	 * Requeue to the end of queue if we (and all of our ancestors) are not
2328 	 * the only element on the queue
2329 	 */
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);
2333 			resched_curr(rq);
2334 			return;
2335 		}
2336 	}
2337 }
2338 
2339 static void set_curr_task_rt(struct rq *rq)
2340 {
2341 	struct task_struct *p = rq->curr;
2342 
2343 	p->se.exec_start = rq_clock_task(rq);
2344 
2345 	/* The running task is never eligible for pushing */
2346 	dequeue_pushable_task(rq, p);
2347 }
2348 
2349 static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
2350 {
2351 	/*
2352 	 * Time slice is 0 for SCHED_FIFO tasks
2353 	 */
2354 	if (task->policy == SCHED_RR)
2355 		return sched_rr_timeslice;
2356 	else
2357 		return 0;
2358 }
2359 
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,
2365 
2366 	.check_preempt_curr	= check_preempt_curr_rt,
2367 
2368 	.pick_next_task		= pick_next_task_rt,
2369 	.put_prev_task		= put_prev_task_rt,
2370 
2371 #ifdef CONFIG_SMP
2372 	.select_task_rq		= select_task_rq_rt,
2373 
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,
2379 #endif
2380 
2381 	.set_curr_task          = set_curr_task_rt,
2382 	.task_tick		= task_tick_rt,
2383 
2384 	.get_rr_interval	= get_rr_interval_rt,
2385 
2386 	.prio_changed		= prio_changed_rt,
2387 	.switched_to		= switched_to_rt,
2388 
2389 	.update_curr		= update_curr_rt,
2390 };
2391 
2392 #ifdef CONFIG_RT_GROUP_SCHED
2393 /*
2394  * Ensure that the real time constraints are schedulable.
2395  */
2396 static DEFINE_MUTEX(rt_constraints_mutex);
2397 
2398 /* Must be called with tasklist_lock held */
2399 static inline int tg_has_rt_tasks(struct task_group *tg)
2400 {
2401 	struct task_struct *g, *p;
2402 
2403 	/*
2404 	 * Autogroups do not have RT tasks; see autogroup_create().
2405 	 */
2406 	if (task_group_is_autogroup(tg))
2407 		return 0;
2408 
2409 	for_each_process_thread(g, p) {
2410 		if (rt_task(p) && task_group(p) == tg)
2411 			return 1;
2412 	}
2413 
2414 	return 0;
2415 }
2416 
2417 struct rt_schedulable_data {
2418 	struct task_group *tg;
2419 	u64 rt_period;
2420 	u64 rt_runtime;
2421 };
2422 
2423 static int tg_rt_schedulable(struct task_group *tg, void *data)
2424 {
2425 	struct rt_schedulable_data *d = data;
2426 	struct task_group *child;
2427 	unsigned long total, sum = 0;
2428 	u64 period, runtime;
2429 
2430 	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
2431 	runtime = tg->rt_bandwidth.rt_runtime;
2432 
2433 	if (tg == d->tg) {
2434 		period = d->rt_period;
2435 		runtime = d->rt_runtime;
2436 	}
2437 
2438 	/*
2439 	 * Cannot have more runtime than the period.
2440 	 */
2441 	if (runtime > period && runtime != RUNTIME_INF)
2442 		return -EINVAL;
2443 
2444 	/*
2445 	 * Ensure we don't starve existing RT tasks.
2446 	 */
2447 	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
2448 		return -EBUSY;
2449 
2450 	total = to_ratio(period, runtime);
2451 
2452 	/*
2453 	 * Nobody can have more than the global setting allows.
2454 	 */
2455 	if (total > to_ratio(global_rt_period(), global_rt_runtime()))
2456 		return -EINVAL;
2457 
2458 	/*
2459 	 * The sum of our children's runtime should not exceed our own.
2460 	 */
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;
2464 
2465 		if (child == d->tg) {
2466 			period = d->rt_period;
2467 			runtime = d->rt_runtime;
2468 		}
2469 
2470 		sum += to_ratio(period, runtime);
2471 	}
2472 
2473 	if (sum > total)
2474 		return -EINVAL;
2475 
2476 	return 0;
2477 }
2478 
2479 static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
2480 {
2481 	int ret;
2482 
2483 	struct rt_schedulable_data data = {
2484 		.tg = tg,
2485 		.rt_period = period,
2486 		.rt_runtime = runtime,
2487 	};
2488 
2489 	rcu_read_lock();
2490 	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
2491 	rcu_read_unlock();
2492 
2493 	return ret;
2494 }
2495 
2496 static int tg_set_rt_bandwidth(struct task_group *tg,
2497 		u64 rt_period, u64 rt_runtime)
2498 {
2499 	int i, err = 0;
2500 
2501 	/*
2502 	 * Disallowing the root group RT runtime is BAD, it would disallow the
2503 	 * kernel creating (and or operating) RT threads.
2504 	 */
2505 	if (tg == &root_task_group && rt_runtime == 0)
2506 		return -EINVAL;
2507 
2508 	/* No period doesn't make any sense. */
2509 	if (rt_period == 0)
2510 		return -EINVAL;
2511 
2512 	mutex_lock(&rt_constraints_mutex);
2513 	read_lock(&tasklist_lock);
2514 	err = __rt_schedulable(tg, rt_period, rt_runtime);
2515 	if (err)
2516 		goto unlock;
2517 
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;
2521 
2522 	for_each_possible_cpu(i) {
2523 		struct rt_rq *rt_rq = tg->rt_rq[i];
2524 
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);
2528 	}
2529 	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
2530 unlock:
2531 	read_unlock(&tasklist_lock);
2532 	mutex_unlock(&rt_constraints_mutex);
2533 
2534 	return err;
2535 }
2536 
2537 int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
2538 {
2539 	u64 rt_runtime, rt_period;
2540 
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 
2546 	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2547 }
2548 
2549 long sched_group_rt_runtime(struct task_group *tg)
2550 {
2551 	u64 rt_runtime_us;
2552 
2553 	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
2554 		return -1;
2555 
2556 	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
2557 	do_div(rt_runtime_us, NSEC_PER_USEC);
2558 	return rt_runtime_us;
2559 }
2560 
2561 int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
2562 {
2563 	u64 rt_runtime, rt_period;
2564 
2565 	rt_period = rt_period_us * NSEC_PER_USEC;
2566 	rt_runtime = tg->rt_bandwidth.rt_runtime;
2567 
2568 	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2569 }
2570 
2571 long sched_group_rt_period(struct task_group *tg)
2572 {
2573 	u64 rt_period_us;
2574 
2575 	rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
2576 	do_div(rt_period_us, NSEC_PER_USEC);
2577 	return rt_period_us;
2578 }
2579 
2580 static int sched_rt_global_constraints(void)
2581 {
2582 	int ret = 0;
2583 
2584 	mutex_lock(&rt_constraints_mutex);
2585 	read_lock(&tasklist_lock);
2586 	ret = __rt_schedulable(NULL, 0, 0);
2587 	read_unlock(&tasklist_lock);
2588 	mutex_unlock(&rt_constraints_mutex);
2589 
2590 	return ret;
2591 }
2592 
2593 int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
2594 {
2595 	/* Don't accept realtime tasks when there is no way for them to run */
2596 	if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
2597 		return 0;
2598 
2599 	return 1;
2600 }
2601 
2602 #else /* !CONFIG_RT_GROUP_SCHED */
2603 static int sched_rt_global_constraints(void)
2604 {
2605 	unsigned long flags;
2606 	int i;
2607 
2608 	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
2609 	for_each_possible_cpu(i) {
2610 		struct rt_rq *rt_rq = &cpu_rq(i)->rt;
2611 
2612 		raw_spin_lock(&rt_rq->rt_runtime_lock);
2613 		rt_rq->rt_runtime = global_rt_runtime();
2614 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
2615 	}
2616 	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
2617 
2618 	return 0;
2619 }
2620 #endif /* CONFIG_RT_GROUP_SCHED */
2621 
2622 static int sched_rt_global_validate(void)
2623 {
2624 	if (sysctl_sched_rt_period <= 0)
2625 		return -EINVAL;
2626 
2627 	if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
2628 		(sysctl_sched_rt_runtime > sysctl_sched_rt_period))
2629 		return -EINVAL;
2630 
2631 	return 0;
2632 }
2633 
2634 static void sched_rt_do_global(void)
2635 {
2636 	def_rt_bandwidth.rt_runtime = global_rt_runtime();
2637 	def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period());
2638 }
2639 
2640 int sched_rt_handler(struct ctl_table *table, int write,
2641 		void __user *buffer, size_t *lenp,
2642 		loff_t *ppos)
2643 {
2644 	int old_period, old_runtime;
2645 	static DEFINE_MUTEX(mutex);
2646 	int ret;
2647 
2648 	mutex_lock(&mutex);
2649 	old_period = sysctl_sched_rt_period;
2650 	old_runtime = sysctl_sched_rt_runtime;
2651 
2652 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
2653 
2654 	if (!ret && write) {
2655 		ret = sched_rt_global_validate();
2656 		if (ret)
2657 			goto undo;
2658 
2659 		ret = sched_dl_global_validate();
2660 		if (ret)
2661 			goto undo;
2662 
2663 		ret = sched_rt_global_constraints();
2664 		if (ret)
2665 			goto undo;
2666 
2667 		sched_rt_do_global();
2668 		sched_dl_do_global();
2669 	}
2670 	if (0) {
2671 undo:
2672 		sysctl_sched_rt_period = old_period;
2673 		sysctl_sched_rt_runtime = old_runtime;
2674 	}
2675 	mutex_unlock(&mutex);
2676 
2677 	return ret;
2678 }
2679 
2680 int sched_rr_handler(struct ctl_table *table, int write,
2681 		void __user *buffer, size_t *lenp,
2682 		loff_t *ppos)
2683 {
2684 	int ret;
2685 	static DEFINE_MUTEX(mutex);
2686 
2687 	mutex_lock(&mutex);
2688 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
2689 	/*
2690 	 * Make sure that internally we keep jiffies.
2691 	 * Also, writing zero resets the timeslice to default:
2692 	 */
2693 	if (!ret && write) {
2694 		sched_rr_timeslice =
2695 			sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE :
2696 			msecs_to_jiffies(sysctl_sched_rr_timeslice);
2697 	}
2698 	mutex_unlock(&mutex);
2699 
2700 	return ret;
2701 }
2702 
2703 #ifdef CONFIG_SCHED_DEBUG
2704 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2705 
2706 void print_rt_stats(struct seq_file *m, int cpu)
2707 {
2708 	rt_rq_iter_t iter;
2709 	struct rt_rq *rt_rq;
2710 
2711 	rcu_read_lock();
2712 	for_each_rt_rq(rt_rq, iter, cpu_rq(cpu))
2713 		print_rt_rq(m, cpu, rt_rq);
2714 	rcu_read_unlock();
2715 }
2716 #endif /* CONFIG_SCHED_DEBUG */
2717