xref: /openbmc/linux/kernel/sched/rt.c (revision 8fdf9062)
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 #include "pelt.h"
9 
10 int sched_rr_timeslice = RR_TIMESLICE;
11 int sysctl_sched_rr_timeslice = (MSEC_PER_SEC / HZ) * RR_TIMESLICE;
12 
13 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
14 
15 struct rt_bandwidth def_rt_bandwidth;
16 
17 static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
18 {
19 	struct rt_bandwidth *rt_b =
20 		container_of(timer, struct rt_bandwidth, rt_period_timer);
21 	int idle = 0;
22 	int overrun;
23 
24 	raw_spin_lock(&rt_b->rt_runtime_lock);
25 	for (;;) {
26 		overrun = hrtimer_forward_now(timer, rt_b->rt_period);
27 		if (!overrun)
28 			break;
29 
30 		raw_spin_unlock(&rt_b->rt_runtime_lock);
31 		idle = do_sched_rt_period_timer(rt_b, overrun);
32 		raw_spin_lock(&rt_b->rt_runtime_lock);
33 	}
34 	if (idle)
35 		rt_b->rt_period_active = 0;
36 	raw_spin_unlock(&rt_b->rt_runtime_lock);
37 
38 	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
39 }
40 
41 void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
42 {
43 	rt_b->rt_period = ns_to_ktime(period);
44 	rt_b->rt_runtime = runtime;
45 
46 	raw_spin_lock_init(&rt_b->rt_runtime_lock);
47 
48 	hrtimer_init(&rt_b->rt_period_timer,
49 			CLOCK_MONOTONIC, HRTIMER_MODE_REL);
50 	rt_b->rt_period_timer.function = sched_rt_period_timer;
51 }
52 
53 static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
54 {
55 	if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
56 		return;
57 
58 	raw_spin_lock(&rt_b->rt_runtime_lock);
59 	if (!rt_b->rt_period_active) {
60 		rt_b->rt_period_active = 1;
61 		/*
62 		 * SCHED_DEADLINE updates the bandwidth, as a run away
63 		 * RT task with a DL task could hog a CPU. But DL does
64 		 * not reset the period. If a deadline task was running
65 		 * without an RT task running, it can cause RT tasks to
66 		 * throttle when they start up. Kick the timer right away
67 		 * to update the period.
68 		 */
69 		hrtimer_forward_now(&rt_b->rt_period_timer, ns_to_ktime(0));
70 		hrtimer_start_expires(&rt_b->rt_period_timer, HRTIMER_MODE_ABS_PINNED);
71 	}
72 	raw_spin_unlock(&rt_b->rt_runtime_lock);
73 }
74 
75 void init_rt_rq(struct rt_rq *rt_rq)
76 {
77 	struct rt_prio_array *array;
78 	int i;
79 
80 	array = &rt_rq->active;
81 	for (i = 0; i < MAX_RT_PRIO; i++) {
82 		INIT_LIST_HEAD(array->queue + i);
83 		__clear_bit(i, array->bitmap);
84 	}
85 	/* delimiter for bitsearch: */
86 	__set_bit(MAX_RT_PRIO, array->bitmap);
87 
88 #if defined CONFIG_SMP
89 	rt_rq->highest_prio.curr = MAX_RT_PRIO;
90 	rt_rq->highest_prio.next = MAX_RT_PRIO;
91 	rt_rq->rt_nr_migratory = 0;
92 	rt_rq->overloaded = 0;
93 	plist_head_init(&rt_rq->pushable_tasks);
94 #endif /* CONFIG_SMP */
95 	/* We start is dequeued state, because no RT tasks are queued */
96 	rt_rq->rt_queued = 0;
97 
98 	rt_rq->rt_time = 0;
99 	rt_rq->rt_throttled = 0;
100 	rt_rq->rt_runtime = 0;
101 	raw_spin_lock_init(&rt_rq->rt_runtime_lock);
102 }
103 
104 #ifdef CONFIG_RT_GROUP_SCHED
105 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
106 {
107 	hrtimer_cancel(&rt_b->rt_period_timer);
108 }
109 
110 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
111 
112 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
113 {
114 #ifdef CONFIG_SCHED_DEBUG
115 	WARN_ON_ONCE(!rt_entity_is_task(rt_se));
116 #endif
117 	return container_of(rt_se, struct task_struct, rt);
118 }
119 
120 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
121 {
122 	return rt_rq->rq;
123 }
124 
125 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
126 {
127 	return rt_se->rt_rq;
128 }
129 
130 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
131 {
132 	struct rt_rq *rt_rq = rt_se->rt_rq;
133 
134 	return rt_rq->rq;
135 }
136 
137 void free_rt_sched_group(struct task_group *tg)
138 {
139 	int i;
140 
141 	if (tg->rt_se)
142 		destroy_rt_bandwidth(&tg->rt_bandwidth);
143 
144 	for_each_possible_cpu(i) {
145 		if (tg->rt_rq)
146 			kfree(tg->rt_rq[i]);
147 		if (tg->rt_se)
148 			kfree(tg->rt_se[i]);
149 	}
150 
151 	kfree(tg->rt_rq);
152 	kfree(tg->rt_se);
153 }
154 
155 void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
156 		struct sched_rt_entity *rt_se, int cpu,
157 		struct sched_rt_entity *parent)
158 {
159 	struct rq *rq = cpu_rq(cpu);
160 
161 	rt_rq->highest_prio.curr = MAX_RT_PRIO;
162 	rt_rq->rt_nr_boosted = 0;
163 	rt_rq->rq = rq;
164 	rt_rq->tg = tg;
165 
166 	tg->rt_rq[cpu] = rt_rq;
167 	tg->rt_se[cpu] = rt_se;
168 
169 	if (!rt_se)
170 		return;
171 
172 	if (!parent)
173 		rt_se->rt_rq = &rq->rt;
174 	else
175 		rt_se->rt_rq = parent->my_q;
176 
177 	rt_se->my_q = rt_rq;
178 	rt_se->parent = parent;
179 	INIT_LIST_HEAD(&rt_se->run_list);
180 }
181 
182 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
183 {
184 	struct rt_rq *rt_rq;
185 	struct sched_rt_entity *rt_se;
186 	int i;
187 
188 	tg->rt_rq = kcalloc(nr_cpu_ids, sizeof(rt_rq), GFP_KERNEL);
189 	if (!tg->rt_rq)
190 		goto err;
191 	tg->rt_se = kcalloc(nr_cpu_ids, sizeof(rt_se), GFP_KERNEL);
192 	if (!tg->rt_se)
193 		goto err;
194 
195 	init_rt_bandwidth(&tg->rt_bandwidth,
196 			ktime_to_ns(def_rt_bandwidth.rt_period), 0);
197 
198 	for_each_possible_cpu(i) {
199 		rt_rq = kzalloc_node(sizeof(struct rt_rq),
200 				     GFP_KERNEL, cpu_to_node(i));
201 		if (!rt_rq)
202 			goto err;
203 
204 		rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
205 				     GFP_KERNEL, cpu_to_node(i));
206 		if (!rt_se)
207 			goto err_free_rq;
208 
209 		init_rt_rq(rt_rq);
210 		rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
211 		init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
212 	}
213 
214 	return 1;
215 
216 err_free_rq:
217 	kfree(rt_rq);
218 err:
219 	return 0;
220 }
221 
222 #else /* CONFIG_RT_GROUP_SCHED */
223 
224 #define rt_entity_is_task(rt_se) (1)
225 
226 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
227 {
228 	return container_of(rt_se, struct task_struct, rt);
229 }
230 
231 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
232 {
233 	return container_of(rt_rq, struct rq, rt);
234 }
235 
236 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
237 {
238 	struct task_struct *p = rt_task_of(rt_se);
239 
240 	return task_rq(p);
241 }
242 
243 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
244 {
245 	struct rq *rq = rq_of_rt_se(rt_se);
246 
247 	return &rq->rt;
248 }
249 
250 void free_rt_sched_group(struct task_group *tg) { }
251 
252 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
253 {
254 	return 1;
255 }
256 #endif /* CONFIG_RT_GROUP_SCHED */
257 
258 #ifdef CONFIG_SMP
259 
260 static void pull_rt_task(struct rq *this_rq);
261 
262 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
263 {
264 	/* Try to pull RT tasks here if we lower this rq's prio */
265 	return rq->rt.highest_prio.curr > prev->prio;
266 }
267 
268 static inline int rt_overloaded(struct rq *rq)
269 {
270 	return atomic_read(&rq->rd->rto_count);
271 }
272 
273 static inline void rt_set_overload(struct rq *rq)
274 {
275 	if (!rq->online)
276 		return;
277 
278 	cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
279 	/*
280 	 * Make sure the mask is visible before we set
281 	 * the overload count. That is checked to determine
282 	 * if we should look at the mask. It would be a shame
283 	 * if we looked at the mask, but the mask was not
284 	 * updated yet.
285 	 *
286 	 * Matched by the barrier in pull_rt_task().
287 	 */
288 	smp_wmb();
289 	atomic_inc(&rq->rd->rto_count);
290 }
291 
292 static inline void rt_clear_overload(struct rq *rq)
293 {
294 	if (!rq->online)
295 		return;
296 
297 	/* the order here really doesn't matter */
298 	atomic_dec(&rq->rd->rto_count);
299 	cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
300 }
301 
302 static void update_rt_migration(struct rt_rq *rt_rq)
303 {
304 	if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
305 		if (!rt_rq->overloaded) {
306 			rt_set_overload(rq_of_rt_rq(rt_rq));
307 			rt_rq->overloaded = 1;
308 		}
309 	} else if (rt_rq->overloaded) {
310 		rt_clear_overload(rq_of_rt_rq(rt_rq));
311 		rt_rq->overloaded = 0;
312 	}
313 }
314 
315 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
316 {
317 	struct task_struct *p;
318 
319 	if (!rt_entity_is_task(rt_se))
320 		return;
321 
322 	p = rt_task_of(rt_se);
323 	rt_rq = &rq_of_rt_rq(rt_rq)->rt;
324 
325 	rt_rq->rt_nr_total++;
326 	if (p->nr_cpus_allowed > 1)
327 		rt_rq->rt_nr_migratory++;
328 
329 	update_rt_migration(rt_rq);
330 }
331 
332 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
333 {
334 	struct task_struct *p;
335 
336 	if (!rt_entity_is_task(rt_se))
337 		return;
338 
339 	p = rt_task_of(rt_se);
340 	rt_rq = &rq_of_rt_rq(rt_rq)->rt;
341 
342 	rt_rq->rt_nr_total--;
343 	if (p->nr_cpus_allowed > 1)
344 		rt_rq->rt_nr_migratory--;
345 
346 	update_rt_migration(rt_rq);
347 }
348 
349 static inline int has_pushable_tasks(struct rq *rq)
350 {
351 	return !plist_head_empty(&rq->rt.pushable_tasks);
352 }
353 
354 static DEFINE_PER_CPU(struct callback_head, rt_push_head);
355 static DEFINE_PER_CPU(struct callback_head, rt_pull_head);
356 
357 static void push_rt_tasks(struct rq *);
358 static void pull_rt_task(struct rq *);
359 
360 static inline void rt_queue_push_tasks(struct rq *rq)
361 {
362 	if (!has_pushable_tasks(rq))
363 		return;
364 
365 	queue_balance_callback(rq, &per_cpu(rt_push_head, rq->cpu), push_rt_tasks);
366 }
367 
368 static inline void rt_queue_pull_task(struct rq *rq)
369 {
370 	queue_balance_callback(rq, &per_cpu(rt_pull_head, rq->cpu), pull_rt_task);
371 }
372 
373 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
374 {
375 	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
376 	plist_node_init(&p->pushable_tasks, p->prio);
377 	plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
378 
379 	/* Update the highest prio pushable task */
380 	if (p->prio < rq->rt.highest_prio.next)
381 		rq->rt.highest_prio.next = p->prio;
382 }
383 
384 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
385 {
386 	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
387 
388 	/* Update the new highest prio pushable task */
389 	if (has_pushable_tasks(rq)) {
390 		p = plist_first_entry(&rq->rt.pushable_tasks,
391 				      struct task_struct, pushable_tasks);
392 		rq->rt.highest_prio.next = p->prio;
393 	} else
394 		rq->rt.highest_prio.next = MAX_RT_PRIO;
395 }
396 
397 #else
398 
399 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
400 {
401 }
402 
403 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
404 {
405 }
406 
407 static inline
408 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
409 {
410 }
411 
412 static inline
413 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
414 {
415 }
416 
417 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
418 {
419 	return false;
420 }
421 
422 static inline void pull_rt_task(struct rq *this_rq)
423 {
424 }
425 
426 static inline void rt_queue_push_tasks(struct rq *rq)
427 {
428 }
429 #endif /* CONFIG_SMP */
430 
431 static void enqueue_top_rt_rq(struct rt_rq *rt_rq);
432 static void dequeue_top_rt_rq(struct rt_rq *rt_rq);
433 
434 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
435 {
436 	return rt_se->on_rq;
437 }
438 
439 #ifdef CONFIG_RT_GROUP_SCHED
440 
441 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
442 {
443 	if (!rt_rq->tg)
444 		return RUNTIME_INF;
445 
446 	return rt_rq->rt_runtime;
447 }
448 
449 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
450 {
451 	return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
452 }
453 
454 typedef struct task_group *rt_rq_iter_t;
455 
456 static inline struct task_group *next_task_group(struct task_group *tg)
457 {
458 	do {
459 		tg = list_entry_rcu(tg->list.next,
460 			typeof(struct task_group), list);
461 	} while (&tg->list != &task_groups && task_group_is_autogroup(tg));
462 
463 	if (&tg->list == &task_groups)
464 		tg = NULL;
465 
466 	return tg;
467 }
468 
469 #define for_each_rt_rq(rt_rq, iter, rq)					\
470 	for (iter = container_of(&task_groups, typeof(*iter), list);	\
471 		(iter = next_task_group(iter)) &&			\
472 		(rt_rq = iter->rt_rq[cpu_of(rq)]);)
473 
474 #define for_each_sched_rt_entity(rt_se) \
475 	for (; rt_se; rt_se = rt_se->parent)
476 
477 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
478 {
479 	return rt_se->my_q;
480 }
481 
482 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
483 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags);
484 
485 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
486 {
487 	struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
488 	struct rq *rq = rq_of_rt_rq(rt_rq);
489 	struct sched_rt_entity *rt_se;
490 
491 	int cpu = cpu_of(rq);
492 
493 	rt_se = rt_rq->tg->rt_se[cpu];
494 
495 	if (rt_rq->rt_nr_running) {
496 		if (!rt_se)
497 			enqueue_top_rt_rq(rt_rq);
498 		else if (!on_rt_rq(rt_se))
499 			enqueue_rt_entity(rt_se, 0);
500 
501 		if (rt_rq->highest_prio.curr < curr->prio)
502 			resched_curr(rq);
503 	}
504 }
505 
506 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
507 {
508 	struct sched_rt_entity *rt_se;
509 	int cpu = cpu_of(rq_of_rt_rq(rt_rq));
510 
511 	rt_se = rt_rq->tg->rt_se[cpu];
512 
513 	if (!rt_se) {
514 		dequeue_top_rt_rq(rt_rq);
515 		/* Kick cpufreq (see the comment in kernel/sched/sched.h). */
516 		cpufreq_update_util(rq_of_rt_rq(rt_rq), 0);
517 	}
518 	else if (on_rt_rq(rt_se))
519 		dequeue_rt_entity(rt_se, 0);
520 }
521 
522 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
523 {
524 	return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
525 }
526 
527 static int rt_se_boosted(struct sched_rt_entity *rt_se)
528 {
529 	struct rt_rq *rt_rq = group_rt_rq(rt_se);
530 	struct task_struct *p;
531 
532 	if (rt_rq)
533 		return !!rt_rq->rt_nr_boosted;
534 
535 	p = rt_task_of(rt_se);
536 	return p->prio != p->normal_prio;
537 }
538 
539 #ifdef CONFIG_SMP
540 static inline const struct cpumask *sched_rt_period_mask(void)
541 {
542 	return this_rq()->rd->span;
543 }
544 #else
545 static inline const struct cpumask *sched_rt_period_mask(void)
546 {
547 	return cpu_online_mask;
548 }
549 #endif
550 
551 static inline
552 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
553 {
554 	return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
555 }
556 
557 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
558 {
559 	return &rt_rq->tg->rt_bandwidth;
560 }
561 
562 #else /* !CONFIG_RT_GROUP_SCHED */
563 
564 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
565 {
566 	return rt_rq->rt_runtime;
567 }
568 
569 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
570 {
571 	return ktime_to_ns(def_rt_bandwidth.rt_period);
572 }
573 
574 typedef struct rt_rq *rt_rq_iter_t;
575 
576 #define for_each_rt_rq(rt_rq, iter, rq) \
577 	for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
578 
579 #define for_each_sched_rt_entity(rt_se) \
580 	for (; rt_se; rt_se = NULL)
581 
582 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
583 {
584 	return NULL;
585 }
586 
587 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
588 {
589 	struct rq *rq = rq_of_rt_rq(rt_rq);
590 
591 	if (!rt_rq->rt_nr_running)
592 		return;
593 
594 	enqueue_top_rt_rq(rt_rq);
595 	resched_curr(rq);
596 }
597 
598 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
599 {
600 	dequeue_top_rt_rq(rt_rq);
601 }
602 
603 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
604 {
605 	return rt_rq->rt_throttled;
606 }
607 
608 static inline const struct cpumask *sched_rt_period_mask(void)
609 {
610 	return cpu_online_mask;
611 }
612 
613 static inline
614 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
615 {
616 	return &cpu_rq(cpu)->rt;
617 }
618 
619 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
620 {
621 	return &def_rt_bandwidth;
622 }
623 
624 #endif /* CONFIG_RT_GROUP_SCHED */
625 
626 bool sched_rt_bandwidth_account(struct rt_rq *rt_rq)
627 {
628 	struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
629 
630 	return (hrtimer_active(&rt_b->rt_period_timer) ||
631 		rt_rq->rt_time < rt_b->rt_runtime);
632 }
633 
634 #ifdef CONFIG_SMP
635 /*
636  * We ran out of runtime, see if we can borrow some from our neighbours.
637  */
638 static void do_balance_runtime(struct rt_rq *rt_rq)
639 {
640 	struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
641 	struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd;
642 	int i, weight;
643 	u64 rt_period;
644 
645 	weight = cpumask_weight(rd->span);
646 
647 	raw_spin_lock(&rt_b->rt_runtime_lock);
648 	rt_period = ktime_to_ns(rt_b->rt_period);
649 	for_each_cpu(i, rd->span) {
650 		struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
651 		s64 diff;
652 
653 		if (iter == rt_rq)
654 			continue;
655 
656 		raw_spin_lock(&iter->rt_runtime_lock);
657 		/*
658 		 * Either all rqs have inf runtime and there's nothing to steal
659 		 * or __disable_runtime() below sets a specific rq to inf to
660 		 * indicate its been disabled and disalow stealing.
661 		 */
662 		if (iter->rt_runtime == RUNTIME_INF)
663 			goto next;
664 
665 		/*
666 		 * From runqueues with spare time, take 1/n part of their
667 		 * spare time, but no more than our period.
668 		 */
669 		diff = iter->rt_runtime - iter->rt_time;
670 		if (diff > 0) {
671 			diff = div_u64((u64)diff, weight);
672 			if (rt_rq->rt_runtime + diff > rt_period)
673 				diff = rt_period - rt_rq->rt_runtime;
674 			iter->rt_runtime -= diff;
675 			rt_rq->rt_runtime += diff;
676 			if (rt_rq->rt_runtime == rt_period) {
677 				raw_spin_unlock(&iter->rt_runtime_lock);
678 				break;
679 			}
680 		}
681 next:
682 		raw_spin_unlock(&iter->rt_runtime_lock);
683 	}
684 	raw_spin_unlock(&rt_b->rt_runtime_lock);
685 }
686 
687 /*
688  * Ensure this RQ takes back all the runtime it lend to its neighbours.
689  */
690 static void __disable_runtime(struct rq *rq)
691 {
692 	struct root_domain *rd = rq->rd;
693 	rt_rq_iter_t iter;
694 	struct rt_rq *rt_rq;
695 
696 	if (unlikely(!scheduler_running))
697 		return;
698 
699 	for_each_rt_rq(rt_rq, iter, rq) {
700 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
701 		s64 want;
702 		int i;
703 
704 		raw_spin_lock(&rt_b->rt_runtime_lock);
705 		raw_spin_lock(&rt_rq->rt_runtime_lock);
706 		/*
707 		 * Either we're all inf and nobody needs to borrow, or we're
708 		 * already disabled and thus have nothing to do, or we have
709 		 * exactly the right amount of runtime to take out.
710 		 */
711 		if (rt_rq->rt_runtime == RUNTIME_INF ||
712 				rt_rq->rt_runtime == rt_b->rt_runtime)
713 			goto balanced;
714 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
715 
716 		/*
717 		 * Calculate the difference between what we started out with
718 		 * and what we current have, that's the amount of runtime
719 		 * we lend and now have to reclaim.
720 		 */
721 		want = rt_b->rt_runtime - rt_rq->rt_runtime;
722 
723 		/*
724 		 * Greedy reclaim, take back as much as we can.
725 		 */
726 		for_each_cpu(i, rd->span) {
727 			struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
728 			s64 diff;
729 
730 			/*
731 			 * Can't reclaim from ourselves or disabled runqueues.
732 			 */
733 			if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
734 				continue;
735 
736 			raw_spin_lock(&iter->rt_runtime_lock);
737 			if (want > 0) {
738 				diff = min_t(s64, iter->rt_runtime, want);
739 				iter->rt_runtime -= diff;
740 				want -= diff;
741 			} else {
742 				iter->rt_runtime -= want;
743 				want -= want;
744 			}
745 			raw_spin_unlock(&iter->rt_runtime_lock);
746 
747 			if (!want)
748 				break;
749 		}
750 
751 		raw_spin_lock(&rt_rq->rt_runtime_lock);
752 		/*
753 		 * We cannot be left wanting - that would mean some runtime
754 		 * leaked out of the system.
755 		 */
756 		BUG_ON(want);
757 balanced:
758 		/*
759 		 * Disable all the borrow logic by pretending we have inf
760 		 * runtime - in which case borrowing doesn't make sense.
761 		 */
762 		rt_rq->rt_runtime = RUNTIME_INF;
763 		rt_rq->rt_throttled = 0;
764 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
765 		raw_spin_unlock(&rt_b->rt_runtime_lock);
766 
767 		/* Make rt_rq available for pick_next_task() */
768 		sched_rt_rq_enqueue(rt_rq);
769 	}
770 }
771 
772 static void __enable_runtime(struct rq *rq)
773 {
774 	rt_rq_iter_t iter;
775 	struct rt_rq *rt_rq;
776 
777 	if (unlikely(!scheduler_running))
778 		return;
779 
780 	/*
781 	 * Reset each runqueue's bandwidth settings
782 	 */
783 	for_each_rt_rq(rt_rq, iter, rq) {
784 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
785 
786 		raw_spin_lock(&rt_b->rt_runtime_lock);
787 		raw_spin_lock(&rt_rq->rt_runtime_lock);
788 		rt_rq->rt_runtime = rt_b->rt_runtime;
789 		rt_rq->rt_time = 0;
790 		rt_rq->rt_throttled = 0;
791 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
792 		raw_spin_unlock(&rt_b->rt_runtime_lock);
793 	}
794 }
795 
796 static void balance_runtime(struct rt_rq *rt_rq)
797 {
798 	if (!sched_feat(RT_RUNTIME_SHARE))
799 		return;
800 
801 	if (rt_rq->rt_time > rt_rq->rt_runtime) {
802 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
803 		do_balance_runtime(rt_rq);
804 		raw_spin_lock(&rt_rq->rt_runtime_lock);
805 	}
806 }
807 #else /* !CONFIG_SMP */
808 static inline void balance_runtime(struct rt_rq *rt_rq) {}
809 #endif /* CONFIG_SMP */
810 
811 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
812 {
813 	int i, idle = 1, throttled = 0;
814 	const struct cpumask *span;
815 
816 	span = sched_rt_period_mask();
817 #ifdef CONFIG_RT_GROUP_SCHED
818 	/*
819 	 * FIXME: isolated CPUs should really leave the root task group,
820 	 * whether they are isolcpus or were isolated via cpusets, lest
821 	 * the timer run on a CPU which does not service all runqueues,
822 	 * potentially leaving other CPUs indefinitely throttled.  If
823 	 * isolation is really required, the user will turn the throttle
824 	 * off to kill the perturbations it causes anyway.  Meanwhile,
825 	 * this maintains functionality for boot and/or troubleshooting.
826 	 */
827 	if (rt_b == &root_task_group.rt_bandwidth)
828 		span = cpu_online_mask;
829 #endif
830 	for_each_cpu(i, span) {
831 		int enqueue = 0;
832 		struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
833 		struct rq *rq = rq_of_rt_rq(rt_rq);
834 		int skip;
835 
836 		/*
837 		 * When span == cpu_online_mask, taking each rq->lock
838 		 * can be time-consuming. Try to avoid it when possible.
839 		 */
840 		raw_spin_lock(&rt_rq->rt_runtime_lock);
841 		if (!sched_feat(RT_RUNTIME_SHARE) && rt_rq->rt_runtime != RUNTIME_INF)
842 			rt_rq->rt_runtime = rt_b->rt_runtime;
843 		skip = !rt_rq->rt_time && !rt_rq->rt_nr_running;
844 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
845 		if (skip)
846 			continue;
847 
848 		raw_spin_lock(&rq->lock);
849 		update_rq_clock(rq);
850 
851 		if (rt_rq->rt_time) {
852 			u64 runtime;
853 
854 			raw_spin_lock(&rt_rq->rt_runtime_lock);
855 			if (rt_rq->rt_throttled)
856 				balance_runtime(rt_rq);
857 			runtime = rt_rq->rt_runtime;
858 			rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
859 			if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
860 				rt_rq->rt_throttled = 0;
861 				enqueue = 1;
862 
863 				/*
864 				 * When we're idle and a woken (rt) task is
865 				 * throttled check_preempt_curr() will set
866 				 * skip_update and the time between the wakeup
867 				 * and this unthrottle will get accounted as
868 				 * 'runtime'.
869 				 */
870 				if (rt_rq->rt_nr_running && rq->curr == rq->idle)
871 					rq_clock_cancel_skipupdate(rq);
872 			}
873 			if (rt_rq->rt_time || rt_rq->rt_nr_running)
874 				idle = 0;
875 			raw_spin_unlock(&rt_rq->rt_runtime_lock);
876 		} else if (rt_rq->rt_nr_running) {
877 			idle = 0;
878 			if (!rt_rq_throttled(rt_rq))
879 				enqueue = 1;
880 		}
881 		if (rt_rq->rt_throttled)
882 			throttled = 1;
883 
884 		if (enqueue)
885 			sched_rt_rq_enqueue(rt_rq);
886 		raw_spin_unlock(&rq->lock);
887 	}
888 
889 	if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
890 		return 1;
891 
892 	return idle;
893 }
894 
895 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
896 {
897 #ifdef CONFIG_RT_GROUP_SCHED
898 	struct rt_rq *rt_rq = group_rt_rq(rt_se);
899 
900 	if (rt_rq)
901 		return rt_rq->highest_prio.curr;
902 #endif
903 
904 	return rt_task_of(rt_se)->prio;
905 }
906 
907 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
908 {
909 	u64 runtime = sched_rt_runtime(rt_rq);
910 
911 	if (rt_rq->rt_throttled)
912 		return rt_rq_throttled(rt_rq);
913 
914 	if (runtime >= sched_rt_period(rt_rq))
915 		return 0;
916 
917 	balance_runtime(rt_rq);
918 	runtime = sched_rt_runtime(rt_rq);
919 	if (runtime == RUNTIME_INF)
920 		return 0;
921 
922 	if (rt_rq->rt_time > runtime) {
923 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
924 
925 		/*
926 		 * Don't actually throttle groups that have no runtime assigned
927 		 * but accrue some time due to boosting.
928 		 */
929 		if (likely(rt_b->rt_runtime)) {
930 			rt_rq->rt_throttled = 1;
931 			printk_deferred_once("sched: RT throttling activated\n");
932 		} else {
933 			/*
934 			 * In case we did anyway, make it go away,
935 			 * replenishment is a joke, since it will replenish us
936 			 * with exactly 0 ns.
937 			 */
938 			rt_rq->rt_time = 0;
939 		}
940 
941 		if (rt_rq_throttled(rt_rq)) {
942 			sched_rt_rq_dequeue(rt_rq);
943 			return 1;
944 		}
945 	}
946 
947 	return 0;
948 }
949 
950 /*
951  * Update the current task's runtime statistics. Skip current tasks that
952  * are not in our scheduling class.
953  */
954 static void update_curr_rt(struct rq *rq)
955 {
956 	struct task_struct *curr = rq->curr;
957 	struct sched_rt_entity *rt_se = &curr->rt;
958 	u64 delta_exec;
959 	u64 now;
960 
961 	if (curr->sched_class != &rt_sched_class)
962 		return;
963 
964 	now = rq_clock_task(rq);
965 	delta_exec = now - curr->se.exec_start;
966 	if (unlikely((s64)delta_exec <= 0))
967 		return;
968 
969 	schedstat_set(curr->se.statistics.exec_max,
970 		      max(curr->se.statistics.exec_max, delta_exec));
971 
972 	curr->se.sum_exec_runtime += delta_exec;
973 	account_group_exec_runtime(curr, delta_exec);
974 
975 	curr->se.exec_start = now;
976 	cgroup_account_cputime(curr, delta_exec);
977 
978 	if (!rt_bandwidth_enabled())
979 		return;
980 
981 	for_each_sched_rt_entity(rt_se) {
982 		struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
983 
984 		if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
985 			raw_spin_lock(&rt_rq->rt_runtime_lock);
986 			rt_rq->rt_time += delta_exec;
987 			if (sched_rt_runtime_exceeded(rt_rq))
988 				resched_curr(rq);
989 			raw_spin_unlock(&rt_rq->rt_runtime_lock);
990 		}
991 	}
992 }
993 
994 static void
995 dequeue_top_rt_rq(struct rt_rq *rt_rq)
996 {
997 	struct rq *rq = rq_of_rt_rq(rt_rq);
998 
999 	BUG_ON(&rq->rt != rt_rq);
1000 
1001 	if (!rt_rq->rt_queued)
1002 		return;
1003 
1004 	BUG_ON(!rq->nr_running);
1005 
1006 	sub_nr_running(rq, rt_rq->rt_nr_running);
1007 	rt_rq->rt_queued = 0;
1008 
1009 }
1010 
1011 static void
1012 enqueue_top_rt_rq(struct rt_rq *rt_rq)
1013 {
1014 	struct rq *rq = rq_of_rt_rq(rt_rq);
1015 
1016 	BUG_ON(&rq->rt != rt_rq);
1017 
1018 	if (rt_rq->rt_queued)
1019 		return;
1020 
1021 	if (rt_rq_throttled(rt_rq))
1022 		return;
1023 
1024 	if (rt_rq->rt_nr_running) {
1025 		add_nr_running(rq, rt_rq->rt_nr_running);
1026 		rt_rq->rt_queued = 1;
1027 	}
1028 
1029 	/* Kick cpufreq (see the comment in kernel/sched/sched.h). */
1030 	cpufreq_update_util(rq, 0);
1031 }
1032 
1033 #if defined CONFIG_SMP
1034 
1035 static void
1036 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1037 {
1038 	struct rq *rq = rq_of_rt_rq(rt_rq);
1039 
1040 #ifdef CONFIG_RT_GROUP_SCHED
1041 	/*
1042 	 * Change rq's cpupri only if rt_rq is the top queue.
1043 	 */
1044 	if (&rq->rt != rt_rq)
1045 		return;
1046 #endif
1047 	if (rq->online && prio < prev_prio)
1048 		cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
1049 }
1050 
1051 static void
1052 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1053 {
1054 	struct rq *rq = rq_of_rt_rq(rt_rq);
1055 
1056 #ifdef CONFIG_RT_GROUP_SCHED
1057 	/*
1058 	 * Change rq's cpupri only if rt_rq is the top queue.
1059 	 */
1060 	if (&rq->rt != rt_rq)
1061 		return;
1062 #endif
1063 	if (rq->online && rt_rq->highest_prio.curr != prev_prio)
1064 		cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
1065 }
1066 
1067 #else /* CONFIG_SMP */
1068 
1069 static inline
1070 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1071 static inline
1072 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1073 
1074 #endif /* CONFIG_SMP */
1075 
1076 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
1077 static void
1078 inc_rt_prio(struct rt_rq *rt_rq, int prio)
1079 {
1080 	int prev_prio = rt_rq->highest_prio.curr;
1081 
1082 	if (prio < prev_prio)
1083 		rt_rq->highest_prio.curr = prio;
1084 
1085 	inc_rt_prio_smp(rt_rq, prio, prev_prio);
1086 }
1087 
1088 static void
1089 dec_rt_prio(struct rt_rq *rt_rq, int prio)
1090 {
1091 	int prev_prio = rt_rq->highest_prio.curr;
1092 
1093 	if (rt_rq->rt_nr_running) {
1094 
1095 		WARN_ON(prio < prev_prio);
1096 
1097 		/*
1098 		 * This may have been our highest task, and therefore
1099 		 * we may have some recomputation to do
1100 		 */
1101 		if (prio == prev_prio) {
1102 			struct rt_prio_array *array = &rt_rq->active;
1103 
1104 			rt_rq->highest_prio.curr =
1105 				sched_find_first_bit(array->bitmap);
1106 		}
1107 
1108 	} else
1109 		rt_rq->highest_prio.curr = MAX_RT_PRIO;
1110 
1111 	dec_rt_prio_smp(rt_rq, prio, prev_prio);
1112 }
1113 
1114 #else
1115 
1116 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
1117 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
1118 
1119 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1120 
1121 #ifdef CONFIG_RT_GROUP_SCHED
1122 
1123 static void
1124 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1125 {
1126 	if (rt_se_boosted(rt_se))
1127 		rt_rq->rt_nr_boosted++;
1128 
1129 	if (rt_rq->tg)
1130 		start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
1131 }
1132 
1133 static void
1134 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1135 {
1136 	if (rt_se_boosted(rt_se))
1137 		rt_rq->rt_nr_boosted--;
1138 
1139 	WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
1140 }
1141 
1142 #else /* CONFIG_RT_GROUP_SCHED */
1143 
1144 static void
1145 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1146 {
1147 	start_rt_bandwidth(&def_rt_bandwidth);
1148 }
1149 
1150 static inline
1151 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
1152 
1153 #endif /* CONFIG_RT_GROUP_SCHED */
1154 
1155 static inline
1156 unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se)
1157 {
1158 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1159 
1160 	if (group_rq)
1161 		return group_rq->rt_nr_running;
1162 	else
1163 		return 1;
1164 }
1165 
1166 static inline
1167 unsigned int rt_se_rr_nr_running(struct sched_rt_entity *rt_se)
1168 {
1169 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1170 	struct task_struct *tsk;
1171 
1172 	if (group_rq)
1173 		return group_rq->rr_nr_running;
1174 
1175 	tsk = rt_task_of(rt_se);
1176 
1177 	return (tsk->policy == SCHED_RR) ? 1 : 0;
1178 }
1179 
1180 static inline
1181 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1182 {
1183 	int prio = rt_se_prio(rt_se);
1184 
1185 	WARN_ON(!rt_prio(prio));
1186 	rt_rq->rt_nr_running += rt_se_nr_running(rt_se);
1187 	rt_rq->rr_nr_running += rt_se_rr_nr_running(rt_se);
1188 
1189 	inc_rt_prio(rt_rq, prio);
1190 	inc_rt_migration(rt_se, rt_rq);
1191 	inc_rt_group(rt_se, rt_rq);
1192 }
1193 
1194 static inline
1195 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1196 {
1197 	WARN_ON(!rt_prio(rt_se_prio(rt_se)));
1198 	WARN_ON(!rt_rq->rt_nr_running);
1199 	rt_rq->rt_nr_running -= rt_se_nr_running(rt_se);
1200 	rt_rq->rr_nr_running -= rt_se_rr_nr_running(rt_se);
1201 
1202 	dec_rt_prio(rt_rq, rt_se_prio(rt_se));
1203 	dec_rt_migration(rt_se, rt_rq);
1204 	dec_rt_group(rt_se, rt_rq);
1205 }
1206 
1207 /*
1208  * Change rt_se->run_list location unless SAVE && !MOVE
1209  *
1210  * assumes ENQUEUE/DEQUEUE flags match
1211  */
1212 static inline bool move_entity(unsigned int flags)
1213 {
1214 	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
1215 		return false;
1216 
1217 	return true;
1218 }
1219 
1220 static void __delist_rt_entity(struct sched_rt_entity *rt_se, struct rt_prio_array *array)
1221 {
1222 	list_del_init(&rt_se->run_list);
1223 
1224 	if (list_empty(array->queue + rt_se_prio(rt_se)))
1225 		__clear_bit(rt_se_prio(rt_se), array->bitmap);
1226 
1227 	rt_se->on_list = 0;
1228 }
1229 
1230 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1231 {
1232 	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1233 	struct rt_prio_array *array = &rt_rq->active;
1234 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1235 	struct list_head *queue = array->queue + rt_se_prio(rt_se);
1236 
1237 	/*
1238 	 * Don't enqueue the group if its throttled, or when empty.
1239 	 * The latter is a consequence of the former when a child group
1240 	 * get throttled and the current group doesn't have any other
1241 	 * active members.
1242 	 */
1243 	if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) {
1244 		if (rt_se->on_list)
1245 			__delist_rt_entity(rt_se, array);
1246 		return;
1247 	}
1248 
1249 	if (move_entity(flags)) {
1250 		WARN_ON_ONCE(rt_se->on_list);
1251 		if (flags & ENQUEUE_HEAD)
1252 			list_add(&rt_se->run_list, queue);
1253 		else
1254 			list_add_tail(&rt_se->run_list, queue);
1255 
1256 		__set_bit(rt_se_prio(rt_se), array->bitmap);
1257 		rt_se->on_list = 1;
1258 	}
1259 	rt_se->on_rq = 1;
1260 
1261 	inc_rt_tasks(rt_se, rt_rq);
1262 }
1263 
1264 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1265 {
1266 	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1267 	struct rt_prio_array *array = &rt_rq->active;
1268 
1269 	if (move_entity(flags)) {
1270 		WARN_ON_ONCE(!rt_se->on_list);
1271 		__delist_rt_entity(rt_se, array);
1272 	}
1273 	rt_se->on_rq = 0;
1274 
1275 	dec_rt_tasks(rt_se, rt_rq);
1276 }
1277 
1278 /*
1279  * Because the prio of an upper entry depends on the lower
1280  * entries, we must remove entries top - down.
1281  */
1282 static void dequeue_rt_stack(struct sched_rt_entity *rt_se, unsigned int flags)
1283 {
1284 	struct sched_rt_entity *back = NULL;
1285 
1286 	for_each_sched_rt_entity(rt_se) {
1287 		rt_se->back = back;
1288 		back = rt_se;
1289 	}
1290 
1291 	dequeue_top_rt_rq(rt_rq_of_se(back));
1292 
1293 	for (rt_se = back; rt_se; rt_se = rt_se->back) {
1294 		if (on_rt_rq(rt_se))
1295 			__dequeue_rt_entity(rt_se, flags);
1296 	}
1297 }
1298 
1299 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1300 {
1301 	struct rq *rq = rq_of_rt_se(rt_se);
1302 
1303 	dequeue_rt_stack(rt_se, flags);
1304 	for_each_sched_rt_entity(rt_se)
1305 		__enqueue_rt_entity(rt_se, flags);
1306 	enqueue_top_rt_rq(&rq->rt);
1307 }
1308 
1309 static void dequeue_rt_entity(struct sched_rt_entity *rt_se, unsigned int flags)
1310 {
1311 	struct rq *rq = rq_of_rt_se(rt_se);
1312 
1313 	dequeue_rt_stack(rt_se, flags);
1314 
1315 	for_each_sched_rt_entity(rt_se) {
1316 		struct rt_rq *rt_rq = group_rt_rq(rt_se);
1317 
1318 		if (rt_rq && rt_rq->rt_nr_running)
1319 			__enqueue_rt_entity(rt_se, flags);
1320 	}
1321 	enqueue_top_rt_rq(&rq->rt);
1322 }
1323 
1324 /*
1325  * Adding/removing a task to/from a priority array:
1326  */
1327 static void
1328 enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1329 {
1330 	struct sched_rt_entity *rt_se = &p->rt;
1331 
1332 	if (flags & ENQUEUE_WAKEUP)
1333 		rt_se->timeout = 0;
1334 
1335 	enqueue_rt_entity(rt_se, flags);
1336 
1337 	if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1338 		enqueue_pushable_task(rq, p);
1339 }
1340 
1341 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1342 {
1343 	struct sched_rt_entity *rt_se = &p->rt;
1344 
1345 	update_curr_rt(rq);
1346 	dequeue_rt_entity(rt_se, flags);
1347 
1348 	dequeue_pushable_task(rq, p);
1349 }
1350 
1351 /*
1352  * Put task to the head or the end of the run list without the overhead of
1353  * dequeue followed by enqueue.
1354  */
1355 static void
1356 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
1357 {
1358 	if (on_rt_rq(rt_se)) {
1359 		struct rt_prio_array *array = &rt_rq->active;
1360 		struct list_head *queue = array->queue + rt_se_prio(rt_se);
1361 
1362 		if (head)
1363 			list_move(&rt_se->run_list, queue);
1364 		else
1365 			list_move_tail(&rt_se->run_list, queue);
1366 	}
1367 }
1368 
1369 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
1370 {
1371 	struct sched_rt_entity *rt_se = &p->rt;
1372 	struct rt_rq *rt_rq;
1373 
1374 	for_each_sched_rt_entity(rt_se) {
1375 		rt_rq = rt_rq_of_se(rt_se);
1376 		requeue_rt_entity(rt_rq, rt_se, head);
1377 	}
1378 }
1379 
1380 static void yield_task_rt(struct rq *rq)
1381 {
1382 	requeue_task_rt(rq, rq->curr, 0);
1383 }
1384 
1385 #ifdef CONFIG_SMP
1386 static int find_lowest_rq(struct task_struct *task);
1387 
1388 static int
1389 select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags)
1390 {
1391 	struct task_struct *curr;
1392 	struct rq *rq;
1393 
1394 	/* For anything but wake ups, just return the task_cpu */
1395 	if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK)
1396 		goto out;
1397 
1398 	rq = cpu_rq(cpu);
1399 
1400 	rcu_read_lock();
1401 	curr = READ_ONCE(rq->curr); /* unlocked access */
1402 
1403 	/*
1404 	 * If the current task on @p's runqueue is an RT task, then
1405 	 * try to see if we can wake this RT task up on another
1406 	 * runqueue. Otherwise simply start this RT task
1407 	 * on its current runqueue.
1408 	 *
1409 	 * We want to avoid overloading runqueues. If the woken
1410 	 * task is a higher priority, then it will stay on this CPU
1411 	 * and the lower prio task should be moved to another CPU.
1412 	 * Even though this will probably make the lower prio task
1413 	 * lose its cache, we do not want to bounce a higher task
1414 	 * around just because it gave up its CPU, perhaps for a
1415 	 * lock?
1416 	 *
1417 	 * For equal prio tasks, we just let the scheduler sort it out.
1418 	 *
1419 	 * Otherwise, just let it ride on the affined RQ and the
1420 	 * post-schedule router will push the preempted task away
1421 	 *
1422 	 * This test is optimistic, if we get it wrong the load-balancer
1423 	 * will have to sort it out.
1424 	 */
1425 	if (curr && unlikely(rt_task(curr)) &&
1426 	    (curr->nr_cpus_allowed < 2 ||
1427 	     curr->prio <= p->prio)) {
1428 		int target = find_lowest_rq(p);
1429 
1430 		/*
1431 		 * Don't bother moving it if the destination CPU is
1432 		 * not running a lower priority task.
1433 		 */
1434 		if (target != -1 &&
1435 		    p->prio < cpu_rq(target)->rt.highest_prio.curr)
1436 			cpu = target;
1437 	}
1438 	rcu_read_unlock();
1439 
1440 out:
1441 	return cpu;
1442 }
1443 
1444 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
1445 {
1446 	/*
1447 	 * Current can't be migrated, useless to reschedule,
1448 	 * let's hope p can move out.
1449 	 */
1450 	if (rq->curr->nr_cpus_allowed == 1 ||
1451 	    !cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
1452 		return;
1453 
1454 	/*
1455 	 * p is migratable, so let's not schedule it and
1456 	 * see if it is pushed or pulled somewhere else.
1457 	 */
1458 	if (p->nr_cpus_allowed != 1
1459 	    && cpupri_find(&rq->rd->cpupri, p, NULL))
1460 		return;
1461 
1462 	/*
1463 	 * There appear to be other CPUs that can accept
1464 	 * the current task but none can run 'p', so lets reschedule
1465 	 * to try and push the current task away:
1466 	 */
1467 	requeue_task_rt(rq, p, 1);
1468 	resched_curr(rq);
1469 }
1470 
1471 #endif /* CONFIG_SMP */
1472 
1473 /*
1474  * Preempt the current task with a newly woken task if needed:
1475  */
1476 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1477 {
1478 	if (p->prio < rq->curr->prio) {
1479 		resched_curr(rq);
1480 		return;
1481 	}
1482 
1483 #ifdef CONFIG_SMP
1484 	/*
1485 	 * If:
1486 	 *
1487 	 * - the newly woken task is of equal priority to the current task
1488 	 * - the newly woken task is non-migratable while current is migratable
1489 	 * - current will be preempted on the next reschedule
1490 	 *
1491 	 * we should check to see if current can readily move to a different
1492 	 * cpu.  If so, we will reschedule to allow the push logic to try
1493 	 * to move current somewhere else, making room for our non-migratable
1494 	 * task.
1495 	 */
1496 	if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr))
1497 		check_preempt_equal_prio(rq, p);
1498 #endif
1499 }
1500 
1501 static inline void set_next_task(struct rq *rq, struct task_struct *p)
1502 {
1503 	p->se.exec_start = rq_clock_task(rq);
1504 
1505 	/* The running task is never eligible for pushing */
1506 	dequeue_pushable_task(rq, p);
1507 }
1508 
1509 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1510 						   struct rt_rq *rt_rq)
1511 {
1512 	struct rt_prio_array *array = &rt_rq->active;
1513 	struct sched_rt_entity *next = NULL;
1514 	struct list_head *queue;
1515 	int idx;
1516 
1517 	idx = sched_find_first_bit(array->bitmap);
1518 	BUG_ON(idx >= MAX_RT_PRIO);
1519 
1520 	queue = array->queue + idx;
1521 	next = list_entry(queue->next, struct sched_rt_entity, run_list);
1522 
1523 	return next;
1524 }
1525 
1526 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1527 {
1528 	struct sched_rt_entity *rt_se;
1529 	struct rt_rq *rt_rq  = &rq->rt;
1530 
1531 	do {
1532 		rt_se = pick_next_rt_entity(rq, rt_rq);
1533 		BUG_ON(!rt_se);
1534 		rt_rq = group_rt_rq(rt_se);
1535 	} while (rt_rq);
1536 
1537 	return rt_task_of(rt_se);
1538 }
1539 
1540 static struct task_struct *
1541 pick_next_task_rt(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
1542 {
1543 	struct task_struct *p;
1544 	struct rt_rq *rt_rq = &rq->rt;
1545 
1546 	if (need_pull_rt_task(rq, prev)) {
1547 		/*
1548 		 * This is OK, because current is on_cpu, which avoids it being
1549 		 * picked for load-balance and preemption/IRQs are still
1550 		 * disabled avoiding further scheduler activity on it and we're
1551 		 * being very careful to re-start the picking loop.
1552 		 */
1553 		rq_unpin_lock(rq, rf);
1554 		pull_rt_task(rq);
1555 		rq_repin_lock(rq, rf);
1556 		/*
1557 		 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1558 		 * means a dl or stop task can slip in, in which case we need
1559 		 * to re-start task selection.
1560 		 */
1561 		if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) ||
1562 			     rq->dl.dl_nr_running))
1563 			return RETRY_TASK;
1564 	}
1565 
1566 	/*
1567 	 * We may dequeue prev's rt_rq in put_prev_task().
1568 	 * So, we update time before rt_queued check.
1569 	 */
1570 	if (prev->sched_class == &rt_sched_class)
1571 		update_curr_rt(rq);
1572 
1573 	if (!rt_rq->rt_queued)
1574 		return NULL;
1575 
1576 	put_prev_task(rq, prev);
1577 
1578 	p = _pick_next_task_rt(rq);
1579 
1580 	set_next_task(rq, p);
1581 
1582 	rt_queue_push_tasks(rq);
1583 
1584 	/*
1585 	 * If prev task was rt, put_prev_task() has already updated the
1586 	 * utilization. We only care of the case where we start to schedule a
1587 	 * rt task
1588 	 */
1589 	if (rq->curr->sched_class != &rt_sched_class)
1590 		update_rt_rq_load_avg(rq_clock_task(rq), rq, 0);
1591 
1592 	return p;
1593 }
1594 
1595 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1596 {
1597 	update_curr_rt(rq);
1598 
1599 	update_rt_rq_load_avg(rq_clock_task(rq), rq, 1);
1600 
1601 	/*
1602 	 * The previous task needs to be made eligible for pushing
1603 	 * if it is still active
1604 	 */
1605 	if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
1606 		enqueue_pushable_task(rq, p);
1607 }
1608 
1609 #ifdef CONFIG_SMP
1610 
1611 /* Only try algorithms three times */
1612 #define RT_MAX_TRIES 3
1613 
1614 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1615 {
1616 	if (!task_running(rq, p) &&
1617 	    cpumask_test_cpu(cpu, &p->cpus_allowed))
1618 		return 1;
1619 
1620 	return 0;
1621 }
1622 
1623 /*
1624  * Return the highest pushable rq's task, which is suitable to be executed
1625  * on the CPU, NULL otherwise
1626  */
1627 static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu)
1628 {
1629 	struct plist_head *head = &rq->rt.pushable_tasks;
1630 	struct task_struct *p;
1631 
1632 	if (!has_pushable_tasks(rq))
1633 		return NULL;
1634 
1635 	plist_for_each_entry(p, head, pushable_tasks) {
1636 		if (pick_rt_task(rq, p, cpu))
1637 			return p;
1638 	}
1639 
1640 	return NULL;
1641 }
1642 
1643 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1644 
1645 static int find_lowest_rq(struct task_struct *task)
1646 {
1647 	struct sched_domain *sd;
1648 	struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask);
1649 	int this_cpu = smp_processor_id();
1650 	int cpu      = task_cpu(task);
1651 
1652 	/* Make sure the mask is initialized first */
1653 	if (unlikely(!lowest_mask))
1654 		return -1;
1655 
1656 	if (task->nr_cpus_allowed == 1)
1657 		return -1; /* No other targets possible */
1658 
1659 	if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1660 		return -1; /* No targets found */
1661 
1662 	/*
1663 	 * At this point we have built a mask of CPUs representing the
1664 	 * lowest priority tasks in the system.  Now we want to elect
1665 	 * the best one based on our affinity and topology.
1666 	 *
1667 	 * We prioritize the last CPU that the task executed on since
1668 	 * it is most likely cache-hot in that location.
1669 	 */
1670 	if (cpumask_test_cpu(cpu, lowest_mask))
1671 		return cpu;
1672 
1673 	/*
1674 	 * Otherwise, we consult the sched_domains span maps to figure
1675 	 * out which CPU is logically closest to our hot cache data.
1676 	 */
1677 	if (!cpumask_test_cpu(this_cpu, lowest_mask))
1678 		this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1679 
1680 	rcu_read_lock();
1681 	for_each_domain(cpu, sd) {
1682 		if (sd->flags & SD_WAKE_AFFINE) {
1683 			int best_cpu;
1684 
1685 			/*
1686 			 * "this_cpu" is cheaper to preempt than a
1687 			 * remote processor.
1688 			 */
1689 			if (this_cpu != -1 &&
1690 			    cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1691 				rcu_read_unlock();
1692 				return this_cpu;
1693 			}
1694 
1695 			best_cpu = cpumask_first_and(lowest_mask,
1696 						     sched_domain_span(sd));
1697 			if (best_cpu < nr_cpu_ids) {
1698 				rcu_read_unlock();
1699 				return best_cpu;
1700 			}
1701 		}
1702 	}
1703 	rcu_read_unlock();
1704 
1705 	/*
1706 	 * And finally, if there were no matches within the domains
1707 	 * just give the caller *something* to work with from the compatible
1708 	 * locations.
1709 	 */
1710 	if (this_cpu != -1)
1711 		return this_cpu;
1712 
1713 	cpu = cpumask_any(lowest_mask);
1714 	if (cpu < nr_cpu_ids)
1715 		return cpu;
1716 
1717 	return -1;
1718 }
1719 
1720 /* Will lock the rq it finds */
1721 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1722 {
1723 	struct rq *lowest_rq = NULL;
1724 	int tries;
1725 	int cpu;
1726 
1727 	for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1728 		cpu = find_lowest_rq(task);
1729 
1730 		if ((cpu == -1) || (cpu == rq->cpu))
1731 			break;
1732 
1733 		lowest_rq = cpu_rq(cpu);
1734 
1735 		if (lowest_rq->rt.highest_prio.curr <= task->prio) {
1736 			/*
1737 			 * Target rq has tasks of equal or higher priority,
1738 			 * retrying does not release any lock and is unlikely
1739 			 * to yield a different result.
1740 			 */
1741 			lowest_rq = NULL;
1742 			break;
1743 		}
1744 
1745 		/* if the prio of this runqueue changed, try again */
1746 		if (double_lock_balance(rq, lowest_rq)) {
1747 			/*
1748 			 * We had to unlock the run queue. In
1749 			 * the mean time, task could have
1750 			 * migrated already or had its affinity changed.
1751 			 * Also make sure that it wasn't scheduled on its rq.
1752 			 */
1753 			if (unlikely(task_rq(task) != rq ||
1754 				     !cpumask_test_cpu(lowest_rq->cpu, &task->cpus_allowed) ||
1755 				     task_running(rq, task) ||
1756 				     !rt_task(task) ||
1757 				     !task_on_rq_queued(task))) {
1758 
1759 				double_unlock_balance(rq, lowest_rq);
1760 				lowest_rq = NULL;
1761 				break;
1762 			}
1763 		}
1764 
1765 		/* If this rq is still suitable use it. */
1766 		if (lowest_rq->rt.highest_prio.curr > task->prio)
1767 			break;
1768 
1769 		/* try again */
1770 		double_unlock_balance(rq, lowest_rq);
1771 		lowest_rq = NULL;
1772 	}
1773 
1774 	return lowest_rq;
1775 }
1776 
1777 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1778 {
1779 	struct task_struct *p;
1780 
1781 	if (!has_pushable_tasks(rq))
1782 		return NULL;
1783 
1784 	p = plist_first_entry(&rq->rt.pushable_tasks,
1785 			      struct task_struct, pushable_tasks);
1786 
1787 	BUG_ON(rq->cpu != task_cpu(p));
1788 	BUG_ON(task_current(rq, p));
1789 	BUG_ON(p->nr_cpus_allowed <= 1);
1790 
1791 	BUG_ON(!task_on_rq_queued(p));
1792 	BUG_ON(!rt_task(p));
1793 
1794 	return p;
1795 }
1796 
1797 /*
1798  * If the current CPU has more than one RT task, see if the non
1799  * running task can migrate over to a CPU that is running a task
1800  * of lesser priority.
1801  */
1802 static int push_rt_task(struct rq *rq)
1803 {
1804 	struct task_struct *next_task;
1805 	struct rq *lowest_rq;
1806 	int ret = 0;
1807 
1808 	if (!rq->rt.overloaded)
1809 		return 0;
1810 
1811 	next_task = pick_next_pushable_task(rq);
1812 	if (!next_task)
1813 		return 0;
1814 
1815 retry:
1816 	if (WARN_ON(next_task == rq->curr))
1817 		return 0;
1818 
1819 	/*
1820 	 * It's possible that the next_task slipped in of
1821 	 * higher priority than current. If that's the case
1822 	 * just reschedule current.
1823 	 */
1824 	if (unlikely(next_task->prio < rq->curr->prio)) {
1825 		resched_curr(rq);
1826 		return 0;
1827 	}
1828 
1829 	/* We might release rq lock */
1830 	get_task_struct(next_task);
1831 
1832 	/* find_lock_lowest_rq locks the rq if found */
1833 	lowest_rq = find_lock_lowest_rq(next_task, rq);
1834 	if (!lowest_rq) {
1835 		struct task_struct *task;
1836 		/*
1837 		 * find_lock_lowest_rq releases rq->lock
1838 		 * so it is possible that next_task has migrated.
1839 		 *
1840 		 * We need to make sure that the task is still on the same
1841 		 * run-queue and is also still the next task eligible for
1842 		 * pushing.
1843 		 */
1844 		task = pick_next_pushable_task(rq);
1845 		if (task == next_task) {
1846 			/*
1847 			 * The task hasn't migrated, and is still the next
1848 			 * eligible task, but we failed to find a run-queue
1849 			 * to push it to.  Do not retry in this case, since
1850 			 * other CPUs will pull from us when ready.
1851 			 */
1852 			goto out;
1853 		}
1854 
1855 		if (!task)
1856 			/* No more tasks, just exit */
1857 			goto out;
1858 
1859 		/*
1860 		 * Something has shifted, try again.
1861 		 */
1862 		put_task_struct(next_task);
1863 		next_task = task;
1864 		goto retry;
1865 	}
1866 
1867 	deactivate_task(rq, next_task, 0);
1868 	set_task_cpu(next_task, lowest_rq->cpu);
1869 	activate_task(lowest_rq, next_task, 0);
1870 	ret = 1;
1871 
1872 	resched_curr(lowest_rq);
1873 
1874 	double_unlock_balance(rq, lowest_rq);
1875 
1876 out:
1877 	put_task_struct(next_task);
1878 
1879 	return ret;
1880 }
1881 
1882 static void push_rt_tasks(struct rq *rq)
1883 {
1884 	/* push_rt_task will return true if it moved an RT */
1885 	while (push_rt_task(rq))
1886 		;
1887 }
1888 
1889 #ifdef HAVE_RT_PUSH_IPI
1890 
1891 /*
1892  * When a high priority task schedules out from a CPU and a lower priority
1893  * task is scheduled in, a check is made to see if there's any RT tasks
1894  * on other CPUs that are waiting to run because a higher priority RT task
1895  * is currently running on its CPU. In this case, the CPU with multiple RT
1896  * tasks queued on it (overloaded) needs to be notified that a CPU has opened
1897  * up that may be able to run one of its non-running queued RT tasks.
1898  *
1899  * All CPUs with overloaded RT tasks need to be notified as there is currently
1900  * no way to know which of these CPUs have the highest priority task waiting
1901  * to run. Instead of trying to take a spinlock on each of these CPUs,
1902  * which has shown to cause large latency when done on machines with many
1903  * CPUs, sending an IPI to the CPUs to have them push off the overloaded
1904  * RT tasks waiting to run.
1905  *
1906  * Just sending an IPI to each of the CPUs is also an issue, as on large
1907  * count CPU machines, this can cause an IPI storm on a CPU, especially
1908  * if its the only CPU with multiple RT tasks queued, and a large number
1909  * of CPUs scheduling a lower priority task at the same time.
1910  *
1911  * Each root domain has its own irq work function that can iterate over
1912  * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT
1913  * tassk must be checked if there's one or many CPUs that are lowering
1914  * their priority, there's a single irq work iterator that will try to
1915  * push off RT tasks that are waiting to run.
1916  *
1917  * When a CPU schedules a lower priority task, it will kick off the
1918  * irq work iterator that will jump to each CPU with overloaded RT tasks.
1919  * As it only takes the first CPU that schedules a lower priority task
1920  * to start the process, the rto_start variable is incremented and if
1921  * the atomic result is one, then that CPU will try to take the rto_lock.
1922  * This prevents high contention on the lock as the process handles all
1923  * CPUs scheduling lower priority tasks.
1924  *
1925  * All CPUs that are scheduling a lower priority task will increment the
1926  * rt_loop_next variable. This will make sure that the irq work iterator
1927  * checks all RT overloaded CPUs whenever a CPU schedules a new lower
1928  * priority task, even if the iterator is in the middle of a scan. Incrementing
1929  * the rt_loop_next will cause the iterator to perform another scan.
1930  *
1931  */
1932 static int rto_next_cpu(struct root_domain *rd)
1933 {
1934 	int next;
1935 	int cpu;
1936 
1937 	/*
1938 	 * When starting the IPI RT pushing, the rto_cpu is set to -1,
1939 	 * rt_next_cpu() will simply return the first CPU found in
1940 	 * the rto_mask.
1941 	 *
1942 	 * If rto_next_cpu() is called with rto_cpu is a valid CPU, it
1943 	 * will return the next CPU found in the rto_mask.
1944 	 *
1945 	 * If there are no more CPUs left in the rto_mask, then a check is made
1946 	 * against rto_loop and rto_loop_next. rto_loop is only updated with
1947 	 * the rto_lock held, but any CPU may increment the rto_loop_next
1948 	 * without any locking.
1949 	 */
1950 	for (;;) {
1951 
1952 		/* When rto_cpu is -1 this acts like cpumask_first() */
1953 		cpu = cpumask_next(rd->rto_cpu, rd->rto_mask);
1954 
1955 		rd->rto_cpu = cpu;
1956 
1957 		if (cpu < nr_cpu_ids)
1958 			return cpu;
1959 
1960 		rd->rto_cpu = -1;
1961 
1962 		/*
1963 		 * ACQUIRE ensures we see the @rto_mask changes
1964 		 * made prior to the @next value observed.
1965 		 *
1966 		 * Matches WMB in rt_set_overload().
1967 		 */
1968 		next = atomic_read_acquire(&rd->rto_loop_next);
1969 
1970 		if (rd->rto_loop == next)
1971 			break;
1972 
1973 		rd->rto_loop = next;
1974 	}
1975 
1976 	return -1;
1977 }
1978 
1979 static inline bool rto_start_trylock(atomic_t *v)
1980 {
1981 	return !atomic_cmpxchg_acquire(v, 0, 1);
1982 }
1983 
1984 static inline void rto_start_unlock(atomic_t *v)
1985 {
1986 	atomic_set_release(v, 0);
1987 }
1988 
1989 static void tell_cpu_to_push(struct rq *rq)
1990 {
1991 	int cpu = -1;
1992 
1993 	/* Keep the loop going if the IPI is currently active */
1994 	atomic_inc(&rq->rd->rto_loop_next);
1995 
1996 	/* Only one CPU can initiate a loop at a time */
1997 	if (!rto_start_trylock(&rq->rd->rto_loop_start))
1998 		return;
1999 
2000 	raw_spin_lock(&rq->rd->rto_lock);
2001 
2002 	/*
2003 	 * The rto_cpu is updated under the lock, if it has a valid CPU
2004 	 * then the IPI is still running and will continue due to the
2005 	 * update to loop_next, and nothing needs to be done here.
2006 	 * Otherwise it is finishing up and an ipi needs to be sent.
2007 	 */
2008 	if (rq->rd->rto_cpu < 0)
2009 		cpu = rto_next_cpu(rq->rd);
2010 
2011 	raw_spin_unlock(&rq->rd->rto_lock);
2012 
2013 	rto_start_unlock(&rq->rd->rto_loop_start);
2014 
2015 	if (cpu >= 0) {
2016 		/* Make sure the rd does not get freed while pushing */
2017 		sched_get_rd(rq->rd);
2018 		irq_work_queue_on(&rq->rd->rto_push_work, cpu);
2019 	}
2020 }
2021 
2022 /* Called from hardirq context */
2023 void rto_push_irq_work_func(struct irq_work *work)
2024 {
2025 	struct root_domain *rd =
2026 		container_of(work, struct root_domain, rto_push_work);
2027 	struct rq *rq;
2028 	int cpu;
2029 
2030 	rq = this_rq();
2031 
2032 	/*
2033 	 * We do not need to grab the lock to check for has_pushable_tasks.
2034 	 * When it gets updated, a check is made if a push is possible.
2035 	 */
2036 	if (has_pushable_tasks(rq)) {
2037 		raw_spin_lock(&rq->lock);
2038 		push_rt_tasks(rq);
2039 		raw_spin_unlock(&rq->lock);
2040 	}
2041 
2042 	raw_spin_lock(&rd->rto_lock);
2043 
2044 	/* Pass the IPI to the next rt overloaded queue */
2045 	cpu = rto_next_cpu(rd);
2046 
2047 	raw_spin_unlock(&rd->rto_lock);
2048 
2049 	if (cpu < 0) {
2050 		sched_put_rd(rd);
2051 		return;
2052 	}
2053 
2054 	/* Try the next RT overloaded CPU */
2055 	irq_work_queue_on(&rd->rto_push_work, cpu);
2056 }
2057 #endif /* HAVE_RT_PUSH_IPI */
2058 
2059 static void pull_rt_task(struct rq *this_rq)
2060 {
2061 	int this_cpu = this_rq->cpu, cpu;
2062 	bool resched = false;
2063 	struct task_struct *p;
2064 	struct rq *src_rq;
2065 	int rt_overload_count = rt_overloaded(this_rq);
2066 
2067 	if (likely(!rt_overload_count))
2068 		return;
2069 
2070 	/*
2071 	 * Match the barrier from rt_set_overloaded; this guarantees that if we
2072 	 * see overloaded we must also see the rto_mask bit.
2073 	 */
2074 	smp_rmb();
2075 
2076 	/* If we are the only overloaded CPU do nothing */
2077 	if (rt_overload_count == 1 &&
2078 	    cpumask_test_cpu(this_rq->cpu, this_rq->rd->rto_mask))
2079 		return;
2080 
2081 #ifdef HAVE_RT_PUSH_IPI
2082 	if (sched_feat(RT_PUSH_IPI)) {
2083 		tell_cpu_to_push(this_rq);
2084 		return;
2085 	}
2086 #endif
2087 
2088 	for_each_cpu(cpu, this_rq->rd->rto_mask) {
2089 		if (this_cpu == cpu)
2090 			continue;
2091 
2092 		src_rq = cpu_rq(cpu);
2093 
2094 		/*
2095 		 * Don't bother taking the src_rq->lock if the next highest
2096 		 * task is known to be lower-priority than our current task.
2097 		 * This may look racy, but if this value is about to go
2098 		 * logically higher, the src_rq will push this task away.
2099 		 * And if its going logically lower, we do not care
2100 		 */
2101 		if (src_rq->rt.highest_prio.next >=
2102 		    this_rq->rt.highest_prio.curr)
2103 			continue;
2104 
2105 		/*
2106 		 * We can potentially drop this_rq's lock in
2107 		 * double_lock_balance, and another CPU could
2108 		 * alter this_rq
2109 		 */
2110 		double_lock_balance(this_rq, src_rq);
2111 
2112 		/*
2113 		 * We can pull only a task, which is pushable
2114 		 * on its rq, and no others.
2115 		 */
2116 		p = pick_highest_pushable_task(src_rq, this_cpu);
2117 
2118 		/*
2119 		 * Do we have an RT task that preempts
2120 		 * the to-be-scheduled task?
2121 		 */
2122 		if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
2123 			WARN_ON(p == src_rq->curr);
2124 			WARN_ON(!task_on_rq_queued(p));
2125 
2126 			/*
2127 			 * There's a chance that p is higher in priority
2128 			 * than what's currently running on its CPU.
2129 			 * This is just that p is wakeing up and hasn't
2130 			 * had a chance to schedule. We only pull
2131 			 * p if it is lower in priority than the
2132 			 * current task on the run queue
2133 			 */
2134 			if (p->prio < src_rq->curr->prio)
2135 				goto skip;
2136 
2137 			resched = true;
2138 
2139 			deactivate_task(src_rq, p, 0);
2140 			set_task_cpu(p, this_cpu);
2141 			activate_task(this_rq, p, 0);
2142 			/*
2143 			 * We continue with the search, just in
2144 			 * case there's an even higher prio task
2145 			 * in another runqueue. (low likelihood
2146 			 * but possible)
2147 			 */
2148 		}
2149 skip:
2150 		double_unlock_balance(this_rq, src_rq);
2151 	}
2152 
2153 	if (resched)
2154 		resched_curr(this_rq);
2155 }
2156 
2157 /*
2158  * If we are not running and we are not going to reschedule soon, we should
2159  * try to push tasks away now
2160  */
2161 static void task_woken_rt(struct rq *rq, struct task_struct *p)
2162 {
2163 	if (!task_running(rq, p) &&
2164 	    !test_tsk_need_resched(rq->curr) &&
2165 	    p->nr_cpus_allowed > 1 &&
2166 	    (dl_task(rq->curr) || rt_task(rq->curr)) &&
2167 	    (rq->curr->nr_cpus_allowed < 2 ||
2168 	     rq->curr->prio <= p->prio))
2169 		push_rt_tasks(rq);
2170 }
2171 
2172 /* Assumes rq->lock is held */
2173 static void rq_online_rt(struct rq *rq)
2174 {
2175 	if (rq->rt.overloaded)
2176 		rt_set_overload(rq);
2177 
2178 	__enable_runtime(rq);
2179 
2180 	cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
2181 }
2182 
2183 /* Assumes rq->lock is held */
2184 static void rq_offline_rt(struct rq *rq)
2185 {
2186 	if (rq->rt.overloaded)
2187 		rt_clear_overload(rq);
2188 
2189 	__disable_runtime(rq);
2190 
2191 	cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
2192 }
2193 
2194 /*
2195  * When switch from the rt queue, we bring ourselves to a position
2196  * that we might want to pull RT tasks from other runqueues.
2197  */
2198 static void switched_from_rt(struct rq *rq, struct task_struct *p)
2199 {
2200 	/*
2201 	 * If there are other RT tasks then we will reschedule
2202 	 * and the scheduling of the other RT tasks will handle
2203 	 * the balancing. But if we are the last RT task
2204 	 * we may need to handle the pulling of RT tasks
2205 	 * now.
2206 	 */
2207 	if (!task_on_rq_queued(p) || rq->rt.rt_nr_running)
2208 		return;
2209 
2210 	rt_queue_pull_task(rq);
2211 }
2212 
2213 void __init init_sched_rt_class(void)
2214 {
2215 	unsigned int i;
2216 
2217 	for_each_possible_cpu(i) {
2218 		zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
2219 					GFP_KERNEL, cpu_to_node(i));
2220 	}
2221 }
2222 #endif /* CONFIG_SMP */
2223 
2224 /*
2225  * When switching a task to RT, we may overload the runqueue
2226  * with RT tasks. In this case we try to push them off to
2227  * other runqueues.
2228  */
2229 static void switched_to_rt(struct rq *rq, struct task_struct *p)
2230 {
2231 	/*
2232 	 * If we are already running, then there's nothing
2233 	 * that needs to be done. But if we are not running
2234 	 * we may need to preempt the current running task.
2235 	 * If that current running task is also an RT task
2236 	 * then see if we can move to another run queue.
2237 	 */
2238 	if (task_on_rq_queued(p) && rq->curr != p) {
2239 #ifdef CONFIG_SMP
2240 		if (p->nr_cpus_allowed > 1 && rq->rt.overloaded)
2241 			rt_queue_push_tasks(rq);
2242 #endif /* CONFIG_SMP */
2243 		if (p->prio < rq->curr->prio && cpu_online(cpu_of(rq)))
2244 			resched_curr(rq);
2245 	}
2246 }
2247 
2248 /*
2249  * Priority of the task has changed. This may cause
2250  * us to initiate a push or pull.
2251  */
2252 static void
2253 prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio)
2254 {
2255 	if (!task_on_rq_queued(p))
2256 		return;
2257 
2258 	if (rq->curr == p) {
2259 #ifdef CONFIG_SMP
2260 		/*
2261 		 * If our priority decreases while running, we
2262 		 * may need to pull tasks to this runqueue.
2263 		 */
2264 		if (oldprio < p->prio)
2265 			rt_queue_pull_task(rq);
2266 
2267 		/*
2268 		 * If there's a higher priority task waiting to run
2269 		 * then reschedule.
2270 		 */
2271 		if (p->prio > rq->rt.highest_prio.curr)
2272 			resched_curr(rq);
2273 #else
2274 		/* For UP simply resched on drop of prio */
2275 		if (oldprio < p->prio)
2276 			resched_curr(rq);
2277 #endif /* CONFIG_SMP */
2278 	} else {
2279 		/*
2280 		 * This task is not running, but if it is
2281 		 * greater than the current running task
2282 		 * then reschedule.
2283 		 */
2284 		if (p->prio < rq->curr->prio)
2285 			resched_curr(rq);
2286 	}
2287 }
2288 
2289 #ifdef CONFIG_POSIX_TIMERS
2290 static void watchdog(struct rq *rq, struct task_struct *p)
2291 {
2292 	unsigned long soft, hard;
2293 
2294 	/* max may change after cur was read, this will be fixed next tick */
2295 	soft = task_rlimit(p, RLIMIT_RTTIME);
2296 	hard = task_rlimit_max(p, RLIMIT_RTTIME);
2297 
2298 	if (soft != RLIM_INFINITY) {
2299 		unsigned long next;
2300 
2301 		if (p->rt.watchdog_stamp != jiffies) {
2302 			p->rt.timeout++;
2303 			p->rt.watchdog_stamp = jiffies;
2304 		}
2305 
2306 		next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
2307 		if (p->rt.timeout > next)
2308 			p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
2309 	}
2310 }
2311 #else
2312 static inline void watchdog(struct rq *rq, struct task_struct *p) { }
2313 #endif
2314 
2315 /*
2316  * scheduler tick hitting a task of our scheduling class.
2317  *
2318  * NOTE: This function can be called remotely by the tick offload that
2319  * goes along full dynticks. Therefore no local assumption can be made
2320  * and everything must be accessed through the @rq and @curr passed in
2321  * parameters.
2322  */
2323 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
2324 {
2325 	struct sched_rt_entity *rt_se = &p->rt;
2326 
2327 	update_curr_rt(rq);
2328 	update_rt_rq_load_avg(rq_clock_task(rq), rq, 1);
2329 
2330 	watchdog(rq, p);
2331 
2332 	/*
2333 	 * RR tasks need a special form of timeslice management.
2334 	 * FIFO tasks have no timeslices.
2335 	 */
2336 	if (p->policy != SCHED_RR)
2337 		return;
2338 
2339 	if (--p->rt.time_slice)
2340 		return;
2341 
2342 	p->rt.time_slice = sched_rr_timeslice;
2343 
2344 	/*
2345 	 * Requeue to the end of queue if we (and all of our ancestors) are not
2346 	 * the only element on the queue
2347 	 */
2348 	for_each_sched_rt_entity(rt_se) {
2349 		if (rt_se->run_list.prev != rt_se->run_list.next) {
2350 			requeue_task_rt(rq, p, 0);
2351 			resched_curr(rq);
2352 			return;
2353 		}
2354 	}
2355 }
2356 
2357 static void set_curr_task_rt(struct rq *rq)
2358 {
2359 	set_next_task(rq, rq->curr);
2360 }
2361 
2362 static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
2363 {
2364 	/*
2365 	 * Time slice is 0 for SCHED_FIFO tasks
2366 	 */
2367 	if (task->policy == SCHED_RR)
2368 		return sched_rr_timeslice;
2369 	else
2370 		return 0;
2371 }
2372 
2373 const struct sched_class rt_sched_class = {
2374 	.next			= &fair_sched_class,
2375 	.enqueue_task		= enqueue_task_rt,
2376 	.dequeue_task		= dequeue_task_rt,
2377 	.yield_task		= yield_task_rt,
2378 
2379 	.check_preempt_curr	= check_preempt_curr_rt,
2380 
2381 	.pick_next_task		= pick_next_task_rt,
2382 	.put_prev_task		= put_prev_task_rt,
2383 
2384 #ifdef CONFIG_SMP
2385 	.select_task_rq		= select_task_rq_rt,
2386 
2387 	.set_cpus_allowed       = set_cpus_allowed_common,
2388 	.rq_online              = rq_online_rt,
2389 	.rq_offline             = rq_offline_rt,
2390 	.task_woken		= task_woken_rt,
2391 	.switched_from		= switched_from_rt,
2392 #endif
2393 
2394 	.set_curr_task          = set_curr_task_rt,
2395 	.task_tick		= task_tick_rt,
2396 
2397 	.get_rr_interval	= get_rr_interval_rt,
2398 
2399 	.prio_changed		= prio_changed_rt,
2400 	.switched_to		= switched_to_rt,
2401 
2402 	.update_curr		= update_curr_rt,
2403 };
2404 
2405 #ifdef CONFIG_RT_GROUP_SCHED
2406 /*
2407  * Ensure that the real time constraints are schedulable.
2408  */
2409 static DEFINE_MUTEX(rt_constraints_mutex);
2410 
2411 /* Must be called with tasklist_lock held */
2412 static inline int tg_has_rt_tasks(struct task_group *tg)
2413 {
2414 	struct task_struct *g, *p;
2415 
2416 	/*
2417 	 * Autogroups do not have RT tasks; see autogroup_create().
2418 	 */
2419 	if (task_group_is_autogroup(tg))
2420 		return 0;
2421 
2422 	for_each_process_thread(g, p) {
2423 		if (rt_task(p) && task_group(p) == tg)
2424 			return 1;
2425 	}
2426 
2427 	return 0;
2428 }
2429 
2430 struct rt_schedulable_data {
2431 	struct task_group *tg;
2432 	u64 rt_period;
2433 	u64 rt_runtime;
2434 };
2435 
2436 static int tg_rt_schedulable(struct task_group *tg, void *data)
2437 {
2438 	struct rt_schedulable_data *d = data;
2439 	struct task_group *child;
2440 	unsigned long total, sum = 0;
2441 	u64 period, runtime;
2442 
2443 	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
2444 	runtime = tg->rt_bandwidth.rt_runtime;
2445 
2446 	if (tg == d->tg) {
2447 		period = d->rt_period;
2448 		runtime = d->rt_runtime;
2449 	}
2450 
2451 	/*
2452 	 * Cannot have more runtime than the period.
2453 	 */
2454 	if (runtime > period && runtime != RUNTIME_INF)
2455 		return -EINVAL;
2456 
2457 	/*
2458 	 * Ensure we don't starve existing RT tasks.
2459 	 */
2460 	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
2461 		return -EBUSY;
2462 
2463 	total = to_ratio(period, runtime);
2464 
2465 	/*
2466 	 * Nobody can have more than the global setting allows.
2467 	 */
2468 	if (total > to_ratio(global_rt_period(), global_rt_runtime()))
2469 		return -EINVAL;
2470 
2471 	/*
2472 	 * The sum of our children's runtime should not exceed our own.
2473 	 */
2474 	list_for_each_entry_rcu(child, &tg->children, siblings) {
2475 		period = ktime_to_ns(child->rt_bandwidth.rt_period);
2476 		runtime = child->rt_bandwidth.rt_runtime;
2477 
2478 		if (child == d->tg) {
2479 			period = d->rt_period;
2480 			runtime = d->rt_runtime;
2481 		}
2482 
2483 		sum += to_ratio(period, runtime);
2484 	}
2485 
2486 	if (sum > total)
2487 		return -EINVAL;
2488 
2489 	return 0;
2490 }
2491 
2492 static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
2493 {
2494 	int ret;
2495 
2496 	struct rt_schedulable_data data = {
2497 		.tg = tg,
2498 		.rt_period = period,
2499 		.rt_runtime = runtime,
2500 	};
2501 
2502 	rcu_read_lock();
2503 	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
2504 	rcu_read_unlock();
2505 
2506 	return ret;
2507 }
2508 
2509 static int tg_set_rt_bandwidth(struct task_group *tg,
2510 		u64 rt_period, u64 rt_runtime)
2511 {
2512 	int i, err = 0;
2513 
2514 	/*
2515 	 * Disallowing the root group RT runtime is BAD, it would disallow the
2516 	 * kernel creating (and or operating) RT threads.
2517 	 */
2518 	if (tg == &root_task_group && rt_runtime == 0)
2519 		return -EINVAL;
2520 
2521 	/* No period doesn't make any sense. */
2522 	if (rt_period == 0)
2523 		return -EINVAL;
2524 
2525 	mutex_lock(&rt_constraints_mutex);
2526 	read_lock(&tasklist_lock);
2527 	err = __rt_schedulable(tg, rt_period, rt_runtime);
2528 	if (err)
2529 		goto unlock;
2530 
2531 	raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
2532 	tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
2533 	tg->rt_bandwidth.rt_runtime = rt_runtime;
2534 
2535 	for_each_possible_cpu(i) {
2536 		struct rt_rq *rt_rq = tg->rt_rq[i];
2537 
2538 		raw_spin_lock(&rt_rq->rt_runtime_lock);
2539 		rt_rq->rt_runtime = rt_runtime;
2540 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
2541 	}
2542 	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
2543 unlock:
2544 	read_unlock(&tasklist_lock);
2545 	mutex_unlock(&rt_constraints_mutex);
2546 
2547 	return err;
2548 }
2549 
2550 int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
2551 {
2552 	u64 rt_runtime, rt_period;
2553 
2554 	rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
2555 	rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
2556 	if (rt_runtime_us < 0)
2557 		rt_runtime = RUNTIME_INF;
2558 
2559 	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2560 }
2561 
2562 long sched_group_rt_runtime(struct task_group *tg)
2563 {
2564 	u64 rt_runtime_us;
2565 
2566 	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
2567 		return -1;
2568 
2569 	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
2570 	do_div(rt_runtime_us, NSEC_PER_USEC);
2571 	return rt_runtime_us;
2572 }
2573 
2574 int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
2575 {
2576 	u64 rt_runtime, rt_period;
2577 
2578 	rt_period = rt_period_us * NSEC_PER_USEC;
2579 	rt_runtime = tg->rt_bandwidth.rt_runtime;
2580 
2581 	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
2582 }
2583 
2584 long sched_group_rt_period(struct task_group *tg)
2585 {
2586 	u64 rt_period_us;
2587 
2588 	rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
2589 	do_div(rt_period_us, NSEC_PER_USEC);
2590 	return rt_period_us;
2591 }
2592 
2593 static int sched_rt_global_constraints(void)
2594 {
2595 	int ret = 0;
2596 
2597 	mutex_lock(&rt_constraints_mutex);
2598 	read_lock(&tasklist_lock);
2599 	ret = __rt_schedulable(NULL, 0, 0);
2600 	read_unlock(&tasklist_lock);
2601 	mutex_unlock(&rt_constraints_mutex);
2602 
2603 	return ret;
2604 }
2605 
2606 int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
2607 {
2608 	/* Don't accept realtime tasks when there is no way for them to run */
2609 	if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
2610 		return 0;
2611 
2612 	return 1;
2613 }
2614 
2615 #else /* !CONFIG_RT_GROUP_SCHED */
2616 static int sched_rt_global_constraints(void)
2617 {
2618 	unsigned long flags;
2619 	int i;
2620 
2621 	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
2622 	for_each_possible_cpu(i) {
2623 		struct rt_rq *rt_rq = &cpu_rq(i)->rt;
2624 
2625 		raw_spin_lock(&rt_rq->rt_runtime_lock);
2626 		rt_rq->rt_runtime = global_rt_runtime();
2627 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
2628 	}
2629 	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
2630 
2631 	return 0;
2632 }
2633 #endif /* CONFIG_RT_GROUP_SCHED */
2634 
2635 static int sched_rt_global_validate(void)
2636 {
2637 	if (sysctl_sched_rt_period <= 0)
2638 		return -EINVAL;
2639 
2640 	if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
2641 		(sysctl_sched_rt_runtime > sysctl_sched_rt_period))
2642 		return -EINVAL;
2643 
2644 	return 0;
2645 }
2646 
2647 static void sched_rt_do_global(void)
2648 {
2649 	def_rt_bandwidth.rt_runtime = global_rt_runtime();
2650 	def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period());
2651 }
2652 
2653 int sched_rt_handler(struct ctl_table *table, int write,
2654 		void __user *buffer, size_t *lenp,
2655 		loff_t *ppos)
2656 {
2657 	int old_period, old_runtime;
2658 	static DEFINE_MUTEX(mutex);
2659 	int ret;
2660 
2661 	mutex_lock(&mutex);
2662 	old_period = sysctl_sched_rt_period;
2663 	old_runtime = sysctl_sched_rt_runtime;
2664 
2665 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
2666 
2667 	if (!ret && write) {
2668 		ret = sched_rt_global_validate();
2669 		if (ret)
2670 			goto undo;
2671 
2672 		ret = sched_dl_global_validate();
2673 		if (ret)
2674 			goto undo;
2675 
2676 		ret = sched_rt_global_constraints();
2677 		if (ret)
2678 			goto undo;
2679 
2680 		sched_rt_do_global();
2681 		sched_dl_do_global();
2682 	}
2683 	if (0) {
2684 undo:
2685 		sysctl_sched_rt_period = old_period;
2686 		sysctl_sched_rt_runtime = old_runtime;
2687 	}
2688 	mutex_unlock(&mutex);
2689 
2690 	return ret;
2691 }
2692 
2693 int sched_rr_handler(struct ctl_table *table, int write,
2694 		void __user *buffer, size_t *lenp,
2695 		loff_t *ppos)
2696 {
2697 	int ret;
2698 	static DEFINE_MUTEX(mutex);
2699 
2700 	mutex_lock(&mutex);
2701 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
2702 	/*
2703 	 * Make sure that internally we keep jiffies.
2704 	 * Also, writing zero resets the timeslice to default:
2705 	 */
2706 	if (!ret && write) {
2707 		sched_rr_timeslice =
2708 			sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE :
2709 			msecs_to_jiffies(sysctl_sched_rr_timeslice);
2710 	}
2711 	mutex_unlock(&mutex);
2712 
2713 	return ret;
2714 }
2715 
2716 #ifdef CONFIG_SCHED_DEBUG
2717 void print_rt_stats(struct seq_file *m, int cpu)
2718 {
2719 	rt_rq_iter_t iter;
2720 	struct rt_rq *rt_rq;
2721 
2722 	rcu_read_lock();
2723 	for_each_rt_rq(rt_rq, iter, cpu_rq(cpu))
2724 		print_rt_rq(m, cpu, rt_rq);
2725 	rcu_read_unlock();
2726 }
2727 #endif /* CONFIG_SCHED_DEBUG */
2728