xref: /openbmc/linux/kernel/sched/rt.c (revision fd7a4bed183523275279c9addbf42fce550c2e90)
1 /*
2  * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
3  * policies)
4  */
5 
6 #include "sched.h"
7 
8 #include <linux/slab.h>
9 #include <linux/irq_work.h>
10 
11 int sched_rr_timeslice = 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 		hrtimer_forward_now(&rt_b->rt_period_timer, rt_b->rt_period);
62 		hrtimer_start_expires(&rt_b->rt_period_timer, HRTIMER_MODE_ABS_PINNED);
63 	}
64 	raw_spin_unlock(&rt_b->rt_runtime_lock);
65 }
66 
67 #ifdef CONFIG_SMP
68 static void push_irq_work_func(struct irq_work *work);
69 #endif
70 
71 void init_rt_rq(struct rt_rq *rt_rq)
72 {
73 	struct rt_prio_array *array;
74 	int i;
75 
76 	array = &rt_rq->active;
77 	for (i = 0; i < MAX_RT_PRIO; i++) {
78 		INIT_LIST_HEAD(array->queue + i);
79 		__clear_bit(i, array->bitmap);
80 	}
81 	/* delimiter for bitsearch: */
82 	__set_bit(MAX_RT_PRIO, array->bitmap);
83 
84 #if defined CONFIG_SMP
85 	rt_rq->highest_prio.curr = MAX_RT_PRIO;
86 	rt_rq->highest_prio.next = MAX_RT_PRIO;
87 	rt_rq->rt_nr_migratory = 0;
88 	rt_rq->overloaded = 0;
89 	plist_head_init(&rt_rq->pushable_tasks);
90 
91 #ifdef HAVE_RT_PUSH_IPI
92 	rt_rq->push_flags = 0;
93 	rt_rq->push_cpu = nr_cpu_ids;
94 	raw_spin_lock_init(&rt_rq->push_lock);
95 	init_irq_work(&rt_rq->push_work, push_irq_work_func);
96 #endif
97 #endif /* CONFIG_SMP */
98 	/* We start is dequeued state, because no RT tasks are queued */
99 	rt_rq->rt_queued = 0;
100 
101 	rt_rq->rt_time = 0;
102 	rt_rq->rt_throttled = 0;
103 	rt_rq->rt_runtime = 0;
104 	raw_spin_lock_init(&rt_rq->rt_runtime_lock);
105 }
106 
107 #ifdef CONFIG_RT_GROUP_SCHED
108 static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
109 {
110 	hrtimer_cancel(&rt_b->rt_period_timer);
111 }
112 
113 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
114 
115 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
116 {
117 #ifdef CONFIG_SCHED_DEBUG
118 	WARN_ON_ONCE(!rt_entity_is_task(rt_se));
119 #endif
120 	return container_of(rt_se, struct task_struct, rt);
121 }
122 
123 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
124 {
125 	return rt_rq->rq;
126 }
127 
128 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
129 {
130 	return rt_se->rt_rq;
131 }
132 
133 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
134 {
135 	struct rt_rq *rt_rq = rt_se->rt_rq;
136 
137 	return rt_rq->rq;
138 }
139 
140 void free_rt_sched_group(struct task_group *tg)
141 {
142 	int i;
143 
144 	if (tg->rt_se)
145 		destroy_rt_bandwidth(&tg->rt_bandwidth);
146 
147 	for_each_possible_cpu(i) {
148 		if (tg->rt_rq)
149 			kfree(tg->rt_rq[i]);
150 		if (tg->rt_se)
151 			kfree(tg->rt_se[i]);
152 	}
153 
154 	kfree(tg->rt_rq);
155 	kfree(tg->rt_se);
156 }
157 
158 void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
159 		struct sched_rt_entity *rt_se, int cpu,
160 		struct sched_rt_entity *parent)
161 {
162 	struct rq *rq = cpu_rq(cpu);
163 
164 	rt_rq->highest_prio.curr = MAX_RT_PRIO;
165 	rt_rq->rt_nr_boosted = 0;
166 	rt_rq->rq = rq;
167 	rt_rq->tg = tg;
168 
169 	tg->rt_rq[cpu] = rt_rq;
170 	tg->rt_se[cpu] = rt_se;
171 
172 	if (!rt_se)
173 		return;
174 
175 	if (!parent)
176 		rt_se->rt_rq = &rq->rt;
177 	else
178 		rt_se->rt_rq = parent->my_q;
179 
180 	rt_se->my_q = rt_rq;
181 	rt_se->parent = parent;
182 	INIT_LIST_HEAD(&rt_se->run_list);
183 }
184 
185 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
186 {
187 	struct rt_rq *rt_rq;
188 	struct sched_rt_entity *rt_se;
189 	int i;
190 
191 	tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
192 	if (!tg->rt_rq)
193 		goto err;
194 	tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
195 	if (!tg->rt_se)
196 		goto err;
197 
198 	init_rt_bandwidth(&tg->rt_bandwidth,
199 			ktime_to_ns(def_rt_bandwidth.rt_period), 0);
200 
201 	for_each_possible_cpu(i) {
202 		rt_rq = kzalloc_node(sizeof(struct rt_rq),
203 				     GFP_KERNEL, cpu_to_node(i));
204 		if (!rt_rq)
205 			goto err;
206 
207 		rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
208 				     GFP_KERNEL, cpu_to_node(i));
209 		if (!rt_se)
210 			goto err_free_rq;
211 
212 		init_rt_rq(rt_rq);
213 		rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
214 		init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
215 	}
216 
217 	return 1;
218 
219 err_free_rq:
220 	kfree(rt_rq);
221 err:
222 	return 0;
223 }
224 
225 #else /* CONFIG_RT_GROUP_SCHED */
226 
227 #define rt_entity_is_task(rt_se) (1)
228 
229 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
230 {
231 	return container_of(rt_se, struct task_struct, rt);
232 }
233 
234 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
235 {
236 	return container_of(rt_rq, struct rq, rt);
237 }
238 
239 static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se)
240 {
241 	struct task_struct *p = rt_task_of(rt_se);
242 
243 	return task_rq(p);
244 }
245 
246 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
247 {
248 	struct rq *rq = rq_of_rt_se(rt_se);
249 
250 	return &rq->rt;
251 }
252 
253 void free_rt_sched_group(struct task_group *tg) { }
254 
255 int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
256 {
257 	return 1;
258 }
259 #endif /* CONFIG_RT_GROUP_SCHED */
260 
261 #ifdef CONFIG_SMP
262 
263 static void pull_rt_task(struct rq *this_rq);
264 
265 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
266 {
267 	/* Try to pull RT tasks here if we lower this rq's prio */
268 	return rq->rt.highest_prio.curr > prev->prio;
269 }
270 
271 static inline int rt_overloaded(struct rq *rq)
272 {
273 	return atomic_read(&rq->rd->rto_count);
274 }
275 
276 static inline void rt_set_overload(struct rq *rq)
277 {
278 	if (!rq->online)
279 		return;
280 
281 	cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
282 	/*
283 	 * Make sure the mask is visible before we set
284 	 * the overload count. That is checked to determine
285 	 * if we should look at the mask. It would be a shame
286 	 * if we looked at the mask, but the mask was not
287 	 * updated yet.
288 	 *
289 	 * Matched by the barrier in pull_rt_task().
290 	 */
291 	smp_wmb();
292 	atomic_inc(&rq->rd->rto_count);
293 }
294 
295 static inline void rt_clear_overload(struct rq *rq)
296 {
297 	if (!rq->online)
298 		return;
299 
300 	/* the order here really doesn't matter */
301 	atomic_dec(&rq->rd->rto_count);
302 	cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
303 }
304 
305 static void update_rt_migration(struct rt_rq *rt_rq)
306 {
307 	if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
308 		if (!rt_rq->overloaded) {
309 			rt_set_overload(rq_of_rt_rq(rt_rq));
310 			rt_rq->overloaded = 1;
311 		}
312 	} else if (rt_rq->overloaded) {
313 		rt_clear_overload(rq_of_rt_rq(rt_rq));
314 		rt_rq->overloaded = 0;
315 	}
316 }
317 
318 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
319 {
320 	struct task_struct *p;
321 
322 	if (!rt_entity_is_task(rt_se))
323 		return;
324 
325 	p = rt_task_of(rt_se);
326 	rt_rq = &rq_of_rt_rq(rt_rq)->rt;
327 
328 	rt_rq->rt_nr_total++;
329 	if (p->nr_cpus_allowed > 1)
330 		rt_rq->rt_nr_migratory++;
331 
332 	update_rt_migration(rt_rq);
333 }
334 
335 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
336 {
337 	struct task_struct *p;
338 
339 	if (!rt_entity_is_task(rt_se))
340 		return;
341 
342 	p = rt_task_of(rt_se);
343 	rt_rq = &rq_of_rt_rq(rt_rq)->rt;
344 
345 	rt_rq->rt_nr_total--;
346 	if (p->nr_cpus_allowed > 1)
347 		rt_rq->rt_nr_migratory--;
348 
349 	update_rt_migration(rt_rq);
350 }
351 
352 static inline int has_pushable_tasks(struct rq *rq)
353 {
354 	return !plist_head_empty(&rq->rt.pushable_tasks);
355 }
356 
357 static DEFINE_PER_CPU(struct callback_head, rt_push_head);
358 static DEFINE_PER_CPU(struct callback_head, rt_pull_head);
359 
360 static void push_rt_tasks(struct rq *);
361 static void pull_rt_task(struct rq *);
362 
363 static inline void queue_push_tasks(struct rq *rq)
364 {
365 	if (!has_pushable_tasks(rq))
366 		return;
367 
368 	queue_balance_callback(rq, &per_cpu(rt_push_head, rq->cpu), push_rt_tasks);
369 }
370 
371 static inline void queue_pull_task(struct rq *rq)
372 {
373 	queue_balance_callback(rq, &per_cpu(rt_pull_head, rq->cpu), pull_rt_task);
374 }
375 
376 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
377 {
378 	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
379 	plist_node_init(&p->pushable_tasks, p->prio);
380 	plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
381 
382 	/* Update the highest prio pushable task */
383 	if (p->prio < rq->rt.highest_prio.next)
384 		rq->rt.highest_prio.next = p->prio;
385 }
386 
387 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
388 {
389 	plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
390 
391 	/* Update the new highest prio pushable task */
392 	if (has_pushable_tasks(rq)) {
393 		p = plist_first_entry(&rq->rt.pushable_tasks,
394 				      struct task_struct, pushable_tasks);
395 		rq->rt.highest_prio.next = p->prio;
396 	} else
397 		rq->rt.highest_prio.next = MAX_RT_PRIO;
398 }
399 
400 #else
401 
402 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
403 {
404 }
405 
406 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
407 {
408 }
409 
410 static inline
411 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
412 {
413 }
414 
415 static inline
416 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
417 {
418 }
419 
420 static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev)
421 {
422 	return false;
423 }
424 
425 static inline void pull_rt_task(struct rq *this_rq)
426 {
427 }
428 
429 static inline void queue_push_tasks(struct rq *rq)
430 {
431 }
432 #endif /* CONFIG_SMP */
433 
434 static void enqueue_top_rt_rq(struct rt_rq *rt_rq);
435 static void dequeue_top_rt_rq(struct rt_rq *rt_rq);
436 
437 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
438 {
439 	return !list_empty(&rt_se->run_list);
440 }
441 
442 #ifdef CONFIG_RT_GROUP_SCHED
443 
444 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
445 {
446 	if (!rt_rq->tg)
447 		return RUNTIME_INF;
448 
449 	return rt_rq->rt_runtime;
450 }
451 
452 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
453 {
454 	return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
455 }
456 
457 typedef struct task_group *rt_rq_iter_t;
458 
459 static inline struct task_group *next_task_group(struct task_group *tg)
460 {
461 	do {
462 		tg = list_entry_rcu(tg->list.next,
463 			typeof(struct task_group), list);
464 	} while (&tg->list != &task_groups && task_group_is_autogroup(tg));
465 
466 	if (&tg->list == &task_groups)
467 		tg = NULL;
468 
469 	return tg;
470 }
471 
472 #define for_each_rt_rq(rt_rq, iter, rq)					\
473 	for (iter = container_of(&task_groups, typeof(*iter), list);	\
474 		(iter = next_task_group(iter)) &&			\
475 		(rt_rq = iter->rt_rq[cpu_of(rq)]);)
476 
477 #define for_each_sched_rt_entity(rt_se) \
478 	for (; rt_se; rt_se = rt_se->parent)
479 
480 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
481 {
482 	return rt_se->my_q;
483 }
484 
485 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head);
486 static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
487 
488 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
489 {
490 	struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
491 	struct rq *rq = rq_of_rt_rq(rt_rq);
492 	struct sched_rt_entity *rt_se;
493 
494 	int cpu = cpu_of(rq);
495 
496 	rt_se = rt_rq->tg->rt_se[cpu];
497 
498 	if (rt_rq->rt_nr_running) {
499 		if (!rt_se)
500 			enqueue_top_rt_rq(rt_rq);
501 		else if (!on_rt_rq(rt_se))
502 			enqueue_rt_entity(rt_se, false);
503 
504 		if (rt_rq->highest_prio.curr < curr->prio)
505 			resched_curr(rq);
506 	}
507 }
508 
509 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
510 {
511 	struct sched_rt_entity *rt_se;
512 	int cpu = cpu_of(rq_of_rt_rq(rt_rq));
513 
514 	rt_se = rt_rq->tg->rt_se[cpu];
515 
516 	if (!rt_se)
517 		dequeue_top_rt_rq(rt_rq);
518 	else if (on_rt_rq(rt_se))
519 		dequeue_rt_entity(rt_se);
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 int 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, more = 0;
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 			more = 1;
677 			if (rt_rq->rt_runtime == rt_period) {
678 				raw_spin_unlock(&iter->rt_runtime_lock);
679 				break;
680 			}
681 		}
682 next:
683 		raw_spin_unlock(&iter->rt_runtime_lock);
684 	}
685 	raw_spin_unlock(&rt_b->rt_runtime_lock);
686 
687 	return more;
688 }
689 
690 /*
691  * Ensure this RQ takes back all the runtime it lend to its neighbours.
692  */
693 static void __disable_runtime(struct rq *rq)
694 {
695 	struct root_domain *rd = rq->rd;
696 	rt_rq_iter_t iter;
697 	struct rt_rq *rt_rq;
698 
699 	if (unlikely(!scheduler_running))
700 		return;
701 
702 	for_each_rt_rq(rt_rq, iter, rq) {
703 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
704 		s64 want;
705 		int i;
706 
707 		raw_spin_lock(&rt_b->rt_runtime_lock);
708 		raw_spin_lock(&rt_rq->rt_runtime_lock);
709 		/*
710 		 * Either we're all inf and nobody needs to borrow, or we're
711 		 * already disabled and thus have nothing to do, or we have
712 		 * exactly the right amount of runtime to take out.
713 		 */
714 		if (rt_rq->rt_runtime == RUNTIME_INF ||
715 				rt_rq->rt_runtime == rt_b->rt_runtime)
716 			goto balanced;
717 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
718 
719 		/*
720 		 * Calculate the difference between what we started out with
721 		 * and what we current have, that's the amount of runtime
722 		 * we lend and now have to reclaim.
723 		 */
724 		want = rt_b->rt_runtime - rt_rq->rt_runtime;
725 
726 		/*
727 		 * Greedy reclaim, take back as much as we can.
728 		 */
729 		for_each_cpu(i, rd->span) {
730 			struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
731 			s64 diff;
732 
733 			/*
734 			 * Can't reclaim from ourselves or disabled runqueues.
735 			 */
736 			if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
737 				continue;
738 
739 			raw_spin_lock(&iter->rt_runtime_lock);
740 			if (want > 0) {
741 				diff = min_t(s64, iter->rt_runtime, want);
742 				iter->rt_runtime -= diff;
743 				want -= diff;
744 			} else {
745 				iter->rt_runtime -= want;
746 				want -= want;
747 			}
748 			raw_spin_unlock(&iter->rt_runtime_lock);
749 
750 			if (!want)
751 				break;
752 		}
753 
754 		raw_spin_lock(&rt_rq->rt_runtime_lock);
755 		/*
756 		 * We cannot be left wanting - that would mean some runtime
757 		 * leaked out of the system.
758 		 */
759 		BUG_ON(want);
760 balanced:
761 		/*
762 		 * Disable all the borrow logic by pretending we have inf
763 		 * runtime - in which case borrowing doesn't make sense.
764 		 */
765 		rt_rq->rt_runtime = RUNTIME_INF;
766 		rt_rq->rt_throttled = 0;
767 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
768 		raw_spin_unlock(&rt_b->rt_runtime_lock);
769 
770 		/* Make rt_rq available for pick_next_task() */
771 		sched_rt_rq_enqueue(rt_rq);
772 	}
773 }
774 
775 static void __enable_runtime(struct rq *rq)
776 {
777 	rt_rq_iter_t iter;
778 	struct rt_rq *rt_rq;
779 
780 	if (unlikely(!scheduler_running))
781 		return;
782 
783 	/*
784 	 * Reset each runqueue's bandwidth settings
785 	 */
786 	for_each_rt_rq(rt_rq, iter, rq) {
787 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
788 
789 		raw_spin_lock(&rt_b->rt_runtime_lock);
790 		raw_spin_lock(&rt_rq->rt_runtime_lock);
791 		rt_rq->rt_runtime = rt_b->rt_runtime;
792 		rt_rq->rt_time = 0;
793 		rt_rq->rt_throttled = 0;
794 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
795 		raw_spin_unlock(&rt_b->rt_runtime_lock);
796 	}
797 }
798 
799 static int balance_runtime(struct rt_rq *rt_rq)
800 {
801 	int more = 0;
802 
803 	if (!sched_feat(RT_RUNTIME_SHARE))
804 		return more;
805 
806 	if (rt_rq->rt_time > rt_rq->rt_runtime) {
807 		raw_spin_unlock(&rt_rq->rt_runtime_lock);
808 		more = do_balance_runtime(rt_rq);
809 		raw_spin_lock(&rt_rq->rt_runtime_lock);
810 	}
811 
812 	return more;
813 }
814 #else /* !CONFIG_SMP */
815 static inline int balance_runtime(struct rt_rq *rt_rq)
816 {
817 	return 0;
818 }
819 #endif /* CONFIG_SMP */
820 
821 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
822 {
823 	int i, idle = 1, throttled = 0;
824 	const struct cpumask *span;
825 
826 	span = sched_rt_period_mask();
827 #ifdef CONFIG_RT_GROUP_SCHED
828 	/*
829 	 * FIXME: isolated CPUs should really leave the root task group,
830 	 * whether they are isolcpus or were isolated via cpusets, lest
831 	 * the timer run on a CPU which does not service all runqueues,
832 	 * potentially leaving other CPUs indefinitely throttled.  If
833 	 * isolation is really required, the user will turn the throttle
834 	 * off to kill the perturbations it causes anyway.  Meanwhile,
835 	 * this maintains functionality for boot and/or troubleshooting.
836 	 */
837 	if (rt_b == &root_task_group.rt_bandwidth)
838 		span = cpu_online_mask;
839 #endif
840 	for_each_cpu(i, span) {
841 		int enqueue = 0;
842 		struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
843 		struct rq *rq = rq_of_rt_rq(rt_rq);
844 
845 		raw_spin_lock(&rq->lock);
846 		if (rt_rq->rt_time) {
847 			u64 runtime;
848 
849 			raw_spin_lock(&rt_rq->rt_runtime_lock);
850 			if (rt_rq->rt_throttled)
851 				balance_runtime(rt_rq);
852 			runtime = rt_rq->rt_runtime;
853 			rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
854 			if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
855 				rt_rq->rt_throttled = 0;
856 				enqueue = 1;
857 
858 				/*
859 				 * When we're idle and a woken (rt) task is
860 				 * throttled check_preempt_curr() will set
861 				 * skip_update and the time between the wakeup
862 				 * and this unthrottle will get accounted as
863 				 * 'runtime'.
864 				 */
865 				if (rt_rq->rt_nr_running && rq->curr == rq->idle)
866 					rq_clock_skip_update(rq, false);
867 			}
868 			if (rt_rq->rt_time || rt_rq->rt_nr_running)
869 				idle = 0;
870 			raw_spin_unlock(&rt_rq->rt_runtime_lock);
871 		} else if (rt_rq->rt_nr_running) {
872 			idle = 0;
873 			if (!rt_rq_throttled(rt_rq))
874 				enqueue = 1;
875 		}
876 		if (rt_rq->rt_throttled)
877 			throttled = 1;
878 
879 		if (enqueue)
880 			sched_rt_rq_enqueue(rt_rq);
881 		raw_spin_unlock(&rq->lock);
882 	}
883 
884 	if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
885 		return 1;
886 
887 	return idle;
888 }
889 
890 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
891 {
892 #ifdef CONFIG_RT_GROUP_SCHED
893 	struct rt_rq *rt_rq = group_rt_rq(rt_se);
894 
895 	if (rt_rq)
896 		return rt_rq->highest_prio.curr;
897 #endif
898 
899 	return rt_task_of(rt_se)->prio;
900 }
901 
902 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
903 {
904 	u64 runtime = sched_rt_runtime(rt_rq);
905 
906 	if (rt_rq->rt_throttled)
907 		return rt_rq_throttled(rt_rq);
908 
909 	if (runtime >= sched_rt_period(rt_rq))
910 		return 0;
911 
912 	balance_runtime(rt_rq);
913 	runtime = sched_rt_runtime(rt_rq);
914 	if (runtime == RUNTIME_INF)
915 		return 0;
916 
917 	if (rt_rq->rt_time > runtime) {
918 		struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
919 
920 		/*
921 		 * Don't actually throttle groups that have no runtime assigned
922 		 * but accrue some time due to boosting.
923 		 */
924 		if (likely(rt_b->rt_runtime)) {
925 			rt_rq->rt_throttled = 1;
926 			printk_deferred_once("sched: RT throttling activated\n");
927 		} else {
928 			/*
929 			 * In case we did anyway, make it go away,
930 			 * replenishment is a joke, since it will replenish us
931 			 * with exactly 0 ns.
932 			 */
933 			rt_rq->rt_time = 0;
934 		}
935 
936 		if (rt_rq_throttled(rt_rq)) {
937 			sched_rt_rq_dequeue(rt_rq);
938 			return 1;
939 		}
940 	}
941 
942 	return 0;
943 }
944 
945 /*
946  * Update the current task's runtime statistics. Skip current tasks that
947  * are not in our scheduling class.
948  */
949 static void update_curr_rt(struct rq *rq)
950 {
951 	struct task_struct *curr = rq->curr;
952 	struct sched_rt_entity *rt_se = &curr->rt;
953 	u64 delta_exec;
954 
955 	if (curr->sched_class != &rt_sched_class)
956 		return;
957 
958 	delta_exec = rq_clock_task(rq) - 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 = rq_clock_task(rq);
969 	cpuacct_charge(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 
1005 static void
1006 enqueue_top_rt_rq(struct rt_rq *rt_rq)
1007 {
1008 	struct rq *rq = rq_of_rt_rq(rt_rq);
1009 
1010 	BUG_ON(&rq->rt != rt_rq);
1011 
1012 	if (rt_rq->rt_queued)
1013 		return;
1014 	if (rt_rq_throttled(rt_rq) || !rt_rq->rt_nr_running)
1015 		return;
1016 
1017 	add_nr_running(rq, rt_rq->rt_nr_running);
1018 	rt_rq->rt_queued = 1;
1019 }
1020 
1021 #if defined CONFIG_SMP
1022 
1023 static void
1024 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1025 {
1026 	struct rq *rq = rq_of_rt_rq(rt_rq);
1027 
1028 #ifdef CONFIG_RT_GROUP_SCHED
1029 	/*
1030 	 * Change rq's cpupri only if rt_rq is the top queue.
1031 	 */
1032 	if (&rq->rt != rt_rq)
1033 		return;
1034 #endif
1035 	if (rq->online && prio < prev_prio)
1036 		cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
1037 }
1038 
1039 static void
1040 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
1041 {
1042 	struct rq *rq = rq_of_rt_rq(rt_rq);
1043 
1044 #ifdef CONFIG_RT_GROUP_SCHED
1045 	/*
1046 	 * Change rq's cpupri only if rt_rq is the top queue.
1047 	 */
1048 	if (&rq->rt != rt_rq)
1049 		return;
1050 #endif
1051 	if (rq->online && rt_rq->highest_prio.curr != prev_prio)
1052 		cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
1053 }
1054 
1055 #else /* CONFIG_SMP */
1056 
1057 static inline
1058 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1059 static inline
1060 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
1061 
1062 #endif /* CONFIG_SMP */
1063 
1064 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
1065 static void
1066 inc_rt_prio(struct rt_rq *rt_rq, int prio)
1067 {
1068 	int prev_prio = rt_rq->highest_prio.curr;
1069 
1070 	if (prio < prev_prio)
1071 		rt_rq->highest_prio.curr = prio;
1072 
1073 	inc_rt_prio_smp(rt_rq, prio, prev_prio);
1074 }
1075 
1076 static void
1077 dec_rt_prio(struct rt_rq *rt_rq, int prio)
1078 {
1079 	int prev_prio = rt_rq->highest_prio.curr;
1080 
1081 	if (rt_rq->rt_nr_running) {
1082 
1083 		WARN_ON(prio < prev_prio);
1084 
1085 		/*
1086 		 * This may have been our highest task, and therefore
1087 		 * we may have some recomputation to do
1088 		 */
1089 		if (prio == prev_prio) {
1090 			struct rt_prio_array *array = &rt_rq->active;
1091 
1092 			rt_rq->highest_prio.curr =
1093 				sched_find_first_bit(array->bitmap);
1094 		}
1095 
1096 	} else
1097 		rt_rq->highest_prio.curr = MAX_RT_PRIO;
1098 
1099 	dec_rt_prio_smp(rt_rq, prio, prev_prio);
1100 }
1101 
1102 #else
1103 
1104 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
1105 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
1106 
1107 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1108 
1109 #ifdef CONFIG_RT_GROUP_SCHED
1110 
1111 static void
1112 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1113 {
1114 	if (rt_se_boosted(rt_se))
1115 		rt_rq->rt_nr_boosted++;
1116 
1117 	if (rt_rq->tg)
1118 		start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
1119 }
1120 
1121 static void
1122 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1123 {
1124 	if (rt_se_boosted(rt_se))
1125 		rt_rq->rt_nr_boosted--;
1126 
1127 	WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
1128 }
1129 
1130 #else /* CONFIG_RT_GROUP_SCHED */
1131 
1132 static void
1133 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1134 {
1135 	start_rt_bandwidth(&def_rt_bandwidth);
1136 }
1137 
1138 static inline
1139 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
1140 
1141 #endif /* CONFIG_RT_GROUP_SCHED */
1142 
1143 static inline
1144 unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se)
1145 {
1146 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1147 
1148 	if (group_rq)
1149 		return group_rq->rt_nr_running;
1150 	else
1151 		return 1;
1152 }
1153 
1154 static inline
1155 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1156 {
1157 	int prio = rt_se_prio(rt_se);
1158 
1159 	WARN_ON(!rt_prio(prio));
1160 	rt_rq->rt_nr_running += rt_se_nr_running(rt_se);
1161 
1162 	inc_rt_prio(rt_rq, prio);
1163 	inc_rt_migration(rt_se, rt_rq);
1164 	inc_rt_group(rt_se, rt_rq);
1165 }
1166 
1167 static inline
1168 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
1169 {
1170 	WARN_ON(!rt_prio(rt_se_prio(rt_se)));
1171 	WARN_ON(!rt_rq->rt_nr_running);
1172 	rt_rq->rt_nr_running -= rt_se_nr_running(rt_se);
1173 
1174 	dec_rt_prio(rt_rq, rt_se_prio(rt_se));
1175 	dec_rt_migration(rt_se, rt_rq);
1176 	dec_rt_group(rt_se, rt_rq);
1177 }
1178 
1179 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
1180 {
1181 	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1182 	struct rt_prio_array *array = &rt_rq->active;
1183 	struct rt_rq *group_rq = group_rt_rq(rt_se);
1184 	struct list_head *queue = array->queue + rt_se_prio(rt_se);
1185 
1186 	/*
1187 	 * Don't enqueue the group if its throttled, or when empty.
1188 	 * The latter is a consequence of the former when a child group
1189 	 * get throttled and the current group doesn't have any other
1190 	 * active members.
1191 	 */
1192 	if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
1193 		return;
1194 
1195 	if (head)
1196 		list_add(&rt_se->run_list, queue);
1197 	else
1198 		list_add_tail(&rt_se->run_list, queue);
1199 	__set_bit(rt_se_prio(rt_se), array->bitmap);
1200 
1201 	inc_rt_tasks(rt_se, rt_rq);
1202 }
1203 
1204 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
1205 {
1206 	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
1207 	struct rt_prio_array *array = &rt_rq->active;
1208 
1209 	list_del_init(&rt_se->run_list);
1210 	if (list_empty(array->queue + rt_se_prio(rt_se)))
1211 		__clear_bit(rt_se_prio(rt_se), array->bitmap);
1212 
1213 	dec_rt_tasks(rt_se, rt_rq);
1214 }
1215 
1216 /*
1217  * Because the prio of an upper entry depends on the lower
1218  * entries, we must remove entries top - down.
1219  */
1220 static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
1221 {
1222 	struct sched_rt_entity *back = NULL;
1223 
1224 	for_each_sched_rt_entity(rt_se) {
1225 		rt_se->back = back;
1226 		back = rt_se;
1227 	}
1228 
1229 	dequeue_top_rt_rq(rt_rq_of_se(back));
1230 
1231 	for (rt_se = back; rt_se; rt_se = rt_se->back) {
1232 		if (on_rt_rq(rt_se))
1233 			__dequeue_rt_entity(rt_se);
1234 	}
1235 }
1236 
1237 static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head)
1238 {
1239 	struct rq *rq = rq_of_rt_se(rt_se);
1240 
1241 	dequeue_rt_stack(rt_se);
1242 	for_each_sched_rt_entity(rt_se)
1243 		__enqueue_rt_entity(rt_se, head);
1244 	enqueue_top_rt_rq(&rq->rt);
1245 }
1246 
1247 static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
1248 {
1249 	struct rq *rq = rq_of_rt_se(rt_se);
1250 
1251 	dequeue_rt_stack(rt_se);
1252 
1253 	for_each_sched_rt_entity(rt_se) {
1254 		struct rt_rq *rt_rq = group_rt_rq(rt_se);
1255 
1256 		if (rt_rq && rt_rq->rt_nr_running)
1257 			__enqueue_rt_entity(rt_se, false);
1258 	}
1259 	enqueue_top_rt_rq(&rq->rt);
1260 }
1261 
1262 /*
1263  * Adding/removing a task to/from a priority array:
1264  */
1265 static void
1266 enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1267 {
1268 	struct sched_rt_entity *rt_se = &p->rt;
1269 
1270 	if (flags & ENQUEUE_WAKEUP)
1271 		rt_se->timeout = 0;
1272 
1273 	enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD);
1274 
1275 	if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1276 		enqueue_pushable_task(rq, p);
1277 }
1278 
1279 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
1280 {
1281 	struct sched_rt_entity *rt_se = &p->rt;
1282 
1283 	update_curr_rt(rq);
1284 	dequeue_rt_entity(rt_se);
1285 
1286 	dequeue_pushable_task(rq, p);
1287 }
1288 
1289 /*
1290  * Put task to the head or the end of the run list without the overhead of
1291  * dequeue followed by enqueue.
1292  */
1293 static void
1294 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
1295 {
1296 	if (on_rt_rq(rt_se)) {
1297 		struct rt_prio_array *array = &rt_rq->active;
1298 		struct list_head *queue = array->queue + rt_se_prio(rt_se);
1299 
1300 		if (head)
1301 			list_move(&rt_se->run_list, queue);
1302 		else
1303 			list_move_tail(&rt_se->run_list, queue);
1304 	}
1305 }
1306 
1307 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
1308 {
1309 	struct sched_rt_entity *rt_se = &p->rt;
1310 	struct rt_rq *rt_rq;
1311 
1312 	for_each_sched_rt_entity(rt_se) {
1313 		rt_rq = rt_rq_of_se(rt_se);
1314 		requeue_rt_entity(rt_rq, rt_se, head);
1315 	}
1316 }
1317 
1318 static void yield_task_rt(struct rq *rq)
1319 {
1320 	requeue_task_rt(rq, rq->curr, 0);
1321 }
1322 
1323 #ifdef CONFIG_SMP
1324 static int find_lowest_rq(struct task_struct *task);
1325 
1326 static int
1327 select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags)
1328 {
1329 	struct task_struct *curr;
1330 	struct rq *rq;
1331 
1332 	/* For anything but wake ups, just return the task_cpu */
1333 	if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK)
1334 		goto out;
1335 
1336 	rq = cpu_rq(cpu);
1337 
1338 	rcu_read_lock();
1339 	curr = READ_ONCE(rq->curr); /* unlocked access */
1340 
1341 	/*
1342 	 * If the current task on @p's runqueue is an RT task, then
1343 	 * try to see if we can wake this RT task up on another
1344 	 * runqueue. Otherwise simply start this RT task
1345 	 * on its current runqueue.
1346 	 *
1347 	 * We want to avoid overloading runqueues. If the woken
1348 	 * task is a higher priority, then it will stay on this CPU
1349 	 * and the lower prio task should be moved to another CPU.
1350 	 * Even though this will probably make the lower prio task
1351 	 * lose its cache, we do not want to bounce a higher task
1352 	 * around just because it gave up its CPU, perhaps for a
1353 	 * lock?
1354 	 *
1355 	 * For equal prio tasks, we just let the scheduler sort it out.
1356 	 *
1357 	 * Otherwise, just let it ride on the affined RQ and the
1358 	 * post-schedule router will push the preempted task away
1359 	 *
1360 	 * This test is optimistic, if we get it wrong the load-balancer
1361 	 * will have to sort it out.
1362 	 */
1363 	if (curr && unlikely(rt_task(curr)) &&
1364 	    (curr->nr_cpus_allowed < 2 ||
1365 	     curr->prio <= p->prio)) {
1366 		int target = find_lowest_rq(p);
1367 
1368 		/*
1369 		 * Don't bother moving it if the destination CPU is
1370 		 * not running a lower priority task.
1371 		 */
1372 		if (target != -1 &&
1373 		    p->prio < cpu_rq(target)->rt.highest_prio.curr)
1374 			cpu = target;
1375 	}
1376 	rcu_read_unlock();
1377 
1378 out:
1379 	return cpu;
1380 }
1381 
1382 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
1383 {
1384 	/*
1385 	 * Current can't be migrated, useless to reschedule,
1386 	 * let's hope p can move out.
1387 	 */
1388 	if (rq->curr->nr_cpus_allowed == 1 ||
1389 	    !cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
1390 		return;
1391 
1392 	/*
1393 	 * p is migratable, so let's not schedule it and
1394 	 * see if it is pushed or pulled somewhere else.
1395 	 */
1396 	if (p->nr_cpus_allowed != 1
1397 	    && cpupri_find(&rq->rd->cpupri, p, NULL))
1398 		return;
1399 
1400 	/*
1401 	 * There appears to be other cpus that can accept
1402 	 * current and none to run 'p', so lets reschedule
1403 	 * to try and push current away:
1404 	 */
1405 	requeue_task_rt(rq, p, 1);
1406 	resched_curr(rq);
1407 }
1408 
1409 #endif /* CONFIG_SMP */
1410 
1411 /*
1412  * Preempt the current task with a newly woken task if needed:
1413  */
1414 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags)
1415 {
1416 	if (p->prio < rq->curr->prio) {
1417 		resched_curr(rq);
1418 		return;
1419 	}
1420 
1421 #ifdef CONFIG_SMP
1422 	/*
1423 	 * If:
1424 	 *
1425 	 * - the newly woken task is of equal priority to the current task
1426 	 * - the newly woken task is non-migratable while current is migratable
1427 	 * - current will be preempted on the next reschedule
1428 	 *
1429 	 * we should check to see if current can readily move to a different
1430 	 * cpu.  If so, we will reschedule to allow the push logic to try
1431 	 * to move current somewhere else, making room for our non-migratable
1432 	 * task.
1433 	 */
1434 	if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr))
1435 		check_preempt_equal_prio(rq, p);
1436 #endif
1437 }
1438 
1439 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1440 						   struct rt_rq *rt_rq)
1441 {
1442 	struct rt_prio_array *array = &rt_rq->active;
1443 	struct sched_rt_entity *next = NULL;
1444 	struct list_head *queue;
1445 	int idx;
1446 
1447 	idx = sched_find_first_bit(array->bitmap);
1448 	BUG_ON(idx >= MAX_RT_PRIO);
1449 
1450 	queue = array->queue + idx;
1451 	next = list_entry(queue->next, struct sched_rt_entity, run_list);
1452 
1453 	return next;
1454 }
1455 
1456 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1457 {
1458 	struct sched_rt_entity *rt_se;
1459 	struct task_struct *p;
1460 	struct rt_rq *rt_rq  = &rq->rt;
1461 
1462 	do {
1463 		rt_se = pick_next_rt_entity(rq, rt_rq);
1464 		BUG_ON(!rt_se);
1465 		rt_rq = group_rt_rq(rt_se);
1466 	} while (rt_rq);
1467 
1468 	p = rt_task_of(rt_se);
1469 	p->se.exec_start = rq_clock_task(rq);
1470 
1471 	return p;
1472 }
1473 
1474 static struct task_struct *
1475 pick_next_task_rt(struct rq *rq, struct task_struct *prev)
1476 {
1477 	struct task_struct *p;
1478 	struct rt_rq *rt_rq = &rq->rt;
1479 
1480 	if (need_pull_rt_task(rq, prev)) {
1481 		pull_rt_task(rq);
1482 		/*
1483 		 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1484 		 * means a dl or stop task can slip in, in which case we need
1485 		 * to re-start task selection.
1486 		 */
1487 		if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) ||
1488 			     rq->dl.dl_nr_running))
1489 			return RETRY_TASK;
1490 	}
1491 
1492 	/*
1493 	 * We may dequeue prev's rt_rq in put_prev_task().
1494 	 * So, we update time before rt_nr_running check.
1495 	 */
1496 	if (prev->sched_class == &rt_sched_class)
1497 		update_curr_rt(rq);
1498 
1499 	if (!rt_rq->rt_queued)
1500 		return NULL;
1501 
1502 	put_prev_task(rq, prev);
1503 
1504 	p = _pick_next_task_rt(rq);
1505 
1506 	/* The running task is never eligible for pushing */
1507 	dequeue_pushable_task(rq, p);
1508 
1509 	queue_push_tasks(rq);
1510 
1511 	return p;
1512 }
1513 
1514 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1515 {
1516 	update_curr_rt(rq);
1517 
1518 	/*
1519 	 * The previous task needs to be made eligible for pushing
1520 	 * if it is still active
1521 	 */
1522 	if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1)
1523 		enqueue_pushable_task(rq, p);
1524 }
1525 
1526 #ifdef CONFIG_SMP
1527 
1528 /* Only try algorithms three times */
1529 #define RT_MAX_TRIES 3
1530 
1531 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1532 {
1533 	if (!task_running(rq, p) &&
1534 	    cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
1535 		return 1;
1536 	return 0;
1537 }
1538 
1539 /*
1540  * Return the highest pushable rq's task, which is suitable to be executed
1541  * on the cpu, NULL otherwise
1542  */
1543 static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu)
1544 {
1545 	struct plist_head *head = &rq->rt.pushable_tasks;
1546 	struct task_struct *p;
1547 
1548 	if (!has_pushable_tasks(rq))
1549 		return NULL;
1550 
1551 	plist_for_each_entry(p, head, pushable_tasks) {
1552 		if (pick_rt_task(rq, p, cpu))
1553 			return p;
1554 	}
1555 
1556 	return NULL;
1557 }
1558 
1559 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1560 
1561 static int find_lowest_rq(struct task_struct *task)
1562 {
1563 	struct sched_domain *sd;
1564 	struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask);
1565 	int this_cpu = smp_processor_id();
1566 	int cpu      = task_cpu(task);
1567 
1568 	/* Make sure the mask is initialized first */
1569 	if (unlikely(!lowest_mask))
1570 		return -1;
1571 
1572 	if (task->nr_cpus_allowed == 1)
1573 		return -1; /* No other targets possible */
1574 
1575 	if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1576 		return -1; /* No targets found */
1577 
1578 	/*
1579 	 * At this point we have built a mask of cpus representing the
1580 	 * lowest priority tasks in the system.  Now we want to elect
1581 	 * the best one based on our affinity and topology.
1582 	 *
1583 	 * We prioritize the last cpu that the task executed on since
1584 	 * it is most likely cache-hot in that location.
1585 	 */
1586 	if (cpumask_test_cpu(cpu, lowest_mask))
1587 		return cpu;
1588 
1589 	/*
1590 	 * Otherwise, we consult the sched_domains span maps to figure
1591 	 * out which cpu is logically closest to our hot cache data.
1592 	 */
1593 	if (!cpumask_test_cpu(this_cpu, lowest_mask))
1594 		this_cpu = -1; /* Skip this_cpu opt if not among lowest */
1595 
1596 	rcu_read_lock();
1597 	for_each_domain(cpu, sd) {
1598 		if (sd->flags & SD_WAKE_AFFINE) {
1599 			int best_cpu;
1600 
1601 			/*
1602 			 * "this_cpu" is cheaper to preempt than a
1603 			 * remote processor.
1604 			 */
1605 			if (this_cpu != -1 &&
1606 			    cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1607 				rcu_read_unlock();
1608 				return this_cpu;
1609 			}
1610 
1611 			best_cpu = cpumask_first_and(lowest_mask,
1612 						     sched_domain_span(sd));
1613 			if (best_cpu < nr_cpu_ids) {
1614 				rcu_read_unlock();
1615 				return best_cpu;
1616 			}
1617 		}
1618 	}
1619 	rcu_read_unlock();
1620 
1621 	/*
1622 	 * And finally, if there were no matches within the domains
1623 	 * just give the caller *something* to work with from the compatible
1624 	 * locations.
1625 	 */
1626 	if (this_cpu != -1)
1627 		return this_cpu;
1628 
1629 	cpu = cpumask_any(lowest_mask);
1630 	if (cpu < nr_cpu_ids)
1631 		return cpu;
1632 	return -1;
1633 }
1634 
1635 /* Will lock the rq it finds */
1636 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1637 {
1638 	struct rq *lowest_rq = NULL;
1639 	int tries;
1640 	int cpu;
1641 
1642 	for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1643 		cpu = find_lowest_rq(task);
1644 
1645 		if ((cpu == -1) || (cpu == rq->cpu))
1646 			break;
1647 
1648 		lowest_rq = cpu_rq(cpu);
1649 
1650 		if (lowest_rq->rt.highest_prio.curr <= task->prio) {
1651 			/*
1652 			 * Target rq has tasks of equal or higher priority,
1653 			 * retrying does not release any lock and is unlikely
1654 			 * to yield a different result.
1655 			 */
1656 			lowest_rq = NULL;
1657 			break;
1658 		}
1659 
1660 		/* if the prio of this runqueue changed, try again */
1661 		if (double_lock_balance(rq, lowest_rq)) {
1662 			/*
1663 			 * We had to unlock the run queue. In
1664 			 * the mean time, task could have
1665 			 * migrated already or had its affinity changed.
1666 			 * Also make sure that it wasn't scheduled on its rq.
1667 			 */
1668 			if (unlikely(task_rq(task) != rq ||
1669 				     !cpumask_test_cpu(lowest_rq->cpu,
1670 						       tsk_cpus_allowed(task)) ||
1671 				     task_running(rq, task) ||
1672 				     !task_on_rq_queued(task))) {
1673 
1674 				double_unlock_balance(rq, lowest_rq);
1675 				lowest_rq = NULL;
1676 				break;
1677 			}
1678 		}
1679 
1680 		/* If this rq is still suitable use it. */
1681 		if (lowest_rq->rt.highest_prio.curr > task->prio)
1682 			break;
1683 
1684 		/* try again */
1685 		double_unlock_balance(rq, lowest_rq);
1686 		lowest_rq = NULL;
1687 	}
1688 
1689 	return lowest_rq;
1690 }
1691 
1692 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1693 {
1694 	struct task_struct *p;
1695 
1696 	if (!has_pushable_tasks(rq))
1697 		return NULL;
1698 
1699 	p = plist_first_entry(&rq->rt.pushable_tasks,
1700 			      struct task_struct, pushable_tasks);
1701 
1702 	BUG_ON(rq->cpu != task_cpu(p));
1703 	BUG_ON(task_current(rq, p));
1704 	BUG_ON(p->nr_cpus_allowed <= 1);
1705 
1706 	BUG_ON(!task_on_rq_queued(p));
1707 	BUG_ON(!rt_task(p));
1708 
1709 	return p;
1710 }
1711 
1712 /*
1713  * If the current CPU has more than one RT task, see if the non
1714  * running task can migrate over to a CPU that is running a task
1715  * of lesser priority.
1716  */
1717 static int push_rt_task(struct rq *rq)
1718 {
1719 	struct task_struct *next_task;
1720 	struct rq *lowest_rq;
1721 	int ret = 0;
1722 
1723 	if (!rq->rt.overloaded)
1724 		return 0;
1725 
1726 	next_task = pick_next_pushable_task(rq);
1727 	if (!next_task)
1728 		return 0;
1729 
1730 retry:
1731 	if (unlikely(next_task == rq->curr)) {
1732 		WARN_ON(1);
1733 		return 0;
1734 	}
1735 
1736 	/*
1737 	 * It's possible that the next_task slipped in of
1738 	 * higher priority than current. If that's the case
1739 	 * just reschedule current.
1740 	 */
1741 	if (unlikely(next_task->prio < rq->curr->prio)) {
1742 		resched_curr(rq);
1743 		return 0;
1744 	}
1745 
1746 	/* We might release rq lock */
1747 	get_task_struct(next_task);
1748 
1749 	/* find_lock_lowest_rq locks the rq if found */
1750 	lowest_rq = find_lock_lowest_rq(next_task, rq);
1751 	if (!lowest_rq) {
1752 		struct task_struct *task;
1753 		/*
1754 		 * find_lock_lowest_rq releases rq->lock
1755 		 * so it is possible that next_task has migrated.
1756 		 *
1757 		 * We need to make sure that the task is still on the same
1758 		 * run-queue and is also still the next task eligible for
1759 		 * pushing.
1760 		 */
1761 		task = pick_next_pushable_task(rq);
1762 		if (task_cpu(next_task) == rq->cpu && task == next_task) {
1763 			/*
1764 			 * The task hasn't migrated, and is still the next
1765 			 * eligible task, but we failed to find a run-queue
1766 			 * to push it to.  Do not retry in this case, since
1767 			 * other cpus will pull from us when ready.
1768 			 */
1769 			goto out;
1770 		}
1771 
1772 		if (!task)
1773 			/* No more tasks, just exit */
1774 			goto out;
1775 
1776 		/*
1777 		 * Something has shifted, try again.
1778 		 */
1779 		put_task_struct(next_task);
1780 		next_task = task;
1781 		goto retry;
1782 	}
1783 
1784 	deactivate_task(rq, next_task, 0);
1785 	set_task_cpu(next_task, lowest_rq->cpu);
1786 	activate_task(lowest_rq, next_task, 0);
1787 	ret = 1;
1788 
1789 	resched_curr(lowest_rq);
1790 
1791 	double_unlock_balance(rq, lowest_rq);
1792 
1793 out:
1794 	put_task_struct(next_task);
1795 
1796 	return ret;
1797 }
1798 
1799 static void push_rt_tasks(struct rq *rq)
1800 {
1801 	/* push_rt_task will return true if it moved an RT */
1802 	while (push_rt_task(rq))
1803 		;
1804 }
1805 
1806 #ifdef HAVE_RT_PUSH_IPI
1807 /*
1808  * The search for the next cpu always starts at rq->cpu and ends
1809  * when we reach rq->cpu again. It will never return rq->cpu.
1810  * This returns the next cpu to check, or nr_cpu_ids if the loop
1811  * is complete.
1812  *
1813  * rq->rt.push_cpu holds the last cpu returned by this function,
1814  * or if this is the first instance, it must hold rq->cpu.
1815  */
1816 static int rto_next_cpu(struct rq *rq)
1817 {
1818 	int prev_cpu = rq->rt.push_cpu;
1819 	int cpu;
1820 
1821 	cpu = cpumask_next(prev_cpu, rq->rd->rto_mask);
1822 
1823 	/*
1824 	 * If the previous cpu is less than the rq's CPU, then it already
1825 	 * passed the end of the mask, and has started from the beginning.
1826 	 * We end if the next CPU is greater or equal to rq's CPU.
1827 	 */
1828 	if (prev_cpu < rq->cpu) {
1829 		if (cpu >= rq->cpu)
1830 			return nr_cpu_ids;
1831 
1832 	} else if (cpu >= nr_cpu_ids) {
1833 		/*
1834 		 * We passed the end of the mask, start at the beginning.
1835 		 * If the result is greater or equal to the rq's CPU, then
1836 		 * the loop is finished.
1837 		 */
1838 		cpu = cpumask_first(rq->rd->rto_mask);
1839 		if (cpu >= rq->cpu)
1840 			return nr_cpu_ids;
1841 	}
1842 	rq->rt.push_cpu = cpu;
1843 
1844 	/* Return cpu to let the caller know if the loop is finished or not */
1845 	return cpu;
1846 }
1847 
1848 static int find_next_push_cpu(struct rq *rq)
1849 {
1850 	struct rq *next_rq;
1851 	int cpu;
1852 
1853 	while (1) {
1854 		cpu = rto_next_cpu(rq);
1855 		if (cpu >= nr_cpu_ids)
1856 			break;
1857 		next_rq = cpu_rq(cpu);
1858 
1859 		/* Make sure the next rq can push to this rq */
1860 		if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr)
1861 			break;
1862 	}
1863 
1864 	return cpu;
1865 }
1866 
1867 #define RT_PUSH_IPI_EXECUTING		1
1868 #define RT_PUSH_IPI_RESTART		2
1869 
1870 static void tell_cpu_to_push(struct rq *rq)
1871 {
1872 	int cpu;
1873 
1874 	if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) {
1875 		raw_spin_lock(&rq->rt.push_lock);
1876 		/* Make sure it's still executing */
1877 		if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) {
1878 			/*
1879 			 * Tell the IPI to restart the loop as things have
1880 			 * changed since it started.
1881 			 */
1882 			rq->rt.push_flags |= RT_PUSH_IPI_RESTART;
1883 			raw_spin_unlock(&rq->rt.push_lock);
1884 			return;
1885 		}
1886 		raw_spin_unlock(&rq->rt.push_lock);
1887 	}
1888 
1889 	/* When here, there's no IPI going around */
1890 
1891 	rq->rt.push_cpu = rq->cpu;
1892 	cpu = find_next_push_cpu(rq);
1893 	if (cpu >= nr_cpu_ids)
1894 		return;
1895 
1896 	rq->rt.push_flags = RT_PUSH_IPI_EXECUTING;
1897 
1898 	irq_work_queue_on(&rq->rt.push_work, cpu);
1899 }
1900 
1901 /* Called from hardirq context */
1902 static void try_to_push_tasks(void *arg)
1903 {
1904 	struct rt_rq *rt_rq = arg;
1905 	struct rq *rq, *src_rq;
1906 	int this_cpu;
1907 	int cpu;
1908 
1909 	this_cpu = rt_rq->push_cpu;
1910 
1911 	/* Paranoid check */
1912 	BUG_ON(this_cpu != smp_processor_id());
1913 
1914 	rq = cpu_rq(this_cpu);
1915 	src_rq = rq_of_rt_rq(rt_rq);
1916 
1917 again:
1918 	if (has_pushable_tasks(rq)) {
1919 		raw_spin_lock(&rq->lock);
1920 		push_rt_task(rq);
1921 		raw_spin_unlock(&rq->lock);
1922 	}
1923 
1924 	/* Pass the IPI to the next rt overloaded queue */
1925 	raw_spin_lock(&rt_rq->push_lock);
1926 	/*
1927 	 * If the source queue changed since the IPI went out,
1928 	 * we need to restart the search from that CPU again.
1929 	 */
1930 	if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) {
1931 		rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART;
1932 		rt_rq->push_cpu = src_rq->cpu;
1933 	}
1934 
1935 	cpu = find_next_push_cpu(src_rq);
1936 
1937 	if (cpu >= nr_cpu_ids)
1938 		rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING;
1939 	raw_spin_unlock(&rt_rq->push_lock);
1940 
1941 	if (cpu >= nr_cpu_ids)
1942 		return;
1943 
1944 	/*
1945 	 * It is possible that a restart caused this CPU to be
1946 	 * chosen again. Don't bother with an IPI, just see if we
1947 	 * have more to push.
1948 	 */
1949 	if (unlikely(cpu == rq->cpu))
1950 		goto again;
1951 
1952 	/* Try the next RT overloaded CPU */
1953 	irq_work_queue_on(&rt_rq->push_work, cpu);
1954 }
1955 
1956 static void push_irq_work_func(struct irq_work *work)
1957 {
1958 	struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work);
1959 
1960 	try_to_push_tasks(rt_rq);
1961 }
1962 #endif /* HAVE_RT_PUSH_IPI */
1963 
1964 static void pull_rt_task(struct rq *this_rq)
1965 {
1966 	int this_cpu = this_rq->cpu, cpu;
1967 	bool resched = false;
1968 	struct task_struct *p;
1969 	struct rq *src_rq;
1970 
1971 	if (likely(!rt_overloaded(this_rq)))
1972 		return;
1973 
1974 	/*
1975 	 * Match the barrier from rt_set_overloaded; this guarantees that if we
1976 	 * see overloaded we must also see the rto_mask bit.
1977 	 */
1978 	smp_rmb();
1979 
1980 #ifdef HAVE_RT_PUSH_IPI
1981 	if (sched_feat(RT_PUSH_IPI)) {
1982 		tell_cpu_to_push(this_rq);
1983 		return;
1984 	}
1985 #endif
1986 
1987 	for_each_cpu(cpu, this_rq->rd->rto_mask) {
1988 		if (this_cpu == cpu)
1989 			continue;
1990 
1991 		src_rq = cpu_rq(cpu);
1992 
1993 		/*
1994 		 * Don't bother taking the src_rq->lock if the next highest
1995 		 * task is known to be lower-priority than our current task.
1996 		 * This may look racy, but if this value is about to go
1997 		 * logically higher, the src_rq will push this task away.
1998 		 * And if its going logically lower, we do not care
1999 		 */
2000 		if (src_rq->rt.highest_prio.next >=
2001 		    this_rq->rt.highest_prio.curr)
2002 			continue;
2003 
2004 		/*
2005 		 * We can potentially drop this_rq's lock in
2006 		 * double_lock_balance, and another CPU could
2007 		 * alter this_rq
2008 		 */
2009 		double_lock_balance(this_rq, src_rq);
2010 
2011 		/*
2012 		 * We can pull only a task, which is pushable
2013 		 * on its rq, and no others.
2014 		 */
2015 		p = pick_highest_pushable_task(src_rq, this_cpu);
2016 
2017 		/*
2018 		 * Do we have an RT task that preempts
2019 		 * the to-be-scheduled task?
2020 		 */
2021 		if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
2022 			WARN_ON(p == src_rq->curr);
2023 			WARN_ON(!task_on_rq_queued(p));
2024 
2025 			/*
2026 			 * There's a chance that p is higher in priority
2027 			 * than what's currently running on its cpu.
2028 			 * This is just that p is wakeing up and hasn't
2029 			 * had a chance to schedule. We only pull
2030 			 * p if it is lower in priority than the
2031 			 * current task on the run queue
2032 			 */
2033 			if (p->prio < src_rq->curr->prio)
2034 				goto skip;
2035 
2036 			resched = true;
2037 
2038 			deactivate_task(src_rq, p, 0);
2039 			set_task_cpu(p, this_cpu);
2040 			activate_task(this_rq, p, 0);
2041 			/*
2042 			 * We continue with the search, just in
2043 			 * case there's an even higher prio task
2044 			 * in another runqueue. (low likelihood
2045 			 * but possible)
2046 			 */
2047 		}
2048 skip:
2049 		double_unlock_balance(this_rq, src_rq);
2050 	}
2051 
2052 	if (resched)
2053 		resched_curr(this_rq);
2054 }
2055 
2056 /*
2057  * If we are not running and we are not going to reschedule soon, we should
2058  * try to push tasks away now
2059  */
2060 static void task_woken_rt(struct rq *rq, struct task_struct *p)
2061 {
2062 	if (!task_running(rq, p) &&
2063 	    !test_tsk_need_resched(rq->curr) &&
2064 	    has_pushable_tasks(rq) &&
2065 	    p->nr_cpus_allowed > 1 &&
2066 	    (dl_task(rq->curr) || rt_task(rq->curr)) &&
2067 	    (rq->curr->nr_cpus_allowed < 2 ||
2068 	     rq->curr->prio <= p->prio))
2069 		push_rt_tasks(rq);
2070 }
2071 
2072 static void set_cpus_allowed_rt(struct task_struct *p,
2073 				const struct cpumask *new_mask)
2074 {
2075 	struct rq *rq;
2076 	int weight;
2077 
2078 	BUG_ON(!rt_task(p));
2079 
2080 	if (!task_on_rq_queued(p))
2081 		return;
2082 
2083 	weight = cpumask_weight(new_mask);
2084 
2085 	/*
2086 	 * Only update if the process changes its state from whether it
2087 	 * can migrate or not.
2088 	 */
2089 	if ((p->nr_cpus_allowed > 1) == (weight > 1))
2090 		return;
2091 
2092 	rq = task_rq(p);
2093 
2094 	/*
2095 	 * The process used to be able to migrate OR it can now migrate
2096 	 */
2097 	if (weight <= 1) {
2098 		if (!task_current(rq, p))
2099 			dequeue_pushable_task(rq, p);
2100 		BUG_ON(!rq->rt.rt_nr_migratory);
2101 		rq->rt.rt_nr_migratory--;
2102 	} else {
2103 		if (!task_current(rq, p))
2104 			enqueue_pushable_task(rq, p);
2105 		rq->rt.rt_nr_migratory++;
2106 	}
2107 
2108 	update_rt_migration(&rq->rt);
2109 }
2110 
2111 /* Assumes rq->lock is held */
2112 static void rq_online_rt(struct rq *rq)
2113 {
2114 	if (rq->rt.overloaded)
2115 		rt_set_overload(rq);
2116 
2117 	__enable_runtime(rq);
2118 
2119 	cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
2120 }
2121 
2122 /* Assumes rq->lock is held */
2123 static void rq_offline_rt(struct rq *rq)
2124 {
2125 	if (rq->rt.overloaded)
2126 		rt_clear_overload(rq);
2127 
2128 	__disable_runtime(rq);
2129 
2130 	cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
2131 }
2132 
2133 /*
2134  * When switch from the rt queue, we bring ourselves to a position
2135  * that we might want to pull RT tasks from other runqueues.
2136  */
2137 static void switched_from_rt(struct rq *rq, struct task_struct *p)
2138 {
2139 	/*
2140 	 * If there are other RT tasks then we will reschedule
2141 	 * and the scheduling of the other RT tasks will handle
2142 	 * the balancing. But if we are the last RT task
2143 	 * we may need to handle the pulling of RT tasks
2144 	 * now.
2145 	 */
2146 	if (!task_on_rq_queued(p) || rq->rt.rt_nr_running)
2147 		return;
2148 
2149 	queue_pull_task(rq);
2150 }
2151 
2152 void __init init_sched_rt_class(void)
2153 {
2154 	unsigned int i;
2155 
2156 	for_each_possible_cpu(i) {
2157 		zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
2158 					GFP_KERNEL, cpu_to_node(i));
2159 	}
2160 }
2161 #endif /* CONFIG_SMP */
2162 
2163 /*
2164  * When switching a task to RT, we may overload the runqueue
2165  * with RT tasks. In this case we try to push them off to
2166  * other runqueues.
2167  */
2168 static void switched_to_rt(struct rq *rq, struct task_struct *p)
2169 {
2170 	/*
2171 	 * If we are already running, then there's nothing
2172 	 * that needs to be done. But if we are not running
2173 	 * we may need to preempt the current running task.
2174 	 * If that current running task is also an RT task
2175 	 * then see if we can move to another run queue.
2176 	 */
2177 	if (task_on_rq_queued(p) && rq->curr != p) {
2178 #ifdef CONFIG_SMP
2179 		if (p->nr_cpus_allowed > 1 && rq->rt.overloaded)
2180 			queue_push_tasks(rq);
2181 #else
2182 		if (p->prio < rq->curr->prio)
2183 			resched_curr(rq);
2184 #endif /* CONFIG_SMP */
2185 	}
2186 }
2187 
2188 /*
2189  * Priority of the task has changed. This may cause
2190  * us to initiate a push or pull.
2191  */
2192 static void
2193 prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio)
2194 {
2195 	if (!task_on_rq_queued(p))
2196 		return;
2197 
2198 	if (rq->curr == p) {
2199 #ifdef CONFIG_SMP
2200 		/*
2201 		 * If our priority decreases while running, we
2202 		 * may need to pull tasks to this runqueue.
2203 		 */
2204 		if (oldprio < p->prio)
2205 			queue_pull_task(rq);
2206 
2207 		/*
2208 		 * If there's a higher priority task waiting to run
2209 		 * then reschedule.
2210 		 */
2211 		if (p->prio > rq->rt.highest_prio.curr)
2212 			resched_curr(rq);
2213 #else
2214 		/* For UP simply resched on drop of prio */
2215 		if (oldprio < p->prio)
2216 			resched_curr(rq);
2217 #endif /* CONFIG_SMP */
2218 	} else {
2219 		/*
2220 		 * This task is not running, but if it is
2221 		 * greater than the current running task
2222 		 * then reschedule.
2223 		 */
2224 		if (p->prio < rq->curr->prio)
2225 			resched_curr(rq);
2226 	}
2227 }
2228 
2229 static void watchdog(struct rq *rq, struct task_struct *p)
2230 {
2231 	unsigned long soft, hard;
2232 
2233 	/* max may change after cur was read, this will be fixed next tick */
2234 	soft = task_rlimit(p, RLIMIT_RTTIME);
2235 	hard = task_rlimit_max(p, RLIMIT_RTTIME);
2236 
2237 	if (soft != RLIM_INFINITY) {
2238 		unsigned long next;
2239 
2240 		if (p->rt.watchdog_stamp != jiffies) {
2241 			p->rt.timeout++;
2242 			p->rt.watchdog_stamp = jiffies;
2243 		}
2244 
2245 		next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
2246 		if (p->rt.timeout > next)
2247 			p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
2248 	}
2249 }
2250 
2251 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
2252 {
2253 	struct sched_rt_entity *rt_se = &p->rt;
2254 
2255 	update_curr_rt(rq);
2256 
2257 	watchdog(rq, p);
2258 
2259 	/*
2260 	 * RR tasks need a special form of timeslice management.
2261 	 * FIFO tasks have no timeslices.
2262 	 */
2263 	if (p->policy != SCHED_RR)
2264 		return;
2265 
2266 	if (--p->rt.time_slice)
2267 		return;
2268 
2269 	p->rt.time_slice = sched_rr_timeslice;
2270 
2271 	/*
2272 	 * Requeue to the end of queue if we (and all of our ancestors) are not
2273 	 * the only element on the queue
2274 	 */
2275 	for_each_sched_rt_entity(rt_se) {
2276 		if (rt_se->run_list.prev != rt_se->run_list.next) {
2277 			requeue_task_rt(rq, p, 0);
2278 			resched_curr(rq);
2279 			return;
2280 		}
2281 	}
2282 }
2283 
2284 static void set_curr_task_rt(struct rq *rq)
2285 {
2286 	struct task_struct *p = rq->curr;
2287 
2288 	p->se.exec_start = rq_clock_task(rq);
2289 
2290 	/* The running task is never eligible for pushing */
2291 	dequeue_pushable_task(rq, p);
2292 }
2293 
2294 static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task)
2295 {
2296 	/*
2297 	 * Time slice is 0 for SCHED_FIFO tasks
2298 	 */
2299 	if (task->policy == SCHED_RR)
2300 		return sched_rr_timeslice;
2301 	else
2302 		return 0;
2303 }
2304 
2305 const struct sched_class rt_sched_class = {
2306 	.next			= &fair_sched_class,
2307 	.enqueue_task		= enqueue_task_rt,
2308 	.dequeue_task		= dequeue_task_rt,
2309 	.yield_task		= yield_task_rt,
2310 
2311 	.check_preempt_curr	= check_preempt_curr_rt,
2312 
2313 	.pick_next_task		= pick_next_task_rt,
2314 	.put_prev_task		= put_prev_task_rt,
2315 
2316 #ifdef CONFIG_SMP
2317 	.select_task_rq		= select_task_rq_rt,
2318 
2319 	.set_cpus_allowed       = set_cpus_allowed_rt,
2320 	.rq_online              = rq_online_rt,
2321 	.rq_offline             = rq_offline_rt,
2322 	.task_woken		= task_woken_rt,
2323 	.switched_from		= switched_from_rt,
2324 #endif
2325 
2326 	.set_curr_task          = set_curr_task_rt,
2327 	.task_tick		= task_tick_rt,
2328 
2329 	.get_rr_interval	= get_rr_interval_rt,
2330 
2331 	.prio_changed		= prio_changed_rt,
2332 	.switched_to		= switched_to_rt,
2333 
2334 	.update_curr		= update_curr_rt,
2335 };
2336 
2337 #ifdef CONFIG_SCHED_DEBUG
2338 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
2339 
2340 void print_rt_stats(struct seq_file *m, int cpu)
2341 {
2342 	rt_rq_iter_t iter;
2343 	struct rt_rq *rt_rq;
2344 
2345 	rcu_read_lock();
2346 	for_each_rt_rq(rt_rq, iter, cpu_rq(cpu))
2347 		print_rt_rq(m, cpu, rt_rq);
2348 	rcu_read_unlock();
2349 }
2350 #endif /* CONFIG_SCHED_DEBUG */
2351