1 /*
2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
3 * policies)
4 */
5
6 static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
7 {
8 return container_of(rt_se, struct task_struct, rt);
9 }
10
11 #ifdef CONFIG_RT_GROUP_SCHED
12
13 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
14
15 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
16 {
17 return rt_rq->rq;
18 }
19
20 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
21 {
22 return rt_se->rt_rq;
23 }
24
25 #else /* CONFIG_RT_GROUP_SCHED */
26
27 #define rt_entity_is_task(rt_se) (1)
28
29 static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
30 {
31 return container_of(rt_rq, struct rq, rt);
32 }
33
34 static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
35 {
36 struct task_struct *p = rt_task_of(rt_se);
37 struct rq *rq = task_rq(p);
38
39 return &rq->rt;
40 }
41
42 #endif /* CONFIG_RT_GROUP_SCHED */
43
44 #ifdef CONFIG_SMP
45
46 static inline int rt_overloaded(struct rq *rq)
47 {
48 return atomic_read(&rq->rd->rto_count);
49 }
50
51 static inline void rt_set_overload(struct rq *rq)
52 {
53 if (!rq->online)
54 return;
55
56 cpumask_set_cpu(rq->cpu, rq->rd->rto_mask);
57 /*
58 * Make sure the mask is visible before we set
59 * the overload count. That is checked to determine
60 * if we should look at the mask. It would be a shame
61 * if we looked at the mask, but the mask was not
62 * updated yet.
63 */
64 wmb();
65 atomic_inc(&rq->rd->rto_count);
66 }
67
68 static inline void rt_clear_overload(struct rq *rq)
69 {
70 if (!rq->online)
71 return;
72
73 /* the order here really doesn't matter */
74 atomic_dec(&rq->rd->rto_count);
75 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
76 }
77
78 static void update_rt_migration(struct rt_rq *rt_rq)
79 {
80 if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) {
81 if (!rt_rq->overloaded) {
82 rt_set_overload(rq_of_rt_rq(rt_rq));
83 rt_rq->overloaded = 1;
84 }
85 } else if (rt_rq->overloaded) {
86 rt_clear_overload(rq_of_rt_rq(rt_rq));
87 rt_rq->overloaded = 0;
88 }
89 }
90
91 static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
92 {
93 if (!rt_entity_is_task(rt_se))
94 return;
95
96 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
97
98 rt_rq->rt_nr_total++;
99 if (rt_se->nr_cpus_allowed > 1)
100 rt_rq->rt_nr_migratory++;
101
102 update_rt_migration(rt_rq);
103 }
104
105 static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
106 {
107 if (!rt_entity_is_task(rt_se))
108 return;
109
110 rt_rq = &rq_of_rt_rq(rt_rq)->rt;
111
112 rt_rq->rt_nr_total--;
113 if (rt_se->nr_cpus_allowed > 1)
114 rt_rq->rt_nr_migratory--;
115
116 update_rt_migration(rt_rq);
117 }
118
119 static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
120 {
121 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
122 plist_node_init(&p->pushable_tasks, p->prio);
123 plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
124 }
125
126 static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
127 {
128 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
129 }
130
131 #else
132
133 static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
134 {
135 }
136
137 static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
138 {
139 }
140
141 static inline
142 void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
143 {
144 }
145
146 static inline
147 void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
148 {
149 }
150
151 #endif /* CONFIG_SMP */
152
153 static inline int on_rt_rq(struct sched_rt_entity *rt_se)
154 {
155 return !list_empty(&rt_se->run_list);
156 }
157
158 #ifdef CONFIG_RT_GROUP_SCHED
159
160 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
161 {
162 if (!rt_rq->tg)
163 return RUNTIME_INF;
164
165 return rt_rq->rt_runtime;
166 }
167
168 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
169 {
170 return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period);
171 }
172
173 #define for_each_leaf_rt_rq(rt_rq, rq) \
174 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
175
176 #define for_each_sched_rt_entity(rt_se) \
177 for (; rt_se; rt_se = rt_se->parent)
178
179 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
180 {
181 return rt_se->my_q;
182 }
183
184 static void enqueue_rt_entity(struct sched_rt_entity *rt_se);
185 static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
186
187 static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
188 {
189 struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr;
190 struct sched_rt_entity *rt_se = rt_rq->rt_se;
191
192 if (rt_rq->rt_nr_running) {
193 if (rt_se && !on_rt_rq(rt_se))
194 enqueue_rt_entity(rt_se);
195 if (rt_rq->highest_prio.curr < curr->prio)
196 resched_task(curr);
197 }
198 }
199
200 static void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
201 {
202 struct sched_rt_entity *rt_se = rt_rq->rt_se;
203
204 if (rt_se && on_rt_rq(rt_se))
205 dequeue_rt_entity(rt_se);
206 }
207
208 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
209 {
210 return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted;
211 }
212
213 static int rt_se_boosted(struct sched_rt_entity *rt_se)
214 {
215 struct rt_rq *rt_rq = group_rt_rq(rt_se);
216 struct task_struct *p;
217
218 if (rt_rq)
219 return !!rt_rq->rt_nr_boosted;
220
221 p = rt_task_of(rt_se);
222 return p->prio != p->normal_prio;
223 }
224
225 #ifdef CONFIG_SMP
226 static inline const struct cpumask *sched_rt_period_mask(void)
227 {
228 return cpu_rq(smp_processor_id())->rd->span;
229 }
230 #else
231 static inline const struct cpumask *sched_rt_period_mask(void)
232 {
233 return cpu_online_mask;
234 }
235 #endif
236
237 static inline
238 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
239 {
240 return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
241 }
242
243 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
244 {
245 return &rt_rq->tg->rt_bandwidth;
246 }
247
248 #else /* !CONFIG_RT_GROUP_SCHED */
249
250 static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
251 {
252 return rt_rq->rt_runtime;
253 }
254
255 static inline u64 sched_rt_period(struct rt_rq *rt_rq)
256 {
257 return ktime_to_ns(def_rt_bandwidth.rt_period);
258 }
259
260 #define for_each_leaf_rt_rq(rt_rq, rq) \
261 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
262
263 #define for_each_sched_rt_entity(rt_se) \
264 for (; rt_se; rt_se = NULL)
265
266 static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
267 {
268 return NULL;
269 }
270
271 static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
272 {
273 if (rt_rq->rt_nr_running)
274 resched_task(rq_of_rt_rq(rt_rq)->curr);
275 }
276
277 static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq)
278 {
279 }
280
281 static inline int rt_rq_throttled(struct rt_rq *rt_rq)
282 {
283 return rt_rq->rt_throttled;
284 }
285
286 static inline const struct cpumask *sched_rt_period_mask(void)
287 {
288 return cpu_online_mask;
289 }
290
291 static inline
292 struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
293 {
294 return &cpu_rq(cpu)->rt;
295 }
296
297 static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
298 {
299 return &def_rt_bandwidth;
300 }
301
302 #endif /* CONFIG_RT_GROUP_SCHED */
303
304 #ifdef CONFIG_SMP
305 /*
306 * We ran out of runtime, see if we can borrow some from our neighbours.
307 */
308 static int do_balance_runtime(struct rt_rq *rt_rq)
309 {
310 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
311 struct root_domain *rd = cpu_rq(smp_processor_id())->rd;
312 int i, weight, more = 0;
313 u64 rt_period;
314
315 weight = cpumask_weight(rd->span);
316
317 spin_lock(&rt_b->rt_runtime_lock);
318 rt_period = ktime_to_ns(rt_b->rt_period);
319 for_each_cpu(i, rd->span) {
320 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
321 s64 diff;
322
323 if (iter == rt_rq)
324 continue;
325
326 spin_lock(&iter->rt_runtime_lock);
327 /*
328 * Either all rqs have inf runtime and there's nothing to steal
329 * or __disable_runtime() below sets a specific rq to inf to
330 * indicate its been disabled and disalow stealing.
331 */
332 if (iter->rt_runtime == RUNTIME_INF)
333 goto next;
334
335 /*
336 * From runqueues with spare time, take 1/n part of their
337 * spare time, but no more than our period.
338 */
339 diff = iter->rt_runtime - iter->rt_time;
340 if (diff > 0) {
341 diff = div_u64((u64)diff, weight);
342 if (rt_rq->rt_runtime + diff > rt_period)
343 diff = rt_period - rt_rq->rt_runtime;
344 iter->rt_runtime -= diff;
345 rt_rq->rt_runtime += diff;
346 more = 1;
347 if (rt_rq->rt_runtime == rt_period) {
348 spin_unlock(&iter->rt_runtime_lock);
349 break;
350 }
351 }
352 next:
353 spin_unlock(&iter->rt_runtime_lock);
354 }
355 spin_unlock(&rt_b->rt_runtime_lock);
356
357 return more;
358 }
359
360 /*
361 * Ensure this RQ takes back all the runtime it lend to its neighbours.
362 */
363 static void __disable_runtime(struct rq *rq)
364 {
365 struct root_domain *rd = rq->rd;
366 struct rt_rq *rt_rq;
367
368 if (unlikely(!scheduler_running))
369 return;
370
371 for_each_leaf_rt_rq(rt_rq, rq) {
372 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
373 s64 want;
374 int i;
375
376 spin_lock(&rt_b->rt_runtime_lock);
377 spin_lock(&rt_rq->rt_runtime_lock);
378 /*
379 * Either we're all inf and nobody needs to borrow, or we're
380 * already disabled and thus have nothing to do, or we have
381 * exactly the right amount of runtime to take out.
382 */
383 if (rt_rq->rt_runtime == RUNTIME_INF ||
384 rt_rq->rt_runtime == rt_b->rt_runtime)
385 goto balanced;
386 spin_unlock(&rt_rq->rt_runtime_lock);
387
388 /*
389 * Calculate the difference between what we started out with
390 * and what we current have, that's the amount of runtime
391 * we lend and now have to reclaim.
392 */
393 want = rt_b->rt_runtime - rt_rq->rt_runtime;
394
395 /*
396 * Greedy reclaim, take back as much as we can.
397 */
398 for_each_cpu(i, rd->span) {
399 struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
400 s64 diff;
401
402 /*
403 * Can't reclaim from ourselves or disabled runqueues.
404 */
405 if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
406 continue;
407
408 spin_lock(&iter->rt_runtime_lock);
409 if (want > 0) {
410 diff = min_t(s64, iter->rt_runtime, want);
411 iter->rt_runtime -= diff;
412 want -= diff;
413 } else {
414 iter->rt_runtime -= want;
415 want -= want;
416 }
417 spin_unlock(&iter->rt_runtime_lock);
418
419 if (!want)
420 break;
421 }
422
423 spin_lock(&rt_rq->rt_runtime_lock);
424 /*
425 * We cannot be left wanting - that would mean some runtime
426 * leaked out of the system.
427 */
428 BUG_ON(want);
429 balanced:
430 /*
431 * Disable all the borrow logic by pretending we have inf
432 * runtime - in which case borrowing doesn't make sense.
433 */
434 rt_rq->rt_runtime = RUNTIME_INF;
435 spin_unlock(&rt_rq->rt_runtime_lock);
436 spin_unlock(&rt_b->rt_runtime_lock);
437 }
438 }
439
440 static void disable_runtime(struct rq *rq)
441 {
442 unsigned long flags;
443
444 spin_lock_irqsave(&rq->lock, flags);
445 __disable_runtime(rq);
446 spin_unlock_irqrestore(&rq->lock, flags);
447 }
448
449 static void __enable_runtime(struct rq *rq)
450 {
451 struct rt_rq *rt_rq;
452
453 if (unlikely(!scheduler_running))
454 return;
455
456 /*
457 * Reset each runqueue's bandwidth settings
458 */
459 for_each_leaf_rt_rq(rt_rq, rq) {
460 struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
461
462 spin_lock(&rt_b->rt_runtime_lock);
463 spin_lock(&rt_rq->rt_runtime_lock);
464 rt_rq->rt_runtime = rt_b->rt_runtime;
465 rt_rq->rt_time = 0;
466 rt_rq->rt_throttled = 0;
467 spin_unlock(&rt_rq->rt_runtime_lock);
468 spin_unlock(&rt_b->rt_runtime_lock);
469 }
470 }
471
472 static void enable_runtime(struct rq *rq)
473 {
474 unsigned long flags;
475
476 spin_lock_irqsave(&rq->lock, flags);
477 __enable_runtime(rq);
478 spin_unlock_irqrestore(&rq->lock, flags);
479 }
480
481 static int balance_runtime(struct rt_rq *rt_rq)
482 {
483 int more = 0;
484
485 if (rt_rq->rt_time > rt_rq->rt_runtime) {
486 spin_unlock(&rt_rq->rt_runtime_lock);
487 more = do_balance_runtime(rt_rq);
488 spin_lock(&rt_rq->rt_runtime_lock);
489 }
490
491 return more;
492 }
493 #else /* !CONFIG_SMP */
494 static inline int balance_runtime(struct rt_rq *rt_rq)
495 {
496 return 0;
497 }
498 #endif /* CONFIG_SMP */
499
500 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
501 {
502 int i, idle = 1;
503 const struct cpumask *span;
504
505 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
506 return 1;
507
508 span = sched_rt_period_mask();
509 for_each_cpu(i, span) {
510 int enqueue = 0;
511 struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
512 struct rq *rq = rq_of_rt_rq(rt_rq);
513
514 spin_lock(&rq->lock);
515 if (rt_rq->rt_time) {
516 u64 runtime;
517
518 spin_lock(&rt_rq->rt_runtime_lock);
519 if (rt_rq->rt_throttled)
520 balance_runtime(rt_rq);
521 runtime = rt_rq->rt_runtime;
522 rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
523 if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
524 rt_rq->rt_throttled = 0;
525 enqueue = 1;
526 }
527 if (rt_rq->rt_time || rt_rq->rt_nr_running)
528 idle = 0;
529 spin_unlock(&rt_rq->rt_runtime_lock);
530 } else if (rt_rq->rt_nr_running)
531 idle = 0;
532
533 if (enqueue)
534 sched_rt_rq_enqueue(rt_rq);
535 spin_unlock(&rq->lock);
536 }
537
538 return idle;
539 }
540
541 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
542 {
543 #ifdef CONFIG_RT_GROUP_SCHED
544 struct rt_rq *rt_rq = group_rt_rq(rt_se);
545
546 if (rt_rq)
547 return rt_rq->highest_prio.curr;
548 #endif
549
550 return rt_task_of(rt_se)->prio;
551 }
552
553 static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
554 {
555 u64 runtime = sched_rt_runtime(rt_rq);
556
557 if (rt_rq->rt_throttled)
558 return rt_rq_throttled(rt_rq);
559
560 if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq))
561 return 0;
562
563 balance_runtime(rt_rq);
564 runtime = sched_rt_runtime(rt_rq);
565 if (runtime == RUNTIME_INF)
566 return 0;
567
568 if (rt_rq->rt_time > runtime) {
569 rt_rq->rt_throttled = 1;
570 if (rt_rq_throttled(rt_rq)) {
571 sched_rt_rq_dequeue(rt_rq);
572 return 1;
573 }
574 }
575
576 return 0;
577 }
578
579 /*
580 * Update the current task's runtime statistics. Skip current tasks that
581 * are not in our scheduling class.
582 */
583 static void update_curr_rt(struct rq *rq)
584 {
585 struct task_struct *curr = rq->curr;
586 struct sched_rt_entity *rt_se = &curr->rt;
587 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
588 u64 delta_exec;
589
590 if (!task_has_rt_policy(curr))
591 return;
592
593 delta_exec = rq->clock - curr->se.exec_start;
594 if (unlikely((s64)delta_exec < 0))
595 delta_exec = 0;
596
597 schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec));
598
599 curr->se.sum_exec_runtime += delta_exec;
600 account_group_exec_runtime(curr, delta_exec);
601
602 curr->se.exec_start = rq->clock;
603 cpuacct_charge(curr, delta_exec);
604
605 if (!rt_bandwidth_enabled())
606 return;
607
608 for_each_sched_rt_entity(rt_se) {
609 rt_rq = rt_rq_of_se(rt_se);
610
611 if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
612 spin_lock(&rt_rq->rt_runtime_lock);
613 rt_rq->rt_time += delta_exec;
614 if (sched_rt_runtime_exceeded(rt_rq))
615 resched_task(curr);
616 spin_unlock(&rt_rq->rt_runtime_lock);
617 }
618 }
619 }
620
621 #if defined CONFIG_SMP
622
623 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu);
624
625 static inline int next_prio(struct rq *rq)
626 {
627 struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu);
628
629 if (next && rt_prio(next->prio))
630 return next->prio;
631 else
632 return MAX_RT_PRIO;
633 }
634
635 static void
636 inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
637 {
638 struct rq *rq = rq_of_rt_rq(rt_rq);
639
640 if (prio < prev_prio) {
641
642 /*
643 * If the new task is higher in priority than anything on the
644 * run-queue, we know that the previous high becomes our
645 * next-highest.
646 */
647 rt_rq->highest_prio.next = prev_prio;
648
649 if (rq->online)
650 cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
651
652 } else if (prio == rt_rq->highest_prio.curr)
653 /*
654 * If the next task is equal in priority to the highest on
655 * the run-queue, then we implicitly know that the next highest
656 * task cannot be any lower than current
657 */
658 rt_rq->highest_prio.next = prio;
659 else if (prio < rt_rq->highest_prio.next)
660 /*
661 * Otherwise, we need to recompute next-highest
662 */
663 rt_rq->highest_prio.next = next_prio(rq);
664 }
665
666 static void
667 dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
668 {
669 struct rq *rq = rq_of_rt_rq(rt_rq);
670
671 if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next))
672 rt_rq->highest_prio.next = next_prio(rq);
673
674 if (rq->online && rt_rq->highest_prio.curr != prev_prio)
675 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
676 }
677
678 #else /* CONFIG_SMP */
679
680 static inline
681 void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
682 static inline
683 void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
684
685 #endif /* CONFIG_SMP */
686
687 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
688 static void
689 inc_rt_prio(struct rt_rq *rt_rq, int prio)
690 {
691 int prev_prio = rt_rq->highest_prio.curr;
692
693 if (prio < prev_prio)
694 rt_rq->highest_prio.curr = prio;
695
696 inc_rt_prio_smp(rt_rq, prio, prev_prio);
697 }
698
699 static void
700 dec_rt_prio(struct rt_rq *rt_rq, int prio)
701 {
702 int prev_prio = rt_rq->highest_prio.curr;
703
704 if (rt_rq->rt_nr_running) {
705
706 WARN_ON(prio < prev_prio);
707
708 /*
709 * This may have been our highest task, and therefore
710 * we may have some recomputation to do
711 */
712 if (prio == prev_prio) {
713 struct rt_prio_array *array = &rt_rq->active;
714
715 rt_rq->highest_prio.curr =
716 sched_find_first_bit(array->bitmap);
717 }
718
719 } else
720 rt_rq->highest_prio.curr = MAX_RT_PRIO;
721
722 dec_rt_prio_smp(rt_rq, prio, prev_prio);
723 }
724
725 #else
726
727 static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
728 static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
729
730 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
731
732 #ifdef CONFIG_RT_GROUP_SCHED
733
734 static void
735 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
736 {
737 if (rt_se_boosted(rt_se))
738 rt_rq->rt_nr_boosted++;
739
740 if (rt_rq->tg)
741 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
742 }
743
744 static void
745 dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
746 {
747 if (rt_se_boosted(rt_se))
748 rt_rq->rt_nr_boosted--;
749
750 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
751 }
752
753 #else /* CONFIG_RT_GROUP_SCHED */
754
755 static void
756 inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
757 {
758 start_rt_bandwidth(&def_rt_bandwidth);
759 }
760
761 static inline
762 void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
763
764 #endif /* CONFIG_RT_GROUP_SCHED */
765
766 static inline
767 void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
768 {
769 int prio = rt_se_prio(rt_se);
770
771 WARN_ON(!rt_prio(prio));
772 rt_rq->rt_nr_running++;
773
774 inc_rt_prio(rt_rq, prio);
775 inc_rt_migration(rt_se, rt_rq);
776 inc_rt_group(rt_se, rt_rq);
777 }
778
779 static inline
780 void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
781 {
782 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
783 WARN_ON(!rt_rq->rt_nr_running);
784 rt_rq->rt_nr_running--;
785
786 dec_rt_prio(rt_rq, rt_se_prio(rt_se));
787 dec_rt_migration(rt_se, rt_rq);
788 dec_rt_group(rt_se, rt_rq);
789 }
790
791 static void __enqueue_rt_entity(struct sched_rt_entity *rt_se)
792 {
793 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
794 struct rt_prio_array *array = &rt_rq->active;
795 struct rt_rq *group_rq = group_rt_rq(rt_se);
796 struct list_head *queue = array->queue + rt_se_prio(rt_se);
797
798 /*
799 * Don't enqueue the group if its throttled, or when empty.
800 * The latter is a consequence of the former when a child group
801 * get throttled and the current group doesn't have any other
802 * active members.
803 */
804 if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
805 return;
806
807 list_add_tail(&rt_se->run_list, queue);
808 __set_bit(rt_se_prio(rt_se), array->bitmap);
809
810 inc_rt_tasks(rt_se, rt_rq);
811 }
812
813 static void __dequeue_rt_entity(struct sched_rt_entity *rt_se)
814 {
815 struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
816 struct rt_prio_array *array = &rt_rq->active;
817
818 list_del_init(&rt_se->run_list);
819 if (list_empty(array->queue + rt_se_prio(rt_se)))
820 __clear_bit(rt_se_prio(rt_se), array->bitmap);
821
822 dec_rt_tasks(rt_se, rt_rq);
823 }
824
825 /*
826 * Because the prio of an upper entry depends on the lower
827 * entries, we must remove entries top - down.
828 */
829 static void dequeue_rt_stack(struct sched_rt_entity *rt_se)
830 {
831 struct sched_rt_entity *back = NULL;
832
833 for_each_sched_rt_entity(rt_se) {
834 rt_se->back = back;
835 back = rt_se;
836 }
837
838 for (rt_se = back; rt_se; rt_se = rt_se->back) {
839 if (on_rt_rq(rt_se))
840 __dequeue_rt_entity(rt_se);
841 }
842 }
843
844 static void enqueue_rt_entity(struct sched_rt_entity *rt_se)
845 {
846 dequeue_rt_stack(rt_se);
847 for_each_sched_rt_entity(rt_se)
848 __enqueue_rt_entity(rt_se);
849 }
850
851 static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
852 {
853 dequeue_rt_stack(rt_se);
854
855 for_each_sched_rt_entity(rt_se) {
856 struct rt_rq *rt_rq = group_rt_rq(rt_se);
857
858 if (rt_rq && rt_rq->rt_nr_running)
859 __enqueue_rt_entity(rt_se);
860 }
861 }
862
863 /*
864 * Adding/removing a task to/from a priority array:
865 */
866 static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
867 {
868 struct sched_rt_entity *rt_se = &p->rt;
869
870 if (wakeup)
871 rt_se->timeout = 0;
872
873 enqueue_rt_entity(rt_se);
874
875 if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1)
876 enqueue_pushable_task(rq, p);
877
878 inc_cpu_load(rq, p->se.load.weight);
879 }
880
881 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
882 {
883 struct sched_rt_entity *rt_se = &p->rt;
884
885 update_curr_rt(rq);
886 dequeue_rt_entity(rt_se);
887
888 dequeue_pushable_task(rq, p);
889
890 dec_cpu_load(rq, p->se.load.weight);
891 }
892
893 /*
894 * Put task to the end of the run list without the overhead of dequeue
895 * followed by enqueue.
896 */
897 static void
898 requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
899 {
900 if (on_rt_rq(rt_se)) {
901 struct rt_prio_array *array = &rt_rq->active;
902 struct list_head *queue = array->queue + rt_se_prio(rt_se);
903
904 if (head)
905 list_move(&rt_se->run_list, queue);
906 else
907 list_move_tail(&rt_se->run_list, queue);
908 }
909 }
910
911 static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
912 {
913 struct sched_rt_entity *rt_se = &p->rt;
914 struct rt_rq *rt_rq;
915
916 for_each_sched_rt_entity(rt_se) {
917 rt_rq = rt_rq_of_se(rt_se);
918 requeue_rt_entity(rt_rq, rt_se, head);
919 }
920 }
921
922 static void yield_task_rt(struct rq *rq)
923 {
924 requeue_task_rt(rq, rq->curr, 0);
925 }
926
927 #ifdef CONFIG_SMP
928 static int find_lowest_rq(struct task_struct *task);
929
930 static int select_task_rq_rt(struct task_struct *p, int sync)
931 {
932 struct rq *rq = task_rq(p);
933
934 /*
935 * If the current task is an RT task, then
936 * try to see if we can wake this RT task up on another
937 * runqueue. Otherwise simply start this RT task
938 * on its current runqueue.
939 *
940 * We want to avoid overloading runqueues. Even if
941 * the RT task is of higher priority than the current RT task.
942 * RT tasks behave differently than other tasks. If
943 * one gets preempted, we try to push it off to another queue.
944 * So trying to keep a preempting RT task on the same
945 * cache hot CPU will force the running RT task to
946 * a cold CPU. So we waste all the cache for the lower
947 * RT task in hopes of saving some of a RT task
948 * that is just being woken and probably will have
949 * cold cache anyway.
950 */
951 if (unlikely(rt_task(rq->curr)) &&
952 (p->rt.nr_cpus_allowed > 1)) {
953 int cpu = find_lowest_rq(p);
954
955 return (cpu == -1) ? task_cpu(p) : cpu;
956 }
957
958 /*
959 * Otherwise, just let it ride on the affined RQ and the
960 * post-schedule router will push the preempted task away
961 */
962 return task_cpu(p);
963 }
964
965 static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
966 {
967 if (rq->curr->rt.nr_cpus_allowed == 1)
968 return;
969
970 if (p->rt.nr_cpus_allowed != 1
971 && cpupri_find(&rq->rd->cpupri, p, NULL))
972 return;
973
974 if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL))
975 return;
976
977 /*
978 * There appears to be other cpus that can accept
979 * current and none to run 'p', so lets reschedule
980 * to try and push current away:
981 */
982 requeue_task_rt(rq, p, 1);
983 resched_task(rq->curr);
984 }
985
986 #endif /* CONFIG_SMP */
987
988 /*
989 * Preempt the current task with a newly woken task if needed:
990 */
991 static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int sync)
992 {
993 if (p->prio < rq->curr->prio) {
994 resched_task(rq->curr);
995 return;
996 }
997
998 #ifdef CONFIG_SMP
999 /*
1000 * If:
1001 *
1002 * - the newly woken task is of equal priority to the current task
1003 * - the newly woken task is non-migratable while current is migratable
1004 * - current will be preempted on the next reschedule
1005 *
1006 * we should check to see if current can readily move to a different
1007 * cpu. If so, we will reschedule to allow the push logic to try
1008 * to move current somewhere else, making room for our non-migratable
1009 * task.
1010 */
1011 if (p->prio == rq->curr->prio && !need_resched())
1012 check_preempt_equal_prio(rq, p);
1013 #endif
1014 }
1015
1016 static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
1017 struct rt_rq *rt_rq)
1018 {
1019 struct rt_prio_array *array = &rt_rq->active;
1020 struct sched_rt_entity *next = NULL;
1021 struct list_head *queue;
1022 int idx;
1023
1024 idx = sched_find_first_bit(array->bitmap);
1025 BUG_ON(idx >= MAX_RT_PRIO);
1026
1027 queue = array->queue + idx;
1028 next = list_entry(queue->next, struct sched_rt_entity, run_list);
1029
1030 return next;
1031 }
1032
1033 static struct task_struct *_pick_next_task_rt(struct rq *rq)
1034 {
1035 struct sched_rt_entity *rt_se;
1036 struct task_struct *p;
1037 struct rt_rq *rt_rq;
1038
1039 rt_rq = &rq->rt;
1040
1041 if (unlikely(!rt_rq->rt_nr_running))
1042 return NULL;
1043
1044 if (rt_rq_throttled(rt_rq))
1045 return NULL;
1046
1047 do {
1048 rt_se = pick_next_rt_entity(rq, rt_rq);
1049 BUG_ON(!rt_se);
1050 rt_rq = group_rt_rq(rt_se);
1051 } while (rt_rq);
1052
1053 p = rt_task_of(rt_se);
1054 p->se.exec_start = rq->clock;
1055
1056 return p;
1057 }
1058
1059 static struct task_struct *pick_next_task_rt(struct rq *rq)
1060 {
1061 struct task_struct *p = _pick_next_task_rt(rq);
1062
1063 /* The running task is never eligible for pushing */
1064 if (p)
1065 dequeue_pushable_task(rq, p);
1066
1067 return p;
1068 }
1069
1070 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
1071 {
1072 update_curr_rt(rq);
1073 p->se.exec_start = 0;
1074
1075 /*
1076 * The previous task needs to be made eligible for pushing
1077 * if it is still active
1078 */
1079 if (p->se.on_rq && p->rt.nr_cpus_allowed > 1)
1080 enqueue_pushable_task(rq, p);
1081 }
1082
1083 #ifdef CONFIG_SMP
1084
1085 /* Only try algorithms three times */
1086 #define RT_MAX_TRIES 3
1087
1088 static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep);
1089
1090 static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
1091 {
1092 if (!task_running(rq, p) &&
1093 (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) &&
1094 (p->rt.nr_cpus_allowed > 1))
1095 return 1;
1096 return 0;
1097 }
1098
1099 /* Return the second highest RT task, NULL otherwise */
1100 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
1101 {
1102 struct task_struct *next = NULL;
1103 struct sched_rt_entity *rt_se;
1104 struct rt_prio_array *array;
1105 struct rt_rq *rt_rq;
1106 int idx;
1107
1108 for_each_leaf_rt_rq(rt_rq, rq) {
1109 array = &rt_rq->active;
1110 idx = sched_find_first_bit(array->bitmap);
1111 next_idx:
1112 if (idx >= MAX_RT_PRIO)
1113 continue;
1114 if (next && next->prio < idx)
1115 continue;
1116 list_for_each_entry(rt_se, array->queue + idx, run_list) {
1117 struct task_struct *p = rt_task_of(rt_se);
1118 if (pick_rt_task(rq, p, cpu)) {
1119 next = p;
1120 break;
1121 }
1122 }
1123 if (!next) {
1124 idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
1125 goto next_idx;
1126 }
1127 }
1128
1129 return next;
1130 }
1131
1132 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask);
1133
1134 static inline int pick_optimal_cpu(int this_cpu,
1135 const struct cpumask *mask)
1136 {
1137 int first;
1138
1139 /* "this_cpu" is cheaper to preempt than a remote processor */
1140 if ((this_cpu != -1) && cpumask_test_cpu(this_cpu, mask))
1141 return this_cpu;
1142
1143 first = cpumask_first(mask);
1144 if (first < nr_cpu_ids)
1145 return first;
1146
1147 return -1;
1148 }
1149
1150 static int find_lowest_rq(struct task_struct *task)
1151 {
1152 struct sched_domain *sd;
1153 struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask);
1154 int this_cpu = smp_processor_id();
1155 int cpu = task_cpu(task);
1156 cpumask_var_t domain_mask;
1157
1158 if (task->rt.nr_cpus_allowed == 1)
1159 return -1; /* No other targets possible */
1160
1161 if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask))
1162 return -1; /* No targets found */
1163
1164 /*
1165 * Only consider CPUs that are usable for migration.
1166 * I guess we might want to change cpupri_find() to ignore those
1167 * in the first place.
1168 */
1169 cpumask_and(lowest_mask, lowest_mask, cpu_active_mask);
1170
1171 /*
1172 * At this point we have built a mask of cpus representing the
1173 * lowest priority tasks in the system. Now we want to elect
1174 * the best one based on our affinity and topology.
1175 *
1176 * We prioritize the last cpu that the task executed on since
1177 * it is most likely cache-hot in that location.
1178 */
1179 if (cpumask_test_cpu(cpu, lowest_mask))
1180 return cpu;
1181
1182 /*
1183 * Otherwise, we consult the sched_domains span maps to figure
1184 * out which cpu is logically closest to our hot cache data.
1185 */
1186 if (this_cpu == cpu)
1187 this_cpu = -1; /* Skip this_cpu opt if the same */
1188
1189 if (alloc_cpumask_var(&domain_mask, GFP_ATOMIC)) {
1190 for_each_domain(cpu, sd) {
1191 if (sd->flags & SD_WAKE_AFFINE) {
1192 int best_cpu;
1193
1194 cpumask_and(domain_mask,
1195 sched_domain_span(sd),
1196 lowest_mask);
1197
1198 best_cpu = pick_optimal_cpu(this_cpu,
1199 domain_mask);
1200
1201 if (best_cpu != -1) {
1202 free_cpumask_var(domain_mask);
1203 return best_cpu;
1204 }
1205 }
1206 }
1207 free_cpumask_var(domain_mask);
1208 }
1209
1210 /*
1211 * And finally, if there were no matches within the domains
1212 * just give the caller *something* to work with from the compatible
1213 * locations.
1214 */
1215 return pick_optimal_cpu(this_cpu, lowest_mask);
1216 }
1217
1218 /* Will lock the rq it finds */
1219 static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1220 {
1221 struct rq *lowest_rq = NULL;
1222 int tries;
1223 int cpu;
1224
1225 for (tries = 0; tries < RT_MAX_TRIES; tries++) {
1226 cpu = find_lowest_rq(task);
1227
1228 if ((cpu == -1) || (cpu == rq->cpu))
1229 break;
1230
1231 lowest_rq = cpu_rq(cpu);
1232
1233 /* if the prio of this runqueue changed, try again */
1234 if (double_lock_balance(rq, lowest_rq)) {
1235 /*
1236 * We had to unlock the run queue. In
1237 * the mean time, task could have
1238 * migrated already or had its affinity changed.
1239 * Also make sure that it wasn't scheduled on its rq.
1240 */
1241 if (unlikely(task_rq(task) != rq ||
1242 !cpumask_test_cpu(lowest_rq->cpu,
1243 &task->cpus_allowed) ||
1244 task_running(rq, task) ||
1245 !task->se.on_rq)) {
1246
1247 spin_unlock(&lowest_rq->lock);
1248 lowest_rq = NULL;
1249 break;
1250 }
1251 }
1252
1253 /* If this rq is still suitable use it. */
1254 if (lowest_rq->rt.highest_prio.curr > task->prio)
1255 break;
1256
1257 /* try again */
1258 double_unlock_balance(rq, lowest_rq);
1259 lowest_rq = NULL;
1260 }
1261
1262 return lowest_rq;
1263 }
1264
1265 static inline int has_pushable_tasks(struct rq *rq)
1266 {
1267 return !plist_head_empty(&rq->rt.pushable_tasks);
1268 }
1269
1270 static struct task_struct *pick_next_pushable_task(struct rq *rq)
1271 {
1272 struct task_struct *p;
1273
1274 if (!has_pushable_tasks(rq))
1275 return NULL;
1276
1277 p = plist_first_entry(&rq->rt.pushable_tasks,
1278 struct task_struct, pushable_tasks);
1279
1280 BUG_ON(rq->cpu != task_cpu(p));
1281 BUG_ON(task_current(rq, p));
1282 BUG_ON(p->rt.nr_cpus_allowed <= 1);
1283
1284 BUG_ON(!p->se.on_rq);
1285 BUG_ON(!rt_task(p));
1286
1287 return p;
1288 }
1289
1290 /*
1291 * If the current CPU has more than one RT task, see if the non
1292 * running task can migrate over to a CPU that is running a task
1293 * of lesser priority.
1294 */
1295 static int push_rt_task(struct rq *rq)
1296 {
1297 struct task_struct *next_task;
1298 struct rq *lowest_rq;
1299
1300 if (!rq->rt.overloaded)
1301 return 0;
1302
1303 next_task = pick_next_pushable_task(rq);
1304 if (!next_task)
1305 return 0;
1306
1307 retry:
1308 if (unlikely(next_task == rq->curr)) {
1309 WARN_ON(1);
1310 return 0;
1311 }
1312
1313 /*
1314 * It's possible that the next_task slipped in of
1315 * higher priority than current. If that's the case
1316 * just reschedule current.
1317 */
1318 if (unlikely(next_task->prio < rq->curr->prio)) {
1319 resched_task(rq->curr);
1320 return 0;
1321 }
1322
1323 /* We might release rq lock */
1324 get_task_struct(next_task);
1325
1326 /* find_lock_lowest_rq locks the rq if found */
1327 lowest_rq = find_lock_lowest_rq(next_task, rq);
1328 if (!lowest_rq) {
1329 struct task_struct *task;
1330 /*
1331 * find lock_lowest_rq releases rq->lock
1332 * so it is possible that next_task has migrated.
1333 *
1334 * We need to make sure that the task is still on the same
1335 * run-queue and is also still the next task eligible for
1336 * pushing.
1337 */
1338 task = pick_next_pushable_task(rq);
1339 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1340 /*
1341 * If we get here, the task hasnt moved at all, but
1342 * it has failed to push. We will not try again,
1343 * since the other cpus will pull from us when they
1344 * are ready.
1345 */
1346 dequeue_pushable_task(rq, next_task);
1347 goto out;
1348 }
1349
1350 if (!task)
1351 /* No more tasks, just exit */
1352 goto out;
1353
1354 /*
1355 * Something has shifted, try again.
1356 */
1357 put_task_struct(next_task);
1358 next_task = task;
1359 goto retry;
1360 }
1361
1362 deactivate_task(rq, next_task, 0);
1363 set_task_cpu(next_task, lowest_rq->cpu);
1364 activate_task(lowest_rq, next_task, 0);
1365
1366 resched_task(lowest_rq->curr);
1367
1368 double_unlock_balance(rq, lowest_rq);
1369
1370 out:
1371 put_task_struct(next_task);
1372
1373 return 1;
1374 }
1375
1376 static void push_rt_tasks(struct rq *rq)
1377 {
1378 /* push_rt_task will return true if it moved an RT */
1379 while (push_rt_task(rq))
1380 ;
1381 }
1382
1383 static int pull_rt_task(struct rq *this_rq)
1384 {
1385 int this_cpu = this_rq->cpu, ret = 0, cpu;
1386 struct task_struct *p;
1387 struct rq *src_rq;
1388
1389 if (likely(!rt_overloaded(this_rq)))
1390 return 0;
1391
1392 for_each_cpu(cpu, this_rq->rd->rto_mask) {
1393 if (this_cpu == cpu)
1394 continue;
1395
1396 src_rq = cpu_rq(cpu);
1397
1398 /*
1399 * Don't bother taking the src_rq->lock if the next highest
1400 * task is known to be lower-priority than our current task.
1401 * This may look racy, but if this value is about to go
1402 * logically higher, the src_rq will push this task away.
1403 * And if its going logically lower, we do not care
1404 */
1405 if (src_rq->rt.highest_prio.next >=
1406 this_rq->rt.highest_prio.curr)
1407 continue;
1408
1409 /*
1410 * We can potentially drop this_rq's lock in
1411 * double_lock_balance, and another CPU could
1412 * alter this_rq
1413 */
1414 double_lock_balance(this_rq, src_rq);
1415
1416 /*
1417 * Are there still pullable RT tasks?
1418 */
1419 if (src_rq->rt.rt_nr_running <= 1)
1420 goto skip;
1421
1422 p = pick_next_highest_task_rt(src_rq, this_cpu);
1423
1424 /*
1425 * Do we have an RT task that preempts
1426 * the to-be-scheduled task?
1427 */
1428 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
1429 WARN_ON(p == src_rq->curr);
1430 WARN_ON(!p->se.on_rq);
1431
1432 /*
1433 * There's a chance that p is higher in priority
1434 * than what's currently running on its cpu.
1435 * This is just that p is wakeing up and hasn't
1436 * had a chance to schedule. We only pull
1437 * p if it is lower in priority than the
1438 * current task on the run queue
1439 */
1440 if (p->prio < src_rq->curr->prio)
1441 goto skip;
1442
1443 ret = 1;
1444
1445 deactivate_task(src_rq, p, 0);
1446 set_task_cpu(p, this_cpu);
1447 activate_task(this_rq, p, 0);
1448 /*
1449 * We continue with the search, just in
1450 * case there's an even higher prio task
1451 * in another runqueue. (low likelyhood
1452 * but possible)
1453 */
1454 }
1455 skip:
1456 double_unlock_balance(this_rq, src_rq);
1457 }
1458
1459 return ret;
1460 }
1461
1462 static void pre_schedule_rt(struct rq *rq, struct task_struct *prev)
1463 {
1464 /* Try to pull RT tasks here if we lower this rq's prio */
1465 if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio)
1466 pull_rt_task(rq);
1467 }
1468
1469 /*
1470 * assumes rq->lock is held
1471 */
1472 static int needs_post_schedule_rt(struct rq *rq)
1473 {
1474 return has_pushable_tasks(rq);
1475 }
1476
1477 static void post_schedule_rt(struct rq *rq)
1478 {
1479 /*
1480 * This is only called if needs_post_schedule_rt() indicates that
1481 * we need to push tasks away
1482 */
1483 spin_lock_irq(&rq->lock);
1484 push_rt_tasks(rq);
1485 spin_unlock_irq(&rq->lock);
1486 }
1487
1488 /*
1489 * If we are not running and we are not going to reschedule soon, we should
1490 * try to push tasks away now
1491 */
1492 static void task_wake_up_rt(struct rq *rq, struct task_struct *p)
1493 {
1494 if (!task_running(rq, p) &&
1495 !test_tsk_need_resched(rq->curr) &&
1496 has_pushable_tasks(rq) &&
1497 p->rt.nr_cpus_allowed > 1)
1498 push_rt_tasks(rq);
1499 }
1500
1501 static unsigned long
1502 load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
1503 unsigned long max_load_move,
1504 struct sched_domain *sd, enum cpu_idle_type idle,
1505 int *all_pinned, int *this_best_prio)
1506 {
1507 /* don't touch RT tasks */
1508 return 0;
1509 }
1510
1511 static int
1512 move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
1513 struct sched_domain *sd, enum cpu_idle_type idle)
1514 {
1515 /* don't touch RT tasks */
1516 return 0;
1517 }
1518
1519 static void set_cpus_allowed_rt(struct task_struct *p,
1520 const struct cpumask *new_mask)
1521 {
1522 int weight = cpumask_weight(new_mask);
1523
1524 BUG_ON(!rt_task(p));
1525
1526 /*
1527 * Update the migration status of the RQ if we have an RT task
1528 * which is running AND changing its weight value.
1529 */
1530 if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) {
1531 struct rq *rq = task_rq(p);
1532
1533 if (!task_current(rq, p)) {
1534 /*
1535 * Make sure we dequeue this task from the pushable list
1536 * before going further. It will either remain off of
1537 * the list because we are no longer pushable, or it
1538 * will be requeued.
1539 */
1540 if (p->rt.nr_cpus_allowed > 1)
1541 dequeue_pushable_task(rq, p);
1542
1543 /*
1544 * Requeue if our weight is changing and still > 1
1545 */
1546 if (weight > 1)
1547 enqueue_pushable_task(rq, p);
1548
1549 }
1550
1551 if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) {
1552 rq->rt.rt_nr_migratory++;
1553 } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) {
1554 BUG_ON(!rq->rt.rt_nr_migratory);
1555 rq->rt.rt_nr_migratory--;
1556 }
1557
1558 update_rt_migration(&rq->rt);
1559 }
1560
1561 cpumask_copy(&p->cpus_allowed, new_mask);
1562 p->rt.nr_cpus_allowed = weight;
1563 }
1564
1565 /* Assumes rq->lock is held */
1566 static void rq_online_rt(struct rq *rq)
1567 {
1568 if (rq->rt.overloaded)
1569 rt_set_overload(rq);
1570
1571 __enable_runtime(rq);
1572
1573 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1574 }
1575
1576 /* Assumes rq->lock is held */
1577 static void rq_offline_rt(struct rq *rq)
1578 {
1579 if (rq->rt.overloaded)
1580 rt_clear_overload(rq);
1581
1582 __disable_runtime(rq);
1583
1584 cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID);
1585 }
1586
1587 /*
1588 * When switch from the rt queue, we bring ourselves to a position
1589 * that we might want to pull RT tasks from other runqueues.
1590 */
1591 static void switched_from_rt(struct rq *rq, struct task_struct *p,
1592 int running)
1593 {
1594 /*
1595 * If there are other RT tasks then we will reschedule
1596 * and the scheduling of the other RT tasks will handle
1597 * the balancing. But if we are the last RT task
1598 * we may need to handle the pulling of RT tasks
1599 * now.
1600 */
1601 if (!rq->rt.rt_nr_running)
1602 pull_rt_task(rq);
1603 }
1604
1605 static inline void init_sched_rt_class(void)
1606 {
1607 unsigned int i;
1608
1609 for_each_possible_cpu(i)
1610 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i),
1611 GFP_KERNEL, cpu_to_node(i));
1612 }
1613 #endif /* CONFIG_SMP */
1614
1615 /*
1616 * When switching a task to RT, we may overload the runqueue
1617 * with RT tasks. In this case we try to push them off to
1618 * other runqueues.
1619 */
1620 static void switched_to_rt(struct rq *rq, struct task_struct *p,
1621 int running)
1622 {
1623 int check_resched = 1;
1624
1625 /*
1626 * If we are already running, then there's nothing
1627 * that needs to be done. But if we are not running
1628 * we may need to preempt the current running task.
1629 * If that current running task is also an RT task
1630 * then see if we can move to another run queue.
1631 */
1632 if (!running) {
1633 #ifdef CONFIG_SMP
1634 if (rq->rt.overloaded && push_rt_task(rq) &&
1635 /* Don't resched if we changed runqueues */
1636 rq != task_rq(p))
1637 check_resched = 0;
1638 #endif /* CONFIG_SMP */
1639 if (check_resched && p->prio < rq->curr->prio)
1640 resched_task(rq->curr);
1641 }
1642 }
1643
1644 /*
1645 * Priority of the task has changed. This may cause
1646 * us to initiate a push or pull.
1647 */
1648 static void prio_changed_rt(struct rq *rq, struct task_struct *p,
1649 int oldprio, int running)
1650 {
1651 if (running) {
1652 #ifdef CONFIG_SMP
1653 /*
1654 * If our priority decreases while running, we
1655 * may need to pull tasks to this runqueue.
1656 */
1657 if (oldprio < p->prio)
1658 pull_rt_task(rq);
1659 /*
1660 * If there's a higher priority task waiting to run
1661 * then reschedule. Note, the above pull_rt_task
1662 * can release the rq lock and p could migrate.
1663 * Only reschedule if p is still on the same runqueue.
1664 */
1665 if (p->prio > rq->rt.highest_prio.curr && rq->curr == p)
1666 resched_task(p);
1667 #else
1668 /* For UP simply resched on drop of prio */
1669 if (oldprio < p->prio)
1670 resched_task(p);
1671 #endif /* CONFIG_SMP */
1672 } else {
1673 /*
1674 * This task is not running, but if it is
1675 * greater than the current running task
1676 * then reschedule.
1677 */
1678 if (p->prio < rq->curr->prio)
1679 resched_task(rq->curr);
1680 }
1681 }
1682
1683 static void watchdog(struct rq *rq, struct task_struct *p)
1684 {
1685 unsigned long soft, hard;
1686
1687 if (!p->signal)
1688 return;
1689
1690 soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur;
1691 hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max;
1692
1693 if (soft != RLIM_INFINITY) {
1694 unsigned long next;
1695
1696 p->rt.timeout++;
1697 next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ);
1698 if (p->rt.timeout > next)
1699 p->cputime_expires.sched_exp = p->se.sum_exec_runtime;
1700 }
1701 }
1702
1703 static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
1704 {
1705 update_curr_rt(rq);
1706
1707 watchdog(rq, p);
1708
1709 /*
1710 * RR tasks need a special form of timeslice management.
1711 * FIFO tasks have no timeslices.
1712 */
1713 if (p->policy != SCHED_RR)
1714 return;
1715
1716 if (--p->rt.time_slice)
1717 return;
1718
1719 p->rt.time_slice = DEF_TIMESLICE;
1720
1721 /*
1722 * Requeue to the end of queue if we are not the only element
1723 * on the queue:
1724 */
1725 if (p->rt.run_list.prev != p->rt.run_list.next) {
1726 requeue_task_rt(rq, p, 0);
1727 set_tsk_need_resched(p);
1728 }
1729 }
1730
1731 static void set_curr_task_rt(struct rq *rq)
1732 {
1733 struct task_struct *p = rq->curr;
1734
1735 p->se.exec_start = rq->clock;
1736
1737 /* The running task is never eligible for pushing */
1738 dequeue_pushable_task(rq, p);
1739 }
1740
1741 static const struct sched_class rt_sched_class = {
1742 .next = &fair_sched_class,
1743 .enqueue_task = enqueue_task_rt,
1744 .dequeue_task = dequeue_task_rt,
1745 .yield_task = yield_task_rt,
1746
1747 .check_preempt_curr = check_preempt_curr_rt,
1748
1749 .pick_next_task = pick_next_task_rt,
1750 .put_prev_task = put_prev_task_rt,
1751
1752 #ifdef CONFIG_SMP
1753 .select_task_rq = select_task_rq_rt,
1754
1755 .load_balance = load_balance_rt,
1756 .move_one_task = move_one_task_rt,
1757 .set_cpus_allowed = set_cpus_allowed_rt,
1758 .rq_online = rq_online_rt,
1759 .rq_offline = rq_offline_rt,
1760 .pre_schedule = pre_schedule_rt,
1761 .needs_post_schedule = needs_post_schedule_rt,
1762 .post_schedule = post_schedule_rt,
1763 .task_wake_up = task_wake_up_rt,
1764 .switched_from = switched_from_rt,
1765 #endif
1766
1767 .set_curr_task = set_curr_task_rt,
1768 .task_tick = task_tick_rt,
1769
1770 .prio_changed = prio_changed_rt,
1771 .switched_to = switched_to_rt,
1772 };
1773
1774 #ifdef CONFIG_SCHED_DEBUG
1775 extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq);
1776
1777 static void print_rt_stats(struct seq_file *m, int cpu)
1778 {
1779 struct rt_rq *rt_rq;
1780
1781 rcu_read_lock();
1782 for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu))
1783 print_rt_rq(m, cpu, rt_rq);
1784 rcu_read_unlock();
1785 }
1786 #endif /* CONFIG_SCHED_DEBUG */
1787
1788
|
This page was automatically generated by the
LXR engine.
|