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