Root/
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 | typedef struct task_group *rt_rq_iter_t; |
187 | |
188 | static inline struct task_group *next_task_group(struct task_group *tg) |
189 | { |
190 | do { |
191 | tg = list_entry_rcu(tg->list.next, |
192 | typeof(struct task_group), list); |
193 | } while (&tg->list != &task_groups && task_group_is_autogroup(tg)); |
194 | |
195 | if (&tg->list == &task_groups) |
196 | tg = NULL; |
197 | |
198 | return tg; |
199 | } |
200 | |
201 | #define for_each_rt_rq(rt_rq, iter, rq) \ |
202 | for (iter = container_of(&task_groups, typeof(*iter), list); \ |
203 | (iter = next_task_group(iter)) && \ |
204 | (rt_rq = iter->rt_rq[cpu_of(rq)]);) |
205 | |
206 | static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq) |
207 | { |
208 | list_add_rcu(&rt_rq->leaf_rt_rq_list, |
209 | &rq_of_rt_rq(rt_rq)->leaf_rt_rq_list); |
210 | } |
211 | |
212 | static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq) |
213 | { |
214 | list_del_rcu(&rt_rq->leaf_rt_rq_list); |
215 | } |
216 | |
217 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
218 | list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) |
219 | |
220 | #define for_each_sched_rt_entity(rt_se) \ |
221 | for (; rt_se; rt_se = rt_se->parent) |
222 | |
223 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
224 | { |
225 | return rt_se->my_q; |
226 | } |
227 | |
228 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
229 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
230 | |
231 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
232 | { |
233 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
234 | struct sched_rt_entity *rt_se; |
235 | |
236 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
237 | |
238 | rt_se = rt_rq->tg->rt_se[cpu]; |
239 | |
240 | if (rt_rq->rt_nr_running) { |
241 | if (rt_se && !on_rt_rq(rt_se)) |
242 | enqueue_rt_entity(rt_se, false); |
243 | if (rt_rq->highest_prio.curr < curr->prio) |
244 | resched_task(curr); |
245 | } |
246 | } |
247 | |
248 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
249 | { |
250 | struct sched_rt_entity *rt_se; |
251 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
252 | |
253 | rt_se = rt_rq->tg->rt_se[cpu]; |
254 | |
255 | if (rt_se && on_rt_rq(rt_se)) |
256 | dequeue_rt_entity(rt_se); |
257 | } |
258 | |
259 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
260 | { |
261 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; |
262 | } |
263 | |
264 | static int rt_se_boosted(struct sched_rt_entity *rt_se) |
265 | { |
266 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
267 | struct task_struct *p; |
268 | |
269 | if (rt_rq) |
270 | return !!rt_rq->rt_nr_boosted; |
271 | |
272 | p = rt_task_of(rt_se); |
273 | return p->prio != p->normal_prio; |
274 | } |
275 | |
276 | #ifdef CONFIG_SMP |
277 | static inline const struct cpumask *sched_rt_period_mask(void) |
278 | { |
279 | return cpu_rq(smp_processor_id())->rd->span; |
280 | } |
281 | #else |
282 | static inline const struct cpumask *sched_rt_period_mask(void) |
283 | { |
284 | return cpu_online_mask; |
285 | } |
286 | #endif |
287 | |
288 | static inline |
289 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) |
290 | { |
291 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
292 | } |
293 | |
294 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
295 | { |
296 | return &rt_rq->tg->rt_bandwidth; |
297 | } |
298 | |
299 | #else /* !CONFIG_RT_GROUP_SCHED */ |
300 | |
301 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
302 | { |
303 | return rt_rq->rt_runtime; |
304 | } |
305 | |
306 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) |
307 | { |
308 | return ktime_to_ns(def_rt_bandwidth.rt_period); |
309 | } |
310 | |
311 | typedef struct rt_rq *rt_rq_iter_t; |
312 | |
313 | #define for_each_rt_rq(rt_rq, iter, rq) \ |
314 | for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL) |
315 | |
316 | static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq) |
317 | { |
318 | } |
319 | |
320 | static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq) |
321 | { |
322 | } |
323 | |
324 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
325 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) |
326 | |
327 | #define for_each_sched_rt_entity(rt_se) \ |
328 | for (; rt_se; rt_se = NULL) |
329 | |
330 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) |
331 | { |
332 | return NULL; |
333 | } |
334 | |
335 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
336 | { |
337 | if (rt_rq->rt_nr_running) |
338 | resched_task(rq_of_rt_rq(rt_rq)->curr); |
339 | } |
340 | |
341 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
342 | { |
343 | } |
344 | |
345 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
346 | { |
347 | return rt_rq->rt_throttled; |
348 | } |
349 | |
350 | static inline const struct cpumask *sched_rt_period_mask(void) |
351 | { |
352 | return cpu_online_mask; |
353 | } |
354 | |
355 | static inline |
356 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) |
357 | { |
358 | return &cpu_rq(cpu)->rt; |
359 | } |
360 | |
361 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
362 | { |
363 | return &def_rt_bandwidth; |
364 | } |
365 | |
366 | #endif /* CONFIG_RT_GROUP_SCHED */ |
367 | |
368 | #ifdef CONFIG_SMP |
369 | /* |
370 | * We ran out of runtime, see if we can borrow some from our neighbours. |
371 | */ |
372 | static int do_balance_runtime(struct rt_rq *rt_rq) |
373 | { |
374 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
375 | struct root_domain *rd = cpu_rq(smp_processor_id())->rd; |
376 | int i, weight, more = 0; |
377 | u64 rt_period; |
378 | |
379 | weight = cpumask_weight(rd->span); |
380 | |
381 | raw_spin_lock(&rt_b->rt_runtime_lock); |
382 | rt_period = ktime_to_ns(rt_b->rt_period); |
383 | for_each_cpu(i, rd->span) { |
384 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
385 | s64 diff; |
386 | |
387 | if (iter == rt_rq) |
388 | continue; |
389 | |
390 | raw_spin_lock(&iter->rt_runtime_lock); |
391 | /* |
392 | * Either all rqs have inf runtime and there's nothing to steal |
393 | * or __disable_runtime() below sets a specific rq to inf to |
394 | * indicate its been disabled and disalow stealing. |
395 | */ |
396 | if (iter->rt_runtime == RUNTIME_INF) |
397 | goto next; |
398 | |
399 | /* |
400 | * From runqueues with spare time, take 1/n part of their |
401 | * spare time, but no more than our period. |
402 | */ |
403 | diff = iter->rt_runtime - iter->rt_time; |
404 | if (diff > 0) { |
405 | diff = div_u64((u64)diff, weight); |
406 | if (rt_rq->rt_runtime + diff > rt_period) |
407 | diff = rt_period - rt_rq->rt_runtime; |
408 | iter->rt_runtime -= diff; |
409 | rt_rq->rt_runtime += diff; |
410 | more = 1; |
411 | if (rt_rq->rt_runtime == rt_period) { |
412 | raw_spin_unlock(&iter->rt_runtime_lock); |
413 | break; |
414 | } |
415 | } |
416 | next: |
417 | raw_spin_unlock(&iter->rt_runtime_lock); |
418 | } |
419 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
420 | |
421 | return more; |
422 | } |
423 | |
424 | /* |
425 | * Ensure this RQ takes back all the runtime it lend to its neighbours. |
426 | */ |
427 | static void __disable_runtime(struct rq *rq) |
428 | { |
429 | struct root_domain *rd = rq->rd; |
430 | rt_rq_iter_t iter; |
431 | struct rt_rq *rt_rq; |
432 | |
433 | if (unlikely(!scheduler_running)) |
434 | return; |
435 | |
436 | for_each_rt_rq(rt_rq, iter, rq) { |
437 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
438 | s64 want; |
439 | int i; |
440 | |
441 | raw_spin_lock(&rt_b->rt_runtime_lock); |
442 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
443 | /* |
444 | * Either we're all inf and nobody needs to borrow, or we're |
445 | * already disabled and thus have nothing to do, or we have |
446 | * exactly the right amount of runtime to take out. |
447 | */ |
448 | if (rt_rq->rt_runtime == RUNTIME_INF || |
449 | rt_rq->rt_runtime == rt_b->rt_runtime) |
450 | goto balanced; |
451 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
452 | |
453 | /* |
454 | * Calculate the difference between what we started out with |
455 | * and what we current have, that's the amount of runtime |
456 | * we lend and now have to reclaim. |
457 | */ |
458 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
459 | |
460 | /* |
461 | * Greedy reclaim, take back as much as we can. |
462 | */ |
463 | for_each_cpu(i, rd->span) { |
464 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
465 | s64 diff; |
466 | |
467 | /* |
468 | * Can't reclaim from ourselves or disabled runqueues. |
469 | */ |
470 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
471 | continue; |
472 | |
473 | raw_spin_lock(&iter->rt_runtime_lock); |
474 | if (want > 0) { |
475 | diff = min_t(s64, iter->rt_runtime, want); |
476 | iter->rt_runtime -= diff; |
477 | want -= diff; |
478 | } else { |
479 | iter->rt_runtime -= want; |
480 | want -= want; |
481 | } |
482 | raw_spin_unlock(&iter->rt_runtime_lock); |
483 | |
484 | if (!want) |
485 | break; |
486 | } |
487 | |
488 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
489 | /* |
490 | * We cannot be left wanting - that would mean some runtime |
491 | * leaked out of the system. |
492 | */ |
493 | BUG_ON(want); |
494 | balanced: |
495 | /* |
496 | * Disable all the borrow logic by pretending we have inf |
497 | * runtime - in which case borrowing doesn't make sense. |
498 | */ |
499 | rt_rq->rt_runtime = RUNTIME_INF; |
500 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
501 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
502 | } |
503 | } |
504 | |
505 | static void disable_runtime(struct rq *rq) |
506 | { |
507 | unsigned long flags; |
508 | |
509 | raw_spin_lock_irqsave(&rq->lock, flags); |
510 | __disable_runtime(rq); |
511 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
512 | } |
513 | |
514 | static void __enable_runtime(struct rq *rq) |
515 | { |
516 | rt_rq_iter_t iter; |
517 | struct rt_rq *rt_rq; |
518 | |
519 | if (unlikely(!scheduler_running)) |
520 | return; |
521 | |
522 | /* |
523 | * Reset each runqueue's bandwidth settings |
524 | */ |
525 | for_each_rt_rq(rt_rq, iter, rq) { |
526 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
527 | |
528 | raw_spin_lock(&rt_b->rt_runtime_lock); |
529 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
530 | rt_rq->rt_runtime = rt_b->rt_runtime; |
531 | rt_rq->rt_time = 0; |
532 | rt_rq->rt_throttled = 0; |
533 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
534 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
535 | } |
536 | } |
537 | |
538 | static void enable_runtime(struct rq *rq) |
539 | { |
540 | unsigned long flags; |
541 | |
542 | raw_spin_lock_irqsave(&rq->lock, flags); |
543 | __enable_runtime(rq); |
544 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
545 | } |
546 | |
547 | static int balance_runtime(struct rt_rq *rt_rq) |
548 | { |
549 | int more = 0; |
550 | |
551 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
552 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
553 | more = do_balance_runtime(rt_rq); |
554 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
555 | } |
556 | |
557 | return more; |
558 | } |
559 | #else /* !CONFIG_SMP */ |
560 | static inline int balance_runtime(struct rt_rq *rt_rq) |
561 | { |
562 | return 0; |
563 | } |
564 | #endif /* CONFIG_SMP */ |
565 | |
566 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
567 | { |
568 | int i, idle = 1; |
569 | const struct cpumask *span; |
570 | |
571 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
572 | return 1; |
573 | |
574 | span = sched_rt_period_mask(); |
575 | for_each_cpu(i, span) { |
576 | int enqueue = 0; |
577 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); |
578 | struct rq *rq = rq_of_rt_rq(rt_rq); |
579 | |
580 | raw_spin_lock(&rq->lock); |
581 | if (rt_rq->rt_time) { |
582 | u64 runtime; |
583 | |
584 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
585 | if (rt_rq->rt_throttled) |
586 | balance_runtime(rt_rq); |
587 | runtime = rt_rq->rt_runtime; |
588 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); |
589 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { |
590 | rt_rq->rt_throttled = 0; |
591 | enqueue = 1; |
592 | |
593 | /* |
594 | * Force a clock update if the CPU was idle, |
595 | * lest wakeup -> unthrottle time accumulate. |
596 | */ |
597 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) |
598 | rq->skip_clock_update = -1; |
599 | } |
600 | if (rt_rq->rt_time || rt_rq->rt_nr_running) |
601 | idle = 0; |
602 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
603 | } else if (rt_rq->rt_nr_running) { |
604 | idle = 0; |
605 | if (!rt_rq_throttled(rt_rq)) |
606 | enqueue = 1; |
607 | } |
608 | |
609 | if (enqueue) |
610 | sched_rt_rq_enqueue(rt_rq); |
611 | raw_spin_unlock(&rq->lock); |
612 | } |
613 | |
614 | return idle; |
615 | } |
616 | |
617 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
618 | { |
619 | #ifdef CONFIG_RT_GROUP_SCHED |
620 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
621 | |
622 | if (rt_rq) |
623 | return rt_rq->highest_prio.curr; |
624 | #endif |
625 | |
626 | return rt_task_of(rt_se)->prio; |
627 | } |
628 | |
629 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
630 | { |
631 | u64 runtime = sched_rt_runtime(rt_rq); |
632 | |
633 | if (rt_rq->rt_throttled) |
634 | return rt_rq_throttled(rt_rq); |
635 | |
636 | if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq)) |
637 | return 0; |
638 | |
639 | balance_runtime(rt_rq); |
640 | runtime = sched_rt_runtime(rt_rq); |
641 | if (runtime == RUNTIME_INF) |
642 | return 0; |
643 | |
644 | if (rt_rq->rt_time > runtime) { |
645 | rt_rq->rt_throttled = 1; |
646 | if (rt_rq_throttled(rt_rq)) { |
647 | sched_rt_rq_dequeue(rt_rq); |
648 | return 1; |
649 | } |
650 | } |
651 | |
652 | return 0; |
653 | } |
654 | |
655 | /* |
656 | * Update the current task's runtime statistics. Skip current tasks that |
657 | * are not in our scheduling class. |
658 | */ |
659 | static void update_curr_rt(struct rq *rq) |
660 | { |
661 | struct task_struct *curr = rq->curr; |
662 | struct sched_rt_entity *rt_se = &curr->rt; |
663 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
664 | u64 delta_exec; |
665 | |
666 | if (curr->sched_class != &rt_sched_class) |
667 | return; |
668 | |
669 | delta_exec = rq->clock_task - curr->se.exec_start; |
670 | if (unlikely((s64)delta_exec < 0)) |
671 | delta_exec = 0; |
672 | |
673 | schedstat_set(curr->se.statistics.exec_max, max(curr->se.statistics.exec_max, delta_exec)); |
674 | |
675 | curr->se.sum_exec_runtime += delta_exec; |
676 | account_group_exec_runtime(curr, delta_exec); |
677 | |
678 | curr->se.exec_start = rq->clock_task; |
679 | cpuacct_charge(curr, delta_exec); |
680 | |
681 | sched_rt_avg_update(rq, delta_exec); |
682 | |
683 | if (!rt_bandwidth_enabled()) |
684 | return; |
685 | |
686 | for_each_sched_rt_entity(rt_se) { |
687 | rt_rq = rt_rq_of_se(rt_se); |
688 | |
689 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
690 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
691 | rt_rq->rt_time += delta_exec; |
692 | if (sched_rt_runtime_exceeded(rt_rq)) |
693 | resched_task(curr); |
694 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
695 | } |
696 | } |
697 | } |
698 | |
699 | #if defined CONFIG_SMP |
700 | |
701 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu); |
702 | |
703 | static inline int next_prio(struct rq *rq) |
704 | { |
705 | struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu); |
706 | |
707 | if (next && rt_prio(next->prio)) |
708 | return next->prio; |
709 | else |
710 | return MAX_RT_PRIO; |
711 | } |
712 | |
713 | static void |
714 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) |
715 | { |
716 | struct rq *rq = rq_of_rt_rq(rt_rq); |
717 | |
718 | if (prio < prev_prio) { |
719 | |
720 | /* |
721 | * If the new task is higher in priority than anything on the |
722 | * run-queue, we know that the previous high becomes our |
723 | * next-highest. |
724 | */ |
725 | rt_rq->highest_prio.next = prev_prio; |
726 | |
727 | if (rq->online) |
728 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); |
729 | |
730 | } else if (prio == rt_rq->highest_prio.curr) |
731 | /* |
732 | * If the next task is equal in priority to the highest on |
733 | * the run-queue, then we implicitly know that the next highest |
734 | * task cannot be any lower than current |
735 | */ |
736 | rt_rq->highest_prio.next = prio; |
737 | else if (prio < rt_rq->highest_prio.next) |
738 | /* |
739 | * Otherwise, we need to recompute next-highest |
740 | */ |
741 | rt_rq->highest_prio.next = next_prio(rq); |
742 | } |
743 | |
744 | static void |
745 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) |
746 | { |
747 | struct rq *rq = rq_of_rt_rq(rt_rq); |
748 | |
749 | if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next)) |
750 | rt_rq->highest_prio.next = next_prio(rq); |
751 | |
752 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
753 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); |
754 | } |
755 | |
756 | #else /* CONFIG_SMP */ |
757 | |
758 | static inline |
759 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
760 | static inline |
761 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
762 | |
763 | #endif /* CONFIG_SMP */ |
764 | |
765 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
766 | static void |
767 | inc_rt_prio(struct rt_rq *rt_rq, int prio) |
768 | { |
769 | int prev_prio = rt_rq->highest_prio.curr; |
770 | |
771 | if (prio < prev_prio) |
772 | rt_rq->highest_prio.curr = prio; |
773 | |
774 | inc_rt_prio_smp(rt_rq, prio, prev_prio); |
775 | } |
776 | |
777 | static void |
778 | dec_rt_prio(struct rt_rq *rt_rq, int prio) |
779 | { |
780 | int prev_prio = rt_rq->highest_prio.curr; |
781 | |
782 | if (rt_rq->rt_nr_running) { |
783 | |
784 | WARN_ON(prio < prev_prio); |
785 | |
786 | /* |
787 | * This may have been our highest task, and therefore |
788 | * we may have some recomputation to do |
789 | */ |
790 | if (prio == prev_prio) { |
791 | struct rt_prio_array *array = &rt_rq->active; |
792 | |
793 | rt_rq->highest_prio.curr = |
794 | sched_find_first_bit(array->bitmap); |
795 | } |
796 | |
797 | } else |
798 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
799 | |
800 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
801 | } |
802 | |
803 | #else |
804 | |
805 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} |
806 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} |
807 | |
808 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ |
809 | |
810 | #ifdef CONFIG_RT_GROUP_SCHED |
811 | |
812 | static void |
813 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
814 | { |
815 | if (rt_se_boosted(rt_se)) |
816 | rt_rq->rt_nr_boosted++; |
817 | |
818 | if (rt_rq->tg) |
819 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); |
820 | } |
821 | |
822 | static void |
823 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
824 | { |
825 | if (rt_se_boosted(rt_se)) |
826 | rt_rq->rt_nr_boosted--; |
827 | |
828 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); |
829 | } |
830 | |
831 | #else /* CONFIG_RT_GROUP_SCHED */ |
832 | |
833 | static void |
834 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
835 | { |
836 | start_rt_bandwidth(&def_rt_bandwidth); |
837 | } |
838 | |
839 | static inline |
840 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} |
841 | |
842 | #endif /* CONFIG_RT_GROUP_SCHED */ |
843 | |
844 | static inline |
845 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
846 | { |
847 | int prio = rt_se_prio(rt_se); |
848 | |
849 | WARN_ON(!rt_prio(prio)); |
850 | rt_rq->rt_nr_running++; |
851 | |
852 | inc_rt_prio(rt_rq, prio); |
853 | inc_rt_migration(rt_se, rt_rq); |
854 | inc_rt_group(rt_se, rt_rq); |
855 | } |
856 | |
857 | static inline |
858 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
859 | { |
860 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
861 | WARN_ON(!rt_rq->rt_nr_running); |
862 | rt_rq->rt_nr_running--; |
863 | |
864 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); |
865 | dec_rt_migration(rt_se, rt_rq); |
866 | dec_rt_group(rt_se, rt_rq); |
867 | } |
868 | |
869 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
870 | { |
871 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
872 | struct rt_prio_array *array = &rt_rq->active; |
873 | struct rt_rq *group_rq = group_rt_rq(rt_se); |
874 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
875 | |
876 | /* |
877 | * Don't enqueue the group if its throttled, or when empty. |
878 | * The latter is a consequence of the former when a child group |
879 | * get throttled and the current group doesn't have any other |
880 | * active members. |
881 | */ |
882 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) |
883 | return; |
884 | |
885 | if (!rt_rq->rt_nr_running) |
886 | list_add_leaf_rt_rq(rt_rq); |
887 | |
888 | if (head) |
889 | list_add(&rt_se->run_list, queue); |
890 | else |
891 | list_add_tail(&rt_se->run_list, queue); |
892 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
893 | |
894 | inc_rt_tasks(rt_se, rt_rq); |
895 | } |
896 | |
897 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
898 | { |
899 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
900 | struct rt_prio_array *array = &rt_rq->active; |
901 | |
902 | list_del_init(&rt_se->run_list); |
903 | if (list_empty(array->queue + rt_se_prio(rt_se))) |
904 | __clear_bit(rt_se_prio(rt_se), array->bitmap); |
905 | |
906 | dec_rt_tasks(rt_se, rt_rq); |
907 | if (!rt_rq->rt_nr_running) |
908 | list_del_leaf_rt_rq(rt_rq); |
909 | } |
910 | |
911 | /* |
912 | * Because the prio of an upper entry depends on the lower |
913 | * entries, we must remove entries top - down. |
914 | */ |
915 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
916 | { |
917 | struct sched_rt_entity *back = NULL; |
918 | |
919 | for_each_sched_rt_entity(rt_se) { |
920 | rt_se->back = back; |
921 | back = rt_se; |
922 | } |
923 | |
924 | for (rt_se = back; rt_se; rt_se = rt_se->back) { |
925 | if (on_rt_rq(rt_se)) |
926 | __dequeue_rt_entity(rt_se); |
927 | } |
928 | } |
929 | |
930 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
931 | { |
932 | dequeue_rt_stack(rt_se); |
933 | for_each_sched_rt_entity(rt_se) |
934 | __enqueue_rt_entity(rt_se, head); |
935 | } |
936 | |
937 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) |
938 | { |
939 | dequeue_rt_stack(rt_se); |
940 | |
941 | for_each_sched_rt_entity(rt_se) { |
942 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
943 | |
944 | if (rt_rq && rt_rq->rt_nr_running) |
945 | __enqueue_rt_entity(rt_se, false); |
946 | } |
947 | } |
948 | |
949 | /* |
950 | * Adding/removing a task to/from a priority array: |
951 | */ |
952 | static void |
953 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
954 | { |
955 | struct sched_rt_entity *rt_se = &p->rt; |
956 | |
957 | if (flags & ENQUEUE_WAKEUP) |
958 | rt_se->timeout = 0; |
959 | |
960 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
961 | |
962 | if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1) |
963 | enqueue_pushable_task(rq, p); |
964 | } |
965 | |
966 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
967 | { |
968 | struct sched_rt_entity *rt_se = &p->rt; |
969 | |
970 | update_curr_rt(rq); |
971 | dequeue_rt_entity(rt_se); |
972 | |
973 | dequeue_pushable_task(rq, p); |
974 | } |
975 | |
976 | /* |
977 | * Put task to the end of the run list without the overhead of dequeue |
978 | * followed by enqueue. |
979 | */ |
980 | static void |
981 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) |
982 | { |
983 | if (on_rt_rq(rt_se)) { |
984 | struct rt_prio_array *array = &rt_rq->active; |
985 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
986 | |
987 | if (head) |
988 | list_move(&rt_se->run_list, queue); |
989 | else |
990 | list_move_tail(&rt_se->run_list, queue); |
991 | } |
992 | } |
993 | |
994 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
995 | { |
996 | struct sched_rt_entity *rt_se = &p->rt; |
997 | struct rt_rq *rt_rq; |
998 | |
999 | for_each_sched_rt_entity(rt_se) { |
1000 | rt_rq = rt_rq_of_se(rt_se); |
1001 | requeue_rt_entity(rt_rq, rt_se, head); |
1002 | } |
1003 | } |
1004 | |
1005 | static void yield_task_rt(struct rq *rq) |
1006 | { |
1007 | requeue_task_rt(rq, rq->curr, 0); |
1008 | } |
1009 | |
1010 | #ifdef CONFIG_SMP |
1011 | static int find_lowest_rq(struct task_struct *task); |
1012 | |
1013 | static int |
1014 | select_task_rq_rt(struct task_struct *p, int sd_flag, int flags) |
1015 | { |
1016 | struct task_struct *curr; |
1017 | struct rq *rq; |
1018 | int cpu; |
1019 | |
1020 | if (sd_flag != SD_BALANCE_WAKE) |
1021 | return smp_processor_id(); |
1022 | |
1023 | cpu = task_cpu(p); |
1024 | rq = cpu_rq(cpu); |
1025 | |
1026 | rcu_read_lock(); |
1027 | curr = ACCESS_ONCE(rq->curr); /* unlocked access */ |
1028 | |
1029 | /* |
1030 | * If the current task on @p's runqueue is an RT task, then |
1031 | * try to see if we can wake this RT task up on another |
1032 | * runqueue. Otherwise simply start this RT task |
1033 | * on its current runqueue. |
1034 | * |
1035 | * We want to avoid overloading runqueues. If the woken |
1036 | * task is a higher priority, then it will stay on this CPU |
1037 | * and the lower prio task should be moved to another CPU. |
1038 | * Even though this will probably make the lower prio task |
1039 | * lose its cache, we do not want to bounce a higher task |
1040 | * around just because it gave up its CPU, perhaps for a |
1041 | * lock? |
1042 | * |
1043 | * For equal prio tasks, we just let the scheduler sort it out. |
1044 | * |
1045 | * Otherwise, just let it ride on the affined RQ and the |
1046 | * post-schedule router will push the preempted task away |
1047 | * |
1048 | * This test is optimistic, if we get it wrong the load-balancer |
1049 | * will have to sort it out. |
1050 | */ |
1051 | if (curr && unlikely(rt_task(curr)) && |
1052 | (curr->rt.nr_cpus_allowed < 2 || |
1053 | curr->prio <= p->prio) && |
1054 | (p->rt.nr_cpus_allowed > 1)) { |
1055 | int target = find_lowest_rq(p); |
1056 | |
1057 | if (target != -1) |
1058 | cpu = target; |
1059 | } |
1060 | rcu_read_unlock(); |
1061 | |
1062 | return cpu; |
1063 | } |
1064 | |
1065 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) |
1066 | { |
1067 | if (rq->curr->rt.nr_cpus_allowed == 1) |
1068 | return; |
1069 | |
1070 | if (p->rt.nr_cpus_allowed != 1 |
1071 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1072 | return; |
1073 | |
1074 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
1075 | return; |
1076 | |
1077 | /* |
1078 | * There appears to be other cpus that can accept |
1079 | * current and none to run 'p', so lets reschedule |
1080 | * to try and push current away: |
1081 | */ |
1082 | requeue_task_rt(rq, p, 1); |
1083 | resched_task(rq->curr); |
1084 | } |
1085 | |
1086 | #endif /* CONFIG_SMP */ |
1087 | |
1088 | /* |
1089 | * Preempt the current task with a newly woken task if needed: |
1090 | */ |
1091 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
1092 | { |
1093 | if (p->prio < rq->curr->prio) { |
1094 | resched_task(rq->curr); |
1095 | return; |
1096 | } |
1097 | |
1098 | #ifdef CONFIG_SMP |
1099 | /* |
1100 | * If: |
1101 | * |
1102 | * - the newly woken task is of equal priority to the current task |
1103 | * - the newly woken task is non-migratable while current is migratable |
1104 | * - current will be preempted on the next reschedule |
1105 | * |
1106 | * we should check to see if current can readily move to a different |
1107 | * cpu. If so, we will reschedule to allow the push logic to try |
1108 | * to move current somewhere else, making room for our non-migratable |
1109 | * task. |
1110 | */ |
1111 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
1112 | check_preempt_equal_prio(rq, p); |
1113 | #endif |
1114 | } |
1115 | |
1116 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1117 | struct rt_rq *rt_rq) |
1118 | { |
1119 | struct rt_prio_array *array = &rt_rq->active; |
1120 | struct sched_rt_entity *next = NULL; |
1121 | struct list_head *queue; |
1122 | int idx; |
1123 | |
1124 | idx = sched_find_first_bit(array->bitmap); |
1125 | BUG_ON(idx >= MAX_RT_PRIO); |
1126 | |
1127 | queue = array->queue + idx; |
1128 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
1129 | |
1130 | return next; |
1131 | } |
1132 | |
1133 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
1134 | { |
1135 | struct sched_rt_entity *rt_se; |
1136 | struct task_struct *p; |
1137 | struct rt_rq *rt_rq; |
1138 | |
1139 | rt_rq = &rq->rt; |
1140 | |
1141 | if (!rt_rq->rt_nr_running) |
1142 | return NULL; |
1143 | |
1144 | if (rt_rq_throttled(rt_rq)) |
1145 | return NULL; |
1146 | |
1147 | do { |
1148 | rt_se = pick_next_rt_entity(rq, rt_rq); |
1149 | BUG_ON(!rt_se); |
1150 | rt_rq = group_rt_rq(rt_se); |
1151 | } while (rt_rq); |
1152 | |
1153 | p = rt_task_of(rt_se); |
1154 | p->se.exec_start = rq->clock_task; |
1155 | |
1156 | return p; |
1157 | } |
1158 | |
1159 | static struct task_struct *pick_next_task_rt(struct rq *rq) |
1160 | { |
1161 | struct task_struct *p = _pick_next_task_rt(rq); |
1162 | |
1163 | /* The running task is never eligible for pushing */ |
1164 | if (p) |
1165 | dequeue_pushable_task(rq, p); |
1166 | |
1167 | #ifdef CONFIG_SMP |
1168 | /* |
1169 | * We detect this state here so that we can avoid taking the RQ |
1170 | * lock again later if there is no need to push |
1171 | */ |
1172 | rq->post_schedule = has_pushable_tasks(rq); |
1173 | #endif |
1174 | |
1175 | return p; |
1176 | } |
1177 | |
1178 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
1179 | { |
1180 | update_curr_rt(rq); |
1181 | p->se.exec_start = 0; |
1182 | |
1183 | /* |
1184 | * The previous task needs to be made eligible for pushing |
1185 | * if it is still active |
1186 | */ |
1187 | if (on_rt_rq(&p->rt) && p->rt.nr_cpus_allowed > 1) |
1188 | enqueue_pushable_task(rq, p); |
1189 | } |
1190 | |
1191 | #ifdef CONFIG_SMP |
1192 | |
1193 | /* Only try algorithms three times */ |
1194 | #define RT_MAX_TRIES 3 |
1195 | |
1196 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); |
1197 | |
1198 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1199 | { |
1200 | if (!task_running(rq, p) && |
1201 | (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) && |
1202 | (p->rt.nr_cpus_allowed > 1)) |
1203 | return 1; |
1204 | return 0; |
1205 | } |
1206 | |
1207 | /* Return the second highest RT task, NULL otherwise */ |
1208 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
1209 | { |
1210 | struct task_struct *next = NULL; |
1211 | struct sched_rt_entity *rt_se; |
1212 | struct rt_prio_array *array; |
1213 | struct rt_rq *rt_rq; |
1214 | int idx; |
1215 | |
1216 | for_each_leaf_rt_rq(rt_rq, rq) { |
1217 | array = &rt_rq->active; |
1218 | idx = sched_find_first_bit(array->bitmap); |
1219 | next_idx: |
1220 | if (idx >= MAX_RT_PRIO) |
1221 | continue; |
1222 | if (next && next->prio < idx) |
1223 | continue; |
1224 | list_for_each_entry(rt_se, array->queue + idx, run_list) { |
1225 | struct task_struct *p; |
1226 | |
1227 | if (!rt_entity_is_task(rt_se)) |
1228 | continue; |
1229 | |
1230 | p = rt_task_of(rt_se); |
1231 | if (pick_rt_task(rq, p, cpu)) { |
1232 | next = p; |
1233 | break; |
1234 | } |
1235 | } |
1236 | if (!next) { |
1237 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); |
1238 | goto next_idx; |
1239 | } |
1240 | } |
1241 | |
1242 | return next; |
1243 | } |
1244 | |
1245 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
1246 | |
1247 | static int find_lowest_rq(struct task_struct *task) |
1248 | { |
1249 | struct sched_domain *sd; |
1250 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
1251 | int this_cpu = smp_processor_id(); |
1252 | int cpu = task_cpu(task); |
1253 | |
1254 | /* Make sure the mask is initialized first */ |
1255 | if (unlikely(!lowest_mask)) |
1256 | return -1; |
1257 | |
1258 | if (task->rt.nr_cpus_allowed == 1) |
1259 | return -1; /* No other targets possible */ |
1260 | |
1261 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1262 | return -1; /* No targets found */ |
1263 | |
1264 | /* |
1265 | * At this point we have built a mask of cpus representing the |
1266 | * lowest priority tasks in the system. Now we want to elect |
1267 | * the best one based on our affinity and topology. |
1268 | * |
1269 | * We prioritize the last cpu that the task executed on since |
1270 | * it is most likely cache-hot in that location. |
1271 | */ |
1272 | if (cpumask_test_cpu(cpu, lowest_mask)) |
1273 | return cpu; |
1274 | |
1275 | /* |
1276 | * Otherwise, we consult the sched_domains span maps to figure |
1277 | * out which cpu is logically closest to our hot cache data. |
1278 | */ |
1279 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1280 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ |
1281 | |
1282 | rcu_read_lock(); |
1283 | for_each_domain(cpu, sd) { |
1284 | if (sd->flags & SD_WAKE_AFFINE) { |
1285 | int best_cpu; |
1286 | |
1287 | /* |
1288 | * "this_cpu" is cheaper to preempt than a |
1289 | * remote processor. |
1290 | */ |
1291 | if (this_cpu != -1 && |
1292 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
1293 | rcu_read_unlock(); |
1294 | return this_cpu; |
1295 | } |
1296 | |
1297 | best_cpu = cpumask_first_and(lowest_mask, |
1298 | sched_domain_span(sd)); |
1299 | if (best_cpu < nr_cpu_ids) { |
1300 | rcu_read_unlock(); |
1301 | return best_cpu; |
1302 | } |
1303 | } |
1304 | } |
1305 | rcu_read_unlock(); |
1306 | |
1307 | /* |
1308 | * And finally, if there were no matches within the domains |
1309 | * just give the caller *something* to work with from the compatible |
1310 | * locations. |
1311 | */ |
1312 | if (this_cpu != -1) |
1313 | return this_cpu; |
1314 | |
1315 | cpu = cpumask_any(lowest_mask); |
1316 | if (cpu < nr_cpu_ids) |
1317 | return cpu; |
1318 | return -1; |
1319 | } |
1320 | |
1321 | /* Will lock the rq it finds */ |
1322 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
1323 | { |
1324 | struct rq *lowest_rq = NULL; |
1325 | int tries; |
1326 | int cpu; |
1327 | |
1328 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1329 | cpu = find_lowest_rq(task); |
1330 | |
1331 | if ((cpu == -1) || (cpu == rq->cpu)) |
1332 | break; |
1333 | |
1334 | lowest_rq = cpu_rq(cpu); |
1335 | |
1336 | /* if the prio of this runqueue changed, try again */ |
1337 | if (double_lock_balance(rq, lowest_rq)) { |
1338 | /* |
1339 | * We had to unlock the run queue. In |
1340 | * the mean time, task could have |
1341 | * migrated already or had its affinity changed. |
1342 | * Also make sure that it wasn't scheduled on its rq. |
1343 | */ |
1344 | if (unlikely(task_rq(task) != rq || |
1345 | !cpumask_test_cpu(lowest_rq->cpu, |
1346 | &task->cpus_allowed) || |
1347 | task_running(rq, task) || |
1348 | !task->on_rq)) { |
1349 | |
1350 | raw_spin_unlock(&lowest_rq->lock); |
1351 | lowest_rq = NULL; |
1352 | break; |
1353 | } |
1354 | } |
1355 | |
1356 | /* If this rq is still suitable use it. */ |
1357 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
1358 | break; |
1359 | |
1360 | /* try again */ |
1361 | double_unlock_balance(rq, lowest_rq); |
1362 | lowest_rq = NULL; |
1363 | } |
1364 | |
1365 | return lowest_rq; |
1366 | } |
1367 | |
1368 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1369 | { |
1370 | struct task_struct *p; |
1371 | |
1372 | if (!has_pushable_tasks(rq)) |
1373 | return NULL; |
1374 | |
1375 | p = plist_first_entry(&rq->rt.pushable_tasks, |
1376 | struct task_struct, pushable_tasks); |
1377 | |
1378 | BUG_ON(rq->cpu != task_cpu(p)); |
1379 | BUG_ON(task_current(rq, p)); |
1380 | BUG_ON(p->rt.nr_cpus_allowed <= 1); |
1381 | |
1382 | BUG_ON(!p->on_rq); |
1383 | BUG_ON(!rt_task(p)); |
1384 | |
1385 | return p; |
1386 | } |
1387 | |
1388 | /* |
1389 | * If the current CPU has more than one RT task, see if the non |
1390 | * running task can migrate over to a CPU that is running a task |
1391 | * of lesser priority. |
1392 | */ |
1393 | static int push_rt_task(struct rq *rq) |
1394 | { |
1395 | struct task_struct *next_task; |
1396 | struct rq *lowest_rq; |
1397 | |
1398 | if (!rq->rt.overloaded) |
1399 | return 0; |
1400 | |
1401 | next_task = pick_next_pushable_task(rq); |
1402 | if (!next_task) |
1403 | return 0; |
1404 | |
1405 | retry: |
1406 | if (unlikely(next_task == rq->curr)) { |
1407 | WARN_ON(1); |
1408 | return 0; |
1409 | } |
1410 | |
1411 | /* |
1412 | * It's possible that the next_task slipped in of |
1413 | * higher priority than current. If that's the case |
1414 | * just reschedule current. |
1415 | */ |
1416 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1417 | resched_task(rq->curr); |
1418 | return 0; |
1419 | } |
1420 | |
1421 | /* We might release rq lock */ |
1422 | get_task_struct(next_task); |
1423 | |
1424 | /* find_lock_lowest_rq locks the rq if found */ |
1425 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
1426 | if (!lowest_rq) { |
1427 | struct task_struct *task; |
1428 | /* |
1429 | * find lock_lowest_rq releases rq->lock |
1430 | * so it is possible that next_task has migrated. |
1431 | * |
1432 | * We need to make sure that the task is still on the same |
1433 | * run-queue and is also still the next task eligible for |
1434 | * pushing. |
1435 | */ |
1436 | task = pick_next_pushable_task(rq); |
1437 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1438 | /* |
1439 | * If we get here, the task hasn't moved at all, but |
1440 | * it has failed to push. We will not try again, |
1441 | * since the other cpus will pull from us when they |
1442 | * are ready. |
1443 | */ |
1444 | dequeue_pushable_task(rq, next_task); |
1445 | goto out; |
1446 | } |
1447 | |
1448 | if (!task) |
1449 | /* No more tasks, just exit */ |
1450 | goto out; |
1451 | |
1452 | /* |
1453 | * Something has shifted, try again. |
1454 | */ |
1455 | put_task_struct(next_task); |
1456 | next_task = task; |
1457 | goto retry; |
1458 | } |
1459 | |
1460 | deactivate_task(rq, next_task, 0); |
1461 | set_task_cpu(next_task, lowest_rq->cpu); |
1462 | activate_task(lowest_rq, next_task, 0); |
1463 | |
1464 | resched_task(lowest_rq->curr); |
1465 | |
1466 | double_unlock_balance(rq, lowest_rq); |
1467 | |
1468 | out: |
1469 | put_task_struct(next_task); |
1470 | |
1471 | return 1; |
1472 | } |
1473 | |
1474 | static void push_rt_tasks(struct rq *rq) |
1475 | { |
1476 | /* push_rt_task will return true if it moved an RT */ |
1477 | while (push_rt_task(rq)) |
1478 | ; |
1479 | } |
1480 | |
1481 | static int pull_rt_task(struct rq *this_rq) |
1482 | { |
1483 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
1484 | struct task_struct *p; |
1485 | struct rq *src_rq; |
1486 | |
1487 | if (likely(!rt_overloaded(this_rq))) |
1488 | return 0; |
1489 | |
1490 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
1491 | if (this_cpu == cpu) |
1492 | continue; |
1493 | |
1494 | src_rq = cpu_rq(cpu); |
1495 | |
1496 | /* |
1497 | * Don't bother taking the src_rq->lock if the next highest |
1498 | * task is known to be lower-priority than our current task. |
1499 | * This may look racy, but if this value is about to go |
1500 | * logically higher, the src_rq will push this task away. |
1501 | * And if its going logically lower, we do not care |
1502 | */ |
1503 | if (src_rq->rt.highest_prio.next >= |
1504 | this_rq->rt.highest_prio.curr) |
1505 | continue; |
1506 | |
1507 | /* |
1508 | * We can potentially drop this_rq's lock in |
1509 | * double_lock_balance, and another CPU could |
1510 | * alter this_rq |
1511 | */ |
1512 | double_lock_balance(this_rq, src_rq); |
1513 | |
1514 | /* |
1515 | * Are there still pullable RT tasks? |
1516 | */ |
1517 | if (src_rq->rt.rt_nr_running <= 1) |
1518 | goto skip; |
1519 | |
1520 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
1521 | |
1522 | /* |
1523 | * Do we have an RT task that preempts |
1524 | * the to-be-scheduled task? |
1525 | */ |
1526 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
1527 | WARN_ON(p == src_rq->curr); |
1528 | WARN_ON(!p->on_rq); |
1529 | |
1530 | /* |
1531 | * There's a chance that p is higher in priority |
1532 | * than what's currently running on its cpu. |
1533 | * This is just that p is wakeing up and hasn't |
1534 | * had a chance to schedule. We only pull |
1535 | * p if it is lower in priority than the |
1536 | * current task on the run queue |
1537 | */ |
1538 | if (p->prio < src_rq->curr->prio) |
1539 | goto skip; |
1540 | |
1541 | ret = 1; |
1542 | |
1543 | deactivate_task(src_rq, p, 0); |
1544 | set_task_cpu(p, this_cpu); |
1545 | activate_task(this_rq, p, 0); |
1546 | /* |
1547 | * We continue with the search, just in |
1548 | * case there's an even higher prio task |
1549 | * in another runqueue. (low likelihood |
1550 | * but possible) |
1551 | */ |
1552 | } |
1553 | skip: |
1554 | double_unlock_balance(this_rq, src_rq); |
1555 | } |
1556 | |
1557 | return ret; |
1558 | } |
1559 | |
1560 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
1561 | { |
1562 | /* Try to pull RT tasks here if we lower this rq's prio */ |
1563 | if (rq->rt.highest_prio.curr > prev->prio) |
1564 | pull_rt_task(rq); |
1565 | } |
1566 | |
1567 | static void post_schedule_rt(struct rq *rq) |
1568 | { |
1569 | push_rt_tasks(rq); |
1570 | } |
1571 | |
1572 | /* |
1573 | * If we are not running and we are not going to reschedule soon, we should |
1574 | * try to push tasks away now |
1575 | */ |
1576 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
1577 | { |
1578 | if (!task_running(rq, p) && |
1579 | !test_tsk_need_resched(rq->curr) && |
1580 | has_pushable_tasks(rq) && |
1581 | p->rt.nr_cpus_allowed > 1 && |
1582 | rt_task(rq->curr) && |
1583 | (rq->curr->rt.nr_cpus_allowed < 2 || |
1584 | rq->curr->prio <= p->prio)) |
1585 | push_rt_tasks(rq); |
1586 | } |
1587 | |
1588 | static void set_cpus_allowed_rt(struct task_struct *p, |
1589 | const struct cpumask *new_mask) |
1590 | { |
1591 | int weight = cpumask_weight(new_mask); |
1592 | |
1593 | BUG_ON(!rt_task(p)); |
1594 | |
1595 | /* |
1596 | * Update the migration status of the RQ if we have an RT task |
1597 | * which is running AND changing its weight value. |
1598 | */ |
1599 | if (p->on_rq && (weight != p->rt.nr_cpus_allowed)) { |
1600 | struct rq *rq = task_rq(p); |
1601 | |
1602 | if (!task_current(rq, p)) { |
1603 | /* |
1604 | * Make sure we dequeue this task from the pushable list |
1605 | * before going further. It will either remain off of |
1606 | * the list because we are no longer pushable, or it |
1607 | * will be requeued. |
1608 | */ |
1609 | if (p->rt.nr_cpus_allowed > 1) |
1610 | dequeue_pushable_task(rq, p); |
1611 | |
1612 | /* |
1613 | * Requeue if our weight is changing and still > 1 |
1614 | */ |
1615 | if (weight > 1) |
1616 | enqueue_pushable_task(rq, p); |
1617 | |
1618 | } |
1619 | |
1620 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
1621 | rq->rt.rt_nr_migratory++; |
1622 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
1623 | BUG_ON(!rq->rt.rt_nr_migratory); |
1624 | rq->rt.rt_nr_migratory--; |
1625 | } |
1626 | |
1627 | update_rt_migration(&rq->rt); |
1628 | } |
1629 | |
1630 | cpumask_copy(&p->cpus_allowed, new_mask); |
1631 | p->rt.nr_cpus_allowed = weight; |
1632 | } |
1633 | |
1634 | /* Assumes rq->lock is held */ |
1635 | static void rq_online_rt(struct rq *rq) |
1636 | { |
1637 | if (rq->rt.overloaded) |
1638 | rt_set_overload(rq); |
1639 | |
1640 | __enable_runtime(rq); |
1641 | |
1642 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
1643 | } |
1644 | |
1645 | /* Assumes rq->lock is held */ |
1646 | static void rq_offline_rt(struct rq *rq) |
1647 | { |
1648 | if (rq->rt.overloaded) |
1649 | rt_clear_overload(rq); |
1650 | |
1651 | __disable_runtime(rq); |
1652 | |
1653 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
1654 | } |
1655 | |
1656 | /* |
1657 | * When switch from the rt queue, we bring ourselves to a position |
1658 | * that we might want to pull RT tasks from other runqueues. |
1659 | */ |
1660 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
1661 | { |
1662 | /* |
1663 | * If there are other RT tasks then we will reschedule |
1664 | * and the scheduling of the other RT tasks will handle |
1665 | * the balancing. But if we are the last RT task |
1666 | * we may need to handle the pulling of RT tasks |
1667 | * now. |
1668 | */ |
1669 | if (p->on_rq && !rq->rt.rt_nr_running) |
1670 | pull_rt_task(rq); |
1671 | } |
1672 | |
1673 | static inline void init_sched_rt_class(void) |
1674 | { |
1675 | unsigned int i; |
1676 | |
1677 | for_each_possible_cpu(i) |
1678 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
1679 | GFP_KERNEL, cpu_to_node(i)); |
1680 | } |
1681 | #endif /* CONFIG_SMP */ |
1682 | |
1683 | /* |
1684 | * When switching a task to RT, we may overload the runqueue |
1685 | * with RT tasks. In this case we try to push them off to |
1686 | * other runqueues. |
1687 | */ |
1688 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
1689 | { |
1690 | int check_resched = 1; |
1691 | |
1692 | /* |
1693 | * If we are already running, then there's nothing |
1694 | * that needs to be done. But if we are not running |
1695 | * we may need to preempt the current running task. |
1696 | * If that current running task is also an RT task |
1697 | * then see if we can move to another run queue. |
1698 | */ |
1699 | if (p->on_rq && rq->curr != p) { |
1700 | #ifdef CONFIG_SMP |
1701 | if (rq->rt.overloaded && push_rt_task(rq) && |
1702 | /* Don't resched if we changed runqueues */ |
1703 | rq != task_rq(p)) |
1704 | check_resched = 0; |
1705 | #endif /* CONFIG_SMP */ |
1706 | if (check_resched && p->prio < rq->curr->prio) |
1707 | resched_task(rq->curr); |
1708 | } |
1709 | } |
1710 | |
1711 | /* |
1712 | * Priority of the task has changed. This may cause |
1713 | * us to initiate a push or pull. |
1714 | */ |
1715 | static void |
1716 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) |
1717 | { |
1718 | if (!p->on_rq) |
1719 | return; |
1720 | |
1721 | if (rq->curr == p) { |
1722 | #ifdef CONFIG_SMP |
1723 | /* |
1724 | * If our priority decreases while running, we |
1725 | * may need to pull tasks to this runqueue. |
1726 | */ |
1727 | if (oldprio < p->prio) |
1728 | pull_rt_task(rq); |
1729 | /* |
1730 | * If there's a higher priority task waiting to run |
1731 | * then reschedule. Note, the above pull_rt_task |
1732 | * can release the rq lock and p could migrate. |
1733 | * Only reschedule if p is still on the same runqueue. |
1734 | */ |
1735 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
1736 | resched_task(p); |
1737 | #else |
1738 | /* For UP simply resched on drop of prio */ |
1739 | if (oldprio < p->prio) |
1740 | resched_task(p); |
1741 | #endif /* CONFIG_SMP */ |
1742 | } else { |
1743 | /* |
1744 | * This task is not running, but if it is |
1745 | * greater than the current running task |
1746 | * then reschedule. |
1747 | */ |
1748 | if (p->prio < rq->curr->prio) |
1749 | resched_task(rq->curr); |
1750 | } |
1751 | } |
1752 | |
1753 | static void watchdog(struct rq *rq, struct task_struct *p) |
1754 | { |
1755 | unsigned long soft, hard; |
1756 | |
1757 | /* max may change after cur was read, this will be fixed next tick */ |
1758 | soft = task_rlimit(p, RLIMIT_RTTIME); |
1759 | hard = task_rlimit_max(p, RLIMIT_RTTIME); |
1760 | |
1761 | if (soft != RLIM_INFINITY) { |
1762 | unsigned long next; |
1763 | |
1764 | p->rt.timeout++; |
1765 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
1766 | if (p->rt.timeout > next) |
1767 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
1768 | } |
1769 | } |
1770 | |
1771 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
1772 | { |
1773 | update_curr_rt(rq); |
1774 | |
1775 | watchdog(rq, p); |
1776 | |
1777 | /* |
1778 | * RR tasks need a special form of timeslice management. |
1779 | * FIFO tasks have no timeslices. |
1780 | */ |
1781 | if (p->policy != SCHED_RR) |
1782 | return; |
1783 | |
1784 | if (--p->rt.time_slice) |
1785 | return; |
1786 | |
1787 | p->rt.time_slice = DEF_TIMESLICE; |
1788 | |
1789 | /* |
1790 | * Requeue to the end of queue if we are not the only element |
1791 | * on the queue: |
1792 | */ |
1793 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
1794 | requeue_task_rt(rq, p, 0); |
1795 | set_tsk_need_resched(p); |
1796 | } |
1797 | } |
1798 | |
1799 | static void set_curr_task_rt(struct rq *rq) |
1800 | { |
1801 | struct task_struct *p = rq->curr; |
1802 | |
1803 | p->se.exec_start = rq->clock_task; |
1804 | |
1805 | /* The running task is never eligible for pushing */ |
1806 | dequeue_pushable_task(rq, p); |
1807 | } |
1808 | |
1809 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
1810 | { |
1811 | /* |
1812 | * Time slice is 0 for SCHED_FIFO tasks |
1813 | */ |
1814 | if (task->policy == SCHED_RR) |
1815 | return DEF_TIMESLICE; |
1816 | else |
1817 | return 0; |
1818 | } |
1819 | |
1820 | static const struct sched_class rt_sched_class = { |
1821 | .next = &fair_sched_class, |
1822 | .enqueue_task = enqueue_task_rt, |
1823 | .dequeue_task = dequeue_task_rt, |
1824 | .yield_task = yield_task_rt, |
1825 | |
1826 | .check_preempt_curr = check_preempt_curr_rt, |
1827 | |
1828 | .pick_next_task = pick_next_task_rt, |
1829 | .put_prev_task = put_prev_task_rt, |
1830 | |
1831 | #ifdef CONFIG_SMP |
1832 | .select_task_rq = select_task_rq_rt, |
1833 | |
1834 | .set_cpus_allowed = set_cpus_allowed_rt, |
1835 | .rq_online = rq_online_rt, |
1836 | .rq_offline = rq_offline_rt, |
1837 | .pre_schedule = pre_schedule_rt, |
1838 | .post_schedule = post_schedule_rt, |
1839 | .task_woken = task_woken_rt, |
1840 | .switched_from = switched_from_rt, |
1841 | #endif |
1842 | |
1843 | .set_curr_task = set_curr_task_rt, |
1844 | .task_tick = task_tick_rt, |
1845 | |
1846 | .get_rr_interval = get_rr_interval_rt, |
1847 | |
1848 | .prio_changed = prio_changed_rt, |
1849 | .switched_to = switched_to_rt, |
1850 | }; |
1851 | |
1852 | #ifdef CONFIG_SCHED_DEBUG |
1853 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); |
1854 | |
1855 | static void print_rt_stats(struct seq_file *m, int cpu) |
1856 | { |
1857 | rt_rq_iter_t iter; |
1858 | struct rt_rq *rt_rq; |
1859 | |
1860 | rcu_read_lock(); |
1861 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
1862 | print_rt_rq(m, cpu, rt_rq); |
1863 | rcu_read_unlock(); |
1864 | } |
1865 | #endif /* CONFIG_SCHED_DEBUG */ |
1866 | |
1867 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9