]>
Commit | Line | Data |
---|---|---|
bb44e5d1 IM |
1 | /* |
2 | * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR | |
3 | * policies) | |
4 | */ | |
5 | ||
8f48894f PZ |
6 | #ifdef CONFIG_RT_GROUP_SCHED |
7 | ||
8 | #define rt_entity_is_task(rt_se) (!(rt_se)->my_q) | |
9 | ||
398a153b GH |
10 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
11 | { | |
8f48894f PZ |
12 | #ifdef CONFIG_SCHED_DEBUG |
13 | WARN_ON_ONCE(!rt_entity_is_task(rt_se)); | |
14 | #endif | |
398a153b GH |
15 | return container_of(rt_se, struct task_struct, rt); |
16 | } | |
17 | ||
398a153b GH |
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 | ||
a1ba4d8b PZ |
30 | #define rt_entity_is_task(rt_se) (1) |
31 | ||
8f48894f PZ |
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 | ||
398a153b GH |
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 | ||
4fd29176 | 52 | #ifdef CONFIG_SMP |
84de4274 | 53 | |
637f5085 | 54 | static inline int rt_overloaded(struct rq *rq) |
4fd29176 | 55 | { |
637f5085 | 56 | return atomic_read(&rq->rd->rto_count); |
4fd29176 | 57 | } |
84de4274 | 58 | |
4fd29176 SR |
59 | static inline void rt_set_overload(struct rq *rq) |
60 | { | |
1f11eb6a GH |
61 | if (!rq->online) |
62 | return; | |
63 | ||
c6c4927b | 64 | cpumask_set_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 SR |
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(); | |
637f5085 | 73 | atomic_inc(&rq->rd->rto_count); |
4fd29176 | 74 | } |
84de4274 | 75 | |
4fd29176 SR |
76 | static inline void rt_clear_overload(struct rq *rq) |
77 | { | |
1f11eb6a GH |
78 | if (!rq->online) |
79 | return; | |
80 | ||
4fd29176 | 81 | /* the order here really doesn't matter */ |
637f5085 | 82 | atomic_dec(&rq->rd->rto_count); |
c6c4927b | 83 | cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 | 84 | } |
73fe6aae | 85 | |
398a153b | 86 | static void update_rt_migration(struct rt_rq *rt_rq) |
73fe6aae | 87 | { |
a1ba4d8b | 88 | if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) { |
398a153b GH |
89 | if (!rt_rq->overloaded) { |
90 | rt_set_overload(rq_of_rt_rq(rt_rq)); | |
91 | rt_rq->overloaded = 1; | |
cdc8eb98 | 92 | } |
398a153b GH |
93 | } else if (rt_rq->overloaded) { |
94 | rt_clear_overload(rq_of_rt_rq(rt_rq)); | |
95 | rt_rq->overloaded = 0; | |
637f5085 | 96 | } |
73fe6aae | 97 | } |
4fd29176 | 98 | |
398a153b GH |
99 | static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
100 | { | |
a1ba4d8b PZ |
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++; | |
398a153b GH |
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 | { | |
a1ba4d8b PZ |
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--; | |
398a153b GH |
121 | if (rt_se->nr_cpus_allowed > 1) |
122 | rt_rq->rt_nr_migratory--; | |
123 | ||
124 | update_rt_migration(rt_rq); | |
125 | } | |
126 | ||
917b627d GH |
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 | ||
bcf08df3 IM |
139 | static inline int has_pushable_tasks(struct rq *rq) |
140 | { | |
141 | return !plist_head_empty(&rq->rt.pushable_tasks); | |
142 | } | |
143 | ||
917b627d GH |
144 | #else |
145 | ||
ceacc2c1 | 146 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
fa85ae24 | 147 | { |
6f505b16 PZ |
148 | } |
149 | ||
ceacc2c1 PZ |
150 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
151 | { | |
152 | } | |
153 | ||
b07430ac | 154 | static inline |
ceacc2c1 PZ |
155 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
156 | { | |
157 | } | |
158 | ||
398a153b | 159 | static inline |
ceacc2c1 PZ |
160 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
161 | { | |
162 | } | |
917b627d | 163 | |
4fd29176 SR |
164 | #endif /* CONFIG_SMP */ |
165 | ||
6f505b16 PZ |
166 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
167 | { | |
168 | return !list_empty(&rt_se->run_list); | |
169 | } | |
170 | ||
052f1dc7 | 171 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 172 | |
9f0c1e56 | 173 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
174 | { |
175 | if (!rt_rq->tg) | |
9f0c1e56 | 176 | return RUNTIME_INF; |
6f505b16 | 177 | |
ac086bc2 PZ |
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); | |
6f505b16 PZ |
184 | } |
185 | ||
186 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
80f40ee4 | 187 | list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) |
6f505b16 | 188 | |
6f505b16 PZ |
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); | |
198 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); | |
199 | ||
9f0c1e56 | 200 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 201 | { |
f6121f4f | 202 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
6f505b16 PZ |
203 | struct sched_rt_entity *rt_se = rt_rq->rt_se; |
204 | ||
f6121f4f DF |
205 | if (rt_rq->rt_nr_running) { |
206 | if (rt_se && !on_rt_rq(rt_se)) | |
207 | enqueue_rt_entity(rt_se); | |
e864c499 | 208 | if (rt_rq->highest_prio.curr < curr->prio) |
1020387f | 209 | resched_task(curr); |
6f505b16 PZ |
210 | } |
211 | } | |
212 | ||
9f0c1e56 | 213 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
214 | { |
215 | struct sched_rt_entity *rt_se = rt_rq->rt_se; | |
216 | ||
217 | if (rt_se && on_rt_rq(rt_se)) | |
218 | dequeue_rt_entity(rt_se); | |
219 | } | |
220 | ||
23b0fdfc PZ |
221 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
222 | { | |
223 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
224 | } | |
225 | ||
226 | static int rt_se_boosted(struct sched_rt_entity *rt_se) | |
227 | { | |
228 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
229 | struct task_struct *p; | |
230 | ||
231 | if (rt_rq) | |
232 | return !!rt_rq->rt_nr_boosted; | |
233 | ||
234 | p = rt_task_of(rt_se); | |
235 | return p->prio != p->normal_prio; | |
236 | } | |
237 | ||
d0b27fa7 | 238 | #ifdef CONFIG_SMP |
c6c4927b | 239 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 PZ |
240 | { |
241 | return cpu_rq(smp_processor_id())->rd->span; | |
242 | } | |
6f505b16 | 243 | #else |
c6c4927b | 244 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 245 | { |
c6c4927b | 246 | return cpu_online_mask; |
d0b27fa7 PZ |
247 | } |
248 | #endif | |
6f505b16 | 249 | |
d0b27fa7 PZ |
250 | static inline |
251 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 252 | { |
d0b27fa7 PZ |
253 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
254 | } | |
9f0c1e56 | 255 | |
ac086bc2 PZ |
256 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
257 | { | |
258 | return &rt_rq->tg->rt_bandwidth; | |
259 | } | |
260 | ||
55e12e5e | 261 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
262 | |
263 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
264 | { | |
ac086bc2 PZ |
265 | return rt_rq->rt_runtime; |
266 | } | |
267 | ||
268 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
269 | { | |
270 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
271 | } |
272 | ||
273 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
274 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
275 | ||
6f505b16 PZ |
276 | #define for_each_sched_rt_entity(rt_se) \ |
277 | for (; rt_se; rt_se = NULL) | |
278 | ||
279 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
280 | { | |
281 | return NULL; | |
282 | } | |
283 | ||
9f0c1e56 | 284 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 285 | { |
f3ade837 JB |
286 | if (rt_rq->rt_nr_running) |
287 | resched_task(rq_of_rt_rq(rt_rq)->curr); | |
6f505b16 PZ |
288 | } |
289 | ||
9f0c1e56 | 290 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
291 | { |
292 | } | |
293 | ||
23b0fdfc PZ |
294 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
295 | { | |
296 | return rt_rq->rt_throttled; | |
297 | } | |
d0b27fa7 | 298 | |
c6c4927b | 299 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 300 | { |
c6c4927b | 301 | return cpu_online_mask; |
d0b27fa7 PZ |
302 | } |
303 | ||
304 | static inline | |
305 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
306 | { | |
307 | return &cpu_rq(cpu)->rt; | |
308 | } | |
309 | ||
ac086bc2 PZ |
310 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
311 | { | |
312 | return &def_rt_bandwidth; | |
313 | } | |
314 | ||
55e12e5e | 315 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 316 | |
ac086bc2 | 317 | #ifdef CONFIG_SMP |
78333cdd PZ |
318 | /* |
319 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
320 | */ | |
b79f3833 | 321 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
322 | { |
323 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
324 | struct root_domain *rd = cpu_rq(smp_processor_id())->rd; | |
325 | int i, weight, more = 0; | |
326 | u64 rt_period; | |
327 | ||
c6c4927b | 328 | weight = cpumask_weight(rd->span); |
ac086bc2 PZ |
329 | |
330 | spin_lock(&rt_b->rt_runtime_lock); | |
331 | rt_period = ktime_to_ns(rt_b->rt_period); | |
c6c4927b | 332 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
333 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
334 | s64 diff; | |
335 | ||
336 | if (iter == rt_rq) | |
337 | continue; | |
338 | ||
339 | spin_lock(&iter->rt_runtime_lock); | |
78333cdd PZ |
340 | /* |
341 | * Either all rqs have inf runtime and there's nothing to steal | |
342 | * or __disable_runtime() below sets a specific rq to inf to | |
343 | * indicate its been disabled and disalow stealing. | |
344 | */ | |
7def2be1 PZ |
345 | if (iter->rt_runtime == RUNTIME_INF) |
346 | goto next; | |
347 | ||
78333cdd PZ |
348 | /* |
349 | * From runqueues with spare time, take 1/n part of their | |
350 | * spare time, but no more than our period. | |
351 | */ | |
ac086bc2 PZ |
352 | diff = iter->rt_runtime - iter->rt_time; |
353 | if (diff > 0) { | |
58838cf3 | 354 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
355 | if (rt_rq->rt_runtime + diff > rt_period) |
356 | diff = rt_period - rt_rq->rt_runtime; | |
357 | iter->rt_runtime -= diff; | |
358 | rt_rq->rt_runtime += diff; | |
359 | more = 1; | |
360 | if (rt_rq->rt_runtime == rt_period) { | |
361 | spin_unlock(&iter->rt_runtime_lock); | |
362 | break; | |
363 | } | |
364 | } | |
7def2be1 | 365 | next: |
ac086bc2 PZ |
366 | spin_unlock(&iter->rt_runtime_lock); |
367 | } | |
368 | spin_unlock(&rt_b->rt_runtime_lock); | |
369 | ||
370 | return more; | |
371 | } | |
7def2be1 | 372 | |
78333cdd PZ |
373 | /* |
374 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
375 | */ | |
7def2be1 PZ |
376 | static void __disable_runtime(struct rq *rq) |
377 | { | |
378 | struct root_domain *rd = rq->rd; | |
379 | struct rt_rq *rt_rq; | |
380 | ||
381 | if (unlikely(!scheduler_running)) | |
382 | return; | |
383 | ||
384 | for_each_leaf_rt_rq(rt_rq, rq) { | |
385 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
386 | s64 want; | |
387 | int i; | |
388 | ||
389 | spin_lock(&rt_b->rt_runtime_lock); | |
390 | spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
391 | /* |
392 | * Either we're all inf and nobody needs to borrow, or we're | |
393 | * already disabled and thus have nothing to do, or we have | |
394 | * exactly the right amount of runtime to take out. | |
395 | */ | |
7def2be1 PZ |
396 | if (rt_rq->rt_runtime == RUNTIME_INF || |
397 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
398 | goto balanced; | |
399 | spin_unlock(&rt_rq->rt_runtime_lock); | |
400 | ||
78333cdd PZ |
401 | /* |
402 | * Calculate the difference between what we started out with | |
403 | * and what we current have, that's the amount of runtime | |
404 | * we lend and now have to reclaim. | |
405 | */ | |
7def2be1 PZ |
406 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
407 | ||
78333cdd PZ |
408 | /* |
409 | * Greedy reclaim, take back as much as we can. | |
410 | */ | |
c6c4927b | 411 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
412 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
413 | s64 diff; | |
414 | ||
78333cdd PZ |
415 | /* |
416 | * Can't reclaim from ourselves or disabled runqueues. | |
417 | */ | |
f1679d08 | 418 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
419 | continue; |
420 | ||
421 | spin_lock(&iter->rt_runtime_lock); | |
422 | if (want > 0) { | |
423 | diff = min_t(s64, iter->rt_runtime, want); | |
424 | iter->rt_runtime -= diff; | |
425 | want -= diff; | |
426 | } else { | |
427 | iter->rt_runtime -= want; | |
428 | want -= want; | |
429 | } | |
430 | spin_unlock(&iter->rt_runtime_lock); | |
431 | ||
432 | if (!want) | |
433 | break; | |
434 | } | |
435 | ||
436 | spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
437 | /* |
438 | * We cannot be left wanting - that would mean some runtime | |
439 | * leaked out of the system. | |
440 | */ | |
7def2be1 PZ |
441 | BUG_ON(want); |
442 | balanced: | |
78333cdd PZ |
443 | /* |
444 | * Disable all the borrow logic by pretending we have inf | |
445 | * runtime - in which case borrowing doesn't make sense. | |
446 | */ | |
7def2be1 PZ |
447 | rt_rq->rt_runtime = RUNTIME_INF; |
448 | spin_unlock(&rt_rq->rt_runtime_lock); | |
449 | spin_unlock(&rt_b->rt_runtime_lock); | |
450 | } | |
451 | } | |
452 | ||
453 | static void disable_runtime(struct rq *rq) | |
454 | { | |
455 | unsigned long flags; | |
456 | ||
457 | spin_lock_irqsave(&rq->lock, flags); | |
458 | __disable_runtime(rq); | |
459 | spin_unlock_irqrestore(&rq->lock, flags); | |
460 | } | |
461 | ||
462 | static void __enable_runtime(struct rq *rq) | |
463 | { | |
7def2be1 PZ |
464 | struct rt_rq *rt_rq; |
465 | ||
466 | if (unlikely(!scheduler_running)) | |
467 | return; | |
468 | ||
78333cdd PZ |
469 | /* |
470 | * Reset each runqueue's bandwidth settings | |
471 | */ | |
7def2be1 PZ |
472 | for_each_leaf_rt_rq(rt_rq, rq) { |
473 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
474 | ||
475 | spin_lock(&rt_b->rt_runtime_lock); | |
476 | spin_lock(&rt_rq->rt_runtime_lock); | |
477 | rt_rq->rt_runtime = rt_b->rt_runtime; | |
478 | rt_rq->rt_time = 0; | |
baf25731 | 479 | rt_rq->rt_throttled = 0; |
7def2be1 PZ |
480 | spin_unlock(&rt_rq->rt_runtime_lock); |
481 | spin_unlock(&rt_b->rt_runtime_lock); | |
482 | } | |
483 | } | |
484 | ||
485 | static void enable_runtime(struct rq *rq) | |
486 | { | |
487 | unsigned long flags; | |
488 | ||
489 | spin_lock_irqsave(&rq->lock, flags); | |
490 | __enable_runtime(rq); | |
491 | spin_unlock_irqrestore(&rq->lock, flags); | |
492 | } | |
493 | ||
eff6549b PZ |
494 | static int balance_runtime(struct rt_rq *rt_rq) |
495 | { | |
496 | int more = 0; | |
497 | ||
498 | if (rt_rq->rt_time > rt_rq->rt_runtime) { | |
499 | spin_unlock(&rt_rq->rt_runtime_lock); | |
500 | more = do_balance_runtime(rt_rq); | |
501 | spin_lock(&rt_rq->rt_runtime_lock); | |
502 | } | |
503 | ||
504 | return more; | |
505 | } | |
55e12e5e | 506 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
507 | static inline int balance_runtime(struct rt_rq *rt_rq) |
508 | { | |
509 | return 0; | |
510 | } | |
55e12e5e | 511 | #endif /* CONFIG_SMP */ |
ac086bc2 | 512 | |
eff6549b PZ |
513 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
514 | { | |
515 | int i, idle = 1; | |
c6c4927b | 516 | const struct cpumask *span; |
eff6549b | 517 | |
0b148fa0 | 518 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
eff6549b PZ |
519 | return 1; |
520 | ||
521 | span = sched_rt_period_mask(); | |
c6c4927b | 522 | for_each_cpu(i, span) { |
eff6549b PZ |
523 | int enqueue = 0; |
524 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
525 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
526 | ||
527 | spin_lock(&rq->lock); | |
528 | if (rt_rq->rt_time) { | |
529 | u64 runtime; | |
530 | ||
531 | spin_lock(&rt_rq->rt_runtime_lock); | |
532 | if (rt_rq->rt_throttled) | |
533 | balance_runtime(rt_rq); | |
534 | runtime = rt_rq->rt_runtime; | |
535 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
536 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
537 | rt_rq->rt_throttled = 0; | |
538 | enqueue = 1; | |
539 | } | |
540 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
541 | idle = 0; | |
542 | spin_unlock(&rt_rq->rt_runtime_lock); | |
6c3df255 PZ |
543 | } else if (rt_rq->rt_nr_running) |
544 | idle = 0; | |
eff6549b PZ |
545 | |
546 | if (enqueue) | |
547 | sched_rt_rq_enqueue(rt_rq); | |
548 | spin_unlock(&rq->lock); | |
549 | } | |
550 | ||
551 | return idle; | |
552 | } | |
ac086bc2 | 553 | |
6f505b16 PZ |
554 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
555 | { | |
052f1dc7 | 556 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
557 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
558 | ||
559 | if (rt_rq) | |
e864c499 | 560 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
561 | #endif |
562 | ||
563 | return rt_task_of(rt_se)->prio; | |
564 | } | |
565 | ||
9f0c1e56 | 566 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 567 | { |
9f0c1e56 | 568 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 569 | |
fa85ae24 | 570 | if (rt_rq->rt_throttled) |
23b0fdfc | 571 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 572 | |
ac086bc2 PZ |
573 | if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq)) |
574 | return 0; | |
575 | ||
b79f3833 PZ |
576 | balance_runtime(rt_rq); |
577 | runtime = sched_rt_runtime(rt_rq); | |
578 | if (runtime == RUNTIME_INF) | |
579 | return 0; | |
ac086bc2 | 580 | |
9f0c1e56 | 581 | if (rt_rq->rt_time > runtime) { |
6f505b16 | 582 | rt_rq->rt_throttled = 1; |
23b0fdfc | 583 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 584 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
585 | return 1; |
586 | } | |
fa85ae24 PZ |
587 | } |
588 | ||
589 | return 0; | |
590 | } | |
591 | ||
bb44e5d1 IM |
592 | /* |
593 | * Update the current task's runtime statistics. Skip current tasks that | |
594 | * are not in our scheduling class. | |
595 | */ | |
a9957449 | 596 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
597 | { |
598 | struct task_struct *curr = rq->curr; | |
6f505b16 PZ |
599 | struct sched_rt_entity *rt_se = &curr->rt; |
600 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
bb44e5d1 IM |
601 | u64 delta_exec; |
602 | ||
603 | if (!task_has_rt_policy(curr)) | |
604 | return; | |
605 | ||
d281918d | 606 | delta_exec = rq->clock - curr->se.exec_start; |
bb44e5d1 IM |
607 | if (unlikely((s64)delta_exec < 0)) |
608 | delta_exec = 0; | |
6cfb0d5d IM |
609 | |
610 | schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec)); | |
bb44e5d1 IM |
611 | |
612 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
613 | account_group_exec_runtime(curr, delta_exec); |
614 | ||
d281918d | 615 | curr->se.exec_start = rq->clock; |
d842de87 | 616 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 617 | |
e9e9250b PZ |
618 | sched_rt_avg_update(rq, delta_exec); |
619 | ||
0b148fa0 PZ |
620 | if (!rt_bandwidth_enabled()) |
621 | return; | |
622 | ||
354d60c2 DG |
623 | for_each_sched_rt_entity(rt_se) { |
624 | rt_rq = rt_rq_of_se(rt_se); | |
625 | ||
cc2991cf | 626 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
e113a745 | 627 | spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
628 | rt_rq->rt_time += delta_exec; |
629 | if (sched_rt_runtime_exceeded(rt_rq)) | |
630 | resched_task(curr); | |
e113a745 | 631 | spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 632 | } |
354d60c2 | 633 | } |
bb44e5d1 IM |
634 | } |
635 | ||
398a153b | 636 | #if defined CONFIG_SMP |
e864c499 GH |
637 | |
638 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu); | |
639 | ||
640 | static inline int next_prio(struct rq *rq) | |
63489e45 | 641 | { |
e864c499 GH |
642 | struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu); |
643 | ||
644 | if (next && rt_prio(next->prio)) | |
645 | return next->prio; | |
646 | else | |
647 | return MAX_RT_PRIO; | |
648 | } | |
e864c499 | 649 | |
398a153b GH |
650 | static void |
651 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 652 | { |
4d984277 | 653 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 654 | |
398a153b | 655 | if (prio < prev_prio) { |
4d984277 | 656 | |
e864c499 GH |
657 | /* |
658 | * If the new task is higher in priority than anything on the | |
398a153b GH |
659 | * run-queue, we know that the previous high becomes our |
660 | * next-highest. | |
e864c499 | 661 | */ |
398a153b | 662 | rt_rq->highest_prio.next = prev_prio; |
1f11eb6a GH |
663 | |
664 | if (rq->online) | |
4d984277 | 665 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); |
1100ac91 | 666 | |
e864c499 GH |
667 | } else if (prio == rt_rq->highest_prio.curr) |
668 | /* | |
669 | * If the next task is equal in priority to the highest on | |
670 | * the run-queue, then we implicitly know that the next highest | |
671 | * task cannot be any lower than current | |
672 | */ | |
673 | rt_rq->highest_prio.next = prio; | |
674 | else if (prio < rt_rq->highest_prio.next) | |
675 | /* | |
676 | * Otherwise, we need to recompute next-highest | |
677 | */ | |
678 | rt_rq->highest_prio.next = next_prio(rq); | |
398a153b | 679 | } |
73fe6aae | 680 | |
398a153b GH |
681 | static void |
682 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
683 | { | |
684 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 685 | |
398a153b GH |
686 | if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next)) |
687 | rt_rq->highest_prio.next = next_prio(rq); | |
688 | ||
689 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) | |
690 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
691 | } |
692 | ||
398a153b GH |
693 | #else /* CONFIG_SMP */ |
694 | ||
6f505b16 | 695 | static inline |
398a153b GH |
696 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
697 | static inline | |
698 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
699 | ||
700 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 701 | |
052f1dc7 | 702 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
703 | static void |
704 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
705 | { | |
706 | int prev_prio = rt_rq->highest_prio.curr; | |
707 | ||
708 | if (prio < prev_prio) | |
709 | rt_rq->highest_prio.curr = prio; | |
710 | ||
711 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
712 | } | |
713 | ||
714 | static void | |
715 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
716 | { | |
717 | int prev_prio = rt_rq->highest_prio.curr; | |
718 | ||
6f505b16 | 719 | if (rt_rq->rt_nr_running) { |
764a9d6f | 720 | |
398a153b | 721 | WARN_ON(prio < prev_prio); |
764a9d6f | 722 | |
e864c499 | 723 | /* |
398a153b GH |
724 | * This may have been our highest task, and therefore |
725 | * we may have some recomputation to do | |
e864c499 | 726 | */ |
398a153b | 727 | if (prio == prev_prio) { |
e864c499 GH |
728 | struct rt_prio_array *array = &rt_rq->active; |
729 | ||
730 | rt_rq->highest_prio.curr = | |
764a9d6f | 731 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
732 | } |
733 | ||
764a9d6f | 734 | } else |
e864c499 | 735 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 736 | |
398a153b GH |
737 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
738 | } | |
1f11eb6a | 739 | |
398a153b GH |
740 | #else |
741 | ||
742 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
743 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
744 | ||
745 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 746 | |
052f1dc7 | 747 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
748 | |
749 | static void | |
750 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
751 | { | |
752 | if (rt_se_boosted(rt_se)) | |
753 | rt_rq->rt_nr_boosted++; | |
754 | ||
755 | if (rt_rq->tg) | |
756 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
757 | } | |
758 | ||
759 | static void | |
760 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
761 | { | |
23b0fdfc PZ |
762 | if (rt_se_boosted(rt_se)) |
763 | rt_rq->rt_nr_boosted--; | |
764 | ||
765 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
766 | } |
767 | ||
768 | #else /* CONFIG_RT_GROUP_SCHED */ | |
769 | ||
770 | static void | |
771 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
772 | { | |
773 | start_rt_bandwidth(&def_rt_bandwidth); | |
774 | } | |
775 | ||
776 | static inline | |
777 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
778 | ||
779 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
780 | ||
781 | static inline | |
782 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
783 | { | |
784 | int prio = rt_se_prio(rt_se); | |
785 | ||
786 | WARN_ON(!rt_prio(prio)); | |
787 | rt_rq->rt_nr_running++; | |
788 | ||
789 | inc_rt_prio(rt_rq, prio); | |
790 | inc_rt_migration(rt_se, rt_rq); | |
791 | inc_rt_group(rt_se, rt_rq); | |
792 | } | |
793 | ||
794 | static inline | |
795 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
796 | { | |
797 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
798 | WARN_ON(!rt_rq->rt_nr_running); | |
799 | rt_rq->rt_nr_running--; | |
800 | ||
801 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
802 | dec_rt_migration(rt_se, rt_rq); | |
803 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
804 | } |
805 | ||
ad2a3f13 | 806 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se) |
bb44e5d1 | 807 | { |
6f505b16 PZ |
808 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
809 | struct rt_prio_array *array = &rt_rq->active; | |
810 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 811 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 812 | |
ad2a3f13 PZ |
813 | /* |
814 | * Don't enqueue the group if its throttled, or when empty. | |
815 | * The latter is a consequence of the former when a child group | |
816 | * get throttled and the current group doesn't have any other | |
817 | * active members. | |
818 | */ | |
819 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 820 | return; |
63489e45 | 821 | |
7ebefa8c | 822 | list_add_tail(&rt_se->run_list, queue); |
6f505b16 | 823 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 824 | |
6f505b16 PZ |
825 | inc_rt_tasks(rt_se, rt_rq); |
826 | } | |
827 | ||
ad2a3f13 | 828 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
829 | { |
830 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
831 | struct rt_prio_array *array = &rt_rq->active; | |
832 | ||
833 | list_del_init(&rt_se->run_list); | |
834 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
835 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
836 | ||
837 | dec_rt_tasks(rt_se, rt_rq); | |
838 | } | |
839 | ||
840 | /* | |
841 | * Because the prio of an upper entry depends on the lower | |
842 | * entries, we must remove entries top - down. | |
6f505b16 | 843 | */ |
ad2a3f13 | 844 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 845 | { |
ad2a3f13 | 846 | struct sched_rt_entity *back = NULL; |
6f505b16 | 847 | |
58d6c2d7 PZ |
848 | for_each_sched_rt_entity(rt_se) { |
849 | rt_se->back = back; | |
850 | back = rt_se; | |
851 | } | |
852 | ||
853 | for (rt_se = back; rt_se; rt_se = rt_se->back) { | |
854 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
855 | __dequeue_rt_entity(rt_se); |
856 | } | |
857 | } | |
858 | ||
859 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se) | |
860 | { | |
861 | dequeue_rt_stack(rt_se); | |
862 | for_each_sched_rt_entity(rt_se) | |
863 | __enqueue_rt_entity(rt_se); | |
864 | } | |
865 | ||
866 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
867 | { | |
868 | dequeue_rt_stack(rt_se); | |
869 | ||
870 | for_each_sched_rt_entity(rt_se) { | |
871 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
872 | ||
873 | if (rt_rq && rt_rq->rt_nr_running) | |
874 | __enqueue_rt_entity(rt_se); | |
58d6c2d7 | 875 | } |
bb44e5d1 IM |
876 | } |
877 | ||
878 | /* | |
879 | * Adding/removing a task to/from a priority array: | |
880 | */ | |
6f505b16 PZ |
881 | static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup) |
882 | { | |
883 | struct sched_rt_entity *rt_se = &p->rt; | |
884 | ||
885 | if (wakeup) | |
886 | rt_se->timeout = 0; | |
887 | ||
ad2a3f13 | 888 | enqueue_rt_entity(rt_se); |
c09595f6 | 889 | |
917b627d GH |
890 | if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1) |
891 | enqueue_pushable_task(rq, p); | |
6f505b16 PZ |
892 | } |
893 | ||
f02231e5 | 894 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep) |
bb44e5d1 | 895 | { |
6f505b16 | 896 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 897 | |
f1e14ef6 | 898 | update_curr_rt(rq); |
ad2a3f13 | 899 | dequeue_rt_entity(rt_se); |
c09595f6 | 900 | |
917b627d | 901 | dequeue_pushable_task(rq, p); |
bb44e5d1 IM |
902 | } |
903 | ||
904 | /* | |
905 | * Put task to the end of the run list without the overhead of dequeue | |
906 | * followed by enqueue. | |
907 | */ | |
7ebefa8c DA |
908 | static void |
909 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 910 | { |
1cdad715 | 911 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
912 | struct rt_prio_array *array = &rt_rq->active; |
913 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
914 | ||
915 | if (head) | |
916 | list_move(&rt_se->run_list, queue); | |
917 | else | |
918 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 919 | } |
6f505b16 PZ |
920 | } |
921 | ||
7ebefa8c | 922 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 923 | { |
6f505b16 PZ |
924 | struct sched_rt_entity *rt_se = &p->rt; |
925 | struct rt_rq *rt_rq; | |
bb44e5d1 | 926 | |
6f505b16 PZ |
927 | for_each_sched_rt_entity(rt_se) { |
928 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 929 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 930 | } |
bb44e5d1 IM |
931 | } |
932 | ||
6f505b16 | 933 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 934 | { |
7ebefa8c | 935 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
936 | } |
937 | ||
e7693a36 | 938 | #ifdef CONFIG_SMP |
318e0893 GH |
939 | static int find_lowest_rq(struct task_struct *task); |
940 | ||
5f3edc1b | 941 | static int select_task_rq_rt(struct task_struct *p, int flag, int sync) |
e7693a36 | 942 | { |
318e0893 GH |
943 | struct rq *rq = task_rq(p); |
944 | ||
5f3edc1b PZ |
945 | if (flag != SD_BALANCE_WAKE) |
946 | return smp_processor_id(); | |
947 | ||
318e0893 | 948 | /* |
e1f47d89 SR |
949 | * If the current task is an RT task, then |
950 | * try to see if we can wake this RT task up on another | |
951 | * runqueue. Otherwise simply start this RT task | |
952 | * on its current runqueue. | |
953 | * | |
954 | * We want to avoid overloading runqueues. Even if | |
955 | * the RT task is of higher priority than the current RT task. | |
956 | * RT tasks behave differently than other tasks. If | |
957 | * one gets preempted, we try to push it off to another queue. | |
958 | * So trying to keep a preempting RT task on the same | |
959 | * cache hot CPU will force the running RT task to | |
960 | * a cold CPU. So we waste all the cache for the lower | |
961 | * RT task in hopes of saving some of a RT task | |
962 | * that is just being woken and probably will have | |
963 | * cold cache anyway. | |
318e0893 | 964 | */ |
17b3279b | 965 | if (unlikely(rt_task(rq->curr)) && |
6f505b16 | 966 | (p->rt.nr_cpus_allowed > 1)) { |
318e0893 GH |
967 | int cpu = find_lowest_rq(p); |
968 | ||
969 | return (cpu == -1) ? task_cpu(p) : cpu; | |
970 | } | |
971 | ||
972 | /* | |
973 | * Otherwise, just let it ride on the affined RQ and the | |
974 | * post-schedule router will push the preempted task away | |
975 | */ | |
e7693a36 GH |
976 | return task_cpu(p); |
977 | } | |
7ebefa8c DA |
978 | |
979 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
980 | { | |
7ebefa8c DA |
981 | if (rq->curr->rt.nr_cpus_allowed == 1) |
982 | return; | |
983 | ||
24600ce8 | 984 | if (p->rt.nr_cpus_allowed != 1 |
13b8bd0a RR |
985 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
986 | return; | |
24600ce8 | 987 | |
13b8bd0a RR |
988 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
989 | return; | |
7ebefa8c DA |
990 | |
991 | /* | |
992 | * There appears to be other cpus that can accept | |
993 | * current and none to run 'p', so lets reschedule | |
994 | * to try and push current away: | |
995 | */ | |
996 | requeue_task_rt(rq, p, 1); | |
997 | resched_task(rq->curr); | |
998 | } | |
999 | ||
e7693a36 GH |
1000 | #endif /* CONFIG_SMP */ |
1001 | ||
bb44e5d1 IM |
1002 | /* |
1003 | * Preempt the current task with a newly woken task if needed: | |
1004 | */ | |
15afe09b | 1005 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int sync) |
bb44e5d1 | 1006 | { |
45c01e82 | 1007 | if (p->prio < rq->curr->prio) { |
bb44e5d1 | 1008 | resched_task(rq->curr); |
45c01e82 GH |
1009 | return; |
1010 | } | |
1011 | ||
1012 | #ifdef CONFIG_SMP | |
1013 | /* | |
1014 | * If: | |
1015 | * | |
1016 | * - the newly woken task is of equal priority to the current task | |
1017 | * - the newly woken task is non-migratable while current is migratable | |
1018 | * - current will be preempted on the next reschedule | |
1019 | * | |
1020 | * we should check to see if current can readily move to a different | |
1021 | * cpu. If so, we will reschedule to allow the push logic to try | |
1022 | * to move current somewhere else, making room for our non-migratable | |
1023 | * task. | |
1024 | */ | |
7ebefa8c DA |
1025 | if (p->prio == rq->curr->prio && !need_resched()) |
1026 | check_preempt_equal_prio(rq, p); | |
45c01e82 | 1027 | #endif |
bb44e5d1 IM |
1028 | } |
1029 | ||
6f505b16 PZ |
1030 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1031 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1032 | { |
6f505b16 PZ |
1033 | struct rt_prio_array *array = &rt_rq->active; |
1034 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1035 | struct list_head *queue; |
1036 | int idx; | |
1037 | ||
1038 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1039 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1040 | |
1041 | queue = array->queue + idx; | |
6f505b16 | 1042 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1043 | |
6f505b16 PZ |
1044 | return next; |
1045 | } | |
bb44e5d1 | 1046 | |
917b627d | 1047 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1048 | { |
1049 | struct sched_rt_entity *rt_se; | |
1050 | struct task_struct *p; | |
1051 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1052 | |
6f505b16 PZ |
1053 | rt_rq = &rq->rt; |
1054 | ||
1055 | if (unlikely(!rt_rq->rt_nr_running)) | |
1056 | return NULL; | |
1057 | ||
23b0fdfc | 1058 | if (rt_rq_throttled(rt_rq)) |
6f505b16 PZ |
1059 | return NULL; |
1060 | ||
1061 | do { | |
1062 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1063 | BUG_ON(!rt_se); |
6f505b16 PZ |
1064 | rt_rq = group_rt_rq(rt_se); |
1065 | } while (rt_rq); | |
1066 | ||
1067 | p = rt_task_of(rt_se); | |
1068 | p->se.exec_start = rq->clock; | |
917b627d GH |
1069 | |
1070 | return p; | |
1071 | } | |
1072 | ||
1073 | static struct task_struct *pick_next_task_rt(struct rq *rq) | |
1074 | { | |
1075 | struct task_struct *p = _pick_next_task_rt(rq); | |
1076 | ||
1077 | /* The running task is never eligible for pushing */ | |
1078 | if (p) | |
1079 | dequeue_pushable_task(rq, p); | |
1080 | ||
bcf08df3 | 1081 | #ifdef CONFIG_SMP |
3f029d3c GH |
1082 | /* |
1083 | * We detect this state here so that we can avoid taking the RQ | |
1084 | * lock again later if there is no need to push | |
1085 | */ | |
1086 | rq->post_schedule = has_pushable_tasks(rq); | |
bcf08df3 | 1087 | #endif |
3f029d3c | 1088 | |
6f505b16 | 1089 | return p; |
bb44e5d1 IM |
1090 | } |
1091 | ||
31ee529c | 1092 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1093 | { |
f1e14ef6 | 1094 | update_curr_rt(rq); |
bb44e5d1 | 1095 | p->se.exec_start = 0; |
917b627d GH |
1096 | |
1097 | /* | |
1098 | * The previous task needs to be made eligible for pushing | |
1099 | * if it is still active | |
1100 | */ | |
1101 | if (p->se.on_rq && p->rt.nr_cpus_allowed > 1) | |
1102 | enqueue_pushable_task(rq, p); | |
bb44e5d1 IM |
1103 | } |
1104 | ||
681f3e68 | 1105 | #ifdef CONFIG_SMP |
6f505b16 | 1106 | |
e8fa1362 SR |
1107 | /* Only try algorithms three times */ |
1108 | #define RT_MAX_TRIES 3 | |
1109 | ||
e8fa1362 SR |
1110 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); |
1111 | ||
f65eda4f SR |
1112 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1113 | { | |
1114 | if (!task_running(rq, p) && | |
96f874e2 | 1115 | (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) && |
6f505b16 | 1116 | (p->rt.nr_cpus_allowed > 1)) |
f65eda4f SR |
1117 | return 1; |
1118 | return 0; | |
1119 | } | |
1120 | ||
e8fa1362 | 1121 | /* Return the second highest RT task, NULL otherwise */ |
79064fbf | 1122 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
e8fa1362 | 1123 | { |
6f505b16 PZ |
1124 | struct task_struct *next = NULL; |
1125 | struct sched_rt_entity *rt_se; | |
1126 | struct rt_prio_array *array; | |
1127 | struct rt_rq *rt_rq; | |
e8fa1362 SR |
1128 | int idx; |
1129 | ||
6f505b16 PZ |
1130 | for_each_leaf_rt_rq(rt_rq, rq) { |
1131 | array = &rt_rq->active; | |
1132 | idx = sched_find_first_bit(array->bitmap); | |
1133 | next_idx: | |
1134 | if (idx >= MAX_RT_PRIO) | |
1135 | continue; | |
1136 | if (next && next->prio < idx) | |
1137 | continue; | |
1138 | list_for_each_entry(rt_se, array->queue + idx, run_list) { | |
1139 | struct task_struct *p = rt_task_of(rt_se); | |
1140 | if (pick_rt_task(rq, p, cpu)) { | |
1141 | next = p; | |
1142 | break; | |
1143 | } | |
1144 | } | |
1145 | if (!next) { | |
1146 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); | |
1147 | goto next_idx; | |
1148 | } | |
f65eda4f SR |
1149 | } |
1150 | ||
e8fa1362 SR |
1151 | return next; |
1152 | } | |
1153 | ||
0e3900e6 | 1154 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1155 | |
d38b223c MT |
1156 | static inline int pick_optimal_cpu(int this_cpu, |
1157 | const struct cpumask *mask) | |
6e1254d2 GH |
1158 | { |
1159 | int first; | |
1160 | ||
1161 | /* "this_cpu" is cheaper to preempt than a remote processor */ | |
d38b223c | 1162 | if ((this_cpu != -1) && cpumask_test_cpu(this_cpu, mask)) |
6e1254d2 GH |
1163 | return this_cpu; |
1164 | ||
3d398703 RR |
1165 | first = cpumask_first(mask); |
1166 | if (first < nr_cpu_ids) | |
6e1254d2 GH |
1167 | return first; |
1168 | ||
1169 | return -1; | |
1170 | } | |
1171 | ||
1172 | static int find_lowest_rq(struct task_struct *task) | |
1173 | { | |
1174 | struct sched_domain *sd; | |
96f874e2 | 1175 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
6e1254d2 GH |
1176 | int this_cpu = smp_processor_id(); |
1177 | int cpu = task_cpu(task); | |
d38b223c | 1178 | cpumask_var_t domain_mask; |
06f90dbd | 1179 | |
6e0534f2 GH |
1180 | if (task->rt.nr_cpus_allowed == 1) |
1181 | return -1; /* No other targets possible */ | |
6e1254d2 | 1182 | |
6e0534f2 GH |
1183 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1184 | return -1; /* No targets found */ | |
6e1254d2 GH |
1185 | |
1186 | /* | |
1187 | * At this point we have built a mask of cpus representing the | |
1188 | * lowest priority tasks in the system. Now we want to elect | |
1189 | * the best one based on our affinity and topology. | |
1190 | * | |
1191 | * We prioritize the last cpu that the task executed on since | |
1192 | * it is most likely cache-hot in that location. | |
1193 | */ | |
96f874e2 | 1194 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1195 | return cpu; |
1196 | ||
1197 | /* | |
1198 | * Otherwise, we consult the sched_domains span maps to figure | |
1199 | * out which cpu is logically closest to our hot cache data. | |
1200 | */ | |
1201 | if (this_cpu == cpu) | |
1202 | this_cpu = -1; /* Skip this_cpu opt if the same */ | |
1203 | ||
d38b223c MT |
1204 | if (alloc_cpumask_var(&domain_mask, GFP_ATOMIC)) { |
1205 | for_each_domain(cpu, sd) { | |
1206 | if (sd->flags & SD_WAKE_AFFINE) { | |
1207 | int best_cpu; | |
6e1254d2 | 1208 | |
d38b223c MT |
1209 | cpumask_and(domain_mask, |
1210 | sched_domain_span(sd), | |
1211 | lowest_mask); | |
6e1254d2 | 1212 | |
d38b223c MT |
1213 | best_cpu = pick_optimal_cpu(this_cpu, |
1214 | domain_mask); | |
6e1254d2 | 1215 | |
d38b223c MT |
1216 | if (best_cpu != -1) { |
1217 | free_cpumask_var(domain_mask); | |
1218 | return best_cpu; | |
1219 | } | |
1220 | } | |
6e1254d2 | 1221 | } |
d38b223c | 1222 | free_cpumask_var(domain_mask); |
6e1254d2 GH |
1223 | } |
1224 | ||
1225 | /* | |
1226 | * And finally, if there were no matches within the domains | |
1227 | * just give the caller *something* to work with from the compatible | |
1228 | * locations. | |
1229 | */ | |
1230 | return pick_optimal_cpu(this_cpu, lowest_mask); | |
07b4032c GH |
1231 | } |
1232 | ||
1233 | /* Will lock the rq it finds */ | |
4df64c0b | 1234 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1235 | { |
1236 | struct rq *lowest_rq = NULL; | |
07b4032c | 1237 | int tries; |
4df64c0b | 1238 | int cpu; |
e8fa1362 | 1239 | |
07b4032c GH |
1240 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1241 | cpu = find_lowest_rq(task); | |
1242 | ||
2de0b463 | 1243 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1244 | break; |
1245 | ||
07b4032c GH |
1246 | lowest_rq = cpu_rq(cpu); |
1247 | ||
e8fa1362 | 1248 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1249 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1250 | /* |
1251 | * We had to unlock the run queue. In | |
1252 | * the mean time, task could have | |
1253 | * migrated already or had its affinity changed. | |
1254 | * Also make sure that it wasn't scheduled on its rq. | |
1255 | */ | |
07b4032c | 1256 | if (unlikely(task_rq(task) != rq || |
96f874e2 RR |
1257 | !cpumask_test_cpu(lowest_rq->cpu, |
1258 | &task->cpus_allowed) || | |
07b4032c | 1259 | task_running(rq, task) || |
e8fa1362 | 1260 | !task->se.on_rq)) { |
4df64c0b | 1261 | |
e8fa1362 SR |
1262 | spin_unlock(&lowest_rq->lock); |
1263 | lowest_rq = NULL; | |
1264 | break; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | /* If this rq is still suitable use it. */ | |
e864c499 | 1269 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1270 | break; |
1271 | ||
1272 | /* try again */ | |
1b12bbc7 | 1273 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1274 | lowest_rq = NULL; |
1275 | } | |
1276 | ||
1277 | return lowest_rq; | |
1278 | } | |
1279 | ||
917b627d GH |
1280 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1281 | { | |
1282 | struct task_struct *p; | |
1283 | ||
1284 | if (!has_pushable_tasks(rq)) | |
1285 | return NULL; | |
1286 | ||
1287 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1288 | struct task_struct, pushable_tasks); | |
1289 | ||
1290 | BUG_ON(rq->cpu != task_cpu(p)); | |
1291 | BUG_ON(task_current(rq, p)); | |
1292 | BUG_ON(p->rt.nr_cpus_allowed <= 1); | |
1293 | ||
1294 | BUG_ON(!p->se.on_rq); | |
1295 | BUG_ON(!rt_task(p)); | |
1296 | ||
1297 | return p; | |
1298 | } | |
1299 | ||
e8fa1362 SR |
1300 | /* |
1301 | * If the current CPU has more than one RT task, see if the non | |
1302 | * running task can migrate over to a CPU that is running a task | |
1303 | * of lesser priority. | |
1304 | */ | |
697f0a48 | 1305 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1306 | { |
1307 | struct task_struct *next_task; | |
1308 | struct rq *lowest_rq; | |
e8fa1362 | 1309 | |
a22d7fc1 GH |
1310 | if (!rq->rt.overloaded) |
1311 | return 0; | |
1312 | ||
917b627d | 1313 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1314 | if (!next_task) |
1315 | return 0; | |
1316 | ||
1317 | retry: | |
697f0a48 | 1318 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1319 | WARN_ON(1); |
e8fa1362 | 1320 | return 0; |
f65eda4f | 1321 | } |
e8fa1362 SR |
1322 | |
1323 | /* | |
1324 | * It's possible that the next_task slipped in of | |
1325 | * higher priority than current. If that's the case | |
1326 | * just reschedule current. | |
1327 | */ | |
697f0a48 GH |
1328 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1329 | resched_task(rq->curr); | |
e8fa1362 SR |
1330 | return 0; |
1331 | } | |
1332 | ||
697f0a48 | 1333 | /* We might release rq lock */ |
e8fa1362 SR |
1334 | get_task_struct(next_task); |
1335 | ||
1336 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1337 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1338 | if (!lowest_rq) { |
1339 | struct task_struct *task; | |
1340 | /* | |
697f0a48 | 1341 | * find lock_lowest_rq releases rq->lock |
1563513d GH |
1342 | * so it is possible that next_task has migrated. |
1343 | * | |
1344 | * We need to make sure that the task is still on the same | |
1345 | * run-queue and is also still the next task eligible for | |
1346 | * pushing. | |
e8fa1362 | 1347 | */ |
917b627d | 1348 | task = pick_next_pushable_task(rq); |
1563513d GH |
1349 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1350 | /* | |
1351 | * If we get here, the task hasnt moved at all, but | |
1352 | * it has failed to push. We will not try again, | |
1353 | * since the other cpus will pull from us when they | |
1354 | * are ready. | |
1355 | */ | |
1356 | dequeue_pushable_task(rq, next_task); | |
1357 | goto out; | |
e8fa1362 | 1358 | } |
917b627d | 1359 | |
1563513d GH |
1360 | if (!task) |
1361 | /* No more tasks, just exit */ | |
1362 | goto out; | |
1363 | ||
917b627d | 1364 | /* |
1563513d | 1365 | * Something has shifted, try again. |
917b627d | 1366 | */ |
1563513d GH |
1367 | put_task_struct(next_task); |
1368 | next_task = task; | |
1369 | goto retry; | |
e8fa1362 SR |
1370 | } |
1371 | ||
697f0a48 | 1372 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1373 | set_task_cpu(next_task, lowest_rq->cpu); |
1374 | activate_task(lowest_rq, next_task, 0); | |
1375 | ||
1376 | resched_task(lowest_rq->curr); | |
1377 | ||
1b12bbc7 | 1378 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1379 | |
e8fa1362 SR |
1380 | out: |
1381 | put_task_struct(next_task); | |
1382 | ||
917b627d | 1383 | return 1; |
e8fa1362 SR |
1384 | } |
1385 | ||
e8fa1362 SR |
1386 | static void push_rt_tasks(struct rq *rq) |
1387 | { | |
1388 | /* push_rt_task will return true if it moved an RT */ | |
1389 | while (push_rt_task(rq)) | |
1390 | ; | |
1391 | } | |
1392 | ||
f65eda4f SR |
1393 | static int pull_rt_task(struct rq *this_rq) |
1394 | { | |
80bf3171 | 1395 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
a8728944 | 1396 | struct task_struct *p; |
f65eda4f | 1397 | struct rq *src_rq; |
f65eda4f | 1398 | |
637f5085 | 1399 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
1400 | return 0; |
1401 | ||
c6c4927b | 1402 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1403 | if (this_cpu == cpu) |
1404 | continue; | |
1405 | ||
1406 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1407 | |
1408 | /* | |
1409 | * Don't bother taking the src_rq->lock if the next highest | |
1410 | * task is known to be lower-priority than our current task. | |
1411 | * This may look racy, but if this value is about to go | |
1412 | * logically higher, the src_rq will push this task away. | |
1413 | * And if its going logically lower, we do not care | |
1414 | */ | |
1415 | if (src_rq->rt.highest_prio.next >= | |
1416 | this_rq->rt.highest_prio.curr) | |
1417 | continue; | |
1418 | ||
f65eda4f SR |
1419 | /* |
1420 | * We can potentially drop this_rq's lock in | |
1421 | * double_lock_balance, and another CPU could | |
a8728944 | 1422 | * alter this_rq |
f65eda4f | 1423 | */ |
a8728944 | 1424 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
1425 | |
1426 | /* | |
1427 | * Are there still pullable RT tasks? | |
1428 | */ | |
614ee1f6 MG |
1429 | if (src_rq->rt.rt_nr_running <= 1) |
1430 | goto skip; | |
f65eda4f | 1431 | |
f65eda4f SR |
1432 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
1433 | ||
1434 | /* | |
1435 | * Do we have an RT task that preempts | |
1436 | * the to-be-scheduled task? | |
1437 | */ | |
a8728944 | 1438 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f SR |
1439 | WARN_ON(p == src_rq->curr); |
1440 | WARN_ON(!p->se.on_rq); | |
1441 | ||
1442 | /* | |
1443 | * There's a chance that p is higher in priority | |
1444 | * than what's currently running on its cpu. | |
1445 | * This is just that p is wakeing up and hasn't | |
1446 | * had a chance to schedule. We only pull | |
1447 | * p if it is lower in priority than the | |
a8728944 | 1448 | * current task on the run queue |
f65eda4f | 1449 | */ |
a8728944 | 1450 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 1451 | goto skip; |
f65eda4f SR |
1452 | |
1453 | ret = 1; | |
1454 | ||
1455 | deactivate_task(src_rq, p, 0); | |
1456 | set_task_cpu(p, this_cpu); | |
1457 | activate_task(this_rq, p, 0); | |
1458 | /* | |
1459 | * We continue with the search, just in | |
1460 | * case there's an even higher prio task | |
1461 | * in another runqueue. (low likelyhood | |
1462 | * but possible) | |
f65eda4f | 1463 | */ |
f65eda4f | 1464 | } |
614ee1f6 | 1465 | skip: |
1b12bbc7 | 1466 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
1467 | } |
1468 | ||
1469 | return ret; | |
1470 | } | |
1471 | ||
9a897c5a | 1472 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
f65eda4f SR |
1473 | { |
1474 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
e864c499 | 1475 | if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio) |
f65eda4f SR |
1476 | pull_rt_task(rq); |
1477 | } | |
1478 | ||
9a897c5a | 1479 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 | 1480 | { |
967fc046 | 1481 | push_rt_tasks(rq); |
e8fa1362 SR |
1482 | } |
1483 | ||
8ae121ac GH |
1484 | /* |
1485 | * If we are not running and we are not going to reschedule soon, we should | |
1486 | * try to push tasks away now | |
1487 | */ | |
9a897c5a | 1488 | static void task_wake_up_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1489 | { |
9a897c5a | 1490 | if (!task_running(rq, p) && |
8ae121ac | 1491 | !test_tsk_need_resched(rq->curr) && |
917b627d | 1492 | has_pushable_tasks(rq) && |
777c2f38 | 1493 | p->rt.nr_cpus_allowed > 1) |
4642dafd SR |
1494 | push_rt_tasks(rq); |
1495 | } | |
1496 | ||
43010659 | 1497 | static unsigned long |
bb44e5d1 | 1498 | load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f PW |
1499 | unsigned long max_load_move, |
1500 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1501 | int *all_pinned, int *this_best_prio) | |
bb44e5d1 | 1502 | { |
c7a1e46a SR |
1503 | /* don't touch RT tasks */ |
1504 | return 0; | |
e1d1484f PW |
1505 | } |
1506 | ||
1507 | static int | |
1508 | move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1509 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1510 | { | |
c7a1e46a SR |
1511 | /* don't touch RT tasks */ |
1512 | return 0; | |
bb44e5d1 | 1513 | } |
deeeccd4 | 1514 | |
cd8ba7cd | 1515 | static void set_cpus_allowed_rt(struct task_struct *p, |
96f874e2 | 1516 | const struct cpumask *new_mask) |
73fe6aae | 1517 | { |
96f874e2 | 1518 | int weight = cpumask_weight(new_mask); |
73fe6aae GH |
1519 | |
1520 | BUG_ON(!rt_task(p)); | |
1521 | ||
1522 | /* | |
1523 | * Update the migration status of the RQ if we have an RT task | |
1524 | * which is running AND changing its weight value. | |
1525 | */ | |
6f505b16 | 1526 | if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) { |
73fe6aae GH |
1527 | struct rq *rq = task_rq(p); |
1528 | ||
917b627d GH |
1529 | if (!task_current(rq, p)) { |
1530 | /* | |
1531 | * Make sure we dequeue this task from the pushable list | |
1532 | * before going further. It will either remain off of | |
1533 | * the list because we are no longer pushable, or it | |
1534 | * will be requeued. | |
1535 | */ | |
1536 | if (p->rt.nr_cpus_allowed > 1) | |
1537 | dequeue_pushable_task(rq, p); | |
1538 | ||
1539 | /* | |
1540 | * Requeue if our weight is changing and still > 1 | |
1541 | */ | |
1542 | if (weight > 1) | |
1543 | enqueue_pushable_task(rq, p); | |
1544 | ||
1545 | } | |
1546 | ||
6f505b16 | 1547 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
73fe6aae | 1548 | rq->rt.rt_nr_migratory++; |
6f505b16 | 1549 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
73fe6aae GH |
1550 | BUG_ON(!rq->rt.rt_nr_migratory); |
1551 | rq->rt.rt_nr_migratory--; | |
1552 | } | |
1553 | ||
398a153b | 1554 | update_rt_migration(&rq->rt); |
73fe6aae GH |
1555 | } |
1556 | ||
96f874e2 | 1557 | cpumask_copy(&p->cpus_allowed, new_mask); |
6f505b16 | 1558 | p->rt.nr_cpus_allowed = weight; |
73fe6aae | 1559 | } |
deeeccd4 | 1560 | |
bdd7c81b | 1561 | /* Assumes rq->lock is held */ |
1f11eb6a | 1562 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
1563 | { |
1564 | if (rq->rt.overloaded) | |
1565 | rt_set_overload(rq); | |
6e0534f2 | 1566 | |
7def2be1 PZ |
1567 | __enable_runtime(rq); |
1568 | ||
e864c499 | 1569 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
1570 | } |
1571 | ||
1572 | /* Assumes rq->lock is held */ | |
1f11eb6a | 1573 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
1574 | { |
1575 | if (rq->rt.overloaded) | |
1576 | rt_clear_overload(rq); | |
6e0534f2 | 1577 | |
7def2be1 PZ |
1578 | __disable_runtime(rq); |
1579 | ||
6e0534f2 | 1580 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 1581 | } |
cb469845 SR |
1582 | |
1583 | /* | |
1584 | * When switch from the rt queue, we bring ourselves to a position | |
1585 | * that we might want to pull RT tasks from other runqueues. | |
1586 | */ | |
1587 | static void switched_from_rt(struct rq *rq, struct task_struct *p, | |
1588 | int running) | |
1589 | { | |
1590 | /* | |
1591 | * If there are other RT tasks then we will reschedule | |
1592 | * and the scheduling of the other RT tasks will handle | |
1593 | * the balancing. But if we are the last RT task | |
1594 | * we may need to handle the pulling of RT tasks | |
1595 | * now. | |
1596 | */ | |
1597 | if (!rq->rt.rt_nr_running) | |
1598 | pull_rt_task(rq); | |
1599 | } | |
3d8cbdf8 RR |
1600 | |
1601 | static inline void init_sched_rt_class(void) | |
1602 | { | |
1603 | unsigned int i; | |
1604 | ||
1605 | for_each_possible_cpu(i) | |
eaa95840 | 1606 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 1607 | GFP_KERNEL, cpu_to_node(i)); |
3d8cbdf8 | 1608 | } |
cb469845 SR |
1609 | #endif /* CONFIG_SMP */ |
1610 | ||
1611 | /* | |
1612 | * When switching a task to RT, we may overload the runqueue | |
1613 | * with RT tasks. In this case we try to push them off to | |
1614 | * other runqueues. | |
1615 | */ | |
1616 | static void switched_to_rt(struct rq *rq, struct task_struct *p, | |
1617 | int running) | |
1618 | { | |
1619 | int check_resched = 1; | |
1620 | ||
1621 | /* | |
1622 | * If we are already running, then there's nothing | |
1623 | * that needs to be done. But if we are not running | |
1624 | * we may need to preempt the current running task. | |
1625 | * If that current running task is also an RT task | |
1626 | * then see if we can move to another run queue. | |
1627 | */ | |
1628 | if (!running) { | |
1629 | #ifdef CONFIG_SMP | |
1630 | if (rq->rt.overloaded && push_rt_task(rq) && | |
1631 | /* Don't resched if we changed runqueues */ | |
1632 | rq != task_rq(p)) | |
1633 | check_resched = 0; | |
1634 | #endif /* CONFIG_SMP */ | |
1635 | if (check_resched && p->prio < rq->curr->prio) | |
1636 | resched_task(rq->curr); | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | /* | |
1641 | * Priority of the task has changed. This may cause | |
1642 | * us to initiate a push or pull. | |
1643 | */ | |
1644 | static void prio_changed_rt(struct rq *rq, struct task_struct *p, | |
1645 | int oldprio, int running) | |
1646 | { | |
1647 | if (running) { | |
1648 | #ifdef CONFIG_SMP | |
1649 | /* | |
1650 | * If our priority decreases while running, we | |
1651 | * may need to pull tasks to this runqueue. | |
1652 | */ | |
1653 | if (oldprio < p->prio) | |
1654 | pull_rt_task(rq); | |
1655 | /* | |
1656 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
1657 | * then reschedule. Note, the above pull_rt_task |
1658 | * can release the rq lock and p could migrate. | |
1659 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 1660 | */ |
e864c499 | 1661 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
cb469845 SR |
1662 | resched_task(p); |
1663 | #else | |
1664 | /* For UP simply resched on drop of prio */ | |
1665 | if (oldprio < p->prio) | |
1666 | resched_task(p); | |
e8fa1362 | 1667 | #endif /* CONFIG_SMP */ |
cb469845 SR |
1668 | } else { |
1669 | /* | |
1670 | * This task is not running, but if it is | |
1671 | * greater than the current running task | |
1672 | * then reschedule. | |
1673 | */ | |
1674 | if (p->prio < rq->curr->prio) | |
1675 | resched_task(rq->curr); | |
1676 | } | |
1677 | } | |
1678 | ||
78f2c7db PZ |
1679 | static void watchdog(struct rq *rq, struct task_struct *p) |
1680 | { | |
1681 | unsigned long soft, hard; | |
1682 | ||
1683 | if (!p->signal) | |
1684 | return; | |
1685 | ||
1686 | soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur; | |
1687 | hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max; | |
1688 | ||
1689 | if (soft != RLIM_INFINITY) { | |
1690 | unsigned long next; | |
1691 | ||
1692 | p->rt.timeout++; | |
1693 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); | |
5a52dd50 | 1694 | if (p->rt.timeout > next) |
f06febc9 | 1695 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
1696 | } |
1697 | } | |
bb44e5d1 | 1698 | |
8f4d37ec | 1699 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 1700 | { |
67e2be02 PZ |
1701 | update_curr_rt(rq); |
1702 | ||
78f2c7db PZ |
1703 | watchdog(rq, p); |
1704 | ||
bb44e5d1 IM |
1705 | /* |
1706 | * RR tasks need a special form of timeslice management. | |
1707 | * FIFO tasks have no timeslices. | |
1708 | */ | |
1709 | if (p->policy != SCHED_RR) | |
1710 | return; | |
1711 | ||
fa717060 | 1712 | if (--p->rt.time_slice) |
bb44e5d1 IM |
1713 | return; |
1714 | ||
fa717060 | 1715 | p->rt.time_slice = DEF_TIMESLICE; |
bb44e5d1 | 1716 | |
98fbc798 DA |
1717 | /* |
1718 | * Requeue to the end of queue if we are not the only element | |
1719 | * on the queue: | |
1720 | */ | |
fa717060 | 1721 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
7ebefa8c | 1722 | requeue_task_rt(rq, p, 0); |
98fbc798 DA |
1723 | set_tsk_need_resched(p); |
1724 | } | |
bb44e5d1 IM |
1725 | } |
1726 | ||
83b699ed SV |
1727 | static void set_curr_task_rt(struct rq *rq) |
1728 | { | |
1729 | struct task_struct *p = rq->curr; | |
1730 | ||
1731 | p->se.exec_start = rq->clock; | |
917b627d GH |
1732 | |
1733 | /* The running task is never eligible for pushing */ | |
1734 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
1735 | } |
1736 | ||
2abdad0a | 1737 | static const struct sched_class rt_sched_class = { |
5522d5d5 | 1738 | .next = &fair_sched_class, |
bb44e5d1 IM |
1739 | .enqueue_task = enqueue_task_rt, |
1740 | .dequeue_task = dequeue_task_rt, | |
1741 | .yield_task = yield_task_rt, | |
1742 | ||
1743 | .check_preempt_curr = check_preempt_curr_rt, | |
1744 | ||
1745 | .pick_next_task = pick_next_task_rt, | |
1746 | .put_prev_task = put_prev_task_rt, | |
1747 | ||
681f3e68 | 1748 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
1749 | .select_task_rq = select_task_rq_rt, |
1750 | ||
bb44e5d1 | 1751 | .load_balance = load_balance_rt, |
e1d1484f | 1752 | .move_one_task = move_one_task_rt, |
73fe6aae | 1753 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
1754 | .rq_online = rq_online_rt, |
1755 | .rq_offline = rq_offline_rt, | |
9a897c5a SR |
1756 | .pre_schedule = pre_schedule_rt, |
1757 | .post_schedule = post_schedule_rt, | |
1758 | .task_wake_up = task_wake_up_rt, | |
cb469845 | 1759 | .switched_from = switched_from_rt, |
681f3e68 | 1760 | #endif |
bb44e5d1 | 1761 | |
83b699ed | 1762 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 1763 | .task_tick = task_tick_rt, |
cb469845 SR |
1764 | |
1765 | .prio_changed = prio_changed_rt, | |
1766 | .switched_to = switched_to_rt, | |
bb44e5d1 | 1767 | }; |
ada18de2 PZ |
1768 | |
1769 | #ifdef CONFIG_SCHED_DEBUG | |
1770 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
1771 | ||
1772 | static void print_rt_stats(struct seq_file *m, int cpu) | |
1773 | { | |
1774 | struct rt_rq *rt_rq; | |
1775 | ||
1776 | rcu_read_lock(); | |
1777 | for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu)) | |
1778 | print_rt_rq(m, cpu, rt_rq); | |
1779 | rcu_read_unlock(); | |
1780 | } | |
55e12e5e | 1781 | #endif /* CONFIG_SCHED_DEBUG */ |
0e3900e6 | 1782 |