Commit | Line | Data |
---|---|---|
bb44e5d1 IM |
1 | /* |
2 | * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR | |
3 | * policies) | |
4 | */ | |
5 | ||
398a153b GH |
6 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
7 | { | |
8 | return container_of(rt_se, struct task_struct, rt); | |
9 | } | |
10 | ||
11 | #ifdef CONFIG_RT_GROUP_SCHED | |
12 | ||
13 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) | |
14 | { | |
15 | return rt_rq->rq; | |
16 | } | |
17 | ||
18 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
19 | { | |
20 | return rt_se->rt_rq; | |
21 | } | |
22 | ||
23 | #else /* CONFIG_RT_GROUP_SCHED */ | |
24 | ||
25 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) | |
26 | { | |
27 | return container_of(rt_rq, struct rq, rt); | |
28 | } | |
29 | ||
30 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
31 | { | |
32 | struct task_struct *p = rt_task_of(rt_se); | |
33 | struct rq *rq = task_rq(p); | |
34 | ||
35 | return &rq->rt; | |
36 | } | |
37 | ||
38 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
39 | ||
4fd29176 | 40 | #ifdef CONFIG_SMP |
84de4274 | 41 | |
637f5085 | 42 | static inline int rt_overloaded(struct rq *rq) |
4fd29176 | 43 | { |
637f5085 | 44 | return atomic_read(&rq->rd->rto_count); |
4fd29176 | 45 | } |
84de4274 | 46 | |
4fd29176 SR |
47 | static inline void rt_set_overload(struct rq *rq) |
48 | { | |
1f11eb6a GH |
49 | if (!rq->online) |
50 | return; | |
51 | ||
c6c4927b | 52 | cpumask_set_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 SR |
53 | /* |
54 | * Make sure the mask is visible before we set | |
55 | * the overload count. That is checked to determine | |
56 | * if we should look at the mask. It would be a shame | |
57 | * if we looked at the mask, but the mask was not | |
58 | * updated yet. | |
59 | */ | |
60 | wmb(); | |
637f5085 | 61 | atomic_inc(&rq->rd->rto_count); |
4fd29176 | 62 | } |
84de4274 | 63 | |
4fd29176 SR |
64 | static inline void rt_clear_overload(struct rq *rq) |
65 | { | |
1f11eb6a GH |
66 | if (!rq->online) |
67 | return; | |
68 | ||
4fd29176 | 69 | /* the order here really doesn't matter */ |
637f5085 | 70 | atomic_dec(&rq->rd->rto_count); |
c6c4927b | 71 | cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 | 72 | } |
73fe6aae | 73 | |
398a153b | 74 | static void update_rt_migration(struct rt_rq *rt_rq) |
73fe6aae | 75 | { |
398a153b GH |
76 | if (rt_rq->rt_nr_migratory && (rt_rq->rt_nr_running > 1)) { |
77 | if (!rt_rq->overloaded) { | |
78 | rt_set_overload(rq_of_rt_rq(rt_rq)); | |
79 | rt_rq->overloaded = 1; | |
cdc8eb98 | 80 | } |
398a153b GH |
81 | } else if (rt_rq->overloaded) { |
82 | rt_clear_overload(rq_of_rt_rq(rt_rq)); | |
83 | rt_rq->overloaded = 0; | |
637f5085 | 84 | } |
73fe6aae | 85 | } |
917b627d | 86 | |
398a153b GH |
87 | static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
88 | { | |
89 | if (rt_se->nr_cpus_allowed > 1) | |
90 | rt_rq->rt_nr_migratory++; | |
91 | ||
92 | update_rt_migration(rt_rq); | |
93 | } | |
94 | ||
95 | static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
96 | { | |
97 | if (rt_se->nr_cpus_allowed > 1) | |
98 | rt_rq->rt_nr_migratory--; | |
99 | ||
100 | update_rt_migration(rt_rq); | |
101 | } | |
102 | ||
917b627d GH |
103 | static void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
104 | { | |
105 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
106 | plist_node_init(&p->pushable_tasks, p->prio); | |
107 | plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
108 | } | |
109 | ||
110 | static void dequeue_pushable_task(struct rq *rq, struct task_struct *p) | |
111 | { | |
112 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
113 | } | |
114 | ||
115 | #else | |
116 | ||
b07430ac GH |
117 | static inline |
118 | void enqueue_pushable_task(struct rq *rq, struct task_struct *p) {} | |
119 | static inline | |
120 | void dequeue_pushable_task(struct rq *rq, struct task_struct *p) {} | |
398a153b GH |
121 | static inline |
122 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
123 | static inline | |
124 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
917b627d | 125 | |
4fd29176 SR |
126 | #endif /* CONFIG_SMP */ |
127 | ||
6f505b16 PZ |
128 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
129 | { | |
130 | return !list_empty(&rt_se->run_list); | |
131 | } | |
132 | ||
052f1dc7 | 133 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 134 | |
9f0c1e56 | 135 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
136 | { |
137 | if (!rt_rq->tg) | |
9f0c1e56 | 138 | return RUNTIME_INF; |
6f505b16 | 139 | |
ac086bc2 PZ |
140 | return rt_rq->rt_runtime; |
141 | } | |
142 | ||
143 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
144 | { | |
145 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); | |
6f505b16 PZ |
146 | } |
147 | ||
148 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
80f40ee4 | 149 | list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) |
6f505b16 | 150 | |
6f505b16 PZ |
151 | #define for_each_sched_rt_entity(rt_se) \ |
152 | for (; rt_se; rt_se = rt_se->parent) | |
153 | ||
154 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
155 | { | |
156 | return rt_se->my_q; | |
157 | } | |
158 | ||
159 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se); | |
160 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); | |
161 | ||
9f0c1e56 | 162 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 163 | { |
f6121f4f | 164 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
6f505b16 PZ |
165 | struct sched_rt_entity *rt_se = rt_rq->rt_se; |
166 | ||
f6121f4f DF |
167 | if (rt_rq->rt_nr_running) { |
168 | if (rt_se && !on_rt_rq(rt_se)) | |
169 | enqueue_rt_entity(rt_se); | |
e864c499 | 170 | if (rt_rq->highest_prio.curr < curr->prio) |
1020387f | 171 | resched_task(curr); |
6f505b16 PZ |
172 | } |
173 | } | |
174 | ||
9f0c1e56 | 175 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
176 | { |
177 | struct sched_rt_entity *rt_se = rt_rq->rt_se; | |
178 | ||
179 | if (rt_se && on_rt_rq(rt_se)) | |
180 | dequeue_rt_entity(rt_se); | |
181 | } | |
182 | ||
23b0fdfc PZ |
183 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
184 | { | |
185 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
186 | } | |
187 | ||
188 | static int rt_se_boosted(struct sched_rt_entity *rt_se) | |
189 | { | |
190 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
191 | struct task_struct *p; | |
192 | ||
193 | if (rt_rq) | |
194 | return !!rt_rq->rt_nr_boosted; | |
195 | ||
196 | p = rt_task_of(rt_se); | |
197 | return p->prio != p->normal_prio; | |
198 | } | |
199 | ||
d0b27fa7 | 200 | #ifdef CONFIG_SMP |
c6c4927b | 201 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 PZ |
202 | { |
203 | return cpu_rq(smp_processor_id())->rd->span; | |
204 | } | |
6f505b16 | 205 | #else |
c6c4927b | 206 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 207 | { |
c6c4927b | 208 | return cpu_online_mask; |
d0b27fa7 PZ |
209 | } |
210 | #endif | |
6f505b16 | 211 | |
d0b27fa7 PZ |
212 | static inline |
213 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 214 | { |
d0b27fa7 PZ |
215 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
216 | } | |
9f0c1e56 | 217 | |
ac086bc2 PZ |
218 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
219 | { | |
220 | return &rt_rq->tg->rt_bandwidth; | |
221 | } | |
222 | ||
55e12e5e | 223 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
224 | |
225 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
226 | { | |
ac086bc2 PZ |
227 | return rt_rq->rt_runtime; |
228 | } | |
229 | ||
230 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
231 | { | |
232 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
233 | } |
234 | ||
235 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
236 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
237 | ||
6f505b16 PZ |
238 | #define for_each_sched_rt_entity(rt_se) \ |
239 | for (; rt_se; rt_se = NULL) | |
240 | ||
241 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
242 | { | |
243 | return NULL; | |
244 | } | |
245 | ||
9f0c1e56 | 246 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 247 | { |
f3ade837 JB |
248 | if (rt_rq->rt_nr_running) |
249 | resched_task(rq_of_rt_rq(rt_rq)->curr); | |
6f505b16 PZ |
250 | } |
251 | ||
9f0c1e56 | 252 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
253 | { |
254 | } | |
255 | ||
23b0fdfc PZ |
256 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
257 | { | |
258 | return rt_rq->rt_throttled; | |
259 | } | |
d0b27fa7 | 260 | |
c6c4927b | 261 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 262 | { |
c6c4927b | 263 | return cpu_online_mask; |
d0b27fa7 PZ |
264 | } |
265 | ||
266 | static inline | |
267 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
268 | { | |
269 | return &cpu_rq(cpu)->rt; | |
270 | } | |
271 | ||
ac086bc2 PZ |
272 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
273 | { | |
274 | return &def_rt_bandwidth; | |
275 | } | |
276 | ||
55e12e5e | 277 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 278 | |
ac086bc2 | 279 | #ifdef CONFIG_SMP |
78333cdd PZ |
280 | /* |
281 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
282 | */ | |
b79f3833 | 283 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
284 | { |
285 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
286 | struct root_domain *rd = cpu_rq(smp_processor_id())->rd; | |
287 | int i, weight, more = 0; | |
288 | u64 rt_period; | |
289 | ||
c6c4927b | 290 | weight = cpumask_weight(rd->span); |
ac086bc2 PZ |
291 | |
292 | spin_lock(&rt_b->rt_runtime_lock); | |
293 | rt_period = ktime_to_ns(rt_b->rt_period); | |
c6c4927b | 294 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
295 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
296 | s64 diff; | |
297 | ||
298 | if (iter == rt_rq) | |
299 | continue; | |
300 | ||
301 | spin_lock(&iter->rt_runtime_lock); | |
78333cdd PZ |
302 | /* |
303 | * Either all rqs have inf runtime and there's nothing to steal | |
304 | * or __disable_runtime() below sets a specific rq to inf to | |
305 | * indicate its been disabled and disalow stealing. | |
306 | */ | |
7def2be1 PZ |
307 | if (iter->rt_runtime == RUNTIME_INF) |
308 | goto next; | |
309 | ||
78333cdd PZ |
310 | /* |
311 | * From runqueues with spare time, take 1/n part of their | |
312 | * spare time, but no more than our period. | |
313 | */ | |
ac086bc2 PZ |
314 | diff = iter->rt_runtime - iter->rt_time; |
315 | if (diff > 0) { | |
58838cf3 | 316 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
317 | if (rt_rq->rt_runtime + diff > rt_period) |
318 | diff = rt_period - rt_rq->rt_runtime; | |
319 | iter->rt_runtime -= diff; | |
320 | rt_rq->rt_runtime += diff; | |
321 | more = 1; | |
322 | if (rt_rq->rt_runtime == rt_period) { | |
323 | spin_unlock(&iter->rt_runtime_lock); | |
324 | break; | |
325 | } | |
326 | } | |
7def2be1 | 327 | next: |
ac086bc2 PZ |
328 | spin_unlock(&iter->rt_runtime_lock); |
329 | } | |
330 | spin_unlock(&rt_b->rt_runtime_lock); | |
331 | ||
332 | return more; | |
333 | } | |
7def2be1 | 334 | |
78333cdd PZ |
335 | /* |
336 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
337 | */ | |
7def2be1 PZ |
338 | static void __disable_runtime(struct rq *rq) |
339 | { | |
340 | struct root_domain *rd = rq->rd; | |
341 | struct rt_rq *rt_rq; | |
342 | ||
343 | if (unlikely(!scheduler_running)) | |
344 | return; | |
345 | ||
346 | for_each_leaf_rt_rq(rt_rq, rq) { | |
347 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
348 | s64 want; | |
349 | int i; | |
350 | ||
351 | spin_lock(&rt_b->rt_runtime_lock); | |
352 | spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
353 | /* |
354 | * Either we're all inf and nobody needs to borrow, or we're | |
355 | * already disabled and thus have nothing to do, or we have | |
356 | * exactly the right amount of runtime to take out. | |
357 | */ | |
7def2be1 PZ |
358 | if (rt_rq->rt_runtime == RUNTIME_INF || |
359 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
360 | goto balanced; | |
361 | spin_unlock(&rt_rq->rt_runtime_lock); | |
362 | ||
78333cdd PZ |
363 | /* |
364 | * Calculate the difference between what we started out with | |
365 | * and what we current have, that's the amount of runtime | |
366 | * we lend and now have to reclaim. | |
367 | */ | |
7def2be1 PZ |
368 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
369 | ||
78333cdd PZ |
370 | /* |
371 | * Greedy reclaim, take back as much as we can. | |
372 | */ | |
c6c4927b | 373 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
374 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
375 | s64 diff; | |
376 | ||
78333cdd PZ |
377 | /* |
378 | * Can't reclaim from ourselves or disabled runqueues. | |
379 | */ | |
f1679d08 | 380 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
381 | continue; |
382 | ||
383 | spin_lock(&iter->rt_runtime_lock); | |
384 | if (want > 0) { | |
385 | diff = min_t(s64, iter->rt_runtime, want); | |
386 | iter->rt_runtime -= diff; | |
387 | want -= diff; | |
388 | } else { | |
389 | iter->rt_runtime -= want; | |
390 | want -= want; | |
391 | } | |
392 | spin_unlock(&iter->rt_runtime_lock); | |
393 | ||
394 | if (!want) | |
395 | break; | |
396 | } | |
397 | ||
398 | spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
399 | /* |
400 | * We cannot be left wanting - that would mean some runtime | |
401 | * leaked out of the system. | |
402 | */ | |
7def2be1 PZ |
403 | BUG_ON(want); |
404 | balanced: | |
78333cdd PZ |
405 | /* |
406 | * Disable all the borrow logic by pretending we have inf | |
407 | * runtime - in which case borrowing doesn't make sense. | |
408 | */ | |
7def2be1 PZ |
409 | rt_rq->rt_runtime = RUNTIME_INF; |
410 | spin_unlock(&rt_rq->rt_runtime_lock); | |
411 | spin_unlock(&rt_b->rt_runtime_lock); | |
412 | } | |
413 | } | |
414 | ||
415 | static void disable_runtime(struct rq *rq) | |
416 | { | |
417 | unsigned long flags; | |
418 | ||
419 | spin_lock_irqsave(&rq->lock, flags); | |
420 | __disable_runtime(rq); | |
421 | spin_unlock_irqrestore(&rq->lock, flags); | |
422 | } | |
423 | ||
424 | static void __enable_runtime(struct rq *rq) | |
425 | { | |
7def2be1 PZ |
426 | struct rt_rq *rt_rq; |
427 | ||
428 | if (unlikely(!scheduler_running)) | |
429 | return; | |
430 | ||
78333cdd PZ |
431 | /* |
432 | * Reset each runqueue's bandwidth settings | |
433 | */ | |
7def2be1 PZ |
434 | for_each_leaf_rt_rq(rt_rq, rq) { |
435 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
436 | ||
437 | spin_lock(&rt_b->rt_runtime_lock); | |
438 | spin_lock(&rt_rq->rt_runtime_lock); | |
439 | rt_rq->rt_runtime = rt_b->rt_runtime; | |
440 | rt_rq->rt_time = 0; | |
baf25731 | 441 | rt_rq->rt_throttled = 0; |
7def2be1 PZ |
442 | spin_unlock(&rt_rq->rt_runtime_lock); |
443 | spin_unlock(&rt_b->rt_runtime_lock); | |
444 | } | |
445 | } | |
446 | ||
447 | static void enable_runtime(struct rq *rq) | |
448 | { | |
449 | unsigned long flags; | |
450 | ||
451 | spin_lock_irqsave(&rq->lock, flags); | |
452 | __enable_runtime(rq); | |
453 | spin_unlock_irqrestore(&rq->lock, flags); | |
454 | } | |
455 | ||
eff6549b PZ |
456 | static int balance_runtime(struct rt_rq *rt_rq) |
457 | { | |
458 | int more = 0; | |
459 | ||
460 | if (rt_rq->rt_time > rt_rq->rt_runtime) { | |
461 | spin_unlock(&rt_rq->rt_runtime_lock); | |
462 | more = do_balance_runtime(rt_rq); | |
463 | spin_lock(&rt_rq->rt_runtime_lock); | |
464 | } | |
465 | ||
466 | return more; | |
467 | } | |
55e12e5e | 468 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
469 | static inline int balance_runtime(struct rt_rq *rt_rq) |
470 | { | |
471 | return 0; | |
472 | } | |
55e12e5e | 473 | #endif /* CONFIG_SMP */ |
ac086bc2 | 474 | |
eff6549b PZ |
475 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
476 | { | |
477 | int i, idle = 1; | |
c6c4927b | 478 | const struct cpumask *span; |
eff6549b | 479 | |
0b148fa0 | 480 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
eff6549b PZ |
481 | return 1; |
482 | ||
483 | span = sched_rt_period_mask(); | |
c6c4927b | 484 | for_each_cpu(i, span) { |
eff6549b PZ |
485 | int enqueue = 0; |
486 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
487 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
488 | ||
489 | spin_lock(&rq->lock); | |
490 | if (rt_rq->rt_time) { | |
491 | u64 runtime; | |
492 | ||
493 | spin_lock(&rt_rq->rt_runtime_lock); | |
494 | if (rt_rq->rt_throttled) | |
495 | balance_runtime(rt_rq); | |
496 | runtime = rt_rq->rt_runtime; | |
497 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
498 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
499 | rt_rq->rt_throttled = 0; | |
500 | enqueue = 1; | |
501 | } | |
502 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
503 | idle = 0; | |
504 | spin_unlock(&rt_rq->rt_runtime_lock); | |
6c3df255 PZ |
505 | } else if (rt_rq->rt_nr_running) |
506 | idle = 0; | |
eff6549b PZ |
507 | |
508 | if (enqueue) | |
509 | sched_rt_rq_enqueue(rt_rq); | |
510 | spin_unlock(&rq->lock); | |
511 | } | |
512 | ||
513 | return idle; | |
514 | } | |
ac086bc2 | 515 | |
6f505b16 PZ |
516 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
517 | { | |
052f1dc7 | 518 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
519 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
520 | ||
521 | if (rt_rq) | |
e864c499 | 522 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
523 | #endif |
524 | ||
525 | return rt_task_of(rt_se)->prio; | |
526 | } | |
527 | ||
9f0c1e56 | 528 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 529 | { |
9f0c1e56 | 530 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 531 | |
fa85ae24 | 532 | if (rt_rq->rt_throttled) |
23b0fdfc | 533 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 534 | |
ac086bc2 PZ |
535 | if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq)) |
536 | return 0; | |
537 | ||
b79f3833 PZ |
538 | balance_runtime(rt_rq); |
539 | runtime = sched_rt_runtime(rt_rq); | |
540 | if (runtime == RUNTIME_INF) | |
541 | return 0; | |
ac086bc2 | 542 | |
9f0c1e56 | 543 | if (rt_rq->rt_time > runtime) { |
6f505b16 | 544 | rt_rq->rt_throttled = 1; |
23b0fdfc | 545 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 546 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
547 | return 1; |
548 | } | |
fa85ae24 PZ |
549 | } |
550 | ||
551 | return 0; | |
552 | } | |
553 | ||
bb44e5d1 IM |
554 | /* |
555 | * Update the current task's runtime statistics. Skip current tasks that | |
556 | * are not in our scheduling class. | |
557 | */ | |
a9957449 | 558 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
559 | { |
560 | struct task_struct *curr = rq->curr; | |
6f505b16 PZ |
561 | struct sched_rt_entity *rt_se = &curr->rt; |
562 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
bb44e5d1 IM |
563 | u64 delta_exec; |
564 | ||
565 | if (!task_has_rt_policy(curr)) | |
566 | return; | |
567 | ||
d281918d | 568 | delta_exec = rq->clock - curr->se.exec_start; |
bb44e5d1 IM |
569 | if (unlikely((s64)delta_exec < 0)) |
570 | delta_exec = 0; | |
6cfb0d5d IM |
571 | |
572 | schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec)); | |
bb44e5d1 IM |
573 | |
574 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
575 | account_group_exec_runtime(curr, delta_exec); |
576 | ||
d281918d | 577 | curr->se.exec_start = rq->clock; |
d842de87 | 578 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 579 | |
0b148fa0 PZ |
580 | if (!rt_bandwidth_enabled()) |
581 | return; | |
582 | ||
354d60c2 DG |
583 | for_each_sched_rt_entity(rt_se) { |
584 | rt_rq = rt_rq_of_se(rt_se); | |
585 | ||
cc2991cf | 586 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
e113a745 | 587 | spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
588 | rt_rq->rt_time += delta_exec; |
589 | if (sched_rt_runtime_exceeded(rt_rq)) | |
590 | resched_task(curr); | |
e113a745 | 591 | spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 592 | } |
354d60c2 | 593 | } |
bb44e5d1 IM |
594 | } |
595 | ||
398a153b | 596 | #if defined CONFIG_SMP |
e864c499 GH |
597 | |
598 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu); | |
599 | ||
600 | static inline int next_prio(struct rq *rq) | |
601 | { | |
602 | struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu); | |
603 | ||
604 | if (next && rt_prio(next->prio)) | |
605 | return next->prio; | |
606 | else | |
607 | return MAX_RT_PRIO; | |
608 | } | |
e864c499 | 609 | |
398a153b GH |
610 | static void |
611 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 612 | { |
4d984277 | 613 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 614 | |
398a153b | 615 | if (prio < prev_prio) { |
4d984277 | 616 | |
e864c499 GH |
617 | /* |
618 | * If the new task is higher in priority than anything on the | |
398a153b GH |
619 | * run-queue, we know that the previous high becomes our |
620 | * next-highest. | |
e864c499 | 621 | */ |
398a153b GH |
622 | rt_rq->highest_prio.next = prev_prio; |
623 | ||
1f11eb6a | 624 | if (rq->online) |
4d984277 | 625 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); |
398a153b | 626 | |
e864c499 GH |
627 | } else if (prio == rt_rq->highest_prio.curr) |
628 | /* | |
629 | * If the next task is equal in priority to the highest on | |
630 | * the run-queue, then we implicitly know that the next highest | |
631 | * task cannot be any lower than current | |
632 | */ | |
633 | rt_rq->highest_prio.next = prio; | |
634 | else if (prio < rt_rq->highest_prio.next) | |
635 | /* | |
636 | * Otherwise, we need to recompute next-highest | |
637 | */ | |
638 | rt_rq->highest_prio.next = next_prio(rq); | |
398a153b | 639 | } |
73fe6aae | 640 | |
398a153b GH |
641 | static void |
642 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
643 | { | |
644 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 645 | |
398a153b GH |
646 | if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next)) |
647 | rt_rq->highest_prio.next = next_prio(rq); | |
648 | ||
649 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) | |
650 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
651 | } |
652 | ||
398a153b GH |
653 | #else /* CONFIG_SMP */ |
654 | ||
6f505b16 | 655 | static inline |
398a153b GH |
656 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
657 | static inline | |
658 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
659 | ||
660 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 661 | |
052f1dc7 | 662 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
663 | static void |
664 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
665 | { | |
666 | int prev_prio = rt_rq->highest_prio.curr; | |
667 | ||
668 | if (prio < prev_prio) | |
669 | rt_rq->highest_prio.curr = prio; | |
670 | ||
671 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
672 | } | |
673 | ||
674 | static void | |
675 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
676 | { | |
677 | int prev_prio = rt_rq->highest_prio.curr; | |
678 | ||
6f505b16 | 679 | if (rt_rq->rt_nr_running) { |
764a9d6f | 680 | |
398a153b | 681 | WARN_ON(prio < prev_prio); |
764a9d6f | 682 | |
e864c499 | 683 | /* |
398a153b GH |
684 | * This may have been our highest task, and therefore |
685 | * we may have some recomputation to do | |
e864c499 | 686 | */ |
398a153b | 687 | if (prio == prev_prio) { |
e864c499 GH |
688 | struct rt_prio_array *array = &rt_rq->active; |
689 | ||
690 | rt_rq->highest_prio.curr = | |
764a9d6f | 691 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
692 | } |
693 | ||
764a9d6f | 694 | } else |
e864c499 | 695 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 696 | |
398a153b GH |
697 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
698 | } | |
699 | ||
700 | #else | |
701 | ||
702 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
703 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
704 | ||
705 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
1f11eb6a | 706 | |
052f1dc7 | 707 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
708 | |
709 | static void | |
710 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
711 | { | |
712 | if (rt_se_boosted(rt_se)) | |
713 | rt_rq->rt_nr_boosted++; | |
714 | ||
715 | if (rt_rq->tg) | |
716 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
717 | } | |
718 | ||
719 | static void | |
720 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
721 | { | |
23b0fdfc PZ |
722 | if (rt_se_boosted(rt_se)) |
723 | rt_rq->rt_nr_boosted--; | |
724 | ||
725 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
726 | } |
727 | ||
728 | #else /* CONFIG_RT_GROUP_SCHED */ | |
729 | ||
730 | static void | |
731 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
732 | { | |
733 | start_rt_bandwidth(&def_rt_bandwidth); | |
734 | } | |
735 | ||
736 | static inline | |
737 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
738 | ||
739 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
740 | ||
741 | static inline | |
742 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
743 | { | |
744 | int prio = rt_se_prio(rt_se); | |
745 | ||
746 | WARN_ON(!rt_prio(prio)); | |
747 | rt_rq->rt_nr_running++; | |
748 | ||
749 | inc_rt_prio(rt_rq, prio); | |
750 | inc_rt_migration(rt_se, rt_rq); | |
751 | inc_rt_group(rt_se, rt_rq); | |
752 | } | |
753 | ||
754 | static inline | |
755 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
756 | { | |
757 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
758 | WARN_ON(!rt_rq->rt_nr_running); | |
759 | rt_rq->rt_nr_running--; | |
760 | ||
761 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
762 | dec_rt_migration(rt_se, rt_rq); | |
763 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
764 | } |
765 | ||
ad2a3f13 | 766 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se) |
bb44e5d1 | 767 | { |
6f505b16 PZ |
768 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
769 | struct rt_prio_array *array = &rt_rq->active; | |
770 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 771 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 772 | |
ad2a3f13 PZ |
773 | /* |
774 | * Don't enqueue the group if its throttled, or when empty. | |
775 | * The latter is a consequence of the former when a child group | |
776 | * get throttled and the current group doesn't have any other | |
777 | * active members. | |
778 | */ | |
779 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 780 | return; |
63489e45 | 781 | |
7ebefa8c | 782 | list_add_tail(&rt_se->run_list, queue); |
6f505b16 | 783 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 784 | |
6f505b16 PZ |
785 | inc_rt_tasks(rt_se, rt_rq); |
786 | } | |
787 | ||
ad2a3f13 | 788 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
789 | { |
790 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
791 | struct rt_prio_array *array = &rt_rq->active; | |
792 | ||
793 | list_del_init(&rt_se->run_list); | |
794 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
795 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
796 | ||
797 | dec_rt_tasks(rt_se, rt_rq); | |
798 | } | |
799 | ||
800 | /* | |
801 | * Because the prio of an upper entry depends on the lower | |
802 | * entries, we must remove entries top - down. | |
6f505b16 | 803 | */ |
ad2a3f13 | 804 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 805 | { |
ad2a3f13 | 806 | struct sched_rt_entity *back = NULL; |
6f505b16 | 807 | |
58d6c2d7 PZ |
808 | for_each_sched_rt_entity(rt_se) { |
809 | rt_se->back = back; | |
810 | back = rt_se; | |
811 | } | |
812 | ||
813 | for (rt_se = back; rt_se; rt_se = rt_se->back) { | |
814 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
815 | __dequeue_rt_entity(rt_se); |
816 | } | |
817 | } | |
818 | ||
819 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se) | |
820 | { | |
821 | dequeue_rt_stack(rt_se); | |
822 | for_each_sched_rt_entity(rt_se) | |
823 | __enqueue_rt_entity(rt_se); | |
824 | } | |
825 | ||
826 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
827 | { | |
828 | dequeue_rt_stack(rt_se); | |
829 | ||
830 | for_each_sched_rt_entity(rt_se) { | |
831 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
832 | ||
833 | if (rt_rq && rt_rq->rt_nr_running) | |
834 | __enqueue_rt_entity(rt_se); | |
58d6c2d7 | 835 | } |
bb44e5d1 IM |
836 | } |
837 | ||
838 | /* | |
839 | * Adding/removing a task to/from a priority array: | |
840 | */ | |
6f505b16 PZ |
841 | static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup) |
842 | { | |
843 | struct sched_rt_entity *rt_se = &p->rt; | |
844 | ||
845 | if (wakeup) | |
846 | rt_se->timeout = 0; | |
847 | ||
ad2a3f13 | 848 | enqueue_rt_entity(rt_se); |
c09595f6 | 849 | |
917b627d GH |
850 | if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1) |
851 | enqueue_pushable_task(rq, p); | |
852 | ||
c09595f6 | 853 | inc_cpu_load(rq, p->se.load.weight); |
6f505b16 PZ |
854 | } |
855 | ||
f02231e5 | 856 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep) |
bb44e5d1 | 857 | { |
6f505b16 | 858 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 859 | |
f1e14ef6 | 860 | update_curr_rt(rq); |
ad2a3f13 | 861 | dequeue_rt_entity(rt_se); |
c09595f6 | 862 | |
917b627d GH |
863 | dequeue_pushable_task(rq, p); |
864 | ||
c09595f6 | 865 | dec_cpu_load(rq, p->se.load.weight); |
bb44e5d1 IM |
866 | } |
867 | ||
868 | /* | |
869 | * Put task to the end of the run list without the overhead of dequeue | |
870 | * followed by enqueue. | |
871 | */ | |
7ebefa8c DA |
872 | static void |
873 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 874 | { |
1cdad715 | 875 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
876 | struct rt_prio_array *array = &rt_rq->active; |
877 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
878 | ||
879 | if (head) | |
880 | list_move(&rt_se->run_list, queue); | |
881 | else | |
882 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 883 | } |
6f505b16 PZ |
884 | } |
885 | ||
7ebefa8c | 886 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 887 | { |
6f505b16 PZ |
888 | struct sched_rt_entity *rt_se = &p->rt; |
889 | struct rt_rq *rt_rq; | |
bb44e5d1 | 890 | |
6f505b16 PZ |
891 | for_each_sched_rt_entity(rt_se) { |
892 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 893 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 894 | } |
bb44e5d1 IM |
895 | } |
896 | ||
6f505b16 | 897 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 898 | { |
7ebefa8c | 899 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
900 | } |
901 | ||
e7693a36 | 902 | #ifdef CONFIG_SMP |
318e0893 GH |
903 | static int find_lowest_rq(struct task_struct *task); |
904 | ||
e7693a36 GH |
905 | static int select_task_rq_rt(struct task_struct *p, int sync) |
906 | { | |
318e0893 GH |
907 | struct rq *rq = task_rq(p); |
908 | ||
909 | /* | |
e1f47d89 SR |
910 | * If the current task is an RT task, then |
911 | * try to see if we can wake this RT task up on another | |
912 | * runqueue. Otherwise simply start this RT task | |
913 | * on its current runqueue. | |
914 | * | |
915 | * We want to avoid overloading runqueues. Even if | |
916 | * the RT task is of higher priority than the current RT task. | |
917 | * RT tasks behave differently than other tasks. If | |
918 | * one gets preempted, we try to push it off to another queue. | |
919 | * So trying to keep a preempting RT task on the same | |
920 | * cache hot CPU will force the running RT task to | |
921 | * a cold CPU. So we waste all the cache for the lower | |
922 | * RT task in hopes of saving some of a RT task | |
923 | * that is just being woken and probably will have | |
924 | * cold cache anyway. | |
318e0893 | 925 | */ |
17b3279b | 926 | if (unlikely(rt_task(rq->curr)) && |
6f505b16 | 927 | (p->rt.nr_cpus_allowed > 1)) { |
318e0893 GH |
928 | int cpu = find_lowest_rq(p); |
929 | ||
930 | return (cpu == -1) ? task_cpu(p) : cpu; | |
931 | } | |
932 | ||
933 | /* | |
934 | * Otherwise, just let it ride on the affined RQ and the | |
935 | * post-schedule router will push the preempted task away | |
936 | */ | |
e7693a36 GH |
937 | return task_cpu(p); |
938 | } | |
7ebefa8c DA |
939 | |
940 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
941 | { | |
24600ce8 | 942 | cpumask_var_t mask; |
7ebefa8c DA |
943 | |
944 | if (rq->curr->rt.nr_cpus_allowed == 1) | |
945 | return; | |
946 | ||
24600ce8 | 947 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) |
7ebefa8c DA |
948 | return; |
949 | ||
24600ce8 RR |
950 | if (p->rt.nr_cpus_allowed != 1 |
951 | && cpupri_find(&rq->rd->cpupri, p, mask)) | |
952 | goto free; | |
953 | ||
954 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, mask)) | |
955 | goto free; | |
7ebefa8c DA |
956 | |
957 | /* | |
958 | * There appears to be other cpus that can accept | |
959 | * current and none to run 'p', so lets reschedule | |
960 | * to try and push current away: | |
961 | */ | |
962 | requeue_task_rt(rq, p, 1); | |
963 | resched_task(rq->curr); | |
24600ce8 RR |
964 | free: |
965 | free_cpumask_var(mask); | |
7ebefa8c DA |
966 | } |
967 | ||
e7693a36 GH |
968 | #endif /* CONFIG_SMP */ |
969 | ||
bb44e5d1 IM |
970 | /* |
971 | * Preempt the current task with a newly woken task if needed: | |
972 | */ | |
15afe09b | 973 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int sync) |
bb44e5d1 | 974 | { |
45c01e82 | 975 | if (p->prio < rq->curr->prio) { |
bb44e5d1 | 976 | resched_task(rq->curr); |
45c01e82 GH |
977 | return; |
978 | } | |
979 | ||
980 | #ifdef CONFIG_SMP | |
981 | /* | |
982 | * If: | |
983 | * | |
984 | * - the newly woken task is of equal priority to the current task | |
985 | * - the newly woken task is non-migratable while current is migratable | |
986 | * - current will be preempted on the next reschedule | |
987 | * | |
988 | * we should check to see if current can readily move to a different | |
989 | * cpu. If so, we will reschedule to allow the push logic to try | |
990 | * to move current somewhere else, making room for our non-migratable | |
991 | * task. | |
992 | */ | |
7ebefa8c DA |
993 | if (p->prio == rq->curr->prio && !need_resched()) |
994 | check_preempt_equal_prio(rq, p); | |
45c01e82 | 995 | #endif |
bb44e5d1 IM |
996 | } |
997 | ||
6f505b16 PZ |
998 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
999 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1000 | { |
6f505b16 PZ |
1001 | struct rt_prio_array *array = &rt_rq->active; |
1002 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1003 | struct list_head *queue; |
1004 | int idx; | |
1005 | ||
1006 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1007 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1008 | |
1009 | queue = array->queue + idx; | |
6f505b16 | 1010 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1011 | |
6f505b16 PZ |
1012 | return next; |
1013 | } | |
bb44e5d1 | 1014 | |
917b627d | 1015 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1016 | { |
1017 | struct sched_rt_entity *rt_se; | |
1018 | struct task_struct *p; | |
1019 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1020 | |
6f505b16 PZ |
1021 | rt_rq = &rq->rt; |
1022 | ||
1023 | if (unlikely(!rt_rq->rt_nr_running)) | |
1024 | return NULL; | |
1025 | ||
23b0fdfc | 1026 | if (rt_rq_throttled(rt_rq)) |
6f505b16 PZ |
1027 | return NULL; |
1028 | ||
1029 | do { | |
1030 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1031 | BUG_ON(!rt_se); |
6f505b16 PZ |
1032 | rt_rq = group_rt_rq(rt_se); |
1033 | } while (rt_rq); | |
1034 | ||
1035 | p = rt_task_of(rt_se); | |
1036 | p->se.exec_start = rq->clock; | |
917b627d GH |
1037 | |
1038 | return p; | |
1039 | } | |
1040 | ||
1041 | static struct task_struct *pick_next_task_rt(struct rq *rq) | |
1042 | { | |
1043 | struct task_struct *p = _pick_next_task_rt(rq); | |
1044 | ||
1045 | /* The running task is never eligible for pushing */ | |
1046 | if (p) | |
1047 | dequeue_pushable_task(rq, p); | |
1048 | ||
6f505b16 | 1049 | return p; |
bb44e5d1 IM |
1050 | } |
1051 | ||
31ee529c | 1052 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1053 | { |
f1e14ef6 | 1054 | update_curr_rt(rq); |
bb44e5d1 | 1055 | p->se.exec_start = 0; |
917b627d GH |
1056 | |
1057 | /* | |
1058 | * The previous task needs to be made eligible for pushing | |
1059 | * if it is still active | |
1060 | */ | |
1061 | if (p->se.on_rq && p->rt.nr_cpus_allowed > 1) | |
1062 | enqueue_pushable_task(rq, p); | |
bb44e5d1 IM |
1063 | } |
1064 | ||
681f3e68 | 1065 | #ifdef CONFIG_SMP |
6f505b16 | 1066 | |
e8fa1362 SR |
1067 | /* Only try algorithms three times */ |
1068 | #define RT_MAX_TRIES 3 | |
1069 | ||
e8fa1362 SR |
1070 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); |
1071 | ||
f65eda4f SR |
1072 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1073 | { | |
1074 | if (!task_running(rq, p) && | |
96f874e2 | 1075 | (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) && |
6f505b16 | 1076 | (p->rt.nr_cpus_allowed > 1)) |
f65eda4f SR |
1077 | return 1; |
1078 | return 0; | |
1079 | } | |
1080 | ||
e8fa1362 | 1081 | /* Return the second highest RT task, NULL otherwise */ |
79064fbf | 1082 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
e8fa1362 | 1083 | { |
6f505b16 PZ |
1084 | struct task_struct *next = NULL; |
1085 | struct sched_rt_entity *rt_se; | |
1086 | struct rt_prio_array *array; | |
1087 | struct rt_rq *rt_rq; | |
e8fa1362 SR |
1088 | int idx; |
1089 | ||
6f505b16 PZ |
1090 | for_each_leaf_rt_rq(rt_rq, rq) { |
1091 | array = &rt_rq->active; | |
1092 | idx = sched_find_first_bit(array->bitmap); | |
1093 | next_idx: | |
1094 | if (idx >= MAX_RT_PRIO) | |
1095 | continue; | |
1096 | if (next && next->prio < idx) | |
1097 | continue; | |
1098 | list_for_each_entry(rt_se, array->queue + idx, run_list) { | |
1099 | struct task_struct *p = rt_task_of(rt_se); | |
1100 | if (pick_rt_task(rq, p, cpu)) { | |
1101 | next = p; | |
1102 | break; | |
1103 | } | |
1104 | } | |
1105 | if (!next) { | |
1106 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); | |
1107 | goto next_idx; | |
1108 | } | |
f65eda4f SR |
1109 | } |
1110 | ||
e8fa1362 SR |
1111 | return next; |
1112 | } | |
1113 | ||
0e3900e6 | 1114 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1115 | |
6e1254d2 GH |
1116 | static inline int pick_optimal_cpu(int this_cpu, cpumask_t *mask) |
1117 | { | |
1118 | int first; | |
1119 | ||
1120 | /* "this_cpu" is cheaper to preempt than a remote processor */ | |
1121 | if ((this_cpu != -1) && cpu_isset(this_cpu, *mask)) | |
1122 | return this_cpu; | |
1123 | ||
1124 | first = first_cpu(*mask); | |
1125 | if (first != NR_CPUS) | |
1126 | return first; | |
1127 | ||
1128 | return -1; | |
1129 | } | |
1130 | ||
1131 | static int find_lowest_rq(struct task_struct *task) | |
1132 | { | |
1133 | struct sched_domain *sd; | |
96f874e2 | 1134 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
6e1254d2 GH |
1135 | int this_cpu = smp_processor_id(); |
1136 | int cpu = task_cpu(task); | |
06f90dbd | 1137 | |
6e0534f2 GH |
1138 | if (task->rt.nr_cpus_allowed == 1) |
1139 | return -1; /* No other targets possible */ | |
6e1254d2 | 1140 | |
6e0534f2 GH |
1141 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1142 | return -1; /* No targets found */ | |
6e1254d2 | 1143 | |
e761b772 MK |
1144 | /* |
1145 | * Only consider CPUs that are usable for migration. | |
1146 | * I guess we might want to change cpupri_find() to ignore those | |
1147 | * in the first place. | |
1148 | */ | |
96f874e2 | 1149 | cpumask_and(lowest_mask, lowest_mask, cpu_active_mask); |
e761b772 | 1150 | |
6e1254d2 GH |
1151 | /* |
1152 | * At this point we have built a mask of cpus representing the | |
1153 | * lowest priority tasks in the system. Now we want to elect | |
1154 | * the best one based on our affinity and topology. | |
1155 | * | |
1156 | * We prioritize the last cpu that the task executed on since | |
1157 | * it is most likely cache-hot in that location. | |
1158 | */ | |
96f874e2 | 1159 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1160 | return cpu; |
1161 | ||
1162 | /* | |
1163 | * Otherwise, we consult the sched_domains span maps to figure | |
1164 | * out which cpu is logically closest to our hot cache data. | |
1165 | */ | |
1166 | if (this_cpu == cpu) | |
1167 | this_cpu = -1; /* Skip this_cpu opt if the same */ | |
1168 | ||
1169 | for_each_domain(cpu, sd) { | |
1170 | if (sd->flags & SD_WAKE_AFFINE) { | |
1171 | cpumask_t domain_mask; | |
1172 | int best_cpu; | |
1173 | ||
758b2cdc RR |
1174 | cpumask_and(&domain_mask, sched_domain_span(sd), |
1175 | lowest_mask); | |
6e1254d2 GH |
1176 | |
1177 | best_cpu = pick_optimal_cpu(this_cpu, | |
1178 | &domain_mask); | |
1179 | if (best_cpu != -1) | |
1180 | return best_cpu; | |
1181 | } | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * And finally, if there were no matches within the domains | |
1186 | * just give the caller *something* to work with from the compatible | |
1187 | * locations. | |
1188 | */ | |
1189 | return pick_optimal_cpu(this_cpu, lowest_mask); | |
07b4032c GH |
1190 | } |
1191 | ||
1192 | /* Will lock the rq it finds */ | |
4df64c0b | 1193 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1194 | { |
1195 | struct rq *lowest_rq = NULL; | |
07b4032c | 1196 | int tries; |
4df64c0b | 1197 | int cpu; |
e8fa1362 | 1198 | |
07b4032c GH |
1199 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1200 | cpu = find_lowest_rq(task); | |
1201 | ||
2de0b463 | 1202 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1203 | break; |
1204 | ||
07b4032c GH |
1205 | lowest_rq = cpu_rq(cpu); |
1206 | ||
e8fa1362 | 1207 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1208 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1209 | /* |
1210 | * We had to unlock the run queue. In | |
1211 | * the mean time, task could have | |
1212 | * migrated already or had its affinity changed. | |
1213 | * Also make sure that it wasn't scheduled on its rq. | |
1214 | */ | |
07b4032c | 1215 | if (unlikely(task_rq(task) != rq || |
96f874e2 RR |
1216 | !cpumask_test_cpu(lowest_rq->cpu, |
1217 | &task->cpus_allowed) || | |
07b4032c | 1218 | task_running(rq, task) || |
e8fa1362 | 1219 | !task->se.on_rq)) { |
4df64c0b | 1220 | |
e8fa1362 SR |
1221 | spin_unlock(&lowest_rq->lock); |
1222 | lowest_rq = NULL; | |
1223 | break; | |
1224 | } | |
1225 | } | |
1226 | ||
1227 | /* If this rq is still suitable use it. */ | |
e864c499 | 1228 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1229 | break; |
1230 | ||
1231 | /* try again */ | |
1b12bbc7 | 1232 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1233 | lowest_rq = NULL; |
1234 | } | |
1235 | ||
1236 | return lowest_rq; | |
1237 | } | |
1238 | ||
917b627d GH |
1239 | static inline int has_pushable_tasks(struct rq *rq) |
1240 | { | |
1241 | return !plist_head_empty(&rq->rt.pushable_tasks); | |
1242 | } | |
1243 | ||
1244 | static struct task_struct *pick_next_pushable_task(struct rq *rq) | |
1245 | { | |
1246 | struct task_struct *p; | |
1247 | ||
1248 | if (!has_pushable_tasks(rq)) | |
1249 | return NULL; | |
1250 | ||
1251 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1252 | struct task_struct, pushable_tasks); | |
1253 | ||
1254 | BUG_ON(rq->cpu != task_cpu(p)); | |
1255 | BUG_ON(task_current(rq, p)); | |
1256 | BUG_ON(p->rt.nr_cpus_allowed <= 1); | |
1257 | ||
1258 | BUG_ON(!p->se.on_rq); | |
1259 | BUG_ON(!rt_task(p)); | |
1260 | ||
1261 | return p; | |
1262 | } | |
1263 | ||
e8fa1362 SR |
1264 | /* |
1265 | * If the current CPU has more than one RT task, see if the non | |
1266 | * running task can migrate over to a CPU that is running a task | |
1267 | * of lesser priority. | |
1268 | */ | |
697f0a48 | 1269 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1270 | { |
1271 | struct task_struct *next_task; | |
1272 | struct rq *lowest_rq; | |
e8fa1362 | 1273 | |
a22d7fc1 GH |
1274 | if (!rq->rt.overloaded) |
1275 | return 0; | |
1276 | ||
917b627d | 1277 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1278 | if (!next_task) |
1279 | return 0; | |
1280 | ||
1281 | retry: | |
697f0a48 | 1282 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1283 | WARN_ON(1); |
e8fa1362 | 1284 | return 0; |
f65eda4f | 1285 | } |
e8fa1362 SR |
1286 | |
1287 | /* | |
1288 | * It's possible that the next_task slipped in of | |
1289 | * higher priority than current. If that's the case | |
1290 | * just reschedule current. | |
1291 | */ | |
697f0a48 GH |
1292 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1293 | resched_task(rq->curr); | |
e8fa1362 SR |
1294 | return 0; |
1295 | } | |
1296 | ||
697f0a48 | 1297 | /* We might release rq lock */ |
e8fa1362 SR |
1298 | get_task_struct(next_task); |
1299 | ||
1300 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1301 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1302 | if (!lowest_rq) { |
1303 | struct task_struct *task; | |
1304 | /* | |
697f0a48 | 1305 | * find lock_lowest_rq releases rq->lock |
1563513d GH |
1306 | * so it is possible that next_task has migrated. |
1307 | * | |
1308 | * We need to make sure that the task is still on the same | |
1309 | * run-queue and is also still the next task eligible for | |
1310 | * pushing. | |
e8fa1362 | 1311 | */ |
917b627d | 1312 | task = pick_next_pushable_task(rq); |
1563513d GH |
1313 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1314 | /* | |
1315 | * If we get here, the task hasnt moved at all, but | |
1316 | * it has failed to push. We will not try again, | |
1317 | * since the other cpus will pull from us when they | |
1318 | * are ready. | |
1319 | */ | |
1320 | dequeue_pushable_task(rq, next_task); | |
1321 | goto out; | |
e8fa1362 | 1322 | } |
917b627d | 1323 | |
1563513d GH |
1324 | if (!task) |
1325 | /* No more tasks, just exit */ | |
1326 | goto out; | |
1327 | ||
917b627d | 1328 | /* |
1563513d | 1329 | * Something has shifted, try again. |
917b627d | 1330 | */ |
1563513d GH |
1331 | put_task_struct(next_task); |
1332 | next_task = task; | |
1333 | goto retry; | |
e8fa1362 SR |
1334 | } |
1335 | ||
697f0a48 | 1336 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1337 | set_task_cpu(next_task, lowest_rq->cpu); |
1338 | activate_task(lowest_rq, next_task, 0); | |
1339 | ||
1340 | resched_task(lowest_rq->curr); | |
1341 | ||
1b12bbc7 | 1342 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1343 | |
e8fa1362 SR |
1344 | out: |
1345 | put_task_struct(next_task); | |
1346 | ||
917b627d | 1347 | return 1; |
e8fa1362 SR |
1348 | } |
1349 | ||
e8fa1362 SR |
1350 | static void push_rt_tasks(struct rq *rq) |
1351 | { | |
1352 | /* push_rt_task will return true if it moved an RT */ | |
1353 | while (push_rt_task(rq)) | |
1354 | ; | |
1355 | } | |
1356 | ||
f65eda4f SR |
1357 | static int pull_rt_task(struct rq *this_rq) |
1358 | { | |
80bf3171 | 1359 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
a8728944 | 1360 | struct task_struct *p; |
f65eda4f | 1361 | struct rq *src_rq; |
f65eda4f | 1362 | |
637f5085 | 1363 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
1364 | return 0; |
1365 | ||
c6c4927b | 1366 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1367 | if (this_cpu == cpu) |
1368 | continue; | |
1369 | ||
1370 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1371 | |
1372 | /* | |
1373 | * Don't bother taking the src_rq->lock if the next highest | |
1374 | * task is known to be lower-priority than our current task. | |
1375 | * This may look racy, but if this value is about to go | |
1376 | * logically higher, the src_rq will push this task away. | |
1377 | * And if its going logically lower, we do not care | |
1378 | */ | |
1379 | if (src_rq->rt.highest_prio.next >= | |
1380 | this_rq->rt.highest_prio.curr) | |
1381 | continue; | |
1382 | ||
f65eda4f SR |
1383 | /* |
1384 | * We can potentially drop this_rq's lock in | |
1385 | * double_lock_balance, and another CPU could | |
a8728944 | 1386 | * alter this_rq |
f65eda4f | 1387 | */ |
a8728944 | 1388 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
1389 | |
1390 | /* | |
1391 | * Are there still pullable RT tasks? | |
1392 | */ | |
614ee1f6 MG |
1393 | if (src_rq->rt.rt_nr_running <= 1) |
1394 | goto skip; | |
f65eda4f | 1395 | |
f65eda4f SR |
1396 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
1397 | ||
1398 | /* | |
1399 | * Do we have an RT task that preempts | |
1400 | * the to-be-scheduled task? | |
1401 | */ | |
a8728944 | 1402 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f SR |
1403 | WARN_ON(p == src_rq->curr); |
1404 | WARN_ON(!p->se.on_rq); | |
1405 | ||
1406 | /* | |
1407 | * There's a chance that p is higher in priority | |
1408 | * than what's currently running on its cpu. | |
1409 | * This is just that p is wakeing up and hasn't | |
1410 | * had a chance to schedule. We only pull | |
1411 | * p if it is lower in priority than the | |
a8728944 | 1412 | * current task on the run queue |
f65eda4f | 1413 | */ |
a8728944 | 1414 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 1415 | goto skip; |
f65eda4f SR |
1416 | |
1417 | ret = 1; | |
1418 | ||
1419 | deactivate_task(src_rq, p, 0); | |
1420 | set_task_cpu(p, this_cpu); | |
1421 | activate_task(this_rq, p, 0); | |
1422 | /* | |
1423 | * We continue with the search, just in | |
1424 | * case there's an even higher prio task | |
1425 | * in another runqueue. (low likelyhood | |
1426 | * but possible) | |
f65eda4f | 1427 | */ |
f65eda4f | 1428 | } |
614ee1f6 | 1429 | skip: |
1b12bbc7 | 1430 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
1431 | } |
1432 | ||
1433 | return ret; | |
1434 | } | |
1435 | ||
9a897c5a | 1436 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
f65eda4f SR |
1437 | { |
1438 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
e864c499 | 1439 | if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio) |
f65eda4f SR |
1440 | pull_rt_task(rq); |
1441 | } | |
1442 | ||
967fc046 GH |
1443 | /* |
1444 | * assumes rq->lock is held | |
1445 | */ | |
1446 | static int needs_post_schedule_rt(struct rq *rq) | |
1447 | { | |
917b627d | 1448 | return has_pushable_tasks(rq); |
967fc046 GH |
1449 | } |
1450 | ||
9a897c5a | 1451 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 SR |
1452 | { |
1453 | /* | |
967fc046 GH |
1454 | * This is only called if needs_post_schedule_rt() indicates that |
1455 | * we need to push tasks away | |
e8fa1362 | 1456 | */ |
967fc046 GH |
1457 | spin_lock_irq(&rq->lock); |
1458 | push_rt_tasks(rq); | |
1459 | spin_unlock_irq(&rq->lock); | |
e8fa1362 SR |
1460 | } |
1461 | ||
8ae121ac GH |
1462 | /* |
1463 | * If we are not running and we are not going to reschedule soon, we should | |
1464 | * try to push tasks away now | |
1465 | */ | |
9a897c5a | 1466 | static void task_wake_up_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1467 | { |
9a897c5a | 1468 | if (!task_running(rq, p) && |
8ae121ac | 1469 | !test_tsk_need_resched(rq->curr) && |
917b627d | 1470 | has_pushable_tasks(rq) && |
777c2f38 | 1471 | p->rt.nr_cpus_allowed > 1) |
4642dafd SR |
1472 | push_rt_tasks(rq); |
1473 | } | |
1474 | ||
43010659 | 1475 | static unsigned long |
bb44e5d1 | 1476 | load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f PW |
1477 | unsigned long max_load_move, |
1478 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1479 | int *all_pinned, int *this_best_prio) | |
bb44e5d1 | 1480 | { |
c7a1e46a SR |
1481 | /* don't touch RT tasks */ |
1482 | return 0; | |
e1d1484f PW |
1483 | } |
1484 | ||
1485 | static int | |
1486 | move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1487 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1488 | { | |
c7a1e46a SR |
1489 | /* don't touch RT tasks */ |
1490 | return 0; | |
bb44e5d1 | 1491 | } |
deeeccd4 | 1492 | |
cd8ba7cd | 1493 | static void set_cpus_allowed_rt(struct task_struct *p, |
96f874e2 | 1494 | const struct cpumask *new_mask) |
73fe6aae | 1495 | { |
96f874e2 | 1496 | int weight = cpumask_weight(new_mask); |
73fe6aae GH |
1497 | |
1498 | BUG_ON(!rt_task(p)); | |
1499 | ||
1500 | /* | |
1501 | * Update the migration status of the RQ if we have an RT task | |
1502 | * which is running AND changing its weight value. | |
1503 | */ | |
6f505b16 | 1504 | if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) { |
73fe6aae GH |
1505 | struct rq *rq = task_rq(p); |
1506 | ||
917b627d GH |
1507 | if (!task_current(rq, p)) { |
1508 | /* | |
1509 | * Make sure we dequeue this task from the pushable list | |
1510 | * before going further. It will either remain off of | |
1511 | * the list because we are no longer pushable, or it | |
1512 | * will be requeued. | |
1513 | */ | |
1514 | if (p->rt.nr_cpus_allowed > 1) | |
1515 | dequeue_pushable_task(rq, p); | |
1516 | ||
1517 | /* | |
1518 | * Requeue if our weight is changing and still > 1 | |
1519 | */ | |
1520 | if (weight > 1) | |
1521 | enqueue_pushable_task(rq, p); | |
1522 | ||
1523 | } | |
1524 | ||
6f505b16 | 1525 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
73fe6aae | 1526 | rq->rt.rt_nr_migratory++; |
6f505b16 | 1527 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
73fe6aae GH |
1528 | BUG_ON(!rq->rt.rt_nr_migratory); |
1529 | rq->rt.rt_nr_migratory--; | |
1530 | } | |
1531 | ||
398a153b | 1532 | update_rt_migration(&rq->rt); |
73fe6aae GH |
1533 | } |
1534 | ||
96f874e2 | 1535 | cpumask_copy(&p->cpus_allowed, new_mask); |
6f505b16 | 1536 | p->rt.nr_cpus_allowed = weight; |
73fe6aae | 1537 | } |
deeeccd4 | 1538 | |
bdd7c81b | 1539 | /* Assumes rq->lock is held */ |
1f11eb6a | 1540 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
1541 | { |
1542 | if (rq->rt.overloaded) | |
1543 | rt_set_overload(rq); | |
6e0534f2 | 1544 | |
7def2be1 PZ |
1545 | __enable_runtime(rq); |
1546 | ||
e864c499 | 1547 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
1548 | } |
1549 | ||
1550 | /* Assumes rq->lock is held */ | |
1f11eb6a | 1551 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
1552 | { |
1553 | if (rq->rt.overloaded) | |
1554 | rt_clear_overload(rq); | |
6e0534f2 | 1555 | |
7def2be1 PZ |
1556 | __disable_runtime(rq); |
1557 | ||
6e0534f2 | 1558 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 1559 | } |
cb469845 SR |
1560 | |
1561 | /* | |
1562 | * When switch from the rt queue, we bring ourselves to a position | |
1563 | * that we might want to pull RT tasks from other runqueues. | |
1564 | */ | |
1565 | static void switched_from_rt(struct rq *rq, struct task_struct *p, | |
1566 | int running) | |
1567 | { | |
1568 | /* | |
1569 | * If there are other RT tasks then we will reschedule | |
1570 | * and the scheduling of the other RT tasks will handle | |
1571 | * the balancing. But if we are the last RT task | |
1572 | * we may need to handle the pulling of RT tasks | |
1573 | * now. | |
1574 | */ | |
1575 | if (!rq->rt.rt_nr_running) | |
1576 | pull_rt_task(rq); | |
1577 | } | |
3d8cbdf8 RR |
1578 | |
1579 | static inline void init_sched_rt_class(void) | |
1580 | { | |
1581 | unsigned int i; | |
1582 | ||
1583 | for_each_possible_cpu(i) | |
6ca09dfc MT |
1584 | alloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
1585 | GFP_KERNEL, cpu_to_node(i)); | |
3d8cbdf8 | 1586 | } |
cb469845 SR |
1587 | #endif /* CONFIG_SMP */ |
1588 | ||
1589 | /* | |
1590 | * When switching a task to RT, we may overload the runqueue | |
1591 | * with RT tasks. In this case we try to push them off to | |
1592 | * other runqueues. | |
1593 | */ | |
1594 | static void switched_to_rt(struct rq *rq, struct task_struct *p, | |
1595 | int running) | |
1596 | { | |
1597 | int check_resched = 1; | |
1598 | ||
1599 | /* | |
1600 | * If we are already running, then there's nothing | |
1601 | * that needs to be done. But if we are not running | |
1602 | * we may need to preempt the current running task. | |
1603 | * If that current running task is also an RT task | |
1604 | * then see if we can move to another run queue. | |
1605 | */ | |
1606 | if (!running) { | |
1607 | #ifdef CONFIG_SMP | |
1608 | if (rq->rt.overloaded && push_rt_task(rq) && | |
1609 | /* Don't resched if we changed runqueues */ | |
1610 | rq != task_rq(p)) | |
1611 | check_resched = 0; | |
1612 | #endif /* CONFIG_SMP */ | |
1613 | if (check_resched && p->prio < rq->curr->prio) | |
1614 | resched_task(rq->curr); | |
1615 | } | |
1616 | } | |
1617 | ||
1618 | /* | |
1619 | * Priority of the task has changed. This may cause | |
1620 | * us to initiate a push or pull. | |
1621 | */ | |
1622 | static void prio_changed_rt(struct rq *rq, struct task_struct *p, | |
1623 | int oldprio, int running) | |
1624 | { | |
1625 | if (running) { | |
1626 | #ifdef CONFIG_SMP | |
1627 | /* | |
1628 | * If our priority decreases while running, we | |
1629 | * may need to pull tasks to this runqueue. | |
1630 | */ | |
1631 | if (oldprio < p->prio) | |
1632 | pull_rt_task(rq); | |
1633 | /* | |
1634 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
1635 | * then reschedule. Note, the above pull_rt_task |
1636 | * can release the rq lock and p could migrate. | |
1637 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 1638 | */ |
e864c499 | 1639 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
cb469845 SR |
1640 | resched_task(p); |
1641 | #else | |
1642 | /* For UP simply resched on drop of prio */ | |
1643 | if (oldprio < p->prio) | |
1644 | resched_task(p); | |
e8fa1362 | 1645 | #endif /* CONFIG_SMP */ |
cb469845 SR |
1646 | } else { |
1647 | /* | |
1648 | * This task is not running, but if it is | |
1649 | * greater than the current running task | |
1650 | * then reschedule. | |
1651 | */ | |
1652 | if (p->prio < rq->curr->prio) | |
1653 | resched_task(rq->curr); | |
1654 | } | |
1655 | } | |
1656 | ||
78f2c7db PZ |
1657 | static void watchdog(struct rq *rq, struct task_struct *p) |
1658 | { | |
1659 | unsigned long soft, hard; | |
1660 | ||
1661 | if (!p->signal) | |
1662 | return; | |
1663 | ||
1664 | soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur; | |
1665 | hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max; | |
1666 | ||
1667 | if (soft != RLIM_INFINITY) { | |
1668 | unsigned long next; | |
1669 | ||
1670 | p->rt.timeout++; | |
1671 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); | |
5a52dd50 | 1672 | if (p->rt.timeout > next) |
f06febc9 | 1673 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
1674 | } |
1675 | } | |
bb44e5d1 | 1676 | |
8f4d37ec | 1677 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 1678 | { |
67e2be02 PZ |
1679 | update_curr_rt(rq); |
1680 | ||
78f2c7db PZ |
1681 | watchdog(rq, p); |
1682 | ||
bb44e5d1 IM |
1683 | /* |
1684 | * RR tasks need a special form of timeslice management. | |
1685 | * FIFO tasks have no timeslices. | |
1686 | */ | |
1687 | if (p->policy != SCHED_RR) | |
1688 | return; | |
1689 | ||
fa717060 | 1690 | if (--p->rt.time_slice) |
bb44e5d1 IM |
1691 | return; |
1692 | ||
fa717060 | 1693 | p->rt.time_slice = DEF_TIMESLICE; |
bb44e5d1 | 1694 | |
98fbc798 DA |
1695 | /* |
1696 | * Requeue to the end of queue if we are not the only element | |
1697 | * on the queue: | |
1698 | */ | |
fa717060 | 1699 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
7ebefa8c | 1700 | requeue_task_rt(rq, p, 0); |
98fbc798 DA |
1701 | set_tsk_need_resched(p); |
1702 | } | |
bb44e5d1 IM |
1703 | } |
1704 | ||
83b699ed SV |
1705 | static void set_curr_task_rt(struct rq *rq) |
1706 | { | |
1707 | struct task_struct *p = rq->curr; | |
1708 | ||
1709 | p->se.exec_start = rq->clock; | |
917b627d GH |
1710 | |
1711 | /* The running task is never eligible for pushing */ | |
1712 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
1713 | } |
1714 | ||
2abdad0a | 1715 | static const struct sched_class rt_sched_class = { |
5522d5d5 | 1716 | .next = &fair_sched_class, |
bb44e5d1 IM |
1717 | .enqueue_task = enqueue_task_rt, |
1718 | .dequeue_task = dequeue_task_rt, | |
1719 | .yield_task = yield_task_rt, | |
1720 | ||
1721 | .check_preempt_curr = check_preempt_curr_rt, | |
1722 | ||
1723 | .pick_next_task = pick_next_task_rt, | |
1724 | .put_prev_task = put_prev_task_rt, | |
1725 | ||
681f3e68 | 1726 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
1727 | .select_task_rq = select_task_rq_rt, |
1728 | ||
bb44e5d1 | 1729 | .load_balance = load_balance_rt, |
e1d1484f | 1730 | .move_one_task = move_one_task_rt, |
73fe6aae | 1731 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
1732 | .rq_online = rq_online_rt, |
1733 | .rq_offline = rq_offline_rt, | |
9a897c5a | 1734 | .pre_schedule = pre_schedule_rt, |
967fc046 | 1735 | .needs_post_schedule = needs_post_schedule_rt, |
9a897c5a SR |
1736 | .post_schedule = post_schedule_rt, |
1737 | .task_wake_up = task_wake_up_rt, | |
cb469845 | 1738 | .switched_from = switched_from_rt, |
681f3e68 | 1739 | #endif |
bb44e5d1 | 1740 | |
83b699ed | 1741 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 1742 | .task_tick = task_tick_rt, |
cb469845 SR |
1743 | |
1744 | .prio_changed = prio_changed_rt, | |
1745 | .switched_to = switched_to_rt, | |
bb44e5d1 | 1746 | }; |
ada18de2 PZ |
1747 | |
1748 | #ifdef CONFIG_SCHED_DEBUG | |
1749 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
1750 | ||
1751 | static void print_rt_stats(struct seq_file *m, int cpu) | |
1752 | { | |
1753 | struct rt_rq *rt_rq; | |
1754 | ||
1755 | rcu_read_lock(); | |
1756 | for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu)) | |
1757 | print_rt_rq(m, cpu, rt_rq); | |
1758 | rcu_read_unlock(); | |
1759 | } | |
55e12e5e | 1760 | #endif /* CONFIG_SCHED_DEBUG */ |
0e3900e6 | 1761 |