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 | ||
3d4b47b4 PZ |
186 | static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq) |
187 | { | |
188 | list_add_rcu(&rt_rq->leaf_rt_rq_list, | |
189 | &rq_of_rt_rq(rt_rq)->leaf_rt_rq_list); | |
190 | } | |
191 | ||
192 | static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq) | |
193 | { | |
194 | list_del_rcu(&rt_rq->leaf_rt_rq_list); | |
195 | } | |
196 | ||
6f505b16 | 197 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
80f40ee4 | 198 | list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) |
6f505b16 | 199 | |
6f505b16 PZ |
200 | #define for_each_sched_rt_entity(rt_se) \ |
201 | for (; rt_se; rt_se = rt_se->parent) | |
202 | ||
203 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
204 | { | |
205 | return rt_se->my_q; | |
206 | } | |
207 | ||
37dad3fc | 208 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
6f505b16 PZ |
209 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
210 | ||
9f0c1e56 | 211 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 212 | { |
f6121f4f | 213 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
74b7eb58 YZ |
214 | struct sched_rt_entity *rt_se; |
215 | ||
0c3b9168 BS |
216 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
217 | ||
218 | rt_se = rt_rq->tg->rt_se[cpu]; | |
6f505b16 | 219 | |
f6121f4f DF |
220 | if (rt_rq->rt_nr_running) { |
221 | if (rt_se && !on_rt_rq(rt_se)) | |
37dad3fc | 222 | enqueue_rt_entity(rt_se, false); |
e864c499 | 223 | if (rt_rq->highest_prio.curr < curr->prio) |
1020387f | 224 | resched_task(curr); |
6f505b16 PZ |
225 | } |
226 | } | |
227 | ||
9f0c1e56 | 228 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 229 | { |
74b7eb58 | 230 | struct sched_rt_entity *rt_se; |
0c3b9168 | 231 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
74b7eb58 | 232 | |
0c3b9168 | 233 | rt_se = rt_rq->tg->rt_se[cpu]; |
6f505b16 PZ |
234 | |
235 | if (rt_se && on_rt_rq(rt_se)) | |
236 | dequeue_rt_entity(rt_se); | |
237 | } | |
238 | ||
23b0fdfc PZ |
239 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
240 | { | |
241 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
242 | } | |
243 | ||
244 | static int rt_se_boosted(struct sched_rt_entity *rt_se) | |
245 | { | |
246 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
247 | struct task_struct *p; | |
248 | ||
249 | if (rt_rq) | |
250 | return !!rt_rq->rt_nr_boosted; | |
251 | ||
252 | p = rt_task_of(rt_se); | |
253 | return p->prio != p->normal_prio; | |
254 | } | |
255 | ||
d0b27fa7 | 256 | #ifdef CONFIG_SMP |
c6c4927b | 257 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 PZ |
258 | { |
259 | return cpu_rq(smp_processor_id())->rd->span; | |
260 | } | |
6f505b16 | 261 | #else |
c6c4927b | 262 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 263 | { |
c6c4927b | 264 | return cpu_online_mask; |
d0b27fa7 PZ |
265 | } |
266 | #endif | |
6f505b16 | 267 | |
d0b27fa7 PZ |
268 | static inline |
269 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 270 | { |
d0b27fa7 PZ |
271 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
272 | } | |
9f0c1e56 | 273 | |
ac086bc2 PZ |
274 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
275 | { | |
276 | return &rt_rq->tg->rt_bandwidth; | |
277 | } | |
278 | ||
55e12e5e | 279 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
280 | |
281 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
282 | { | |
ac086bc2 PZ |
283 | return rt_rq->rt_runtime; |
284 | } | |
285 | ||
286 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
287 | { | |
288 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
289 | } |
290 | ||
3d4b47b4 PZ |
291 | static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq) |
292 | { | |
293 | } | |
294 | ||
295 | static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq) | |
296 | { | |
297 | } | |
298 | ||
6f505b16 PZ |
299 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
300 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
301 | ||
6f505b16 PZ |
302 | #define for_each_sched_rt_entity(rt_se) \ |
303 | for (; rt_se; rt_se = NULL) | |
304 | ||
305 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
306 | { | |
307 | return NULL; | |
308 | } | |
309 | ||
9f0c1e56 | 310 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 311 | { |
f3ade837 JB |
312 | if (rt_rq->rt_nr_running) |
313 | resched_task(rq_of_rt_rq(rt_rq)->curr); | |
6f505b16 PZ |
314 | } |
315 | ||
9f0c1e56 | 316 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
317 | { |
318 | } | |
319 | ||
23b0fdfc PZ |
320 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
321 | { | |
322 | return rt_rq->rt_throttled; | |
323 | } | |
d0b27fa7 | 324 | |
c6c4927b | 325 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 326 | { |
c6c4927b | 327 | return cpu_online_mask; |
d0b27fa7 PZ |
328 | } |
329 | ||
330 | static inline | |
331 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
332 | { | |
333 | return &cpu_rq(cpu)->rt; | |
334 | } | |
335 | ||
ac086bc2 PZ |
336 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
337 | { | |
338 | return &def_rt_bandwidth; | |
339 | } | |
340 | ||
55e12e5e | 341 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 342 | |
ac086bc2 | 343 | #ifdef CONFIG_SMP |
78333cdd PZ |
344 | /* |
345 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
346 | */ | |
b79f3833 | 347 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
348 | { |
349 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
350 | struct root_domain *rd = cpu_rq(smp_processor_id())->rd; | |
351 | int i, weight, more = 0; | |
352 | u64 rt_period; | |
353 | ||
c6c4927b | 354 | weight = cpumask_weight(rd->span); |
ac086bc2 | 355 | |
0986b11b | 356 | raw_spin_lock(&rt_b->rt_runtime_lock); |
ac086bc2 | 357 | rt_period = ktime_to_ns(rt_b->rt_period); |
c6c4927b | 358 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
359 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
360 | s64 diff; | |
361 | ||
362 | if (iter == rt_rq) | |
363 | continue; | |
364 | ||
0986b11b | 365 | raw_spin_lock(&iter->rt_runtime_lock); |
78333cdd PZ |
366 | /* |
367 | * Either all rqs have inf runtime and there's nothing to steal | |
368 | * or __disable_runtime() below sets a specific rq to inf to | |
369 | * indicate its been disabled and disalow stealing. | |
370 | */ | |
7def2be1 PZ |
371 | if (iter->rt_runtime == RUNTIME_INF) |
372 | goto next; | |
373 | ||
78333cdd PZ |
374 | /* |
375 | * From runqueues with spare time, take 1/n part of their | |
376 | * spare time, but no more than our period. | |
377 | */ | |
ac086bc2 PZ |
378 | diff = iter->rt_runtime - iter->rt_time; |
379 | if (diff > 0) { | |
58838cf3 | 380 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
381 | if (rt_rq->rt_runtime + diff > rt_period) |
382 | diff = rt_period - rt_rq->rt_runtime; | |
383 | iter->rt_runtime -= diff; | |
384 | rt_rq->rt_runtime += diff; | |
385 | more = 1; | |
386 | if (rt_rq->rt_runtime == rt_period) { | |
0986b11b | 387 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 PZ |
388 | break; |
389 | } | |
390 | } | |
7def2be1 | 391 | next: |
0986b11b | 392 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 | 393 | } |
0986b11b | 394 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
ac086bc2 PZ |
395 | |
396 | return more; | |
397 | } | |
7def2be1 | 398 | |
78333cdd PZ |
399 | /* |
400 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
401 | */ | |
7def2be1 PZ |
402 | static void __disable_runtime(struct rq *rq) |
403 | { | |
404 | struct root_domain *rd = rq->rd; | |
405 | struct rt_rq *rt_rq; | |
406 | ||
407 | if (unlikely(!scheduler_running)) | |
408 | return; | |
409 | ||
410 | for_each_leaf_rt_rq(rt_rq, rq) { | |
411 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
412 | s64 want; | |
413 | int i; | |
414 | ||
0986b11b TG |
415 | raw_spin_lock(&rt_b->rt_runtime_lock); |
416 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
417 | /* |
418 | * Either we're all inf and nobody needs to borrow, or we're | |
419 | * already disabled and thus have nothing to do, or we have | |
420 | * exactly the right amount of runtime to take out. | |
421 | */ | |
7def2be1 PZ |
422 | if (rt_rq->rt_runtime == RUNTIME_INF || |
423 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
424 | goto balanced; | |
0986b11b | 425 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
7def2be1 | 426 | |
78333cdd PZ |
427 | /* |
428 | * Calculate the difference between what we started out with | |
429 | * and what we current have, that's the amount of runtime | |
430 | * we lend and now have to reclaim. | |
431 | */ | |
7def2be1 PZ |
432 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
433 | ||
78333cdd PZ |
434 | /* |
435 | * Greedy reclaim, take back as much as we can. | |
436 | */ | |
c6c4927b | 437 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
438 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
439 | s64 diff; | |
440 | ||
78333cdd PZ |
441 | /* |
442 | * Can't reclaim from ourselves or disabled runqueues. | |
443 | */ | |
f1679d08 | 444 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
445 | continue; |
446 | ||
0986b11b | 447 | raw_spin_lock(&iter->rt_runtime_lock); |
7def2be1 PZ |
448 | if (want > 0) { |
449 | diff = min_t(s64, iter->rt_runtime, want); | |
450 | iter->rt_runtime -= diff; | |
451 | want -= diff; | |
452 | } else { | |
453 | iter->rt_runtime -= want; | |
454 | want -= want; | |
455 | } | |
0986b11b | 456 | raw_spin_unlock(&iter->rt_runtime_lock); |
7def2be1 PZ |
457 | |
458 | if (!want) | |
459 | break; | |
460 | } | |
461 | ||
0986b11b | 462 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
78333cdd PZ |
463 | /* |
464 | * We cannot be left wanting - that would mean some runtime | |
465 | * leaked out of the system. | |
466 | */ | |
7def2be1 PZ |
467 | BUG_ON(want); |
468 | balanced: | |
78333cdd PZ |
469 | /* |
470 | * Disable all the borrow logic by pretending we have inf | |
471 | * runtime - in which case borrowing doesn't make sense. | |
472 | */ | |
7def2be1 | 473 | rt_rq->rt_runtime = RUNTIME_INF; |
0986b11b TG |
474 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
475 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
476 | } |
477 | } | |
478 | ||
479 | static void disable_runtime(struct rq *rq) | |
480 | { | |
481 | unsigned long flags; | |
482 | ||
05fa785c | 483 | raw_spin_lock_irqsave(&rq->lock, flags); |
7def2be1 | 484 | __disable_runtime(rq); |
05fa785c | 485 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
7def2be1 PZ |
486 | } |
487 | ||
488 | static void __enable_runtime(struct rq *rq) | |
489 | { | |
7def2be1 PZ |
490 | struct rt_rq *rt_rq; |
491 | ||
492 | if (unlikely(!scheduler_running)) | |
493 | return; | |
494 | ||
78333cdd PZ |
495 | /* |
496 | * Reset each runqueue's bandwidth settings | |
497 | */ | |
7def2be1 PZ |
498 | for_each_leaf_rt_rq(rt_rq, rq) { |
499 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
500 | ||
0986b11b TG |
501 | raw_spin_lock(&rt_b->rt_runtime_lock); |
502 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
7def2be1 PZ |
503 | rt_rq->rt_runtime = rt_b->rt_runtime; |
504 | rt_rq->rt_time = 0; | |
baf25731 | 505 | rt_rq->rt_throttled = 0; |
0986b11b TG |
506 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
507 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
508 | } |
509 | } | |
510 | ||
511 | static void enable_runtime(struct rq *rq) | |
512 | { | |
513 | unsigned long flags; | |
514 | ||
05fa785c | 515 | raw_spin_lock_irqsave(&rq->lock, flags); |
7def2be1 | 516 | __enable_runtime(rq); |
05fa785c | 517 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
7def2be1 PZ |
518 | } |
519 | ||
eff6549b PZ |
520 | static int balance_runtime(struct rt_rq *rt_rq) |
521 | { | |
522 | int more = 0; | |
523 | ||
524 | if (rt_rq->rt_time > rt_rq->rt_runtime) { | |
0986b11b | 525 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
eff6549b | 526 | more = do_balance_runtime(rt_rq); |
0986b11b | 527 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
528 | } |
529 | ||
530 | return more; | |
531 | } | |
55e12e5e | 532 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
533 | static inline int balance_runtime(struct rt_rq *rt_rq) |
534 | { | |
535 | return 0; | |
536 | } | |
55e12e5e | 537 | #endif /* CONFIG_SMP */ |
ac086bc2 | 538 | |
eff6549b PZ |
539 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
540 | { | |
541 | int i, idle = 1; | |
c6c4927b | 542 | const struct cpumask *span; |
eff6549b | 543 | |
0b148fa0 | 544 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
eff6549b PZ |
545 | return 1; |
546 | ||
547 | span = sched_rt_period_mask(); | |
c6c4927b | 548 | for_each_cpu(i, span) { |
eff6549b PZ |
549 | int enqueue = 0; |
550 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
551 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
552 | ||
05fa785c | 553 | raw_spin_lock(&rq->lock); |
eff6549b PZ |
554 | if (rt_rq->rt_time) { |
555 | u64 runtime; | |
556 | ||
0986b11b | 557 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
558 | if (rt_rq->rt_throttled) |
559 | balance_runtime(rt_rq); | |
560 | runtime = rt_rq->rt_runtime; | |
561 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
562 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
563 | rt_rq->rt_throttled = 0; | |
564 | enqueue = 1; | |
565 | } | |
566 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
567 | idle = 0; | |
0986b11b | 568 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
0c3b9168 | 569 | } else if (rt_rq->rt_nr_running) { |
6c3df255 | 570 | idle = 0; |
0c3b9168 BS |
571 | if (!rt_rq_throttled(rt_rq)) |
572 | enqueue = 1; | |
573 | } | |
eff6549b PZ |
574 | |
575 | if (enqueue) | |
576 | sched_rt_rq_enqueue(rt_rq); | |
05fa785c | 577 | raw_spin_unlock(&rq->lock); |
eff6549b PZ |
578 | } |
579 | ||
580 | return idle; | |
581 | } | |
ac086bc2 | 582 | |
6f505b16 PZ |
583 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
584 | { | |
052f1dc7 | 585 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
586 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
587 | ||
588 | if (rt_rq) | |
e864c499 | 589 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
590 | #endif |
591 | ||
592 | return rt_task_of(rt_se)->prio; | |
593 | } | |
594 | ||
9f0c1e56 | 595 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 596 | { |
9f0c1e56 | 597 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 598 | |
fa85ae24 | 599 | if (rt_rq->rt_throttled) |
23b0fdfc | 600 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 601 | |
ac086bc2 PZ |
602 | if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq)) |
603 | return 0; | |
604 | ||
b79f3833 PZ |
605 | balance_runtime(rt_rq); |
606 | runtime = sched_rt_runtime(rt_rq); | |
607 | if (runtime == RUNTIME_INF) | |
608 | return 0; | |
ac086bc2 | 609 | |
9f0c1e56 | 610 | if (rt_rq->rt_time > runtime) { |
6f505b16 | 611 | rt_rq->rt_throttled = 1; |
23b0fdfc | 612 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 613 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
614 | return 1; |
615 | } | |
fa85ae24 PZ |
616 | } |
617 | ||
618 | return 0; | |
619 | } | |
620 | ||
bb44e5d1 IM |
621 | /* |
622 | * Update the current task's runtime statistics. Skip current tasks that | |
623 | * are not in our scheduling class. | |
624 | */ | |
a9957449 | 625 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
626 | { |
627 | struct task_struct *curr = rq->curr; | |
6f505b16 PZ |
628 | struct sched_rt_entity *rt_se = &curr->rt; |
629 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
bb44e5d1 IM |
630 | u64 delta_exec; |
631 | ||
06c3bc65 | 632 | if (curr->sched_class != &rt_sched_class) |
bb44e5d1 IM |
633 | return; |
634 | ||
305e6835 | 635 | delta_exec = rq->clock_task - curr->se.exec_start; |
bb44e5d1 IM |
636 | if (unlikely((s64)delta_exec < 0)) |
637 | delta_exec = 0; | |
6cfb0d5d | 638 | |
41acab88 | 639 | schedstat_set(curr->se.statistics.exec_max, max(curr->se.statistics.exec_max, delta_exec)); |
bb44e5d1 IM |
640 | |
641 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
642 | account_group_exec_runtime(curr, delta_exec); |
643 | ||
305e6835 | 644 | curr->se.exec_start = rq->clock_task; |
d842de87 | 645 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 646 | |
e9e9250b PZ |
647 | sched_rt_avg_update(rq, delta_exec); |
648 | ||
0b148fa0 PZ |
649 | if (!rt_bandwidth_enabled()) |
650 | return; | |
651 | ||
354d60c2 DG |
652 | for_each_sched_rt_entity(rt_se) { |
653 | rt_rq = rt_rq_of_se(rt_se); | |
654 | ||
cc2991cf | 655 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
0986b11b | 656 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
657 | rt_rq->rt_time += delta_exec; |
658 | if (sched_rt_runtime_exceeded(rt_rq)) | |
659 | resched_task(curr); | |
0986b11b | 660 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 661 | } |
354d60c2 | 662 | } |
bb44e5d1 IM |
663 | } |
664 | ||
398a153b | 665 | #if defined CONFIG_SMP |
e864c499 GH |
666 | |
667 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu); | |
668 | ||
669 | static inline int next_prio(struct rq *rq) | |
63489e45 | 670 | { |
e864c499 GH |
671 | struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu); |
672 | ||
673 | if (next && rt_prio(next->prio)) | |
674 | return next->prio; | |
675 | else | |
676 | return MAX_RT_PRIO; | |
677 | } | |
e864c499 | 678 | |
398a153b GH |
679 | static void |
680 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 681 | { |
4d984277 | 682 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 683 | |
398a153b | 684 | if (prio < prev_prio) { |
4d984277 | 685 | |
e864c499 GH |
686 | /* |
687 | * If the new task is higher in priority than anything on the | |
398a153b GH |
688 | * run-queue, we know that the previous high becomes our |
689 | * next-highest. | |
e864c499 | 690 | */ |
398a153b | 691 | rt_rq->highest_prio.next = prev_prio; |
1f11eb6a GH |
692 | |
693 | if (rq->online) | |
4d984277 | 694 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); |
1100ac91 | 695 | |
e864c499 GH |
696 | } else if (prio == rt_rq->highest_prio.curr) |
697 | /* | |
698 | * If the next task is equal in priority to the highest on | |
699 | * the run-queue, then we implicitly know that the next highest | |
700 | * task cannot be any lower than current | |
701 | */ | |
702 | rt_rq->highest_prio.next = prio; | |
703 | else if (prio < rt_rq->highest_prio.next) | |
704 | /* | |
705 | * Otherwise, we need to recompute next-highest | |
706 | */ | |
707 | rt_rq->highest_prio.next = next_prio(rq); | |
398a153b | 708 | } |
73fe6aae | 709 | |
398a153b GH |
710 | static void |
711 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
712 | { | |
713 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 714 | |
398a153b GH |
715 | if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next)) |
716 | rt_rq->highest_prio.next = next_prio(rq); | |
717 | ||
718 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) | |
719 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
720 | } |
721 | ||
398a153b GH |
722 | #else /* CONFIG_SMP */ |
723 | ||
6f505b16 | 724 | static inline |
398a153b GH |
725 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
726 | static inline | |
727 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
728 | ||
729 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 730 | |
052f1dc7 | 731 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
732 | static void |
733 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
734 | { | |
735 | int prev_prio = rt_rq->highest_prio.curr; | |
736 | ||
737 | if (prio < prev_prio) | |
738 | rt_rq->highest_prio.curr = prio; | |
739 | ||
740 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
741 | } | |
742 | ||
743 | static void | |
744 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
745 | { | |
746 | int prev_prio = rt_rq->highest_prio.curr; | |
747 | ||
6f505b16 | 748 | if (rt_rq->rt_nr_running) { |
764a9d6f | 749 | |
398a153b | 750 | WARN_ON(prio < prev_prio); |
764a9d6f | 751 | |
e864c499 | 752 | /* |
398a153b GH |
753 | * This may have been our highest task, and therefore |
754 | * we may have some recomputation to do | |
e864c499 | 755 | */ |
398a153b | 756 | if (prio == prev_prio) { |
e864c499 GH |
757 | struct rt_prio_array *array = &rt_rq->active; |
758 | ||
759 | rt_rq->highest_prio.curr = | |
764a9d6f | 760 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
761 | } |
762 | ||
764a9d6f | 763 | } else |
e864c499 | 764 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 765 | |
398a153b GH |
766 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
767 | } | |
1f11eb6a | 768 | |
398a153b GH |
769 | #else |
770 | ||
771 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
772 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
773 | ||
774 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 775 | |
052f1dc7 | 776 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
777 | |
778 | static void | |
779 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
780 | { | |
781 | if (rt_se_boosted(rt_se)) | |
782 | rt_rq->rt_nr_boosted++; | |
783 | ||
784 | if (rt_rq->tg) | |
785 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
786 | } | |
787 | ||
788 | static void | |
789 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
790 | { | |
23b0fdfc PZ |
791 | if (rt_se_boosted(rt_se)) |
792 | rt_rq->rt_nr_boosted--; | |
793 | ||
794 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
795 | } |
796 | ||
797 | #else /* CONFIG_RT_GROUP_SCHED */ | |
798 | ||
799 | static void | |
800 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
801 | { | |
802 | start_rt_bandwidth(&def_rt_bandwidth); | |
803 | } | |
804 | ||
805 | static inline | |
806 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
807 | ||
808 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
809 | ||
810 | static inline | |
811 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
812 | { | |
813 | int prio = rt_se_prio(rt_se); | |
814 | ||
815 | WARN_ON(!rt_prio(prio)); | |
816 | rt_rq->rt_nr_running++; | |
817 | ||
818 | inc_rt_prio(rt_rq, prio); | |
819 | inc_rt_migration(rt_se, rt_rq); | |
820 | inc_rt_group(rt_se, rt_rq); | |
821 | } | |
822 | ||
823 | static inline | |
824 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
825 | { | |
826 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
827 | WARN_ON(!rt_rq->rt_nr_running); | |
828 | rt_rq->rt_nr_running--; | |
829 | ||
830 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
831 | dec_rt_migration(rt_se, rt_rq); | |
832 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
833 | } |
834 | ||
37dad3fc | 835 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
bb44e5d1 | 836 | { |
6f505b16 PZ |
837 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
838 | struct rt_prio_array *array = &rt_rq->active; | |
839 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 840 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 841 | |
ad2a3f13 PZ |
842 | /* |
843 | * Don't enqueue the group if its throttled, or when empty. | |
844 | * The latter is a consequence of the former when a child group | |
845 | * get throttled and the current group doesn't have any other | |
846 | * active members. | |
847 | */ | |
848 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 849 | return; |
63489e45 | 850 | |
3d4b47b4 PZ |
851 | if (!rt_rq->rt_nr_running) |
852 | list_add_leaf_rt_rq(rt_rq); | |
853 | ||
37dad3fc TG |
854 | if (head) |
855 | list_add(&rt_se->run_list, queue); | |
856 | else | |
857 | list_add_tail(&rt_se->run_list, queue); | |
6f505b16 | 858 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 859 | |
6f505b16 PZ |
860 | inc_rt_tasks(rt_se, rt_rq); |
861 | } | |
862 | ||
ad2a3f13 | 863 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
864 | { |
865 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
866 | struct rt_prio_array *array = &rt_rq->active; | |
867 | ||
868 | list_del_init(&rt_se->run_list); | |
869 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
870 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
871 | ||
872 | dec_rt_tasks(rt_se, rt_rq); | |
3d4b47b4 PZ |
873 | if (!rt_rq->rt_nr_running) |
874 | list_del_leaf_rt_rq(rt_rq); | |
6f505b16 PZ |
875 | } |
876 | ||
877 | /* | |
878 | * Because the prio of an upper entry depends on the lower | |
879 | * entries, we must remove entries top - down. | |
6f505b16 | 880 | */ |
ad2a3f13 | 881 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 882 | { |
ad2a3f13 | 883 | struct sched_rt_entity *back = NULL; |
6f505b16 | 884 | |
58d6c2d7 PZ |
885 | for_each_sched_rt_entity(rt_se) { |
886 | rt_se->back = back; | |
887 | back = rt_se; | |
888 | } | |
889 | ||
890 | for (rt_se = back; rt_se; rt_se = rt_se->back) { | |
891 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
892 | __dequeue_rt_entity(rt_se); |
893 | } | |
894 | } | |
895 | ||
37dad3fc | 896 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
ad2a3f13 PZ |
897 | { |
898 | dequeue_rt_stack(rt_se); | |
899 | for_each_sched_rt_entity(rt_se) | |
37dad3fc | 900 | __enqueue_rt_entity(rt_se, head); |
ad2a3f13 PZ |
901 | } |
902 | ||
903 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
904 | { | |
905 | dequeue_rt_stack(rt_se); | |
906 | ||
907 | for_each_sched_rt_entity(rt_se) { | |
908 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
909 | ||
910 | if (rt_rq && rt_rq->rt_nr_running) | |
37dad3fc | 911 | __enqueue_rt_entity(rt_se, false); |
58d6c2d7 | 912 | } |
bb44e5d1 IM |
913 | } |
914 | ||
915 | /* | |
916 | * Adding/removing a task to/from a priority array: | |
917 | */ | |
ea87bb78 | 918 | static void |
371fd7e7 | 919 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
6f505b16 PZ |
920 | { |
921 | struct sched_rt_entity *rt_se = &p->rt; | |
922 | ||
371fd7e7 | 923 | if (flags & ENQUEUE_WAKEUP) |
6f505b16 PZ |
924 | rt_se->timeout = 0; |
925 | ||
371fd7e7 | 926 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
c09595f6 | 927 | |
917b627d GH |
928 | if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1) |
929 | enqueue_pushable_task(rq, p); | |
6f505b16 PZ |
930 | } |
931 | ||
371fd7e7 | 932 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 933 | { |
6f505b16 | 934 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 935 | |
f1e14ef6 | 936 | update_curr_rt(rq); |
ad2a3f13 | 937 | dequeue_rt_entity(rt_se); |
c09595f6 | 938 | |
917b627d | 939 | dequeue_pushable_task(rq, p); |
bb44e5d1 IM |
940 | } |
941 | ||
942 | /* | |
943 | * Put task to the end of the run list without the overhead of dequeue | |
944 | * followed by enqueue. | |
945 | */ | |
7ebefa8c DA |
946 | static void |
947 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 948 | { |
1cdad715 | 949 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
950 | struct rt_prio_array *array = &rt_rq->active; |
951 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
952 | ||
953 | if (head) | |
954 | list_move(&rt_se->run_list, queue); | |
955 | else | |
956 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 957 | } |
6f505b16 PZ |
958 | } |
959 | ||
7ebefa8c | 960 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 961 | { |
6f505b16 PZ |
962 | struct sched_rt_entity *rt_se = &p->rt; |
963 | struct rt_rq *rt_rq; | |
bb44e5d1 | 964 | |
6f505b16 PZ |
965 | for_each_sched_rt_entity(rt_se) { |
966 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 967 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 968 | } |
bb44e5d1 IM |
969 | } |
970 | ||
6f505b16 | 971 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 972 | { |
7ebefa8c | 973 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
974 | } |
975 | ||
e7693a36 | 976 | #ifdef CONFIG_SMP |
318e0893 GH |
977 | static int find_lowest_rq(struct task_struct *task); |
978 | ||
0017d735 | 979 | static int |
7608dec2 | 980 | select_task_rq_rt(struct task_struct *p, int sd_flag, int flags) |
e7693a36 | 981 | { |
7608dec2 PZ |
982 | struct task_struct *curr; |
983 | struct rq *rq; | |
984 | int cpu; | |
985 | ||
0763a660 | 986 | if (sd_flag != SD_BALANCE_WAKE) |
5f3edc1b PZ |
987 | return smp_processor_id(); |
988 | ||
7608dec2 PZ |
989 | cpu = task_cpu(p); |
990 | rq = cpu_rq(cpu); | |
991 | ||
992 | rcu_read_lock(); | |
993 | curr = ACCESS_ONCE(rq->curr); /* unlocked access */ | |
994 | ||
318e0893 | 995 | /* |
7608dec2 | 996 | * If the current task on @p's runqueue is an RT task, then |
e1f47d89 SR |
997 | * try to see if we can wake this RT task up on another |
998 | * runqueue. Otherwise simply start this RT task | |
999 | * on its current runqueue. | |
1000 | * | |
43fa5460 SR |
1001 | * We want to avoid overloading runqueues. If the woken |
1002 | * task is a higher priority, then it will stay on this CPU | |
1003 | * and the lower prio task should be moved to another CPU. | |
1004 | * Even though this will probably make the lower prio task | |
1005 | * lose its cache, we do not want to bounce a higher task | |
1006 | * around just because it gave up its CPU, perhaps for a | |
1007 | * lock? | |
1008 | * | |
1009 | * For equal prio tasks, we just let the scheduler sort it out. | |
7608dec2 PZ |
1010 | * |
1011 | * Otherwise, just let it ride on the affined RQ and the | |
1012 | * post-schedule router will push the preempted task away | |
1013 | * | |
1014 | * This test is optimistic, if we get it wrong the load-balancer | |
1015 | * will have to sort it out. | |
318e0893 | 1016 | */ |
7608dec2 PZ |
1017 | if (curr && unlikely(rt_task(curr)) && |
1018 | (curr->rt.nr_cpus_allowed < 2 || | |
1019 | curr->prio < p->prio) && | |
6f505b16 | 1020 | (p->rt.nr_cpus_allowed > 1)) { |
7608dec2 | 1021 | int target = find_lowest_rq(p); |
318e0893 | 1022 | |
7608dec2 PZ |
1023 | if (target != -1) |
1024 | cpu = target; | |
318e0893 | 1025 | } |
7608dec2 | 1026 | rcu_read_unlock(); |
318e0893 | 1027 | |
7608dec2 | 1028 | return cpu; |
e7693a36 | 1029 | } |
7ebefa8c DA |
1030 | |
1031 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
1032 | { | |
7ebefa8c DA |
1033 | if (rq->curr->rt.nr_cpus_allowed == 1) |
1034 | return; | |
1035 | ||
24600ce8 | 1036 | if (p->rt.nr_cpus_allowed != 1 |
13b8bd0a RR |
1037 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1038 | return; | |
24600ce8 | 1039 | |
13b8bd0a RR |
1040 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
1041 | return; | |
7ebefa8c DA |
1042 | |
1043 | /* | |
1044 | * There appears to be other cpus that can accept | |
1045 | * current and none to run 'p', so lets reschedule | |
1046 | * to try and push current away: | |
1047 | */ | |
1048 | requeue_task_rt(rq, p, 1); | |
1049 | resched_task(rq->curr); | |
1050 | } | |
1051 | ||
e7693a36 GH |
1052 | #endif /* CONFIG_SMP */ |
1053 | ||
bb44e5d1 IM |
1054 | /* |
1055 | * Preempt the current task with a newly woken task if needed: | |
1056 | */ | |
7d478721 | 1057 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1058 | { |
45c01e82 | 1059 | if (p->prio < rq->curr->prio) { |
bb44e5d1 | 1060 | resched_task(rq->curr); |
45c01e82 GH |
1061 | return; |
1062 | } | |
1063 | ||
1064 | #ifdef CONFIG_SMP | |
1065 | /* | |
1066 | * If: | |
1067 | * | |
1068 | * - the newly woken task is of equal priority to the current task | |
1069 | * - the newly woken task is non-migratable while current is migratable | |
1070 | * - current will be preempted on the next reschedule | |
1071 | * | |
1072 | * we should check to see if current can readily move to a different | |
1073 | * cpu. If so, we will reschedule to allow the push logic to try | |
1074 | * to move current somewhere else, making room for our non-migratable | |
1075 | * task. | |
1076 | */ | |
7ebefa8c DA |
1077 | if (p->prio == rq->curr->prio && !need_resched()) |
1078 | check_preempt_equal_prio(rq, p); | |
45c01e82 | 1079 | #endif |
bb44e5d1 IM |
1080 | } |
1081 | ||
6f505b16 PZ |
1082 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1083 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1084 | { |
6f505b16 PZ |
1085 | struct rt_prio_array *array = &rt_rq->active; |
1086 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1087 | struct list_head *queue; |
1088 | int idx; | |
1089 | ||
1090 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1091 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1092 | |
1093 | queue = array->queue + idx; | |
6f505b16 | 1094 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1095 | |
6f505b16 PZ |
1096 | return next; |
1097 | } | |
bb44e5d1 | 1098 | |
917b627d | 1099 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1100 | { |
1101 | struct sched_rt_entity *rt_se; | |
1102 | struct task_struct *p; | |
1103 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1104 | |
6f505b16 PZ |
1105 | rt_rq = &rq->rt; |
1106 | ||
1107 | if (unlikely(!rt_rq->rt_nr_running)) | |
1108 | return NULL; | |
1109 | ||
23b0fdfc | 1110 | if (rt_rq_throttled(rt_rq)) |
6f505b16 PZ |
1111 | return NULL; |
1112 | ||
1113 | do { | |
1114 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1115 | BUG_ON(!rt_se); |
6f505b16 PZ |
1116 | rt_rq = group_rt_rq(rt_se); |
1117 | } while (rt_rq); | |
1118 | ||
1119 | p = rt_task_of(rt_se); | |
305e6835 | 1120 | p->se.exec_start = rq->clock_task; |
917b627d GH |
1121 | |
1122 | return p; | |
1123 | } | |
1124 | ||
1125 | static struct task_struct *pick_next_task_rt(struct rq *rq) | |
1126 | { | |
1127 | struct task_struct *p = _pick_next_task_rt(rq); | |
1128 | ||
1129 | /* The running task is never eligible for pushing */ | |
1130 | if (p) | |
1131 | dequeue_pushable_task(rq, p); | |
1132 | ||
bcf08df3 | 1133 | #ifdef CONFIG_SMP |
3f029d3c GH |
1134 | /* |
1135 | * We detect this state here so that we can avoid taking the RQ | |
1136 | * lock again later if there is no need to push | |
1137 | */ | |
1138 | rq->post_schedule = has_pushable_tasks(rq); | |
bcf08df3 | 1139 | #endif |
3f029d3c | 1140 | |
6f505b16 | 1141 | return p; |
bb44e5d1 IM |
1142 | } |
1143 | ||
31ee529c | 1144 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1145 | { |
f1e14ef6 | 1146 | update_curr_rt(rq); |
bb44e5d1 | 1147 | p->se.exec_start = 0; |
917b627d GH |
1148 | |
1149 | /* | |
1150 | * The previous task needs to be made eligible for pushing | |
1151 | * if it is still active | |
1152 | */ | |
fd2f4419 | 1153 | if (on_rt_rq(&p->rt) && p->rt.nr_cpus_allowed > 1) |
917b627d | 1154 | enqueue_pushable_task(rq, p); |
bb44e5d1 IM |
1155 | } |
1156 | ||
681f3e68 | 1157 | #ifdef CONFIG_SMP |
6f505b16 | 1158 | |
e8fa1362 SR |
1159 | /* Only try algorithms three times */ |
1160 | #define RT_MAX_TRIES 3 | |
1161 | ||
e8fa1362 SR |
1162 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); |
1163 | ||
f65eda4f SR |
1164 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1165 | { | |
1166 | if (!task_running(rq, p) && | |
96f874e2 | 1167 | (cpu < 0 || cpumask_test_cpu(cpu, &p->cpus_allowed)) && |
6f505b16 | 1168 | (p->rt.nr_cpus_allowed > 1)) |
f65eda4f SR |
1169 | return 1; |
1170 | return 0; | |
1171 | } | |
1172 | ||
e8fa1362 | 1173 | /* Return the second highest RT task, NULL otherwise */ |
79064fbf | 1174 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
e8fa1362 | 1175 | { |
6f505b16 PZ |
1176 | struct task_struct *next = NULL; |
1177 | struct sched_rt_entity *rt_se; | |
1178 | struct rt_prio_array *array; | |
1179 | struct rt_rq *rt_rq; | |
e8fa1362 SR |
1180 | int idx; |
1181 | ||
6f505b16 PZ |
1182 | for_each_leaf_rt_rq(rt_rq, rq) { |
1183 | array = &rt_rq->active; | |
1184 | idx = sched_find_first_bit(array->bitmap); | |
49246274 | 1185 | next_idx: |
6f505b16 PZ |
1186 | if (idx >= MAX_RT_PRIO) |
1187 | continue; | |
1188 | if (next && next->prio < idx) | |
1189 | continue; | |
1190 | list_for_each_entry(rt_se, array->queue + idx, run_list) { | |
3d07467b PZ |
1191 | struct task_struct *p; |
1192 | ||
1193 | if (!rt_entity_is_task(rt_se)) | |
1194 | continue; | |
1195 | ||
1196 | p = rt_task_of(rt_se); | |
6f505b16 PZ |
1197 | if (pick_rt_task(rq, p, cpu)) { |
1198 | next = p; | |
1199 | break; | |
1200 | } | |
1201 | } | |
1202 | if (!next) { | |
1203 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); | |
1204 | goto next_idx; | |
1205 | } | |
f65eda4f SR |
1206 | } |
1207 | ||
e8fa1362 SR |
1208 | return next; |
1209 | } | |
1210 | ||
0e3900e6 | 1211 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1212 | |
6e1254d2 GH |
1213 | static int find_lowest_rq(struct task_struct *task) |
1214 | { | |
1215 | struct sched_domain *sd; | |
96f874e2 | 1216 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
6e1254d2 GH |
1217 | int this_cpu = smp_processor_id(); |
1218 | int cpu = task_cpu(task); | |
06f90dbd | 1219 | |
6e0534f2 GH |
1220 | if (task->rt.nr_cpus_allowed == 1) |
1221 | return -1; /* No other targets possible */ | |
6e1254d2 | 1222 | |
6e0534f2 GH |
1223 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1224 | return -1; /* No targets found */ | |
6e1254d2 GH |
1225 | |
1226 | /* | |
1227 | * At this point we have built a mask of cpus representing the | |
1228 | * lowest priority tasks in the system. Now we want to elect | |
1229 | * the best one based on our affinity and topology. | |
1230 | * | |
1231 | * We prioritize the last cpu that the task executed on since | |
1232 | * it is most likely cache-hot in that location. | |
1233 | */ | |
96f874e2 | 1234 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1235 | return cpu; |
1236 | ||
1237 | /* | |
1238 | * Otherwise, we consult the sched_domains span maps to figure | |
1239 | * out which cpu is logically closest to our hot cache data. | |
1240 | */ | |
e2c88063 RR |
1241 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1242 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ | |
6e1254d2 | 1243 | |
e2c88063 RR |
1244 | for_each_domain(cpu, sd) { |
1245 | if (sd->flags & SD_WAKE_AFFINE) { | |
1246 | int best_cpu; | |
6e1254d2 | 1247 | |
e2c88063 RR |
1248 | /* |
1249 | * "this_cpu" is cheaper to preempt than a | |
1250 | * remote processor. | |
1251 | */ | |
1252 | if (this_cpu != -1 && | |
1253 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) | |
1254 | return this_cpu; | |
1255 | ||
1256 | best_cpu = cpumask_first_and(lowest_mask, | |
1257 | sched_domain_span(sd)); | |
1258 | if (best_cpu < nr_cpu_ids) | |
1259 | return best_cpu; | |
6e1254d2 GH |
1260 | } |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * And finally, if there were no matches within the domains | |
1265 | * just give the caller *something* to work with from the compatible | |
1266 | * locations. | |
1267 | */ | |
e2c88063 RR |
1268 | if (this_cpu != -1) |
1269 | return this_cpu; | |
1270 | ||
1271 | cpu = cpumask_any(lowest_mask); | |
1272 | if (cpu < nr_cpu_ids) | |
1273 | return cpu; | |
1274 | return -1; | |
07b4032c GH |
1275 | } |
1276 | ||
1277 | /* Will lock the rq it finds */ | |
4df64c0b | 1278 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1279 | { |
1280 | struct rq *lowest_rq = NULL; | |
07b4032c | 1281 | int tries; |
4df64c0b | 1282 | int cpu; |
e8fa1362 | 1283 | |
07b4032c GH |
1284 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1285 | cpu = find_lowest_rq(task); | |
1286 | ||
2de0b463 | 1287 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1288 | break; |
1289 | ||
07b4032c GH |
1290 | lowest_rq = cpu_rq(cpu); |
1291 | ||
e8fa1362 | 1292 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1293 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1294 | /* |
1295 | * We had to unlock the run queue. In | |
1296 | * the mean time, task could have | |
1297 | * migrated already or had its affinity changed. | |
1298 | * Also make sure that it wasn't scheduled on its rq. | |
1299 | */ | |
07b4032c | 1300 | if (unlikely(task_rq(task) != rq || |
96f874e2 RR |
1301 | !cpumask_test_cpu(lowest_rq->cpu, |
1302 | &task->cpus_allowed) || | |
07b4032c | 1303 | task_running(rq, task) || |
fd2f4419 | 1304 | !task->on_rq)) { |
4df64c0b | 1305 | |
05fa785c | 1306 | raw_spin_unlock(&lowest_rq->lock); |
e8fa1362 SR |
1307 | lowest_rq = NULL; |
1308 | break; | |
1309 | } | |
1310 | } | |
1311 | ||
1312 | /* If this rq is still suitable use it. */ | |
e864c499 | 1313 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1314 | break; |
1315 | ||
1316 | /* try again */ | |
1b12bbc7 | 1317 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1318 | lowest_rq = NULL; |
1319 | } | |
1320 | ||
1321 | return lowest_rq; | |
1322 | } | |
1323 | ||
917b627d GH |
1324 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1325 | { | |
1326 | struct task_struct *p; | |
1327 | ||
1328 | if (!has_pushable_tasks(rq)) | |
1329 | return NULL; | |
1330 | ||
1331 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1332 | struct task_struct, pushable_tasks); | |
1333 | ||
1334 | BUG_ON(rq->cpu != task_cpu(p)); | |
1335 | BUG_ON(task_current(rq, p)); | |
1336 | BUG_ON(p->rt.nr_cpus_allowed <= 1); | |
1337 | ||
fd2f4419 | 1338 | BUG_ON(!p->on_rq); |
917b627d GH |
1339 | BUG_ON(!rt_task(p)); |
1340 | ||
1341 | return p; | |
1342 | } | |
1343 | ||
e8fa1362 SR |
1344 | /* |
1345 | * If the current CPU has more than one RT task, see if the non | |
1346 | * running task can migrate over to a CPU that is running a task | |
1347 | * of lesser priority. | |
1348 | */ | |
697f0a48 | 1349 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1350 | { |
1351 | struct task_struct *next_task; | |
1352 | struct rq *lowest_rq; | |
e8fa1362 | 1353 | |
a22d7fc1 GH |
1354 | if (!rq->rt.overloaded) |
1355 | return 0; | |
1356 | ||
917b627d | 1357 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1358 | if (!next_task) |
1359 | return 0; | |
1360 | ||
49246274 | 1361 | retry: |
697f0a48 | 1362 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1363 | WARN_ON(1); |
e8fa1362 | 1364 | return 0; |
f65eda4f | 1365 | } |
e8fa1362 SR |
1366 | |
1367 | /* | |
1368 | * It's possible that the next_task slipped in of | |
1369 | * higher priority than current. If that's the case | |
1370 | * just reschedule current. | |
1371 | */ | |
697f0a48 GH |
1372 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1373 | resched_task(rq->curr); | |
e8fa1362 SR |
1374 | return 0; |
1375 | } | |
1376 | ||
697f0a48 | 1377 | /* We might release rq lock */ |
e8fa1362 SR |
1378 | get_task_struct(next_task); |
1379 | ||
1380 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1381 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1382 | if (!lowest_rq) { |
1383 | struct task_struct *task; | |
1384 | /* | |
697f0a48 | 1385 | * find lock_lowest_rq releases rq->lock |
1563513d GH |
1386 | * so it is possible that next_task has migrated. |
1387 | * | |
1388 | * We need to make sure that the task is still on the same | |
1389 | * run-queue and is also still the next task eligible for | |
1390 | * pushing. | |
e8fa1362 | 1391 | */ |
917b627d | 1392 | task = pick_next_pushable_task(rq); |
1563513d GH |
1393 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1394 | /* | |
25985edc | 1395 | * If we get here, the task hasn't moved at all, but |
1563513d GH |
1396 | * it has failed to push. We will not try again, |
1397 | * since the other cpus will pull from us when they | |
1398 | * are ready. | |
1399 | */ | |
1400 | dequeue_pushable_task(rq, next_task); | |
1401 | goto out; | |
e8fa1362 | 1402 | } |
917b627d | 1403 | |
1563513d GH |
1404 | if (!task) |
1405 | /* No more tasks, just exit */ | |
1406 | goto out; | |
1407 | ||
917b627d | 1408 | /* |
1563513d | 1409 | * Something has shifted, try again. |
917b627d | 1410 | */ |
1563513d GH |
1411 | put_task_struct(next_task); |
1412 | next_task = task; | |
1413 | goto retry; | |
e8fa1362 SR |
1414 | } |
1415 | ||
697f0a48 | 1416 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1417 | set_task_cpu(next_task, lowest_rq->cpu); |
1418 | activate_task(lowest_rq, next_task, 0); | |
1419 | ||
1420 | resched_task(lowest_rq->curr); | |
1421 | ||
1b12bbc7 | 1422 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1423 | |
e8fa1362 SR |
1424 | out: |
1425 | put_task_struct(next_task); | |
1426 | ||
917b627d | 1427 | return 1; |
e8fa1362 SR |
1428 | } |
1429 | ||
e8fa1362 SR |
1430 | static void push_rt_tasks(struct rq *rq) |
1431 | { | |
1432 | /* push_rt_task will return true if it moved an RT */ | |
1433 | while (push_rt_task(rq)) | |
1434 | ; | |
1435 | } | |
1436 | ||
f65eda4f SR |
1437 | static int pull_rt_task(struct rq *this_rq) |
1438 | { | |
80bf3171 | 1439 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
a8728944 | 1440 | struct task_struct *p; |
f65eda4f | 1441 | struct rq *src_rq; |
f65eda4f | 1442 | |
637f5085 | 1443 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
1444 | return 0; |
1445 | ||
c6c4927b | 1446 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1447 | if (this_cpu == cpu) |
1448 | continue; | |
1449 | ||
1450 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1451 | |
1452 | /* | |
1453 | * Don't bother taking the src_rq->lock if the next highest | |
1454 | * task is known to be lower-priority than our current task. | |
1455 | * This may look racy, but if this value is about to go | |
1456 | * logically higher, the src_rq will push this task away. | |
1457 | * And if its going logically lower, we do not care | |
1458 | */ | |
1459 | if (src_rq->rt.highest_prio.next >= | |
1460 | this_rq->rt.highest_prio.curr) | |
1461 | continue; | |
1462 | ||
f65eda4f SR |
1463 | /* |
1464 | * We can potentially drop this_rq's lock in | |
1465 | * double_lock_balance, and another CPU could | |
a8728944 | 1466 | * alter this_rq |
f65eda4f | 1467 | */ |
a8728944 | 1468 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
1469 | |
1470 | /* | |
1471 | * Are there still pullable RT tasks? | |
1472 | */ | |
614ee1f6 MG |
1473 | if (src_rq->rt.rt_nr_running <= 1) |
1474 | goto skip; | |
f65eda4f | 1475 | |
f65eda4f SR |
1476 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
1477 | ||
1478 | /* | |
1479 | * Do we have an RT task that preempts | |
1480 | * the to-be-scheduled task? | |
1481 | */ | |
a8728944 | 1482 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f | 1483 | WARN_ON(p == src_rq->curr); |
fd2f4419 | 1484 | WARN_ON(!p->on_rq); |
f65eda4f SR |
1485 | |
1486 | /* | |
1487 | * There's a chance that p is higher in priority | |
1488 | * than what's currently running on its cpu. | |
1489 | * This is just that p is wakeing up and hasn't | |
1490 | * had a chance to schedule. We only pull | |
1491 | * p if it is lower in priority than the | |
a8728944 | 1492 | * current task on the run queue |
f65eda4f | 1493 | */ |
a8728944 | 1494 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 1495 | goto skip; |
f65eda4f SR |
1496 | |
1497 | ret = 1; | |
1498 | ||
1499 | deactivate_task(src_rq, p, 0); | |
1500 | set_task_cpu(p, this_cpu); | |
1501 | activate_task(this_rq, p, 0); | |
1502 | /* | |
1503 | * We continue with the search, just in | |
1504 | * case there's an even higher prio task | |
25985edc | 1505 | * in another runqueue. (low likelihood |
f65eda4f | 1506 | * but possible) |
f65eda4f | 1507 | */ |
f65eda4f | 1508 | } |
49246274 | 1509 | skip: |
1b12bbc7 | 1510 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
1511 | } |
1512 | ||
1513 | return ret; | |
1514 | } | |
1515 | ||
9a897c5a | 1516 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
f65eda4f SR |
1517 | { |
1518 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
e864c499 | 1519 | if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio) |
f65eda4f SR |
1520 | pull_rt_task(rq); |
1521 | } | |
1522 | ||
9a897c5a | 1523 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 | 1524 | { |
967fc046 | 1525 | push_rt_tasks(rq); |
e8fa1362 SR |
1526 | } |
1527 | ||
8ae121ac GH |
1528 | /* |
1529 | * If we are not running and we are not going to reschedule soon, we should | |
1530 | * try to push tasks away now | |
1531 | */ | |
efbbd05a | 1532 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1533 | { |
9a897c5a | 1534 | if (!task_running(rq, p) && |
8ae121ac | 1535 | !test_tsk_need_resched(rq->curr) && |
917b627d | 1536 | has_pushable_tasks(rq) && |
b3bc211c | 1537 | p->rt.nr_cpus_allowed > 1 && |
43fa5460 | 1538 | rt_task(rq->curr) && |
b3bc211c SR |
1539 | (rq->curr->rt.nr_cpus_allowed < 2 || |
1540 | rq->curr->prio < p->prio)) | |
4642dafd SR |
1541 | push_rt_tasks(rq); |
1542 | } | |
1543 | ||
cd8ba7cd | 1544 | static void set_cpus_allowed_rt(struct task_struct *p, |
96f874e2 | 1545 | const struct cpumask *new_mask) |
73fe6aae | 1546 | { |
96f874e2 | 1547 | int weight = cpumask_weight(new_mask); |
73fe6aae GH |
1548 | |
1549 | BUG_ON(!rt_task(p)); | |
1550 | ||
1551 | /* | |
1552 | * Update the migration status of the RQ if we have an RT task | |
1553 | * which is running AND changing its weight value. | |
1554 | */ | |
fd2f4419 | 1555 | if (p->on_rq && (weight != p->rt.nr_cpus_allowed)) { |
73fe6aae GH |
1556 | struct rq *rq = task_rq(p); |
1557 | ||
917b627d GH |
1558 | if (!task_current(rq, p)) { |
1559 | /* | |
1560 | * Make sure we dequeue this task from the pushable list | |
1561 | * before going further. It will either remain off of | |
1562 | * the list because we are no longer pushable, or it | |
1563 | * will be requeued. | |
1564 | */ | |
1565 | if (p->rt.nr_cpus_allowed > 1) | |
1566 | dequeue_pushable_task(rq, p); | |
1567 | ||
1568 | /* | |
1569 | * Requeue if our weight is changing and still > 1 | |
1570 | */ | |
1571 | if (weight > 1) | |
1572 | enqueue_pushable_task(rq, p); | |
1573 | ||
1574 | } | |
1575 | ||
6f505b16 | 1576 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
73fe6aae | 1577 | rq->rt.rt_nr_migratory++; |
6f505b16 | 1578 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
73fe6aae GH |
1579 | BUG_ON(!rq->rt.rt_nr_migratory); |
1580 | rq->rt.rt_nr_migratory--; | |
1581 | } | |
1582 | ||
398a153b | 1583 | update_rt_migration(&rq->rt); |
73fe6aae GH |
1584 | } |
1585 | ||
96f874e2 | 1586 | cpumask_copy(&p->cpus_allowed, new_mask); |
6f505b16 | 1587 | p->rt.nr_cpus_allowed = weight; |
73fe6aae | 1588 | } |
deeeccd4 | 1589 | |
bdd7c81b | 1590 | /* Assumes rq->lock is held */ |
1f11eb6a | 1591 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
1592 | { |
1593 | if (rq->rt.overloaded) | |
1594 | rt_set_overload(rq); | |
6e0534f2 | 1595 | |
7def2be1 PZ |
1596 | __enable_runtime(rq); |
1597 | ||
e864c499 | 1598 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
1599 | } |
1600 | ||
1601 | /* Assumes rq->lock is held */ | |
1f11eb6a | 1602 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
1603 | { |
1604 | if (rq->rt.overloaded) | |
1605 | rt_clear_overload(rq); | |
6e0534f2 | 1606 | |
7def2be1 PZ |
1607 | __disable_runtime(rq); |
1608 | ||
6e0534f2 | 1609 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 1610 | } |
cb469845 SR |
1611 | |
1612 | /* | |
1613 | * When switch from the rt queue, we bring ourselves to a position | |
1614 | * that we might want to pull RT tasks from other runqueues. | |
1615 | */ | |
da7a735e | 1616 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1617 | { |
1618 | /* | |
1619 | * If there are other RT tasks then we will reschedule | |
1620 | * and the scheduling of the other RT tasks will handle | |
1621 | * the balancing. But if we are the last RT task | |
1622 | * we may need to handle the pulling of RT tasks | |
1623 | * now. | |
1624 | */ | |
fd2f4419 | 1625 | if (p->on_rq && !rq->rt.rt_nr_running) |
cb469845 SR |
1626 | pull_rt_task(rq); |
1627 | } | |
3d8cbdf8 RR |
1628 | |
1629 | static inline void init_sched_rt_class(void) | |
1630 | { | |
1631 | unsigned int i; | |
1632 | ||
1633 | for_each_possible_cpu(i) | |
eaa95840 | 1634 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 1635 | GFP_KERNEL, cpu_to_node(i)); |
3d8cbdf8 | 1636 | } |
cb469845 SR |
1637 | #endif /* CONFIG_SMP */ |
1638 | ||
1639 | /* | |
1640 | * When switching a task to RT, we may overload the runqueue | |
1641 | * with RT tasks. In this case we try to push them off to | |
1642 | * other runqueues. | |
1643 | */ | |
da7a735e | 1644 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1645 | { |
1646 | int check_resched = 1; | |
1647 | ||
1648 | /* | |
1649 | * If we are already running, then there's nothing | |
1650 | * that needs to be done. But if we are not running | |
1651 | * we may need to preempt the current running task. | |
1652 | * If that current running task is also an RT task | |
1653 | * then see if we can move to another run queue. | |
1654 | */ | |
fd2f4419 | 1655 | if (p->on_rq && rq->curr != p) { |
cb469845 SR |
1656 | #ifdef CONFIG_SMP |
1657 | if (rq->rt.overloaded && push_rt_task(rq) && | |
1658 | /* Don't resched if we changed runqueues */ | |
1659 | rq != task_rq(p)) | |
1660 | check_resched = 0; | |
1661 | #endif /* CONFIG_SMP */ | |
1662 | if (check_resched && p->prio < rq->curr->prio) | |
1663 | resched_task(rq->curr); | |
1664 | } | |
1665 | } | |
1666 | ||
1667 | /* | |
1668 | * Priority of the task has changed. This may cause | |
1669 | * us to initiate a push or pull. | |
1670 | */ | |
da7a735e PZ |
1671 | static void |
1672 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) | |
cb469845 | 1673 | { |
fd2f4419 | 1674 | if (!p->on_rq) |
da7a735e PZ |
1675 | return; |
1676 | ||
1677 | if (rq->curr == p) { | |
cb469845 SR |
1678 | #ifdef CONFIG_SMP |
1679 | /* | |
1680 | * If our priority decreases while running, we | |
1681 | * may need to pull tasks to this runqueue. | |
1682 | */ | |
1683 | if (oldprio < p->prio) | |
1684 | pull_rt_task(rq); | |
1685 | /* | |
1686 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
1687 | * then reschedule. Note, the above pull_rt_task |
1688 | * can release the rq lock and p could migrate. | |
1689 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 1690 | */ |
e864c499 | 1691 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
cb469845 SR |
1692 | resched_task(p); |
1693 | #else | |
1694 | /* For UP simply resched on drop of prio */ | |
1695 | if (oldprio < p->prio) | |
1696 | resched_task(p); | |
e8fa1362 | 1697 | #endif /* CONFIG_SMP */ |
cb469845 SR |
1698 | } else { |
1699 | /* | |
1700 | * This task is not running, but if it is | |
1701 | * greater than the current running task | |
1702 | * then reschedule. | |
1703 | */ | |
1704 | if (p->prio < rq->curr->prio) | |
1705 | resched_task(rq->curr); | |
1706 | } | |
1707 | } | |
1708 | ||
78f2c7db PZ |
1709 | static void watchdog(struct rq *rq, struct task_struct *p) |
1710 | { | |
1711 | unsigned long soft, hard; | |
1712 | ||
78d7d407 JS |
1713 | /* max may change after cur was read, this will be fixed next tick */ |
1714 | soft = task_rlimit(p, RLIMIT_RTTIME); | |
1715 | hard = task_rlimit_max(p, RLIMIT_RTTIME); | |
78f2c7db PZ |
1716 | |
1717 | if (soft != RLIM_INFINITY) { | |
1718 | unsigned long next; | |
1719 | ||
1720 | p->rt.timeout++; | |
1721 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); | |
5a52dd50 | 1722 | if (p->rt.timeout > next) |
f06febc9 | 1723 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
1724 | } |
1725 | } | |
bb44e5d1 | 1726 | |
8f4d37ec | 1727 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 1728 | { |
67e2be02 PZ |
1729 | update_curr_rt(rq); |
1730 | ||
78f2c7db PZ |
1731 | watchdog(rq, p); |
1732 | ||
bb44e5d1 IM |
1733 | /* |
1734 | * RR tasks need a special form of timeslice management. | |
1735 | * FIFO tasks have no timeslices. | |
1736 | */ | |
1737 | if (p->policy != SCHED_RR) | |
1738 | return; | |
1739 | ||
fa717060 | 1740 | if (--p->rt.time_slice) |
bb44e5d1 IM |
1741 | return; |
1742 | ||
fa717060 | 1743 | p->rt.time_slice = DEF_TIMESLICE; |
bb44e5d1 | 1744 | |
98fbc798 DA |
1745 | /* |
1746 | * Requeue to the end of queue if we are not the only element | |
1747 | * on the queue: | |
1748 | */ | |
fa717060 | 1749 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
7ebefa8c | 1750 | requeue_task_rt(rq, p, 0); |
98fbc798 DA |
1751 | set_tsk_need_resched(p); |
1752 | } | |
bb44e5d1 IM |
1753 | } |
1754 | ||
83b699ed SV |
1755 | static void set_curr_task_rt(struct rq *rq) |
1756 | { | |
1757 | struct task_struct *p = rq->curr; | |
1758 | ||
305e6835 | 1759 | p->se.exec_start = rq->clock_task; |
917b627d GH |
1760 | |
1761 | /* The running task is never eligible for pushing */ | |
1762 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
1763 | } |
1764 | ||
6d686f45 | 1765 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
0d721cea PW |
1766 | { |
1767 | /* | |
1768 | * Time slice is 0 for SCHED_FIFO tasks | |
1769 | */ | |
1770 | if (task->policy == SCHED_RR) | |
1771 | return DEF_TIMESLICE; | |
1772 | else | |
1773 | return 0; | |
1774 | } | |
1775 | ||
2abdad0a | 1776 | static const struct sched_class rt_sched_class = { |
5522d5d5 | 1777 | .next = &fair_sched_class, |
bb44e5d1 IM |
1778 | .enqueue_task = enqueue_task_rt, |
1779 | .dequeue_task = dequeue_task_rt, | |
1780 | .yield_task = yield_task_rt, | |
1781 | ||
1782 | .check_preempt_curr = check_preempt_curr_rt, | |
1783 | ||
1784 | .pick_next_task = pick_next_task_rt, | |
1785 | .put_prev_task = put_prev_task_rt, | |
1786 | ||
681f3e68 | 1787 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
1788 | .select_task_rq = select_task_rq_rt, |
1789 | ||
73fe6aae | 1790 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
1791 | .rq_online = rq_online_rt, |
1792 | .rq_offline = rq_offline_rt, | |
9a897c5a SR |
1793 | .pre_schedule = pre_schedule_rt, |
1794 | .post_schedule = post_schedule_rt, | |
efbbd05a | 1795 | .task_woken = task_woken_rt, |
cb469845 | 1796 | .switched_from = switched_from_rt, |
681f3e68 | 1797 | #endif |
bb44e5d1 | 1798 | |
83b699ed | 1799 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 1800 | .task_tick = task_tick_rt, |
cb469845 | 1801 | |
0d721cea PW |
1802 | .get_rr_interval = get_rr_interval_rt, |
1803 | ||
cb469845 SR |
1804 | .prio_changed = prio_changed_rt, |
1805 | .switched_to = switched_to_rt, | |
bb44e5d1 | 1806 | }; |
ada18de2 PZ |
1807 | |
1808 | #ifdef CONFIG_SCHED_DEBUG | |
1809 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
1810 | ||
1811 | static void print_rt_stats(struct seq_file *m, int cpu) | |
1812 | { | |
1813 | struct rt_rq *rt_rq; | |
1814 | ||
1815 | rcu_read_lock(); | |
1816 | for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu)) | |
1817 | print_rt_rq(m, cpu, rt_rq); | |
1818 | rcu_read_unlock(); | |
1819 | } | |
55e12e5e | 1820 | #endif /* CONFIG_SCHED_DEBUG */ |
0e3900e6 | 1821 |