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 | ||
5181f4a4 SR |
127 | static inline int has_pushable_tasks(struct rq *rq) |
128 | { | |
129 | return !plist_head_empty(&rq->rt.pushable_tasks); | |
130 | } | |
131 | ||
917b627d GH |
132 | static void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
133 | { | |
134 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
135 | plist_node_init(&p->pushable_tasks, p->prio); | |
136 | plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
5181f4a4 SR |
137 | |
138 | /* Update the highest prio pushable task */ | |
139 | if (p->prio < rq->rt.highest_prio.next) | |
140 | rq->rt.highest_prio.next = p->prio; | |
917b627d GH |
141 | } |
142 | ||
143 | static void dequeue_pushable_task(struct rq *rq, struct task_struct *p) | |
144 | { | |
145 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
917b627d | 146 | |
5181f4a4 SR |
147 | /* Update the new highest prio pushable task */ |
148 | if (has_pushable_tasks(rq)) { | |
149 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
150 | struct task_struct, pushable_tasks); | |
151 | rq->rt.highest_prio.next = p->prio; | |
152 | } else | |
153 | rq->rt.highest_prio.next = MAX_RT_PRIO; | |
bcf08df3 IM |
154 | } |
155 | ||
917b627d GH |
156 | #else |
157 | ||
ceacc2c1 | 158 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
fa85ae24 | 159 | { |
6f505b16 PZ |
160 | } |
161 | ||
ceacc2c1 PZ |
162 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
163 | { | |
164 | } | |
165 | ||
b07430ac | 166 | static inline |
ceacc2c1 PZ |
167 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
168 | { | |
169 | } | |
170 | ||
398a153b | 171 | static inline |
ceacc2c1 PZ |
172 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
173 | { | |
174 | } | |
917b627d | 175 | |
4fd29176 SR |
176 | #endif /* CONFIG_SMP */ |
177 | ||
6f505b16 PZ |
178 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
179 | { | |
180 | return !list_empty(&rt_se->run_list); | |
181 | } | |
182 | ||
052f1dc7 | 183 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 184 | |
9f0c1e56 | 185 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
186 | { |
187 | if (!rt_rq->tg) | |
9f0c1e56 | 188 | return RUNTIME_INF; |
6f505b16 | 189 | |
ac086bc2 PZ |
190 | return rt_rq->rt_runtime; |
191 | } | |
192 | ||
193 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
194 | { | |
195 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); | |
6f505b16 PZ |
196 | } |
197 | ||
ec514c48 CX |
198 | typedef struct task_group *rt_rq_iter_t; |
199 | ||
1c09ab0d YZ |
200 | static inline struct task_group *next_task_group(struct task_group *tg) |
201 | { | |
202 | do { | |
203 | tg = list_entry_rcu(tg->list.next, | |
204 | typeof(struct task_group), list); | |
205 | } while (&tg->list != &task_groups && task_group_is_autogroup(tg)); | |
206 | ||
207 | if (&tg->list == &task_groups) | |
208 | tg = NULL; | |
209 | ||
210 | return tg; | |
211 | } | |
212 | ||
213 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
214 | for (iter = container_of(&task_groups, typeof(*iter), list); \ | |
215 | (iter = next_task_group(iter)) && \ | |
216 | (rt_rq = iter->rt_rq[cpu_of(rq)]);) | |
ec514c48 | 217 | |
3d4b47b4 PZ |
218 | static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq) |
219 | { | |
220 | list_add_rcu(&rt_rq->leaf_rt_rq_list, | |
221 | &rq_of_rt_rq(rt_rq)->leaf_rt_rq_list); | |
222 | } | |
223 | ||
224 | static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq) | |
225 | { | |
226 | list_del_rcu(&rt_rq->leaf_rt_rq_list); | |
227 | } | |
228 | ||
6f505b16 | 229 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
80f40ee4 | 230 | list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) |
6f505b16 | 231 | |
6f505b16 PZ |
232 | #define for_each_sched_rt_entity(rt_se) \ |
233 | for (; rt_se; rt_se = rt_se->parent) | |
234 | ||
235 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
236 | { | |
237 | return rt_se->my_q; | |
238 | } | |
239 | ||
37dad3fc | 240 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
6f505b16 PZ |
241 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
242 | ||
9f0c1e56 | 243 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 244 | { |
f6121f4f | 245 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
74b7eb58 YZ |
246 | struct sched_rt_entity *rt_se; |
247 | ||
0c3b9168 BS |
248 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
249 | ||
250 | rt_se = rt_rq->tg->rt_se[cpu]; | |
6f505b16 | 251 | |
f6121f4f DF |
252 | if (rt_rq->rt_nr_running) { |
253 | if (rt_se && !on_rt_rq(rt_se)) | |
37dad3fc | 254 | enqueue_rt_entity(rt_se, false); |
e864c499 | 255 | if (rt_rq->highest_prio.curr < curr->prio) |
1020387f | 256 | resched_task(curr); |
6f505b16 PZ |
257 | } |
258 | } | |
259 | ||
9f0c1e56 | 260 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 261 | { |
74b7eb58 | 262 | struct sched_rt_entity *rt_se; |
0c3b9168 | 263 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
74b7eb58 | 264 | |
0c3b9168 | 265 | rt_se = rt_rq->tg->rt_se[cpu]; |
6f505b16 PZ |
266 | |
267 | if (rt_se && on_rt_rq(rt_se)) | |
268 | dequeue_rt_entity(rt_se); | |
269 | } | |
270 | ||
23b0fdfc PZ |
271 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
272 | { | |
273 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
274 | } | |
275 | ||
276 | static int rt_se_boosted(struct sched_rt_entity *rt_se) | |
277 | { | |
278 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
279 | struct task_struct *p; | |
280 | ||
281 | if (rt_rq) | |
282 | return !!rt_rq->rt_nr_boosted; | |
283 | ||
284 | p = rt_task_of(rt_se); | |
285 | return p->prio != p->normal_prio; | |
286 | } | |
287 | ||
d0b27fa7 | 288 | #ifdef CONFIG_SMP |
c6c4927b | 289 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 PZ |
290 | { |
291 | return cpu_rq(smp_processor_id())->rd->span; | |
292 | } | |
6f505b16 | 293 | #else |
c6c4927b | 294 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 295 | { |
c6c4927b | 296 | return cpu_online_mask; |
d0b27fa7 PZ |
297 | } |
298 | #endif | |
6f505b16 | 299 | |
d0b27fa7 PZ |
300 | static inline |
301 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 302 | { |
d0b27fa7 PZ |
303 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
304 | } | |
9f0c1e56 | 305 | |
ac086bc2 PZ |
306 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
307 | { | |
308 | return &rt_rq->tg->rt_bandwidth; | |
309 | } | |
310 | ||
55e12e5e | 311 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
312 | |
313 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
314 | { | |
ac086bc2 PZ |
315 | return rt_rq->rt_runtime; |
316 | } | |
317 | ||
318 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
319 | { | |
320 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
321 | } |
322 | ||
ec514c48 CX |
323 | typedef struct rt_rq *rt_rq_iter_t; |
324 | ||
325 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
326 | for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
327 | ||
3d4b47b4 PZ |
328 | static inline void list_add_leaf_rt_rq(struct rt_rq *rt_rq) |
329 | { | |
330 | } | |
331 | ||
332 | static inline void list_del_leaf_rt_rq(struct rt_rq *rt_rq) | |
333 | { | |
334 | } | |
335 | ||
6f505b16 PZ |
336 | #define for_each_leaf_rt_rq(rt_rq, rq) \ |
337 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
338 | ||
6f505b16 PZ |
339 | #define for_each_sched_rt_entity(rt_se) \ |
340 | for (; rt_se; rt_se = NULL) | |
341 | ||
342 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
343 | { | |
344 | return NULL; | |
345 | } | |
346 | ||
9f0c1e56 | 347 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 348 | { |
f3ade837 JB |
349 | if (rt_rq->rt_nr_running) |
350 | resched_task(rq_of_rt_rq(rt_rq)->curr); | |
6f505b16 PZ |
351 | } |
352 | ||
9f0c1e56 | 353 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
354 | { |
355 | } | |
356 | ||
23b0fdfc PZ |
357 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
358 | { | |
359 | return rt_rq->rt_throttled; | |
360 | } | |
d0b27fa7 | 361 | |
c6c4927b | 362 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 363 | { |
c6c4927b | 364 | return cpu_online_mask; |
d0b27fa7 PZ |
365 | } |
366 | ||
367 | static inline | |
368 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
369 | { | |
370 | return &cpu_rq(cpu)->rt; | |
371 | } | |
372 | ||
ac086bc2 PZ |
373 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
374 | { | |
375 | return &def_rt_bandwidth; | |
376 | } | |
377 | ||
55e12e5e | 378 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 379 | |
ac086bc2 | 380 | #ifdef CONFIG_SMP |
78333cdd PZ |
381 | /* |
382 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
383 | */ | |
b79f3833 | 384 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
385 | { |
386 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
387 | struct root_domain *rd = cpu_rq(smp_processor_id())->rd; | |
388 | int i, weight, more = 0; | |
389 | u64 rt_period; | |
390 | ||
c6c4927b | 391 | weight = cpumask_weight(rd->span); |
ac086bc2 | 392 | |
0986b11b | 393 | raw_spin_lock(&rt_b->rt_runtime_lock); |
ac086bc2 | 394 | rt_period = ktime_to_ns(rt_b->rt_period); |
c6c4927b | 395 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
396 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
397 | s64 diff; | |
398 | ||
399 | if (iter == rt_rq) | |
400 | continue; | |
401 | ||
0986b11b | 402 | raw_spin_lock(&iter->rt_runtime_lock); |
78333cdd PZ |
403 | /* |
404 | * Either all rqs have inf runtime and there's nothing to steal | |
405 | * or __disable_runtime() below sets a specific rq to inf to | |
406 | * indicate its been disabled and disalow stealing. | |
407 | */ | |
7def2be1 PZ |
408 | if (iter->rt_runtime == RUNTIME_INF) |
409 | goto next; | |
410 | ||
78333cdd PZ |
411 | /* |
412 | * From runqueues with spare time, take 1/n part of their | |
413 | * spare time, but no more than our period. | |
414 | */ | |
ac086bc2 PZ |
415 | diff = iter->rt_runtime - iter->rt_time; |
416 | if (diff > 0) { | |
58838cf3 | 417 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
418 | if (rt_rq->rt_runtime + diff > rt_period) |
419 | diff = rt_period - rt_rq->rt_runtime; | |
420 | iter->rt_runtime -= diff; | |
421 | rt_rq->rt_runtime += diff; | |
422 | more = 1; | |
423 | if (rt_rq->rt_runtime == rt_period) { | |
0986b11b | 424 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 PZ |
425 | break; |
426 | } | |
427 | } | |
7def2be1 | 428 | next: |
0986b11b | 429 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 | 430 | } |
0986b11b | 431 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
ac086bc2 PZ |
432 | |
433 | return more; | |
434 | } | |
7def2be1 | 435 | |
78333cdd PZ |
436 | /* |
437 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
438 | */ | |
7def2be1 PZ |
439 | static void __disable_runtime(struct rq *rq) |
440 | { | |
441 | struct root_domain *rd = rq->rd; | |
ec514c48 | 442 | rt_rq_iter_t iter; |
7def2be1 PZ |
443 | struct rt_rq *rt_rq; |
444 | ||
445 | if (unlikely(!scheduler_running)) | |
446 | return; | |
447 | ||
ec514c48 | 448 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
449 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
450 | s64 want; | |
451 | int i; | |
452 | ||
0986b11b TG |
453 | raw_spin_lock(&rt_b->rt_runtime_lock); |
454 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
455 | /* |
456 | * Either we're all inf and nobody needs to borrow, or we're | |
457 | * already disabled and thus have nothing to do, or we have | |
458 | * exactly the right amount of runtime to take out. | |
459 | */ | |
7def2be1 PZ |
460 | if (rt_rq->rt_runtime == RUNTIME_INF || |
461 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
462 | goto balanced; | |
0986b11b | 463 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
7def2be1 | 464 | |
78333cdd PZ |
465 | /* |
466 | * Calculate the difference between what we started out with | |
467 | * and what we current have, that's the amount of runtime | |
468 | * we lend and now have to reclaim. | |
469 | */ | |
7def2be1 PZ |
470 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
471 | ||
78333cdd PZ |
472 | /* |
473 | * Greedy reclaim, take back as much as we can. | |
474 | */ | |
c6c4927b | 475 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
476 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
477 | s64 diff; | |
478 | ||
78333cdd PZ |
479 | /* |
480 | * Can't reclaim from ourselves or disabled runqueues. | |
481 | */ | |
f1679d08 | 482 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
483 | continue; |
484 | ||
0986b11b | 485 | raw_spin_lock(&iter->rt_runtime_lock); |
7def2be1 PZ |
486 | if (want > 0) { |
487 | diff = min_t(s64, iter->rt_runtime, want); | |
488 | iter->rt_runtime -= diff; | |
489 | want -= diff; | |
490 | } else { | |
491 | iter->rt_runtime -= want; | |
492 | want -= want; | |
493 | } | |
0986b11b | 494 | raw_spin_unlock(&iter->rt_runtime_lock); |
7def2be1 PZ |
495 | |
496 | if (!want) | |
497 | break; | |
498 | } | |
499 | ||
0986b11b | 500 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
78333cdd PZ |
501 | /* |
502 | * We cannot be left wanting - that would mean some runtime | |
503 | * leaked out of the system. | |
504 | */ | |
7def2be1 PZ |
505 | BUG_ON(want); |
506 | balanced: | |
78333cdd PZ |
507 | /* |
508 | * Disable all the borrow logic by pretending we have inf | |
509 | * runtime - in which case borrowing doesn't make sense. | |
510 | */ | |
7def2be1 | 511 | rt_rq->rt_runtime = RUNTIME_INF; |
0986b11b TG |
512 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
513 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
514 | } |
515 | } | |
516 | ||
517 | static void disable_runtime(struct rq *rq) | |
518 | { | |
519 | unsigned long flags; | |
520 | ||
05fa785c | 521 | raw_spin_lock_irqsave(&rq->lock, flags); |
7def2be1 | 522 | __disable_runtime(rq); |
05fa785c | 523 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
7def2be1 PZ |
524 | } |
525 | ||
526 | static void __enable_runtime(struct rq *rq) | |
527 | { | |
ec514c48 | 528 | rt_rq_iter_t iter; |
7def2be1 PZ |
529 | struct rt_rq *rt_rq; |
530 | ||
531 | if (unlikely(!scheduler_running)) | |
532 | return; | |
533 | ||
78333cdd PZ |
534 | /* |
535 | * Reset each runqueue's bandwidth settings | |
536 | */ | |
ec514c48 | 537 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
538 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
539 | ||
0986b11b TG |
540 | raw_spin_lock(&rt_b->rt_runtime_lock); |
541 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
7def2be1 PZ |
542 | rt_rq->rt_runtime = rt_b->rt_runtime; |
543 | rt_rq->rt_time = 0; | |
baf25731 | 544 | rt_rq->rt_throttled = 0; |
0986b11b TG |
545 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
546 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
547 | } |
548 | } | |
549 | ||
550 | static void enable_runtime(struct rq *rq) | |
551 | { | |
552 | unsigned long flags; | |
553 | ||
05fa785c | 554 | raw_spin_lock_irqsave(&rq->lock, flags); |
7def2be1 | 555 | __enable_runtime(rq); |
05fa785c | 556 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
7def2be1 PZ |
557 | } |
558 | ||
eff6549b PZ |
559 | static int balance_runtime(struct rt_rq *rt_rq) |
560 | { | |
561 | int more = 0; | |
562 | ||
4a6184ce PZ |
563 | if (!sched_feat(RT_RUNTIME_SHARE)) |
564 | return more; | |
565 | ||
eff6549b | 566 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
0986b11b | 567 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
eff6549b | 568 | more = do_balance_runtime(rt_rq); |
0986b11b | 569 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
570 | } |
571 | ||
572 | return more; | |
573 | } | |
55e12e5e | 574 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
575 | static inline int balance_runtime(struct rt_rq *rt_rq) |
576 | { | |
577 | return 0; | |
578 | } | |
55e12e5e | 579 | #endif /* CONFIG_SMP */ |
ac086bc2 | 580 | |
eff6549b PZ |
581 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
582 | { | |
583 | int i, idle = 1; | |
c6c4927b | 584 | const struct cpumask *span; |
eff6549b | 585 | |
0b148fa0 | 586 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
eff6549b PZ |
587 | return 1; |
588 | ||
589 | span = sched_rt_period_mask(); | |
c6c4927b | 590 | for_each_cpu(i, span) { |
eff6549b PZ |
591 | int enqueue = 0; |
592 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
593 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
594 | ||
05fa785c | 595 | raw_spin_lock(&rq->lock); |
eff6549b PZ |
596 | if (rt_rq->rt_time) { |
597 | u64 runtime; | |
598 | ||
0986b11b | 599 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
600 | if (rt_rq->rt_throttled) |
601 | balance_runtime(rt_rq); | |
602 | runtime = rt_rq->rt_runtime; | |
603 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
604 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
605 | rt_rq->rt_throttled = 0; | |
606 | enqueue = 1; | |
61eadef6 MG |
607 | |
608 | /* | |
609 | * Force a clock update if the CPU was idle, | |
610 | * lest wakeup -> unthrottle time accumulate. | |
611 | */ | |
612 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) | |
613 | rq->skip_clock_update = -1; | |
eff6549b PZ |
614 | } |
615 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
616 | idle = 0; | |
0986b11b | 617 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
0c3b9168 | 618 | } else if (rt_rq->rt_nr_running) { |
6c3df255 | 619 | idle = 0; |
0c3b9168 BS |
620 | if (!rt_rq_throttled(rt_rq)) |
621 | enqueue = 1; | |
622 | } | |
eff6549b PZ |
623 | |
624 | if (enqueue) | |
625 | sched_rt_rq_enqueue(rt_rq); | |
05fa785c | 626 | raw_spin_unlock(&rq->lock); |
eff6549b PZ |
627 | } |
628 | ||
629 | return idle; | |
630 | } | |
ac086bc2 | 631 | |
6f505b16 PZ |
632 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
633 | { | |
052f1dc7 | 634 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
635 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
636 | ||
637 | if (rt_rq) | |
e864c499 | 638 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
639 | #endif |
640 | ||
641 | return rt_task_of(rt_se)->prio; | |
642 | } | |
643 | ||
9f0c1e56 | 644 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 645 | { |
9f0c1e56 | 646 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 647 | |
fa85ae24 | 648 | if (rt_rq->rt_throttled) |
23b0fdfc | 649 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 650 | |
ac086bc2 PZ |
651 | if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq)) |
652 | return 0; | |
653 | ||
b79f3833 PZ |
654 | balance_runtime(rt_rq); |
655 | runtime = sched_rt_runtime(rt_rq); | |
656 | if (runtime == RUNTIME_INF) | |
657 | return 0; | |
ac086bc2 | 658 | |
9f0c1e56 | 659 | if (rt_rq->rt_time > runtime) { |
6f505b16 | 660 | rt_rq->rt_throttled = 1; |
1c83437e | 661 | printk_once(KERN_WARNING "sched: RT throttling activated\n"); |
23b0fdfc | 662 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 663 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
664 | return 1; |
665 | } | |
fa85ae24 PZ |
666 | } |
667 | ||
668 | return 0; | |
669 | } | |
670 | ||
bb44e5d1 IM |
671 | /* |
672 | * Update the current task's runtime statistics. Skip current tasks that | |
673 | * are not in our scheduling class. | |
674 | */ | |
a9957449 | 675 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
676 | { |
677 | struct task_struct *curr = rq->curr; | |
6f505b16 PZ |
678 | struct sched_rt_entity *rt_se = &curr->rt; |
679 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
bb44e5d1 IM |
680 | u64 delta_exec; |
681 | ||
06c3bc65 | 682 | if (curr->sched_class != &rt_sched_class) |
bb44e5d1 IM |
683 | return; |
684 | ||
305e6835 | 685 | delta_exec = rq->clock_task - curr->se.exec_start; |
bb44e5d1 IM |
686 | if (unlikely((s64)delta_exec < 0)) |
687 | delta_exec = 0; | |
6cfb0d5d | 688 | |
41acab88 | 689 | schedstat_set(curr->se.statistics.exec_max, max(curr->se.statistics.exec_max, delta_exec)); |
bb44e5d1 IM |
690 | |
691 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
692 | account_group_exec_runtime(curr, delta_exec); |
693 | ||
305e6835 | 694 | curr->se.exec_start = rq->clock_task; |
d842de87 | 695 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 696 | |
e9e9250b PZ |
697 | sched_rt_avg_update(rq, delta_exec); |
698 | ||
0b148fa0 PZ |
699 | if (!rt_bandwidth_enabled()) |
700 | return; | |
701 | ||
354d60c2 DG |
702 | for_each_sched_rt_entity(rt_se) { |
703 | rt_rq = rt_rq_of_se(rt_se); | |
704 | ||
cc2991cf | 705 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
0986b11b | 706 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
707 | rt_rq->rt_time += delta_exec; |
708 | if (sched_rt_runtime_exceeded(rt_rq)) | |
709 | resched_task(curr); | |
0986b11b | 710 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 711 | } |
354d60c2 | 712 | } |
bb44e5d1 IM |
713 | } |
714 | ||
398a153b | 715 | #if defined CONFIG_SMP |
e864c499 | 716 | |
398a153b GH |
717 | static void |
718 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 719 | { |
4d984277 | 720 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 721 | |
5181f4a4 SR |
722 | if (rq->online && prio < prev_prio) |
723 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); | |
398a153b | 724 | } |
73fe6aae | 725 | |
398a153b GH |
726 | static void |
727 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
728 | { | |
729 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 730 | |
398a153b GH |
731 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
732 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
733 | } |
734 | ||
398a153b GH |
735 | #else /* CONFIG_SMP */ |
736 | ||
6f505b16 | 737 | static inline |
398a153b GH |
738 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
739 | static inline | |
740 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
741 | ||
742 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 743 | |
052f1dc7 | 744 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
745 | static void |
746 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
747 | { | |
748 | int prev_prio = rt_rq->highest_prio.curr; | |
749 | ||
750 | if (prio < prev_prio) | |
751 | rt_rq->highest_prio.curr = prio; | |
752 | ||
753 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
754 | } | |
755 | ||
756 | static void | |
757 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
758 | { | |
759 | int prev_prio = rt_rq->highest_prio.curr; | |
760 | ||
6f505b16 | 761 | if (rt_rq->rt_nr_running) { |
764a9d6f | 762 | |
398a153b | 763 | WARN_ON(prio < prev_prio); |
764a9d6f | 764 | |
e864c499 | 765 | /* |
398a153b GH |
766 | * This may have been our highest task, and therefore |
767 | * we may have some recomputation to do | |
e864c499 | 768 | */ |
398a153b | 769 | if (prio == prev_prio) { |
e864c499 GH |
770 | struct rt_prio_array *array = &rt_rq->active; |
771 | ||
772 | rt_rq->highest_prio.curr = | |
764a9d6f | 773 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
774 | } |
775 | ||
764a9d6f | 776 | } else |
e864c499 | 777 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 778 | |
398a153b GH |
779 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
780 | } | |
1f11eb6a | 781 | |
398a153b GH |
782 | #else |
783 | ||
784 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
785 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
786 | ||
787 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 788 | |
052f1dc7 | 789 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
790 | |
791 | static void | |
792 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
793 | { | |
794 | if (rt_se_boosted(rt_se)) | |
795 | rt_rq->rt_nr_boosted++; | |
796 | ||
797 | if (rt_rq->tg) | |
798 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
799 | } | |
800 | ||
801 | static void | |
802 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
803 | { | |
23b0fdfc PZ |
804 | if (rt_se_boosted(rt_se)) |
805 | rt_rq->rt_nr_boosted--; | |
806 | ||
807 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
808 | } |
809 | ||
810 | #else /* CONFIG_RT_GROUP_SCHED */ | |
811 | ||
812 | static void | |
813 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
814 | { | |
815 | start_rt_bandwidth(&def_rt_bandwidth); | |
816 | } | |
817 | ||
818 | static inline | |
819 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
820 | ||
821 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
822 | ||
823 | static inline | |
824 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
825 | { | |
826 | int prio = rt_se_prio(rt_se); | |
827 | ||
828 | WARN_ON(!rt_prio(prio)); | |
829 | rt_rq->rt_nr_running++; | |
830 | ||
831 | inc_rt_prio(rt_rq, prio); | |
832 | inc_rt_migration(rt_se, rt_rq); | |
833 | inc_rt_group(rt_se, rt_rq); | |
834 | } | |
835 | ||
836 | static inline | |
837 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
838 | { | |
839 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
840 | WARN_ON(!rt_rq->rt_nr_running); | |
841 | rt_rq->rt_nr_running--; | |
842 | ||
843 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
844 | dec_rt_migration(rt_se, rt_rq); | |
845 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
846 | } |
847 | ||
37dad3fc | 848 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
bb44e5d1 | 849 | { |
6f505b16 PZ |
850 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
851 | struct rt_prio_array *array = &rt_rq->active; | |
852 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 853 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 854 | |
ad2a3f13 PZ |
855 | /* |
856 | * Don't enqueue the group if its throttled, or when empty. | |
857 | * The latter is a consequence of the former when a child group | |
858 | * get throttled and the current group doesn't have any other | |
859 | * active members. | |
860 | */ | |
861 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 862 | return; |
63489e45 | 863 | |
3d4b47b4 PZ |
864 | if (!rt_rq->rt_nr_running) |
865 | list_add_leaf_rt_rq(rt_rq); | |
866 | ||
37dad3fc TG |
867 | if (head) |
868 | list_add(&rt_se->run_list, queue); | |
869 | else | |
870 | list_add_tail(&rt_se->run_list, queue); | |
6f505b16 | 871 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 872 | |
6f505b16 PZ |
873 | inc_rt_tasks(rt_se, rt_rq); |
874 | } | |
875 | ||
ad2a3f13 | 876 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
877 | { |
878 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
879 | struct rt_prio_array *array = &rt_rq->active; | |
880 | ||
881 | list_del_init(&rt_se->run_list); | |
882 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
883 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
884 | ||
885 | dec_rt_tasks(rt_se, rt_rq); | |
3d4b47b4 PZ |
886 | if (!rt_rq->rt_nr_running) |
887 | list_del_leaf_rt_rq(rt_rq); | |
6f505b16 PZ |
888 | } |
889 | ||
890 | /* | |
891 | * Because the prio of an upper entry depends on the lower | |
892 | * entries, we must remove entries top - down. | |
6f505b16 | 893 | */ |
ad2a3f13 | 894 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 895 | { |
ad2a3f13 | 896 | struct sched_rt_entity *back = NULL; |
6f505b16 | 897 | |
58d6c2d7 PZ |
898 | for_each_sched_rt_entity(rt_se) { |
899 | rt_se->back = back; | |
900 | back = rt_se; | |
901 | } | |
902 | ||
903 | for (rt_se = back; rt_se; rt_se = rt_se->back) { | |
904 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
905 | __dequeue_rt_entity(rt_se); |
906 | } | |
907 | } | |
908 | ||
37dad3fc | 909 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
ad2a3f13 PZ |
910 | { |
911 | dequeue_rt_stack(rt_se); | |
912 | for_each_sched_rt_entity(rt_se) | |
37dad3fc | 913 | __enqueue_rt_entity(rt_se, head); |
ad2a3f13 PZ |
914 | } |
915 | ||
916 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
917 | { | |
918 | dequeue_rt_stack(rt_se); | |
919 | ||
920 | for_each_sched_rt_entity(rt_se) { | |
921 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
922 | ||
923 | if (rt_rq && rt_rq->rt_nr_running) | |
37dad3fc | 924 | __enqueue_rt_entity(rt_se, false); |
58d6c2d7 | 925 | } |
bb44e5d1 IM |
926 | } |
927 | ||
928 | /* | |
929 | * Adding/removing a task to/from a priority array: | |
930 | */ | |
ea87bb78 | 931 | static void |
371fd7e7 | 932 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
6f505b16 PZ |
933 | { |
934 | struct sched_rt_entity *rt_se = &p->rt; | |
935 | ||
371fd7e7 | 936 | if (flags & ENQUEUE_WAKEUP) |
6f505b16 PZ |
937 | rt_se->timeout = 0; |
938 | ||
371fd7e7 | 939 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
c09595f6 | 940 | |
917b627d GH |
941 | if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1) |
942 | enqueue_pushable_task(rq, p); | |
953bfcd1 PT |
943 | |
944 | inc_nr_running(rq); | |
6f505b16 PZ |
945 | } |
946 | ||
371fd7e7 | 947 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 948 | { |
6f505b16 | 949 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 950 | |
f1e14ef6 | 951 | update_curr_rt(rq); |
ad2a3f13 | 952 | dequeue_rt_entity(rt_se); |
c09595f6 | 953 | |
917b627d | 954 | dequeue_pushable_task(rq, p); |
953bfcd1 PT |
955 | |
956 | dec_nr_running(rq); | |
bb44e5d1 IM |
957 | } |
958 | ||
959 | /* | |
960 | * Put task to the end of the run list without the overhead of dequeue | |
961 | * followed by enqueue. | |
962 | */ | |
7ebefa8c DA |
963 | static void |
964 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 965 | { |
1cdad715 | 966 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
967 | struct rt_prio_array *array = &rt_rq->active; |
968 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
969 | ||
970 | if (head) | |
971 | list_move(&rt_se->run_list, queue); | |
972 | else | |
973 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 974 | } |
6f505b16 PZ |
975 | } |
976 | ||
7ebefa8c | 977 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 978 | { |
6f505b16 PZ |
979 | struct sched_rt_entity *rt_se = &p->rt; |
980 | struct rt_rq *rt_rq; | |
bb44e5d1 | 981 | |
6f505b16 PZ |
982 | for_each_sched_rt_entity(rt_se) { |
983 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 984 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 985 | } |
bb44e5d1 IM |
986 | } |
987 | ||
6f505b16 | 988 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 989 | { |
7ebefa8c | 990 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
991 | } |
992 | ||
e7693a36 | 993 | #ifdef CONFIG_SMP |
318e0893 GH |
994 | static int find_lowest_rq(struct task_struct *task); |
995 | ||
0017d735 | 996 | static int |
7608dec2 | 997 | select_task_rq_rt(struct task_struct *p, int sd_flag, int flags) |
e7693a36 | 998 | { |
7608dec2 PZ |
999 | struct task_struct *curr; |
1000 | struct rq *rq; | |
1001 | int cpu; | |
1002 | ||
7608dec2 | 1003 | cpu = task_cpu(p); |
c37495fd SR |
1004 | |
1005 | /* For anything but wake ups, just return the task_cpu */ | |
1006 | if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK) | |
1007 | goto out; | |
1008 | ||
7608dec2 PZ |
1009 | rq = cpu_rq(cpu); |
1010 | ||
1011 | rcu_read_lock(); | |
1012 | curr = ACCESS_ONCE(rq->curr); /* unlocked access */ | |
1013 | ||
318e0893 | 1014 | /* |
7608dec2 | 1015 | * If the current task on @p's runqueue is an RT task, then |
e1f47d89 SR |
1016 | * try to see if we can wake this RT task up on another |
1017 | * runqueue. Otherwise simply start this RT task | |
1018 | * on its current runqueue. | |
1019 | * | |
43fa5460 SR |
1020 | * We want to avoid overloading runqueues. If the woken |
1021 | * task is a higher priority, then it will stay on this CPU | |
1022 | * and the lower prio task should be moved to another CPU. | |
1023 | * Even though this will probably make the lower prio task | |
1024 | * lose its cache, we do not want to bounce a higher task | |
1025 | * around just because it gave up its CPU, perhaps for a | |
1026 | * lock? | |
1027 | * | |
1028 | * For equal prio tasks, we just let the scheduler sort it out. | |
7608dec2 PZ |
1029 | * |
1030 | * Otherwise, just let it ride on the affined RQ and the | |
1031 | * post-schedule router will push the preempted task away | |
1032 | * | |
1033 | * This test is optimistic, if we get it wrong the load-balancer | |
1034 | * will have to sort it out. | |
318e0893 | 1035 | */ |
7608dec2 PZ |
1036 | if (curr && unlikely(rt_task(curr)) && |
1037 | (curr->rt.nr_cpus_allowed < 2 || | |
3be209a8 | 1038 | curr->prio <= p->prio) && |
6f505b16 | 1039 | (p->rt.nr_cpus_allowed > 1)) { |
7608dec2 | 1040 | int target = find_lowest_rq(p); |
318e0893 | 1041 | |
7608dec2 PZ |
1042 | if (target != -1) |
1043 | cpu = target; | |
318e0893 | 1044 | } |
7608dec2 | 1045 | rcu_read_unlock(); |
318e0893 | 1046 | |
c37495fd | 1047 | out: |
7608dec2 | 1048 | return cpu; |
e7693a36 | 1049 | } |
7ebefa8c DA |
1050 | |
1051 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
1052 | { | |
7ebefa8c DA |
1053 | if (rq->curr->rt.nr_cpus_allowed == 1) |
1054 | return; | |
1055 | ||
24600ce8 | 1056 | if (p->rt.nr_cpus_allowed != 1 |
13b8bd0a RR |
1057 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1058 | return; | |
24600ce8 | 1059 | |
13b8bd0a RR |
1060 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
1061 | return; | |
7ebefa8c DA |
1062 | |
1063 | /* | |
1064 | * There appears to be other cpus that can accept | |
1065 | * current and none to run 'p', so lets reschedule | |
1066 | * to try and push current away: | |
1067 | */ | |
1068 | requeue_task_rt(rq, p, 1); | |
1069 | resched_task(rq->curr); | |
1070 | } | |
1071 | ||
e7693a36 GH |
1072 | #endif /* CONFIG_SMP */ |
1073 | ||
bb44e5d1 IM |
1074 | /* |
1075 | * Preempt the current task with a newly woken task if needed: | |
1076 | */ | |
7d478721 | 1077 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1078 | { |
45c01e82 | 1079 | if (p->prio < rq->curr->prio) { |
bb44e5d1 | 1080 | resched_task(rq->curr); |
45c01e82 GH |
1081 | return; |
1082 | } | |
1083 | ||
1084 | #ifdef CONFIG_SMP | |
1085 | /* | |
1086 | * If: | |
1087 | * | |
1088 | * - the newly woken task is of equal priority to the current task | |
1089 | * - the newly woken task is non-migratable while current is migratable | |
1090 | * - current will be preempted on the next reschedule | |
1091 | * | |
1092 | * we should check to see if current can readily move to a different | |
1093 | * cpu. If so, we will reschedule to allow the push logic to try | |
1094 | * to move current somewhere else, making room for our non-migratable | |
1095 | * task. | |
1096 | */ | |
8dd0de8b | 1097 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
7ebefa8c | 1098 | check_preempt_equal_prio(rq, p); |
45c01e82 | 1099 | #endif |
bb44e5d1 IM |
1100 | } |
1101 | ||
6f505b16 PZ |
1102 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1103 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1104 | { |
6f505b16 PZ |
1105 | struct rt_prio_array *array = &rt_rq->active; |
1106 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1107 | struct list_head *queue; |
1108 | int idx; | |
1109 | ||
1110 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1111 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1112 | |
1113 | queue = array->queue + idx; | |
6f505b16 | 1114 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1115 | |
6f505b16 PZ |
1116 | return next; |
1117 | } | |
bb44e5d1 | 1118 | |
917b627d | 1119 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1120 | { |
1121 | struct sched_rt_entity *rt_se; | |
1122 | struct task_struct *p; | |
1123 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1124 | |
6f505b16 PZ |
1125 | rt_rq = &rq->rt; |
1126 | ||
8e54a2c0 | 1127 | if (!rt_rq->rt_nr_running) |
6f505b16 PZ |
1128 | return NULL; |
1129 | ||
23b0fdfc | 1130 | if (rt_rq_throttled(rt_rq)) |
6f505b16 PZ |
1131 | return NULL; |
1132 | ||
1133 | do { | |
1134 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1135 | BUG_ON(!rt_se); |
6f505b16 PZ |
1136 | rt_rq = group_rt_rq(rt_se); |
1137 | } while (rt_rq); | |
1138 | ||
1139 | p = rt_task_of(rt_se); | |
305e6835 | 1140 | p->se.exec_start = rq->clock_task; |
917b627d GH |
1141 | |
1142 | return p; | |
1143 | } | |
1144 | ||
1145 | static struct task_struct *pick_next_task_rt(struct rq *rq) | |
1146 | { | |
1147 | struct task_struct *p = _pick_next_task_rt(rq); | |
1148 | ||
1149 | /* The running task is never eligible for pushing */ | |
1150 | if (p) | |
1151 | dequeue_pushable_task(rq, p); | |
1152 | ||
bcf08df3 | 1153 | #ifdef CONFIG_SMP |
3f029d3c GH |
1154 | /* |
1155 | * We detect this state here so that we can avoid taking the RQ | |
1156 | * lock again later if there is no need to push | |
1157 | */ | |
1158 | rq->post_schedule = has_pushable_tasks(rq); | |
bcf08df3 | 1159 | #endif |
3f029d3c | 1160 | |
6f505b16 | 1161 | return p; |
bb44e5d1 IM |
1162 | } |
1163 | ||
31ee529c | 1164 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1165 | { |
f1e14ef6 | 1166 | update_curr_rt(rq); |
917b627d GH |
1167 | |
1168 | /* | |
1169 | * The previous task needs to be made eligible for pushing | |
1170 | * if it is still active | |
1171 | */ | |
fd2f4419 | 1172 | if (on_rt_rq(&p->rt) && p->rt.nr_cpus_allowed > 1) |
917b627d | 1173 | enqueue_pushable_task(rq, p); |
bb44e5d1 IM |
1174 | } |
1175 | ||
681f3e68 | 1176 | #ifdef CONFIG_SMP |
6f505b16 | 1177 | |
e8fa1362 SR |
1178 | /* Only try algorithms three times */ |
1179 | #define RT_MAX_TRIES 3 | |
1180 | ||
e8fa1362 SR |
1181 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); |
1182 | ||
f65eda4f SR |
1183 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1184 | { | |
1185 | if (!task_running(rq, p) && | |
fa17b507 | 1186 | (cpu < 0 || cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) && |
6f505b16 | 1187 | (p->rt.nr_cpus_allowed > 1)) |
f65eda4f SR |
1188 | return 1; |
1189 | return 0; | |
1190 | } | |
1191 | ||
e8fa1362 | 1192 | /* Return the second highest RT task, NULL otherwise */ |
79064fbf | 1193 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
e8fa1362 | 1194 | { |
6f505b16 PZ |
1195 | struct task_struct *next = NULL; |
1196 | struct sched_rt_entity *rt_se; | |
1197 | struct rt_prio_array *array; | |
1198 | struct rt_rq *rt_rq; | |
e8fa1362 SR |
1199 | int idx; |
1200 | ||
6f505b16 PZ |
1201 | for_each_leaf_rt_rq(rt_rq, rq) { |
1202 | array = &rt_rq->active; | |
1203 | idx = sched_find_first_bit(array->bitmap); | |
49246274 | 1204 | next_idx: |
6f505b16 PZ |
1205 | if (idx >= MAX_RT_PRIO) |
1206 | continue; | |
1207 | if (next && next->prio < idx) | |
1208 | continue; | |
1209 | list_for_each_entry(rt_se, array->queue + idx, run_list) { | |
3d07467b PZ |
1210 | struct task_struct *p; |
1211 | ||
1212 | if (!rt_entity_is_task(rt_se)) | |
1213 | continue; | |
1214 | ||
1215 | p = rt_task_of(rt_se); | |
6f505b16 PZ |
1216 | if (pick_rt_task(rq, p, cpu)) { |
1217 | next = p; | |
1218 | break; | |
1219 | } | |
1220 | } | |
1221 | if (!next) { | |
1222 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); | |
1223 | goto next_idx; | |
1224 | } | |
f65eda4f SR |
1225 | } |
1226 | ||
e8fa1362 SR |
1227 | return next; |
1228 | } | |
1229 | ||
0e3900e6 | 1230 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1231 | |
6e1254d2 GH |
1232 | static int find_lowest_rq(struct task_struct *task) |
1233 | { | |
1234 | struct sched_domain *sd; | |
96f874e2 | 1235 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
6e1254d2 GH |
1236 | int this_cpu = smp_processor_id(); |
1237 | int cpu = task_cpu(task); | |
06f90dbd | 1238 | |
0da938c4 SR |
1239 | /* Make sure the mask is initialized first */ |
1240 | if (unlikely(!lowest_mask)) | |
1241 | return -1; | |
1242 | ||
6e0534f2 GH |
1243 | if (task->rt.nr_cpus_allowed == 1) |
1244 | return -1; /* No other targets possible */ | |
6e1254d2 | 1245 | |
6e0534f2 GH |
1246 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1247 | return -1; /* No targets found */ | |
6e1254d2 GH |
1248 | |
1249 | /* | |
1250 | * At this point we have built a mask of cpus representing the | |
1251 | * lowest priority tasks in the system. Now we want to elect | |
1252 | * the best one based on our affinity and topology. | |
1253 | * | |
1254 | * We prioritize the last cpu that the task executed on since | |
1255 | * it is most likely cache-hot in that location. | |
1256 | */ | |
96f874e2 | 1257 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1258 | return cpu; |
1259 | ||
1260 | /* | |
1261 | * Otherwise, we consult the sched_domains span maps to figure | |
1262 | * out which cpu is logically closest to our hot cache data. | |
1263 | */ | |
e2c88063 RR |
1264 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1265 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ | |
6e1254d2 | 1266 | |
cd4ae6ad | 1267 | rcu_read_lock(); |
e2c88063 RR |
1268 | for_each_domain(cpu, sd) { |
1269 | if (sd->flags & SD_WAKE_AFFINE) { | |
1270 | int best_cpu; | |
6e1254d2 | 1271 | |
e2c88063 RR |
1272 | /* |
1273 | * "this_cpu" is cheaper to preempt than a | |
1274 | * remote processor. | |
1275 | */ | |
1276 | if (this_cpu != -1 && | |
cd4ae6ad XF |
1277 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
1278 | rcu_read_unlock(); | |
e2c88063 | 1279 | return this_cpu; |
cd4ae6ad | 1280 | } |
e2c88063 RR |
1281 | |
1282 | best_cpu = cpumask_first_and(lowest_mask, | |
1283 | sched_domain_span(sd)); | |
cd4ae6ad XF |
1284 | if (best_cpu < nr_cpu_ids) { |
1285 | rcu_read_unlock(); | |
e2c88063 | 1286 | return best_cpu; |
cd4ae6ad | 1287 | } |
6e1254d2 GH |
1288 | } |
1289 | } | |
cd4ae6ad | 1290 | rcu_read_unlock(); |
6e1254d2 GH |
1291 | |
1292 | /* | |
1293 | * And finally, if there were no matches within the domains | |
1294 | * just give the caller *something* to work with from the compatible | |
1295 | * locations. | |
1296 | */ | |
e2c88063 RR |
1297 | if (this_cpu != -1) |
1298 | return this_cpu; | |
1299 | ||
1300 | cpu = cpumask_any(lowest_mask); | |
1301 | if (cpu < nr_cpu_ids) | |
1302 | return cpu; | |
1303 | return -1; | |
07b4032c GH |
1304 | } |
1305 | ||
1306 | /* Will lock the rq it finds */ | |
4df64c0b | 1307 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1308 | { |
1309 | struct rq *lowest_rq = NULL; | |
07b4032c | 1310 | int tries; |
4df64c0b | 1311 | int cpu; |
e8fa1362 | 1312 | |
07b4032c GH |
1313 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1314 | cpu = find_lowest_rq(task); | |
1315 | ||
2de0b463 | 1316 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1317 | break; |
1318 | ||
07b4032c GH |
1319 | lowest_rq = cpu_rq(cpu); |
1320 | ||
e8fa1362 | 1321 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1322 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1323 | /* |
1324 | * We had to unlock the run queue. In | |
1325 | * the mean time, task could have | |
1326 | * migrated already or had its affinity changed. | |
1327 | * Also make sure that it wasn't scheduled on its rq. | |
1328 | */ | |
07b4032c | 1329 | if (unlikely(task_rq(task) != rq || |
96f874e2 | 1330 | !cpumask_test_cpu(lowest_rq->cpu, |
fa17b507 | 1331 | tsk_cpus_allowed(task)) || |
07b4032c | 1332 | task_running(rq, task) || |
fd2f4419 | 1333 | !task->on_rq)) { |
4df64c0b | 1334 | |
05fa785c | 1335 | raw_spin_unlock(&lowest_rq->lock); |
e8fa1362 SR |
1336 | lowest_rq = NULL; |
1337 | break; | |
1338 | } | |
1339 | } | |
1340 | ||
1341 | /* If this rq is still suitable use it. */ | |
e864c499 | 1342 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1343 | break; |
1344 | ||
1345 | /* try again */ | |
1b12bbc7 | 1346 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1347 | lowest_rq = NULL; |
1348 | } | |
1349 | ||
1350 | return lowest_rq; | |
1351 | } | |
1352 | ||
917b627d GH |
1353 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1354 | { | |
1355 | struct task_struct *p; | |
1356 | ||
1357 | if (!has_pushable_tasks(rq)) | |
1358 | return NULL; | |
1359 | ||
1360 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1361 | struct task_struct, pushable_tasks); | |
1362 | ||
1363 | BUG_ON(rq->cpu != task_cpu(p)); | |
1364 | BUG_ON(task_current(rq, p)); | |
1365 | BUG_ON(p->rt.nr_cpus_allowed <= 1); | |
1366 | ||
fd2f4419 | 1367 | BUG_ON(!p->on_rq); |
917b627d GH |
1368 | BUG_ON(!rt_task(p)); |
1369 | ||
1370 | return p; | |
1371 | } | |
1372 | ||
e8fa1362 SR |
1373 | /* |
1374 | * If the current CPU has more than one RT task, see if the non | |
1375 | * running task can migrate over to a CPU that is running a task | |
1376 | * of lesser priority. | |
1377 | */ | |
697f0a48 | 1378 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1379 | { |
1380 | struct task_struct *next_task; | |
1381 | struct rq *lowest_rq; | |
311e800e | 1382 | int ret = 0; |
e8fa1362 | 1383 | |
a22d7fc1 GH |
1384 | if (!rq->rt.overloaded) |
1385 | return 0; | |
1386 | ||
917b627d | 1387 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1388 | if (!next_task) |
1389 | return 0; | |
1390 | ||
49246274 | 1391 | retry: |
697f0a48 | 1392 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1393 | WARN_ON(1); |
e8fa1362 | 1394 | return 0; |
f65eda4f | 1395 | } |
e8fa1362 SR |
1396 | |
1397 | /* | |
1398 | * It's possible that the next_task slipped in of | |
1399 | * higher priority than current. If that's the case | |
1400 | * just reschedule current. | |
1401 | */ | |
697f0a48 GH |
1402 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1403 | resched_task(rq->curr); | |
e8fa1362 SR |
1404 | return 0; |
1405 | } | |
1406 | ||
697f0a48 | 1407 | /* We might release rq lock */ |
e8fa1362 SR |
1408 | get_task_struct(next_task); |
1409 | ||
1410 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1411 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1412 | if (!lowest_rq) { |
1413 | struct task_struct *task; | |
1414 | /* | |
311e800e | 1415 | * find_lock_lowest_rq releases rq->lock |
1563513d GH |
1416 | * so it is possible that next_task has migrated. |
1417 | * | |
1418 | * We need to make sure that the task is still on the same | |
1419 | * run-queue and is also still the next task eligible for | |
1420 | * pushing. | |
e8fa1362 | 1421 | */ |
917b627d | 1422 | task = pick_next_pushable_task(rq); |
1563513d GH |
1423 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1424 | /* | |
311e800e HD |
1425 | * The task hasn't migrated, and is still the next |
1426 | * eligible task, but we failed to find a run-queue | |
1427 | * to push it to. Do not retry in this case, since | |
1428 | * other cpus will pull from us when ready. | |
1563513d | 1429 | */ |
1563513d | 1430 | goto out; |
e8fa1362 | 1431 | } |
917b627d | 1432 | |
1563513d GH |
1433 | if (!task) |
1434 | /* No more tasks, just exit */ | |
1435 | goto out; | |
1436 | ||
917b627d | 1437 | /* |
1563513d | 1438 | * Something has shifted, try again. |
917b627d | 1439 | */ |
1563513d GH |
1440 | put_task_struct(next_task); |
1441 | next_task = task; | |
1442 | goto retry; | |
e8fa1362 SR |
1443 | } |
1444 | ||
697f0a48 | 1445 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1446 | set_task_cpu(next_task, lowest_rq->cpu); |
1447 | activate_task(lowest_rq, next_task, 0); | |
311e800e | 1448 | ret = 1; |
e8fa1362 SR |
1449 | |
1450 | resched_task(lowest_rq->curr); | |
1451 | ||
1b12bbc7 | 1452 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1453 | |
e8fa1362 SR |
1454 | out: |
1455 | put_task_struct(next_task); | |
1456 | ||
311e800e | 1457 | return ret; |
e8fa1362 SR |
1458 | } |
1459 | ||
e8fa1362 SR |
1460 | static void push_rt_tasks(struct rq *rq) |
1461 | { | |
1462 | /* push_rt_task will return true if it moved an RT */ | |
1463 | while (push_rt_task(rq)) | |
1464 | ; | |
1465 | } | |
1466 | ||
f65eda4f SR |
1467 | static int pull_rt_task(struct rq *this_rq) |
1468 | { | |
80bf3171 | 1469 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
a8728944 | 1470 | struct task_struct *p; |
f65eda4f | 1471 | struct rq *src_rq; |
f65eda4f | 1472 | |
637f5085 | 1473 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
1474 | return 0; |
1475 | ||
c6c4927b | 1476 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1477 | if (this_cpu == cpu) |
1478 | continue; | |
1479 | ||
1480 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1481 | |
1482 | /* | |
1483 | * Don't bother taking the src_rq->lock if the next highest | |
1484 | * task is known to be lower-priority than our current task. | |
1485 | * This may look racy, but if this value is about to go | |
1486 | * logically higher, the src_rq will push this task away. | |
1487 | * And if its going logically lower, we do not care | |
1488 | */ | |
1489 | if (src_rq->rt.highest_prio.next >= | |
1490 | this_rq->rt.highest_prio.curr) | |
1491 | continue; | |
1492 | ||
f65eda4f SR |
1493 | /* |
1494 | * We can potentially drop this_rq's lock in | |
1495 | * double_lock_balance, and another CPU could | |
a8728944 | 1496 | * alter this_rq |
f65eda4f | 1497 | */ |
a8728944 | 1498 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
1499 | |
1500 | /* | |
1501 | * Are there still pullable RT tasks? | |
1502 | */ | |
614ee1f6 MG |
1503 | if (src_rq->rt.rt_nr_running <= 1) |
1504 | goto skip; | |
f65eda4f | 1505 | |
f65eda4f SR |
1506 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
1507 | ||
1508 | /* | |
1509 | * Do we have an RT task that preempts | |
1510 | * the to-be-scheduled task? | |
1511 | */ | |
a8728944 | 1512 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f | 1513 | WARN_ON(p == src_rq->curr); |
fd2f4419 | 1514 | WARN_ON(!p->on_rq); |
f65eda4f SR |
1515 | |
1516 | /* | |
1517 | * There's a chance that p is higher in priority | |
1518 | * than what's currently running on its cpu. | |
1519 | * This is just that p is wakeing up and hasn't | |
1520 | * had a chance to schedule. We only pull | |
1521 | * p if it is lower in priority than the | |
a8728944 | 1522 | * current task on the run queue |
f65eda4f | 1523 | */ |
a8728944 | 1524 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 1525 | goto skip; |
f65eda4f SR |
1526 | |
1527 | ret = 1; | |
1528 | ||
1529 | deactivate_task(src_rq, p, 0); | |
1530 | set_task_cpu(p, this_cpu); | |
1531 | activate_task(this_rq, p, 0); | |
1532 | /* | |
1533 | * We continue with the search, just in | |
1534 | * case there's an even higher prio task | |
25985edc | 1535 | * in another runqueue. (low likelihood |
f65eda4f | 1536 | * but possible) |
f65eda4f | 1537 | */ |
f65eda4f | 1538 | } |
49246274 | 1539 | skip: |
1b12bbc7 | 1540 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
1541 | } |
1542 | ||
1543 | return ret; | |
1544 | } | |
1545 | ||
9a897c5a | 1546 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
f65eda4f SR |
1547 | { |
1548 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
33c3d6c6 | 1549 | if (rq->rt.highest_prio.curr > prev->prio) |
f65eda4f SR |
1550 | pull_rt_task(rq); |
1551 | } | |
1552 | ||
9a897c5a | 1553 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 | 1554 | { |
967fc046 | 1555 | push_rt_tasks(rq); |
e8fa1362 SR |
1556 | } |
1557 | ||
8ae121ac GH |
1558 | /* |
1559 | * If we are not running and we are not going to reschedule soon, we should | |
1560 | * try to push tasks away now | |
1561 | */ | |
efbbd05a | 1562 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1563 | { |
9a897c5a | 1564 | if (!task_running(rq, p) && |
8ae121ac | 1565 | !test_tsk_need_resched(rq->curr) && |
917b627d | 1566 | has_pushable_tasks(rq) && |
b3bc211c | 1567 | p->rt.nr_cpus_allowed > 1 && |
43fa5460 | 1568 | rt_task(rq->curr) && |
b3bc211c | 1569 | (rq->curr->rt.nr_cpus_allowed < 2 || |
3be209a8 | 1570 | rq->curr->prio <= p->prio)) |
4642dafd SR |
1571 | push_rt_tasks(rq); |
1572 | } | |
1573 | ||
cd8ba7cd | 1574 | static void set_cpus_allowed_rt(struct task_struct *p, |
96f874e2 | 1575 | const struct cpumask *new_mask) |
73fe6aae | 1576 | { |
96f874e2 | 1577 | int weight = cpumask_weight(new_mask); |
73fe6aae GH |
1578 | |
1579 | BUG_ON(!rt_task(p)); | |
1580 | ||
1581 | /* | |
1582 | * Update the migration status of the RQ if we have an RT task | |
1583 | * which is running AND changing its weight value. | |
1584 | */ | |
fd2f4419 | 1585 | if (p->on_rq && (weight != p->rt.nr_cpus_allowed)) { |
73fe6aae GH |
1586 | struct rq *rq = task_rq(p); |
1587 | ||
917b627d GH |
1588 | if (!task_current(rq, p)) { |
1589 | /* | |
1590 | * Make sure we dequeue this task from the pushable list | |
1591 | * before going further. It will either remain off of | |
1592 | * the list because we are no longer pushable, or it | |
1593 | * will be requeued. | |
1594 | */ | |
1595 | if (p->rt.nr_cpus_allowed > 1) | |
1596 | dequeue_pushable_task(rq, p); | |
1597 | ||
1598 | /* | |
1599 | * Requeue if our weight is changing and still > 1 | |
1600 | */ | |
1601 | if (weight > 1) | |
1602 | enqueue_pushable_task(rq, p); | |
1603 | ||
1604 | } | |
1605 | ||
6f505b16 | 1606 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
73fe6aae | 1607 | rq->rt.rt_nr_migratory++; |
6f505b16 | 1608 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
73fe6aae GH |
1609 | BUG_ON(!rq->rt.rt_nr_migratory); |
1610 | rq->rt.rt_nr_migratory--; | |
1611 | } | |
1612 | ||
398a153b | 1613 | update_rt_migration(&rq->rt); |
73fe6aae | 1614 | } |
73fe6aae | 1615 | } |
deeeccd4 | 1616 | |
bdd7c81b | 1617 | /* Assumes rq->lock is held */ |
1f11eb6a | 1618 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
1619 | { |
1620 | if (rq->rt.overloaded) | |
1621 | rt_set_overload(rq); | |
6e0534f2 | 1622 | |
7def2be1 PZ |
1623 | __enable_runtime(rq); |
1624 | ||
e864c499 | 1625 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
1626 | } |
1627 | ||
1628 | /* Assumes rq->lock is held */ | |
1f11eb6a | 1629 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
1630 | { |
1631 | if (rq->rt.overloaded) | |
1632 | rt_clear_overload(rq); | |
6e0534f2 | 1633 | |
7def2be1 PZ |
1634 | __disable_runtime(rq); |
1635 | ||
6e0534f2 | 1636 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 1637 | } |
cb469845 SR |
1638 | |
1639 | /* | |
1640 | * When switch from the rt queue, we bring ourselves to a position | |
1641 | * that we might want to pull RT tasks from other runqueues. | |
1642 | */ | |
da7a735e | 1643 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1644 | { |
1645 | /* | |
1646 | * If there are other RT tasks then we will reschedule | |
1647 | * and the scheduling of the other RT tasks will handle | |
1648 | * the balancing. But if we are the last RT task | |
1649 | * we may need to handle the pulling of RT tasks | |
1650 | * now. | |
1651 | */ | |
fd2f4419 | 1652 | if (p->on_rq && !rq->rt.rt_nr_running) |
cb469845 SR |
1653 | pull_rt_task(rq); |
1654 | } | |
3d8cbdf8 RR |
1655 | |
1656 | static inline void init_sched_rt_class(void) | |
1657 | { | |
1658 | unsigned int i; | |
1659 | ||
1660 | for_each_possible_cpu(i) | |
eaa95840 | 1661 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 1662 | GFP_KERNEL, cpu_to_node(i)); |
3d8cbdf8 | 1663 | } |
cb469845 SR |
1664 | #endif /* CONFIG_SMP */ |
1665 | ||
1666 | /* | |
1667 | * When switching a task to RT, we may overload the runqueue | |
1668 | * with RT tasks. In this case we try to push them off to | |
1669 | * other runqueues. | |
1670 | */ | |
da7a735e | 1671 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1672 | { |
1673 | int check_resched = 1; | |
1674 | ||
1675 | /* | |
1676 | * If we are already running, then there's nothing | |
1677 | * that needs to be done. But if we are not running | |
1678 | * we may need to preempt the current running task. | |
1679 | * If that current running task is also an RT task | |
1680 | * then see if we can move to another run queue. | |
1681 | */ | |
fd2f4419 | 1682 | if (p->on_rq && rq->curr != p) { |
cb469845 SR |
1683 | #ifdef CONFIG_SMP |
1684 | if (rq->rt.overloaded && push_rt_task(rq) && | |
1685 | /* Don't resched if we changed runqueues */ | |
1686 | rq != task_rq(p)) | |
1687 | check_resched = 0; | |
1688 | #endif /* CONFIG_SMP */ | |
1689 | if (check_resched && p->prio < rq->curr->prio) | |
1690 | resched_task(rq->curr); | |
1691 | } | |
1692 | } | |
1693 | ||
1694 | /* | |
1695 | * Priority of the task has changed. This may cause | |
1696 | * us to initiate a push or pull. | |
1697 | */ | |
da7a735e PZ |
1698 | static void |
1699 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) | |
cb469845 | 1700 | { |
fd2f4419 | 1701 | if (!p->on_rq) |
da7a735e PZ |
1702 | return; |
1703 | ||
1704 | if (rq->curr == p) { | |
cb469845 SR |
1705 | #ifdef CONFIG_SMP |
1706 | /* | |
1707 | * If our priority decreases while running, we | |
1708 | * may need to pull tasks to this runqueue. | |
1709 | */ | |
1710 | if (oldprio < p->prio) | |
1711 | pull_rt_task(rq); | |
1712 | /* | |
1713 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
1714 | * then reschedule. Note, the above pull_rt_task |
1715 | * can release the rq lock and p could migrate. | |
1716 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 1717 | */ |
e864c499 | 1718 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
cb469845 SR |
1719 | resched_task(p); |
1720 | #else | |
1721 | /* For UP simply resched on drop of prio */ | |
1722 | if (oldprio < p->prio) | |
1723 | resched_task(p); | |
e8fa1362 | 1724 | #endif /* CONFIG_SMP */ |
cb469845 SR |
1725 | } else { |
1726 | /* | |
1727 | * This task is not running, but if it is | |
1728 | * greater than the current running task | |
1729 | * then reschedule. | |
1730 | */ | |
1731 | if (p->prio < rq->curr->prio) | |
1732 | resched_task(rq->curr); | |
1733 | } | |
1734 | } | |
1735 | ||
78f2c7db PZ |
1736 | static void watchdog(struct rq *rq, struct task_struct *p) |
1737 | { | |
1738 | unsigned long soft, hard; | |
1739 | ||
78d7d407 JS |
1740 | /* max may change after cur was read, this will be fixed next tick */ |
1741 | soft = task_rlimit(p, RLIMIT_RTTIME); | |
1742 | hard = task_rlimit_max(p, RLIMIT_RTTIME); | |
78f2c7db PZ |
1743 | |
1744 | if (soft != RLIM_INFINITY) { | |
1745 | unsigned long next; | |
1746 | ||
1747 | p->rt.timeout++; | |
1748 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); | |
5a52dd50 | 1749 | if (p->rt.timeout > next) |
f06febc9 | 1750 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
1751 | } |
1752 | } | |
bb44e5d1 | 1753 | |
8f4d37ec | 1754 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 1755 | { |
67e2be02 PZ |
1756 | update_curr_rt(rq); |
1757 | ||
78f2c7db PZ |
1758 | watchdog(rq, p); |
1759 | ||
bb44e5d1 IM |
1760 | /* |
1761 | * RR tasks need a special form of timeslice management. | |
1762 | * FIFO tasks have no timeslices. | |
1763 | */ | |
1764 | if (p->policy != SCHED_RR) | |
1765 | return; | |
1766 | ||
fa717060 | 1767 | if (--p->rt.time_slice) |
bb44e5d1 IM |
1768 | return; |
1769 | ||
fa717060 | 1770 | p->rt.time_slice = DEF_TIMESLICE; |
bb44e5d1 | 1771 | |
98fbc798 DA |
1772 | /* |
1773 | * Requeue to the end of queue if we are not the only element | |
1774 | * on the queue: | |
1775 | */ | |
fa717060 | 1776 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
7ebefa8c | 1777 | requeue_task_rt(rq, p, 0); |
98fbc798 DA |
1778 | set_tsk_need_resched(p); |
1779 | } | |
bb44e5d1 IM |
1780 | } |
1781 | ||
83b699ed SV |
1782 | static void set_curr_task_rt(struct rq *rq) |
1783 | { | |
1784 | struct task_struct *p = rq->curr; | |
1785 | ||
305e6835 | 1786 | p->se.exec_start = rq->clock_task; |
917b627d GH |
1787 | |
1788 | /* The running task is never eligible for pushing */ | |
1789 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
1790 | } |
1791 | ||
6d686f45 | 1792 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
0d721cea PW |
1793 | { |
1794 | /* | |
1795 | * Time slice is 0 for SCHED_FIFO tasks | |
1796 | */ | |
1797 | if (task->policy == SCHED_RR) | |
1798 | return DEF_TIMESLICE; | |
1799 | else | |
1800 | return 0; | |
1801 | } | |
1802 | ||
2abdad0a | 1803 | static const struct sched_class rt_sched_class = { |
5522d5d5 | 1804 | .next = &fair_sched_class, |
bb44e5d1 IM |
1805 | .enqueue_task = enqueue_task_rt, |
1806 | .dequeue_task = dequeue_task_rt, | |
1807 | .yield_task = yield_task_rt, | |
1808 | ||
1809 | .check_preempt_curr = check_preempt_curr_rt, | |
1810 | ||
1811 | .pick_next_task = pick_next_task_rt, | |
1812 | .put_prev_task = put_prev_task_rt, | |
1813 | ||
681f3e68 | 1814 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
1815 | .select_task_rq = select_task_rq_rt, |
1816 | ||
73fe6aae | 1817 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
1818 | .rq_online = rq_online_rt, |
1819 | .rq_offline = rq_offline_rt, | |
9a897c5a SR |
1820 | .pre_schedule = pre_schedule_rt, |
1821 | .post_schedule = post_schedule_rt, | |
efbbd05a | 1822 | .task_woken = task_woken_rt, |
cb469845 | 1823 | .switched_from = switched_from_rt, |
681f3e68 | 1824 | #endif |
bb44e5d1 | 1825 | |
83b699ed | 1826 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 1827 | .task_tick = task_tick_rt, |
cb469845 | 1828 | |
0d721cea PW |
1829 | .get_rr_interval = get_rr_interval_rt, |
1830 | ||
cb469845 SR |
1831 | .prio_changed = prio_changed_rt, |
1832 | .switched_to = switched_to_rt, | |
bb44e5d1 | 1833 | }; |
ada18de2 PZ |
1834 | |
1835 | #ifdef CONFIG_SCHED_DEBUG | |
1836 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
1837 | ||
1838 | static void print_rt_stats(struct seq_file *m, int cpu) | |
1839 | { | |
ec514c48 | 1840 | rt_rq_iter_t iter; |
ada18de2 PZ |
1841 | struct rt_rq *rt_rq; |
1842 | ||
1843 | rcu_read_lock(); | |
ec514c48 | 1844 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
ada18de2 PZ |
1845 | print_rt_rq(m, cpu, rt_rq); |
1846 | rcu_read_unlock(); | |
1847 | } | |
55e12e5e | 1848 | #endif /* CONFIG_SCHED_DEBUG */ |