Commit | Line | Data |
---|---|---|
bb44e5d1 IM |
1 | /* |
2 | * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR | |
3 | * policies) | |
4 | */ | |
5 | ||
029632fb PZ |
6 | #include "sched.h" |
7 | ||
8 | #include <linux/slab.h> | |
9 | ||
ce0dbbbb CW |
10 | int sched_rr_timeslice = RR_TIMESLICE; |
11 | ||
029632fb PZ |
12 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); |
13 | ||
14 | struct rt_bandwidth def_rt_bandwidth; | |
15 | ||
16 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
17 | { | |
18 | struct rt_bandwidth *rt_b = | |
19 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
20 | ktime_t now; | |
21 | int overrun; | |
22 | int idle = 0; | |
23 | ||
24 | for (;;) { | |
25 | now = hrtimer_cb_get_time(timer); | |
26 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
27 | ||
28 | if (!overrun) | |
29 | break; | |
30 | ||
31 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
32 | } | |
33 | ||
34 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
35 | } | |
36 | ||
37 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
38 | { | |
39 | rt_b->rt_period = ns_to_ktime(period); | |
40 | rt_b->rt_runtime = runtime; | |
41 | ||
42 | raw_spin_lock_init(&rt_b->rt_runtime_lock); | |
43 | ||
44 | hrtimer_init(&rt_b->rt_period_timer, | |
45 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
46 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
47 | } | |
48 | ||
49 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
50 | { | |
51 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) | |
52 | return; | |
53 | ||
54 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
55 | return; | |
56 | ||
57 | raw_spin_lock(&rt_b->rt_runtime_lock); | |
58 | start_bandwidth_timer(&rt_b->rt_period_timer, rt_b->rt_period); | |
59 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
60 | } | |
61 | ||
62 | void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) | |
63 | { | |
64 | struct rt_prio_array *array; | |
65 | int i; | |
66 | ||
67 | array = &rt_rq->active; | |
68 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
69 | INIT_LIST_HEAD(array->queue + i); | |
70 | __clear_bit(i, array->bitmap); | |
71 | } | |
72 | /* delimiter for bitsearch: */ | |
73 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
74 | ||
75 | #if defined CONFIG_SMP | |
76 | rt_rq->highest_prio.curr = MAX_RT_PRIO; | |
77 | rt_rq->highest_prio.next = MAX_RT_PRIO; | |
78 | rt_rq->rt_nr_migratory = 0; | |
79 | rt_rq->overloaded = 0; | |
80 | plist_head_init(&rt_rq->pushable_tasks); | |
81 | #endif | |
82 | ||
83 | rt_rq->rt_time = 0; | |
84 | rt_rq->rt_throttled = 0; | |
85 | rt_rq->rt_runtime = 0; | |
86 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); | |
87 | } | |
88 | ||
8f48894f | 89 | #ifdef CONFIG_RT_GROUP_SCHED |
029632fb PZ |
90 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) |
91 | { | |
92 | hrtimer_cancel(&rt_b->rt_period_timer); | |
93 | } | |
8f48894f PZ |
94 | |
95 | #define rt_entity_is_task(rt_se) (!(rt_se)->my_q) | |
96 | ||
398a153b GH |
97 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
98 | { | |
8f48894f PZ |
99 | #ifdef CONFIG_SCHED_DEBUG |
100 | WARN_ON_ONCE(!rt_entity_is_task(rt_se)); | |
101 | #endif | |
398a153b GH |
102 | return container_of(rt_se, struct task_struct, rt); |
103 | } | |
104 | ||
398a153b GH |
105 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
106 | { | |
107 | return rt_rq->rq; | |
108 | } | |
109 | ||
110 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
111 | { | |
112 | return rt_se->rt_rq; | |
113 | } | |
114 | ||
029632fb PZ |
115 | void free_rt_sched_group(struct task_group *tg) |
116 | { | |
117 | int i; | |
118 | ||
119 | if (tg->rt_se) | |
120 | destroy_rt_bandwidth(&tg->rt_bandwidth); | |
121 | ||
122 | for_each_possible_cpu(i) { | |
123 | if (tg->rt_rq) | |
124 | kfree(tg->rt_rq[i]); | |
125 | if (tg->rt_se) | |
126 | kfree(tg->rt_se[i]); | |
127 | } | |
128 | ||
129 | kfree(tg->rt_rq); | |
130 | kfree(tg->rt_se); | |
131 | } | |
132 | ||
133 | void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
134 | struct sched_rt_entity *rt_se, int cpu, | |
135 | struct sched_rt_entity *parent) | |
136 | { | |
137 | struct rq *rq = cpu_rq(cpu); | |
138 | ||
139 | rt_rq->highest_prio.curr = MAX_RT_PRIO; | |
140 | rt_rq->rt_nr_boosted = 0; | |
141 | rt_rq->rq = rq; | |
142 | rt_rq->tg = tg; | |
143 | ||
144 | tg->rt_rq[cpu] = rt_rq; | |
145 | tg->rt_se[cpu] = rt_se; | |
146 | ||
147 | if (!rt_se) | |
148 | return; | |
149 | ||
150 | if (!parent) | |
151 | rt_se->rt_rq = &rq->rt; | |
152 | else | |
153 | rt_se->rt_rq = parent->my_q; | |
154 | ||
155 | rt_se->my_q = rt_rq; | |
156 | rt_se->parent = parent; | |
157 | INIT_LIST_HEAD(&rt_se->run_list); | |
158 | } | |
159 | ||
160 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
161 | { | |
162 | struct rt_rq *rt_rq; | |
163 | struct sched_rt_entity *rt_se; | |
164 | int i; | |
165 | ||
166 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); | |
167 | if (!tg->rt_rq) | |
168 | goto err; | |
169 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); | |
170 | if (!tg->rt_se) | |
171 | goto err; | |
172 | ||
173 | init_rt_bandwidth(&tg->rt_bandwidth, | |
174 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
175 | ||
176 | for_each_possible_cpu(i) { | |
177 | rt_rq = kzalloc_node(sizeof(struct rt_rq), | |
178 | GFP_KERNEL, cpu_to_node(i)); | |
179 | if (!rt_rq) | |
180 | goto err; | |
181 | ||
182 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), | |
183 | GFP_KERNEL, cpu_to_node(i)); | |
184 | if (!rt_se) | |
185 | goto err_free_rq; | |
186 | ||
187 | init_rt_rq(rt_rq, cpu_rq(i)); | |
188 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; | |
189 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); | |
190 | } | |
191 | ||
192 | return 1; | |
193 | ||
194 | err_free_rq: | |
195 | kfree(rt_rq); | |
196 | err: | |
197 | return 0; | |
198 | } | |
199 | ||
398a153b GH |
200 | #else /* CONFIG_RT_GROUP_SCHED */ |
201 | ||
a1ba4d8b PZ |
202 | #define rt_entity_is_task(rt_se) (1) |
203 | ||
8f48894f PZ |
204 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
205 | { | |
206 | return container_of(rt_se, struct task_struct, rt); | |
207 | } | |
208 | ||
398a153b GH |
209 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
210 | { | |
211 | return container_of(rt_rq, struct rq, rt); | |
212 | } | |
213 | ||
214 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
215 | { | |
216 | struct task_struct *p = rt_task_of(rt_se); | |
217 | struct rq *rq = task_rq(p); | |
218 | ||
219 | return &rq->rt; | |
220 | } | |
221 | ||
029632fb PZ |
222 | void free_rt_sched_group(struct task_group *tg) { } |
223 | ||
224 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
225 | { | |
226 | return 1; | |
227 | } | |
398a153b GH |
228 | #endif /* CONFIG_RT_GROUP_SCHED */ |
229 | ||
4fd29176 | 230 | #ifdef CONFIG_SMP |
84de4274 | 231 | |
38033c37 PZ |
232 | static int pull_rt_task(struct rq *this_rq); |
233 | ||
dc877341 PZ |
234 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
235 | { | |
236 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
237 | return rq->rt.highest_prio.curr > prev->prio; | |
238 | } | |
239 | ||
637f5085 | 240 | static inline int rt_overloaded(struct rq *rq) |
4fd29176 | 241 | { |
637f5085 | 242 | return atomic_read(&rq->rd->rto_count); |
4fd29176 | 243 | } |
84de4274 | 244 | |
4fd29176 SR |
245 | static inline void rt_set_overload(struct rq *rq) |
246 | { | |
1f11eb6a GH |
247 | if (!rq->online) |
248 | return; | |
249 | ||
c6c4927b | 250 | cpumask_set_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 SR |
251 | /* |
252 | * Make sure the mask is visible before we set | |
253 | * the overload count. That is checked to determine | |
254 | * if we should look at the mask. It would be a shame | |
255 | * if we looked at the mask, but the mask was not | |
256 | * updated yet. | |
7c3f2ab7 PZ |
257 | * |
258 | * Matched by the barrier in pull_rt_task(). | |
4fd29176 | 259 | */ |
7c3f2ab7 | 260 | smp_wmb(); |
637f5085 | 261 | atomic_inc(&rq->rd->rto_count); |
4fd29176 | 262 | } |
84de4274 | 263 | |
4fd29176 SR |
264 | static inline void rt_clear_overload(struct rq *rq) |
265 | { | |
1f11eb6a GH |
266 | if (!rq->online) |
267 | return; | |
268 | ||
4fd29176 | 269 | /* the order here really doesn't matter */ |
637f5085 | 270 | atomic_dec(&rq->rd->rto_count); |
c6c4927b | 271 | cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 | 272 | } |
73fe6aae | 273 | |
398a153b | 274 | static void update_rt_migration(struct rt_rq *rt_rq) |
73fe6aae | 275 | { |
a1ba4d8b | 276 | if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) { |
398a153b GH |
277 | if (!rt_rq->overloaded) { |
278 | rt_set_overload(rq_of_rt_rq(rt_rq)); | |
279 | rt_rq->overloaded = 1; | |
cdc8eb98 | 280 | } |
398a153b GH |
281 | } else if (rt_rq->overloaded) { |
282 | rt_clear_overload(rq_of_rt_rq(rt_rq)); | |
283 | rt_rq->overloaded = 0; | |
637f5085 | 284 | } |
73fe6aae | 285 | } |
4fd29176 | 286 | |
398a153b GH |
287 | static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
288 | { | |
29baa747 PZ |
289 | struct task_struct *p; |
290 | ||
a1ba4d8b PZ |
291 | if (!rt_entity_is_task(rt_se)) |
292 | return; | |
293 | ||
29baa747 | 294 | p = rt_task_of(rt_se); |
a1ba4d8b PZ |
295 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
296 | ||
297 | rt_rq->rt_nr_total++; | |
29baa747 | 298 | if (p->nr_cpus_allowed > 1) |
398a153b GH |
299 | rt_rq->rt_nr_migratory++; |
300 | ||
301 | update_rt_migration(rt_rq); | |
302 | } | |
303 | ||
304 | static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
305 | { | |
29baa747 PZ |
306 | struct task_struct *p; |
307 | ||
a1ba4d8b PZ |
308 | if (!rt_entity_is_task(rt_se)) |
309 | return; | |
310 | ||
29baa747 | 311 | p = rt_task_of(rt_se); |
a1ba4d8b PZ |
312 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
313 | ||
314 | rt_rq->rt_nr_total--; | |
29baa747 | 315 | if (p->nr_cpus_allowed > 1) |
398a153b GH |
316 | rt_rq->rt_nr_migratory--; |
317 | ||
318 | update_rt_migration(rt_rq); | |
319 | } | |
320 | ||
5181f4a4 SR |
321 | static inline int has_pushable_tasks(struct rq *rq) |
322 | { | |
323 | return !plist_head_empty(&rq->rt.pushable_tasks); | |
324 | } | |
325 | ||
dc877341 PZ |
326 | static inline void set_post_schedule(struct rq *rq) |
327 | { | |
328 | /* | |
329 | * We detect this state here so that we can avoid taking the RQ | |
330 | * lock again later if there is no need to push | |
331 | */ | |
332 | rq->post_schedule = has_pushable_tasks(rq); | |
333 | } | |
334 | ||
917b627d GH |
335 | static void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
336 | { | |
337 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
338 | plist_node_init(&p->pushable_tasks, p->prio); | |
339 | plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
5181f4a4 SR |
340 | |
341 | /* Update the highest prio pushable task */ | |
342 | if (p->prio < rq->rt.highest_prio.next) | |
343 | rq->rt.highest_prio.next = p->prio; | |
917b627d GH |
344 | } |
345 | ||
346 | static void dequeue_pushable_task(struct rq *rq, struct task_struct *p) | |
347 | { | |
348 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
917b627d | 349 | |
5181f4a4 SR |
350 | /* Update the new highest prio pushable task */ |
351 | if (has_pushable_tasks(rq)) { | |
352 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
353 | struct task_struct, pushable_tasks); | |
354 | rq->rt.highest_prio.next = p->prio; | |
355 | } else | |
356 | rq->rt.highest_prio.next = MAX_RT_PRIO; | |
bcf08df3 IM |
357 | } |
358 | ||
917b627d GH |
359 | #else |
360 | ||
ceacc2c1 | 361 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
fa85ae24 | 362 | { |
6f505b16 PZ |
363 | } |
364 | ||
ceacc2c1 PZ |
365 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
366 | { | |
367 | } | |
368 | ||
b07430ac | 369 | static inline |
ceacc2c1 PZ |
370 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
371 | { | |
372 | } | |
373 | ||
398a153b | 374 | static inline |
ceacc2c1 PZ |
375 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
376 | { | |
377 | } | |
917b627d | 378 | |
dc877341 PZ |
379 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
380 | { | |
381 | return false; | |
382 | } | |
383 | ||
384 | static inline int pull_rt_task(struct rq *this_rq) | |
385 | { | |
386 | return 0; | |
387 | } | |
388 | ||
389 | static inline void set_post_schedule(struct rq *rq) | |
390 | { | |
391 | } | |
4fd29176 SR |
392 | #endif /* CONFIG_SMP */ |
393 | ||
6f505b16 PZ |
394 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
395 | { | |
396 | return !list_empty(&rt_se->run_list); | |
397 | } | |
398 | ||
052f1dc7 | 399 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 400 | |
9f0c1e56 | 401 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
402 | { |
403 | if (!rt_rq->tg) | |
9f0c1e56 | 404 | return RUNTIME_INF; |
6f505b16 | 405 | |
ac086bc2 PZ |
406 | return rt_rq->rt_runtime; |
407 | } | |
408 | ||
409 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
410 | { | |
411 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); | |
6f505b16 PZ |
412 | } |
413 | ||
ec514c48 CX |
414 | typedef struct task_group *rt_rq_iter_t; |
415 | ||
1c09ab0d YZ |
416 | static inline struct task_group *next_task_group(struct task_group *tg) |
417 | { | |
418 | do { | |
419 | tg = list_entry_rcu(tg->list.next, | |
420 | typeof(struct task_group), list); | |
421 | } while (&tg->list != &task_groups && task_group_is_autogroup(tg)); | |
422 | ||
423 | if (&tg->list == &task_groups) | |
424 | tg = NULL; | |
425 | ||
426 | return tg; | |
427 | } | |
428 | ||
429 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
430 | for (iter = container_of(&task_groups, typeof(*iter), list); \ | |
431 | (iter = next_task_group(iter)) && \ | |
432 | (rt_rq = iter->rt_rq[cpu_of(rq)]);) | |
ec514c48 | 433 | |
6f505b16 PZ |
434 | #define for_each_sched_rt_entity(rt_se) \ |
435 | for (; rt_se; rt_se = rt_se->parent) | |
436 | ||
437 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
438 | { | |
439 | return rt_se->my_q; | |
440 | } | |
441 | ||
37dad3fc | 442 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
6f505b16 PZ |
443 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
444 | ||
9f0c1e56 | 445 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 446 | { |
f6121f4f | 447 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
74b7eb58 YZ |
448 | struct sched_rt_entity *rt_se; |
449 | ||
0c3b9168 BS |
450 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
451 | ||
452 | rt_se = rt_rq->tg->rt_se[cpu]; | |
6f505b16 | 453 | |
f6121f4f DF |
454 | if (rt_rq->rt_nr_running) { |
455 | if (rt_se && !on_rt_rq(rt_se)) | |
37dad3fc | 456 | enqueue_rt_entity(rt_se, false); |
e864c499 | 457 | if (rt_rq->highest_prio.curr < curr->prio) |
1020387f | 458 | resched_task(curr); |
6f505b16 PZ |
459 | } |
460 | } | |
461 | ||
9f0c1e56 | 462 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 463 | { |
74b7eb58 | 464 | struct sched_rt_entity *rt_se; |
0c3b9168 | 465 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
74b7eb58 | 466 | |
0c3b9168 | 467 | rt_se = rt_rq->tg->rt_se[cpu]; |
6f505b16 PZ |
468 | |
469 | if (rt_se && on_rt_rq(rt_se)) | |
470 | dequeue_rt_entity(rt_se); | |
471 | } | |
472 | ||
23b0fdfc PZ |
473 | static int rt_se_boosted(struct sched_rt_entity *rt_se) |
474 | { | |
475 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
476 | struct task_struct *p; | |
477 | ||
478 | if (rt_rq) | |
479 | return !!rt_rq->rt_nr_boosted; | |
480 | ||
481 | p = rt_task_of(rt_se); | |
482 | return p->prio != p->normal_prio; | |
483 | } | |
484 | ||
d0b27fa7 | 485 | #ifdef CONFIG_SMP |
c6c4927b | 486 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 487 | { |
424c93fe | 488 | return this_rq()->rd->span; |
d0b27fa7 | 489 | } |
6f505b16 | 490 | #else |
c6c4927b | 491 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 492 | { |
c6c4927b | 493 | return cpu_online_mask; |
d0b27fa7 PZ |
494 | } |
495 | #endif | |
6f505b16 | 496 | |
d0b27fa7 PZ |
497 | static inline |
498 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 499 | { |
d0b27fa7 PZ |
500 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
501 | } | |
9f0c1e56 | 502 | |
ac086bc2 PZ |
503 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
504 | { | |
505 | return &rt_rq->tg->rt_bandwidth; | |
506 | } | |
507 | ||
55e12e5e | 508 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
509 | |
510 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
511 | { | |
ac086bc2 PZ |
512 | return rt_rq->rt_runtime; |
513 | } | |
514 | ||
515 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
516 | { | |
517 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
518 | } |
519 | ||
ec514c48 CX |
520 | typedef struct rt_rq *rt_rq_iter_t; |
521 | ||
522 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
523 | for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
524 | ||
6f505b16 PZ |
525 | #define for_each_sched_rt_entity(rt_se) \ |
526 | for (; rt_se; rt_se = NULL) | |
527 | ||
528 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
529 | { | |
530 | return NULL; | |
531 | } | |
532 | ||
9f0c1e56 | 533 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 534 | { |
f3ade837 JB |
535 | if (rt_rq->rt_nr_running) |
536 | resched_task(rq_of_rt_rq(rt_rq)->curr); | |
6f505b16 PZ |
537 | } |
538 | ||
9f0c1e56 | 539 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
540 | { |
541 | } | |
542 | ||
c6c4927b | 543 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 544 | { |
c6c4927b | 545 | return cpu_online_mask; |
d0b27fa7 PZ |
546 | } |
547 | ||
548 | static inline | |
549 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
550 | { | |
551 | return &cpu_rq(cpu)->rt; | |
552 | } | |
553 | ||
ac086bc2 PZ |
554 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
555 | { | |
556 | return &def_rt_bandwidth; | |
557 | } | |
558 | ||
55e12e5e | 559 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 560 | |
faa59937 JL |
561 | bool sched_rt_bandwidth_account(struct rt_rq *rt_rq) |
562 | { | |
563 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
564 | ||
565 | return (hrtimer_active(&rt_b->rt_period_timer) || | |
566 | rt_rq->rt_time < rt_b->rt_runtime); | |
567 | } | |
568 | ||
ac086bc2 | 569 | #ifdef CONFIG_SMP |
78333cdd PZ |
570 | /* |
571 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
572 | */ | |
b79f3833 | 573 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
574 | { |
575 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
aa7f6730 | 576 | struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd; |
ac086bc2 PZ |
577 | int i, weight, more = 0; |
578 | u64 rt_period; | |
579 | ||
c6c4927b | 580 | weight = cpumask_weight(rd->span); |
ac086bc2 | 581 | |
0986b11b | 582 | raw_spin_lock(&rt_b->rt_runtime_lock); |
ac086bc2 | 583 | rt_period = ktime_to_ns(rt_b->rt_period); |
c6c4927b | 584 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
585 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
586 | s64 diff; | |
587 | ||
588 | if (iter == rt_rq) | |
589 | continue; | |
590 | ||
0986b11b | 591 | raw_spin_lock(&iter->rt_runtime_lock); |
78333cdd PZ |
592 | /* |
593 | * Either all rqs have inf runtime and there's nothing to steal | |
594 | * or __disable_runtime() below sets a specific rq to inf to | |
595 | * indicate its been disabled and disalow stealing. | |
596 | */ | |
7def2be1 PZ |
597 | if (iter->rt_runtime == RUNTIME_INF) |
598 | goto next; | |
599 | ||
78333cdd PZ |
600 | /* |
601 | * From runqueues with spare time, take 1/n part of their | |
602 | * spare time, but no more than our period. | |
603 | */ | |
ac086bc2 PZ |
604 | diff = iter->rt_runtime - iter->rt_time; |
605 | if (diff > 0) { | |
58838cf3 | 606 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
607 | if (rt_rq->rt_runtime + diff > rt_period) |
608 | diff = rt_period - rt_rq->rt_runtime; | |
609 | iter->rt_runtime -= diff; | |
610 | rt_rq->rt_runtime += diff; | |
611 | more = 1; | |
612 | if (rt_rq->rt_runtime == rt_period) { | |
0986b11b | 613 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 PZ |
614 | break; |
615 | } | |
616 | } | |
7def2be1 | 617 | next: |
0986b11b | 618 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 | 619 | } |
0986b11b | 620 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
ac086bc2 PZ |
621 | |
622 | return more; | |
623 | } | |
7def2be1 | 624 | |
78333cdd PZ |
625 | /* |
626 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
627 | */ | |
7def2be1 PZ |
628 | static void __disable_runtime(struct rq *rq) |
629 | { | |
630 | struct root_domain *rd = rq->rd; | |
ec514c48 | 631 | rt_rq_iter_t iter; |
7def2be1 PZ |
632 | struct rt_rq *rt_rq; |
633 | ||
634 | if (unlikely(!scheduler_running)) | |
635 | return; | |
636 | ||
ec514c48 | 637 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
638 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
639 | s64 want; | |
640 | int i; | |
641 | ||
0986b11b TG |
642 | raw_spin_lock(&rt_b->rt_runtime_lock); |
643 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
644 | /* |
645 | * Either we're all inf and nobody needs to borrow, or we're | |
646 | * already disabled and thus have nothing to do, or we have | |
647 | * exactly the right amount of runtime to take out. | |
648 | */ | |
7def2be1 PZ |
649 | if (rt_rq->rt_runtime == RUNTIME_INF || |
650 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
651 | goto balanced; | |
0986b11b | 652 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
7def2be1 | 653 | |
78333cdd PZ |
654 | /* |
655 | * Calculate the difference between what we started out with | |
656 | * and what we current have, that's the amount of runtime | |
657 | * we lend and now have to reclaim. | |
658 | */ | |
7def2be1 PZ |
659 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
660 | ||
78333cdd PZ |
661 | /* |
662 | * Greedy reclaim, take back as much as we can. | |
663 | */ | |
c6c4927b | 664 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
665 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
666 | s64 diff; | |
667 | ||
78333cdd PZ |
668 | /* |
669 | * Can't reclaim from ourselves or disabled runqueues. | |
670 | */ | |
f1679d08 | 671 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
672 | continue; |
673 | ||
0986b11b | 674 | raw_spin_lock(&iter->rt_runtime_lock); |
7def2be1 PZ |
675 | if (want > 0) { |
676 | diff = min_t(s64, iter->rt_runtime, want); | |
677 | iter->rt_runtime -= diff; | |
678 | want -= diff; | |
679 | } else { | |
680 | iter->rt_runtime -= want; | |
681 | want -= want; | |
682 | } | |
0986b11b | 683 | raw_spin_unlock(&iter->rt_runtime_lock); |
7def2be1 PZ |
684 | |
685 | if (!want) | |
686 | break; | |
687 | } | |
688 | ||
0986b11b | 689 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
78333cdd PZ |
690 | /* |
691 | * We cannot be left wanting - that would mean some runtime | |
692 | * leaked out of the system. | |
693 | */ | |
7def2be1 PZ |
694 | BUG_ON(want); |
695 | balanced: | |
78333cdd PZ |
696 | /* |
697 | * Disable all the borrow logic by pretending we have inf | |
698 | * runtime - in which case borrowing doesn't make sense. | |
699 | */ | |
7def2be1 | 700 | rt_rq->rt_runtime = RUNTIME_INF; |
a4c96ae3 | 701 | rt_rq->rt_throttled = 0; |
0986b11b TG |
702 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
703 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
704 | } |
705 | } | |
706 | ||
7def2be1 PZ |
707 | static void __enable_runtime(struct rq *rq) |
708 | { | |
ec514c48 | 709 | rt_rq_iter_t iter; |
7def2be1 PZ |
710 | struct rt_rq *rt_rq; |
711 | ||
712 | if (unlikely(!scheduler_running)) | |
713 | return; | |
714 | ||
78333cdd PZ |
715 | /* |
716 | * Reset each runqueue's bandwidth settings | |
717 | */ | |
ec514c48 | 718 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
719 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
720 | ||
0986b11b TG |
721 | raw_spin_lock(&rt_b->rt_runtime_lock); |
722 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
7def2be1 PZ |
723 | rt_rq->rt_runtime = rt_b->rt_runtime; |
724 | rt_rq->rt_time = 0; | |
baf25731 | 725 | rt_rq->rt_throttled = 0; |
0986b11b TG |
726 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
727 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
728 | } |
729 | } | |
730 | ||
eff6549b PZ |
731 | static int balance_runtime(struct rt_rq *rt_rq) |
732 | { | |
733 | int more = 0; | |
734 | ||
4a6184ce PZ |
735 | if (!sched_feat(RT_RUNTIME_SHARE)) |
736 | return more; | |
737 | ||
eff6549b | 738 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
0986b11b | 739 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
eff6549b | 740 | more = do_balance_runtime(rt_rq); |
0986b11b | 741 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
742 | } |
743 | ||
744 | return more; | |
745 | } | |
55e12e5e | 746 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
747 | static inline int balance_runtime(struct rt_rq *rt_rq) |
748 | { | |
749 | return 0; | |
750 | } | |
55e12e5e | 751 | #endif /* CONFIG_SMP */ |
ac086bc2 | 752 | |
eff6549b PZ |
753 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
754 | { | |
42c62a58 | 755 | int i, idle = 1, throttled = 0; |
c6c4927b | 756 | const struct cpumask *span; |
eff6549b | 757 | |
eff6549b | 758 | span = sched_rt_period_mask(); |
e221d028 MG |
759 | #ifdef CONFIG_RT_GROUP_SCHED |
760 | /* | |
761 | * FIXME: isolated CPUs should really leave the root task group, | |
762 | * whether they are isolcpus or were isolated via cpusets, lest | |
763 | * the timer run on a CPU which does not service all runqueues, | |
764 | * potentially leaving other CPUs indefinitely throttled. If | |
765 | * isolation is really required, the user will turn the throttle | |
766 | * off to kill the perturbations it causes anyway. Meanwhile, | |
767 | * this maintains functionality for boot and/or troubleshooting. | |
768 | */ | |
769 | if (rt_b == &root_task_group.rt_bandwidth) | |
770 | span = cpu_online_mask; | |
771 | #endif | |
c6c4927b | 772 | for_each_cpu(i, span) { |
eff6549b PZ |
773 | int enqueue = 0; |
774 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
775 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
776 | ||
05fa785c | 777 | raw_spin_lock(&rq->lock); |
eff6549b PZ |
778 | if (rt_rq->rt_time) { |
779 | u64 runtime; | |
780 | ||
0986b11b | 781 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
782 | if (rt_rq->rt_throttled) |
783 | balance_runtime(rt_rq); | |
784 | runtime = rt_rq->rt_runtime; | |
785 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
786 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
787 | rt_rq->rt_throttled = 0; | |
788 | enqueue = 1; | |
61eadef6 MG |
789 | |
790 | /* | |
791 | * Force a clock update if the CPU was idle, | |
792 | * lest wakeup -> unthrottle time accumulate. | |
793 | */ | |
794 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) | |
795 | rq->skip_clock_update = -1; | |
eff6549b PZ |
796 | } |
797 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
798 | idle = 0; | |
0986b11b | 799 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
0c3b9168 | 800 | } else if (rt_rq->rt_nr_running) { |
6c3df255 | 801 | idle = 0; |
0c3b9168 BS |
802 | if (!rt_rq_throttled(rt_rq)) |
803 | enqueue = 1; | |
804 | } | |
42c62a58 PZ |
805 | if (rt_rq->rt_throttled) |
806 | throttled = 1; | |
eff6549b PZ |
807 | |
808 | if (enqueue) | |
809 | sched_rt_rq_enqueue(rt_rq); | |
05fa785c | 810 | raw_spin_unlock(&rq->lock); |
eff6549b PZ |
811 | } |
812 | ||
42c62a58 PZ |
813 | if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)) |
814 | return 1; | |
815 | ||
eff6549b PZ |
816 | return idle; |
817 | } | |
ac086bc2 | 818 | |
6f505b16 PZ |
819 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
820 | { | |
052f1dc7 | 821 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
822 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
823 | ||
824 | if (rt_rq) | |
e864c499 | 825 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
826 | #endif |
827 | ||
828 | return rt_task_of(rt_se)->prio; | |
829 | } | |
830 | ||
9f0c1e56 | 831 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 832 | { |
9f0c1e56 | 833 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 834 | |
fa85ae24 | 835 | if (rt_rq->rt_throttled) |
23b0fdfc | 836 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 837 | |
5b680fd6 | 838 | if (runtime >= sched_rt_period(rt_rq)) |
ac086bc2 PZ |
839 | return 0; |
840 | ||
b79f3833 PZ |
841 | balance_runtime(rt_rq); |
842 | runtime = sched_rt_runtime(rt_rq); | |
843 | if (runtime == RUNTIME_INF) | |
844 | return 0; | |
ac086bc2 | 845 | |
9f0c1e56 | 846 | if (rt_rq->rt_time > runtime) { |
7abc63b1 PZ |
847 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
848 | ||
849 | /* | |
850 | * Don't actually throttle groups that have no runtime assigned | |
851 | * but accrue some time due to boosting. | |
852 | */ | |
853 | if (likely(rt_b->rt_runtime)) { | |
3ccf3e83 PZ |
854 | static bool once = false; |
855 | ||
7abc63b1 | 856 | rt_rq->rt_throttled = 1; |
3ccf3e83 PZ |
857 | |
858 | if (!once) { | |
859 | once = true; | |
860 | printk_sched("sched: RT throttling activated\n"); | |
861 | } | |
7abc63b1 PZ |
862 | } else { |
863 | /* | |
864 | * In case we did anyway, make it go away, | |
865 | * replenishment is a joke, since it will replenish us | |
866 | * with exactly 0 ns. | |
867 | */ | |
868 | rt_rq->rt_time = 0; | |
869 | } | |
870 | ||
23b0fdfc | 871 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 872 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
873 | return 1; |
874 | } | |
fa85ae24 PZ |
875 | } |
876 | ||
877 | return 0; | |
878 | } | |
879 | ||
bb44e5d1 IM |
880 | /* |
881 | * Update the current task's runtime statistics. Skip current tasks that | |
882 | * are not in our scheduling class. | |
883 | */ | |
a9957449 | 884 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
885 | { |
886 | struct task_struct *curr = rq->curr; | |
6f505b16 PZ |
887 | struct sched_rt_entity *rt_se = &curr->rt; |
888 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
bb44e5d1 IM |
889 | u64 delta_exec; |
890 | ||
06c3bc65 | 891 | if (curr->sched_class != &rt_sched_class) |
bb44e5d1 IM |
892 | return; |
893 | ||
78becc27 | 894 | delta_exec = rq_clock_task(rq) - curr->se.exec_start; |
fc79e240 KT |
895 | if (unlikely((s64)delta_exec <= 0)) |
896 | return; | |
6cfb0d5d | 897 | |
42c62a58 PZ |
898 | schedstat_set(curr->se.statistics.exec_max, |
899 | max(curr->se.statistics.exec_max, delta_exec)); | |
bb44e5d1 IM |
900 | |
901 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
902 | account_group_exec_runtime(curr, delta_exec); |
903 | ||
78becc27 | 904 | curr->se.exec_start = rq_clock_task(rq); |
d842de87 | 905 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 906 | |
e9e9250b PZ |
907 | sched_rt_avg_update(rq, delta_exec); |
908 | ||
0b148fa0 PZ |
909 | if (!rt_bandwidth_enabled()) |
910 | return; | |
911 | ||
354d60c2 DG |
912 | for_each_sched_rt_entity(rt_se) { |
913 | rt_rq = rt_rq_of_se(rt_se); | |
914 | ||
cc2991cf | 915 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
0986b11b | 916 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
917 | rt_rq->rt_time += delta_exec; |
918 | if (sched_rt_runtime_exceeded(rt_rq)) | |
919 | resched_task(curr); | |
0986b11b | 920 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 921 | } |
354d60c2 | 922 | } |
bb44e5d1 IM |
923 | } |
924 | ||
398a153b | 925 | #if defined CONFIG_SMP |
e864c499 | 926 | |
398a153b GH |
927 | static void |
928 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 929 | { |
4d984277 | 930 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 931 | |
757dfcaa KT |
932 | #ifdef CONFIG_RT_GROUP_SCHED |
933 | /* | |
934 | * Change rq's cpupri only if rt_rq is the top queue. | |
935 | */ | |
936 | if (&rq->rt != rt_rq) | |
937 | return; | |
938 | #endif | |
5181f4a4 SR |
939 | if (rq->online && prio < prev_prio) |
940 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); | |
398a153b | 941 | } |
73fe6aae | 942 | |
398a153b GH |
943 | static void |
944 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
945 | { | |
946 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 947 | |
757dfcaa KT |
948 | #ifdef CONFIG_RT_GROUP_SCHED |
949 | /* | |
950 | * Change rq's cpupri only if rt_rq is the top queue. | |
951 | */ | |
952 | if (&rq->rt != rt_rq) | |
953 | return; | |
954 | #endif | |
398a153b GH |
955 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
956 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
957 | } |
958 | ||
398a153b GH |
959 | #else /* CONFIG_SMP */ |
960 | ||
6f505b16 | 961 | static inline |
398a153b GH |
962 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
963 | static inline | |
964 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
965 | ||
966 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 967 | |
052f1dc7 | 968 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
969 | static void |
970 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
971 | { | |
972 | int prev_prio = rt_rq->highest_prio.curr; | |
973 | ||
974 | if (prio < prev_prio) | |
975 | rt_rq->highest_prio.curr = prio; | |
976 | ||
977 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
978 | } | |
979 | ||
980 | static void | |
981 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
982 | { | |
983 | int prev_prio = rt_rq->highest_prio.curr; | |
984 | ||
6f505b16 | 985 | if (rt_rq->rt_nr_running) { |
764a9d6f | 986 | |
398a153b | 987 | WARN_ON(prio < prev_prio); |
764a9d6f | 988 | |
e864c499 | 989 | /* |
398a153b GH |
990 | * This may have been our highest task, and therefore |
991 | * we may have some recomputation to do | |
e864c499 | 992 | */ |
398a153b | 993 | if (prio == prev_prio) { |
e864c499 GH |
994 | struct rt_prio_array *array = &rt_rq->active; |
995 | ||
996 | rt_rq->highest_prio.curr = | |
764a9d6f | 997 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
998 | } |
999 | ||
764a9d6f | 1000 | } else |
e864c499 | 1001 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 1002 | |
398a153b GH |
1003 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
1004 | } | |
1f11eb6a | 1005 | |
398a153b GH |
1006 | #else |
1007 | ||
1008 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1009 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1010 | ||
1011 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 1012 | |
052f1dc7 | 1013 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
1014 | |
1015 | static void | |
1016 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1017 | { | |
1018 | if (rt_se_boosted(rt_se)) | |
1019 | rt_rq->rt_nr_boosted++; | |
1020 | ||
1021 | if (rt_rq->tg) | |
1022 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
1023 | } | |
1024 | ||
1025 | static void | |
1026 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1027 | { | |
23b0fdfc PZ |
1028 | if (rt_se_boosted(rt_se)) |
1029 | rt_rq->rt_nr_boosted--; | |
1030 | ||
1031 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
1032 | } |
1033 | ||
1034 | #else /* CONFIG_RT_GROUP_SCHED */ | |
1035 | ||
1036 | static void | |
1037 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1038 | { | |
1039 | start_rt_bandwidth(&def_rt_bandwidth); | |
1040 | } | |
1041 | ||
1042 | static inline | |
1043 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
1044 | ||
1045 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
1046 | ||
22abdef3 KT |
1047 | static inline |
1048 | unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se) | |
1049 | { | |
1050 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
1051 | ||
1052 | if (group_rq) | |
1053 | return group_rq->rt_nr_running; | |
1054 | else | |
1055 | return 1; | |
1056 | } | |
1057 | ||
398a153b GH |
1058 | static inline |
1059 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1060 | { | |
1061 | int prio = rt_se_prio(rt_se); | |
1062 | ||
1063 | WARN_ON(!rt_prio(prio)); | |
22abdef3 | 1064 | rt_rq->rt_nr_running += rt_se_nr_running(rt_se); |
398a153b GH |
1065 | |
1066 | inc_rt_prio(rt_rq, prio); | |
1067 | inc_rt_migration(rt_se, rt_rq); | |
1068 | inc_rt_group(rt_se, rt_rq); | |
1069 | } | |
1070 | ||
1071 | static inline | |
1072 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1073 | { | |
1074 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
1075 | WARN_ON(!rt_rq->rt_nr_running); | |
22abdef3 | 1076 | rt_rq->rt_nr_running -= rt_se_nr_running(rt_se); |
398a153b GH |
1077 | |
1078 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
1079 | dec_rt_migration(rt_se, rt_rq); | |
1080 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
1081 | } |
1082 | ||
37dad3fc | 1083 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
bb44e5d1 | 1084 | { |
6f505b16 PZ |
1085 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
1086 | struct rt_prio_array *array = &rt_rq->active; | |
1087 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 1088 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 1089 | |
ad2a3f13 PZ |
1090 | /* |
1091 | * Don't enqueue the group if its throttled, or when empty. | |
1092 | * The latter is a consequence of the former when a child group | |
1093 | * get throttled and the current group doesn't have any other | |
1094 | * active members. | |
1095 | */ | |
1096 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 1097 | return; |
63489e45 | 1098 | |
37dad3fc TG |
1099 | if (head) |
1100 | list_add(&rt_se->run_list, queue); | |
1101 | else | |
1102 | list_add_tail(&rt_se->run_list, queue); | |
6f505b16 | 1103 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 1104 | |
6f505b16 PZ |
1105 | inc_rt_tasks(rt_se, rt_rq); |
1106 | } | |
1107 | ||
ad2a3f13 | 1108 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
1109 | { |
1110 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
1111 | struct rt_prio_array *array = &rt_rq->active; | |
1112 | ||
1113 | list_del_init(&rt_se->run_list); | |
1114 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
1115 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
1116 | ||
1117 | dec_rt_tasks(rt_se, rt_rq); | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * Because the prio of an upper entry depends on the lower | |
1122 | * entries, we must remove entries top - down. | |
6f505b16 | 1123 | */ |
ad2a3f13 | 1124 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 1125 | { |
ad2a3f13 | 1126 | struct sched_rt_entity *back = NULL; |
6f505b16 | 1127 | |
58d6c2d7 PZ |
1128 | for_each_sched_rt_entity(rt_se) { |
1129 | rt_se->back = back; | |
1130 | back = rt_se; | |
1131 | } | |
1132 | ||
1133 | for (rt_se = back; rt_se; rt_se = rt_se->back) { | |
1134 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
1135 | __dequeue_rt_entity(rt_se); |
1136 | } | |
1137 | } | |
1138 | ||
37dad3fc | 1139 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
ad2a3f13 PZ |
1140 | { |
1141 | dequeue_rt_stack(rt_se); | |
1142 | for_each_sched_rt_entity(rt_se) | |
37dad3fc | 1143 | __enqueue_rt_entity(rt_se, head); |
ad2a3f13 PZ |
1144 | } |
1145 | ||
1146 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
1147 | { | |
1148 | dequeue_rt_stack(rt_se); | |
1149 | ||
1150 | for_each_sched_rt_entity(rt_se) { | |
1151 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
1152 | ||
1153 | if (rt_rq && rt_rq->rt_nr_running) | |
37dad3fc | 1154 | __enqueue_rt_entity(rt_se, false); |
58d6c2d7 | 1155 | } |
bb44e5d1 IM |
1156 | } |
1157 | ||
1158 | /* | |
1159 | * Adding/removing a task to/from a priority array: | |
1160 | */ | |
ea87bb78 | 1161 | static void |
371fd7e7 | 1162 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
6f505b16 PZ |
1163 | { |
1164 | struct sched_rt_entity *rt_se = &p->rt; | |
1165 | ||
371fd7e7 | 1166 | if (flags & ENQUEUE_WAKEUP) |
6f505b16 PZ |
1167 | rt_se->timeout = 0; |
1168 | ||
371fd7e7 | 1169 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
c09595f6 | 1170 | |
29baa747 | 1171 | if (!task_current(rq, p) && p->nr_cpus_allowed > 1) |
917b627d | 1172 | enqueue_pushable_task(rq, p); |
953bfcd1 PT |
1173 | |
1174 | inc_nr_running(rq); | |
6f505b16 PZ |
1175 | } |
1176 | ||
371fd7e7 | 1177 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1178 | { |
6f505b16 | 1179 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 1180 | |
f1e14ef6 | 1181 | update_curr_rt(rq); |
ad2a3f13 | 1182 | dequeue_rt_entity(rt_se); |
c09595f6 | 1183 | |
917b627d | 1184 | dequeue_pushable_task(rq, p); |
953bfcd1 PT |
1185 | |
1186 | dec_nr_running(rq); | |
bb44e5d1 IM |
1187 | } |
1188 | ||
1189 | /* | |
60686317 RW |
1190 | * Put task to the head or the end of the run list without the overhead of |
1191 | * dequeue followed by enqueue. | |
bb44e5d1 | 1192 | */ |
7ebefa8c DA |
1193 | static void |
1194 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 1195 | { |
1cdad715 | 1196 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
1197 | struct rt_prio_array *array = &rt_rq->active; |
1198 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
1199 | ||
1200 | if (head) | |
1201 | list_move(&rt_se->run_list, queue); | |
1202 | else | |
1203 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 1204 | } |
6f505b16 PZ |
1205 | } |
1206 | ||
7ebefa8c | 1207 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 1208 | { |
6f505b16 PZ |
1209 | struct sched_rt_entity *rt_se = &p->rt; |
1210 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1211 | |
6f505b16 PZ |
1212 | for_each_sched_rt_entity(rt_se) { |
1213 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 1214 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 1215 | } |
bb44e5d1 IM |
1216 | } |
1217 | ||
6f505b16 | 1218 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 1219 | { |
7ebefa8c | 1220 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
1221 | } |
1222 | ||
e7693a36 | 1223 | #ifdef CONFIG_SMP |
318e0893 GH |
1224 | static int find_lowest_rq(struct task_struct *task); |
1225 | ||
0017d735 | 1226 | static int |
ac66f547 | 1227 | select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags) |
e7693a36 | 1228 | { |
7608dec2 PZ |
1229 | struct task_struct *curr; |
1230 | struct rq *rq; | |
c37495fd | 1231 | |
29baa747 | 1232 | if (p->nr_cpus_allowed == 1) |
76854c7e MG |
1233 | goto out; |
1234 | ||
c37495fd SR |
1235 | /* For anything but wake ups, just return the task_cpu */ |
1236 | if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK) | |
1237 | goto out; | |
1238 | ||
7608dec2 PZ |
1239 | rq = cpu_rq(cpu); |
1240 | ||
1241 | rcu_read_lock(); | |
1242 | curr = ACCESS_ONCE(rq->curr); /* unlocked access */ | |
1243 | ||
318e0893 | 1244 | /* |
7608dec2 | 1245 | * If the current task on @p's runqueue is an RT task, then |
e1f47d89 SR |
1246 | * try to see if we can wake this RT task up on another |
1247 | * runqueue. Otherwise simply start this RT task | |
1248 | * on its current runqueue. | |
1249 | * | |
43fa5460 SR |
1250 | * We want to avoid overloading runqueues. If the woken |
1251 | * task is a higher priority, then it will stay on this CPU | |
1252 | * and the lower prio task should be moved to another CPU. | |
1253 | * Even though this will probably make the lower prio task | |
1254 | * lose its cache, we do not want to bounce a higher task | |
1255 | * around just because it gave up its CPU, perhaps for a | |
1256 | * lock? | |
1257 | * | |
1258 | * For equal prio tasks, we just let the scheduler sort it out. | |
7608dec2 PZ |
1259 | * |
1260 | * Otherwise, just let it ride on the affined RQ and the | |
1261 | * post-schedule router will push the preempted task away | |
1262 | * | |
1263 | * This test is optimistic, if we get it wrong the load-balancer | |
1264 | * will have to sort it out. | |
318e0893 | 1265 | */ |
7608dec2 | 1266 | if (curr && unlikely(rt_task(curr)) && |
29baa747 | 1267 | (curr->nr_cpus_allowed < 2 || |
6bfa687c | 1268 | curr->prio <= p->prio)) { |
7608dec2 | 1269 | int target = find_lowest_rq(p); |
318e0893 | 1270 | |
7608dec2 PZ |
1271 | if (target != -1) |
1272 | cpu = target; | |
318e0893 | 1273 | } |
7608dec2 | 1274 | rcu_read_unlock(); |
318e0893 | 1275 | |
c37495fd | 1276 | out: |
7608dec2 | 1277 | return cpu; |
e7693a36 | 1278 | } |
7ebefa8c DA |
1279 | |
1280 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
1281 | { | |
29baa747 | 1282 | if (rq->curr->nr_cpus_allowed == 1) |
7ebefa8c DA |
1283 | return; |
1284 | ||
29baa747 | 1285 | if (p->nr_cpus_allowed != 1 |
13b8bd0a RR |
1286 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1287 | return; | |
24600ce8 | 1288 | |
13b8bd0a RR |
1289 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
1290 | return; | |
7ebefa8c DA |
1291 | |
1292 | /* | |
1293 | * There appears to be other cpus that can accept | |
1294 | * current and none to run 'p', so lets reschedule | |
1295 | * to try and push current away: | |
1296 | */ | |
1297 | requeue_task_rt(rq, p, 1); | |
1298 | resched_task(rq->curr); | |
1299 | } | |
1300 | ||
e7693a36 GH |
1301 | #endif /* CONFIG_SMP */ |
1302 | ||
bb44e5d1 IM |
1303 | /* |
1304 | * Preempt the current task with a newly woken task if needed: | |
1305 | */ | |
7d478721 | 1306 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1307 | { |
45c01e82 | 1308 | if (p->prio < rq->curr->prio) { |
bb44e5d1 | 1309 | resched_task(rq->curr); |
45c01e82 GH |
1310 | return; |
1311 | } | |
1312 | ||
1313 | #ifdef CONFIG_SMP | |
1314 | /* | |
1315 | * If: | |
1316 | * | |
1317 | * - the newly woken task is of equal priority to the current task | |
1318 | * - the newly woken task is non-migratable while current is migratable | |
1319 | * - current will be preempted on the next reschedule | |
1320 | * | |
1321 | * we should check to see if current can readily move to a different | |
1322 | * cpu. If so, we will reschedule to allow the push logic to try | |
1323 | * to move current somewhere else, making room for our non-migratable | |
1324 | * task. | |
1325 | */ | |
8dd0de8b | 1326 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
7ebefa8c | 1327 | check_preempt_equal_prio(rq, p); |
45c01e82 | 1328 | #endif |
bb44e5d1 IM |
1329 | } |
1330 | ||
6f505b16 PZ |
1331 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1332 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1333 | { |
6f505b16 PZ |
1334 | struct rt_prio_array *array = &rt_rq->active; |
1335 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1336 | struct list_head *queue; |
1337 | int idx; | |
1338 | ||
1339 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1340 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1341 | |
1342 | queue = array->queue + idx; | |
6f505b16 | 1343 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1344 | |
6f505b16 PZ |
1345 | return next; |
1346 | } | |
bb44e5d1 | 1347 | |
917b627d | 1348 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1349 | { |
1350 | struct sched_rt_entity *rt_se; | |
1351 | struct task_struct *p; | |
606dba2e | 1352 | struct rt_rq *rt_rq = &rq->rt; |
6f505b16 PZ |
1353 | |
1354 | do { | |
1355 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1356 | BUG_ON(!rt_se); |
6f505b16 PZ |
1357 | rt_rq = group_rt_rq(rt_se); |
1358 | } while (rt_rq); | |
1359 | ||
1360 | p = rt_task_of(rt_se); | |
78becc27 | 1361 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
1362 | |
1363 | return p; | |
1364 | } | |
1365 | ||
606dba2e PZ |
1366 | static struct task_struct * |
1367 | pick_next_task_rt(struct rq *rq, struct task_struct *prev) | |
917b627d | 1368 | { |
606dba2e PZ |
1369 | struct task_struct *p; |
1370 | struct rt_rq *rt_rq = &rq->rt; | |
1371 | ||
37e117c0 | 1372 | if (need_pull_rt_task(rq, prev)) { |
38033c37 | 1373 | pull_rt_task(rq); |
37e117c0 PZ |
1374 | /* |
1375 | * pull_rt_task() can drop (and re-acquire) rq->lock; this | |
a1d9a323 KT |
1376 | * means a dl or stop task can slip in, in which case we need |
1377 | * to re-start task selection. | |
37e117c0 | 1378 | */ |
a1d9a323 KT |
1379 | if (unlikely((rq->stop && rq->stop->on_rq) || |
1380 | rq->dl.dl_nr_running)) | |
37e117c0 PZ |
1381 | return RETRY_TASK; |
1382 | } | |
38033c37 | 1383 | |
734ff2a7 KT |
1384 | /* |
1385 | * We may dequeue prev's rt_rq in put_prev_task(). | |
1386 | * So, we update time before rt_nr_running check. | |
1387 | */ | |
1388 | if (prev->sched_class == &rt_sched_class) | |
1389 | update_curr_rt(rq); | |
1390 | ||
606dba2e PZ |
1391 | if (!rt_rq->rt_nr_running) |
1392 | return NULL; | |
1393 | ||
1394 | if (rt_rq_throttled(rt_rq)) | |
1395 | return NULL; | |
1396 | ||
3f1d2a31 | 1397 | put_prev_task(rq, prev); |
606dba2e PZ |
1398 | |
1399 | p = _pick_next_task_rt(rq); | |
917b627d GH |
1400 | |
1401 | /* The running task is never eligible for pushing */ | |
1402 | if (p) | |
1403 | dequeue_pushable_task(rq, p); | |
1404 | ||
dc877341 | 1405 | set_post_schedule(rq); |
3f029d3c | 1406 | |
6f505b16 | 1407 | return p; |
bb44e5d1 IM |
1408 | } |
1409 | ||
31ee529c | 1410 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1411 | { |
f1e14ef6 | 1412 | update_curr_rt(rq); |
917b627d GH |
1413 | |
1414 | /* | |
1415 | * The previous task needs to be made eligible for pushing | |
1416 | * if it is still active | |
1417 | */ | |
29baa747 | 1418 | if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1) |
917b627d | 1419 | enqueue_pushable_task(rq, p); |
bb44e5d1 IM |
1420 | } |
1421 | ||
681f3e68 | 1422 | #ifdef CONFIG_SMP |
6f505b16 | 1423 | |
e8fa1362 SR |
1424 | /* Only try algorithms three times */ |
1425 | #define RT_MAX_TRIES 3 | |
1426 | ||
f65eda4f SR |
1427 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1428 | { | |
1429 | if (!task_running(rq, p) && | |
60334caf | 1430 | cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) |
f65eda4f SR |
1431 | return 1; |
1432 | return 0; | |
1433 | } | |
1434 | ||
e23ee747 KT |
1435 | /* |
1436 | * Return the highest pushable rq's task, which is suitable to be executed | |
1437 | * on the cpu, NULL otherwise | |
1438 | */ | |
1439 | static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu) | |
e8fa1362 | 1440 | { |
e23ee747 KT |
1441 | struct plist_head *head = &rq->rt.pushable_tasks; |
1442 | struct task_struct *p; | |
3d07467b | 1443 | |
e23ee747 KT |
1444 | if (!has_pushable_tasks(rq)) |
1445 | return NULL; | |
3d07467b | 1446 | |
e23ee747 KT |
1447 | plist_for_each_entry(p, head, pushable_tasks) { |
1448 | if (pick_rt_task(rq, p, cpu)) | |
1449 | return p; | |
f65eda4f SR |
1450 | } |
1451 | ||
e23ee747 | 1452 | return NULL; |
e8fa1362 SR |
1453 | } |
1454 | ||
0e3900e6 | 1455 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1456 | |
6e1254d2 GH |
1457 | static int find_lowest_rq(struct task_struct *task) |
1458 | { | |
1459 | struct sched_domain *sd; | |
96f874e2 | 1460 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
6e1254d2 GH |
1461 | int this_cpu = smp_processor_id(); |
1462 | int cpu = task_cpu(task); | |
06f90dbd | 1463 | |
0da938c4 SR |
1464 | /* Make sure the mask is initialized first */ |
1465 | if (unlikely(!lowest_mask)) | |
1466 | return -1; | |
1467 | ||
29baa747 | 1468 | if (task->nr_cpus_allowed == 1) |
6e0534f2 | 1469 | return -1; /* No other targets possible */ |
6e1254d2 | 1470 | |
6e0534f2 GH |
1471 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1472 | return -1; /* No targets found */ | |
6e1254d2 GH |
1473 | |
1474 | /* | |
1475 | * At this point we have built a mask of cpus representing the | |
1476 | * lowest priority tasks in the system. Now we want to elect | |
1477 | * the best one based on our affinity and topology. | |
1478 | * | |
1479 | * We prioritize the last cpu that the task executed on since | |
1480 | * it is most likely cache-hot in that location. | |
1481 | */ | |
96f874e2 | 1482 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1483 | return cpu; |
1484 | ||
1485 | /* | |
1486 | * Otherwise, we consult the sched_domains span maps to figure | |
1487 | * out which cpu is logically closest to our hot cache data. | |
1488 | */ | |
e2c88063 RR |
1489 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1490 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ | |
6e1254d2 | 1491 | |
cd4ae6ad | 1492 | rcu_read_lock(); |
e2c88063 RR |
1493 | for_each_domain(cpu, sd) { |
1494 | if (sd->flags & SD_WAKE_AFFINE) { | |
1495 | int best_cpu; | |
6e1254d2 | 1496 | |
e2c88063 RR |
1497 | /* |
1498 | * "this_cpu" is cheaper to preempt than a | |
1499 | * remote processor. | |
1500 | */ | |
1501 | if (this_cpu != -1 && | |
cd4ae6ad XF |
1502 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
1503 | rcu_read_unlock(); | |
e2c88063 | 1504 | return this_cpu; |
cd4ae6ad | 1505 | } |
e2c88063 RR |
1506 | |
1507 | best_cpu = cpumask_first_and(lowest_mask, | |
1508 | sched_domain_span(sd)); | |
cd4ae6ad XF |
1509 | if (best_cpu < nr_cpu_ids) { |
1510 | rcu_read_unlock(); | |
e2c88063 | 1511 | return best_cpu; |
cd4ae6ad | 1512 | } |
6e1254d2 GH |
1513 | } |
1514 | } | |
cd4ae6ad | 1515 | rcu_read_unlock(); |
6e1254d2 GH |
1516 | |
1517 | /* | |
1518 | * And finally, if there were no matches within the domains | |
1519 | * just give the caller *something* to work with from the compatible | |
1520 | * locations. | |
1521 | */ | |
e2c88063 RR |
1522 | if (this_cpu != -1) |
1523 | return this_cpu; | |
1524 | ||
1525 | cpu = cpumask_any(lowest_mask); | |
1526 | if (cpu < nr_cpu_ids) | |
1527 | return cpu; | |
1528 | return -1; | |
07b4032c GH |
1529 | } |
1530 | ||
1531 | /* Will lock the rq it finds */ | |
4df64c0b | 1532 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1533 | { |
1534 | struct rq *lowest_rq = NULL; | |
07b4032c | 1535 | int tries; |
4df64c0b | 1536 | int cpu; |
e8fa1362 | 1537 | |
07b4032c GH |
1538 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1539 | cpu = find_lowest_rq(task); | |
1540 | ||
2de0b463 | 1541 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1542 | break; |
1543 | ||
07b4032c GH |
1544 | lowest_rq = cpu_rq(cpu); |
1545 | ||
e8fa1362 | 1546 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1547 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1548 | /* |
1549 | * We had to unlock the run queue. In | |
1550 | * the mean time, task could have | |
1551 | * migrated already or had its affinity changed. | |
1552 | * Also make sure that it wasn't scheduled on its rq. | |
1553 | */ | |
07b4032c | 1554 | if (unlikely(task_rq(task) != rq || |
96f874e2 | 1555 | !cpumask_test_cpu(lowest_rq->cpu, |
fa17b507 | 1556 | tsk_cpus_allowed(task)) || |
07b4032c | 1557 | task_running(rq, task) || |
fd2f4419 | 1558 | !task->on_rq)) { |
4df64c0b | 1559 | |
7f1b4393 | 1560 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1561 | lowest_rq = NULL; |
1562 | break; | |
1563 | } | |
1564 | } | |
1565 | ||
1566 | /* If this rq is still suitable use it. */ | |
e864c499 | 1567 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1568 | break; |
1569 | ||
1570 | /* try again */ | |
1b12bbc7 | 1571 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1572 | lowest_rq = NULL; |
1573 | } | |
1574 | ||
1575 | return lowest_rq; | |
1576 | } | |
1577 | ||
917b627d GH |
1578 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1579 | { | |
1580 | struct task_struct *p; | |
1581 | ||
1582 | if (!has_pushable_tasks(rq)) | |
1583 | return NULL; | |
1584 | ||
1585 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1586 | struct task_struct, pushable_tasks); | |
1587 | ||
1588 | BUG_ON(rq->cpu != task_cpu(p)); | |
1589 | BUG_ON(task_current(rq, p)); | |
29baa747 | 1590 | BUG_ON(p->nr_cpus_allowed <= 1); |
917b627d | 1591 | |
fd2f4419 | 1592 | BUG_ON(!p->on_rq); |
917b627d GH |
1593 | BUG_ON(!rt_task(p)); |
1594 | ||
1595 | return p; | |
1596 | } | |
1597 | ||
e8fa1362 SR |
1598 | /* |
1599 | * If the current CPU has more than one RT task, see if the non | |
1600 | * running task can migrate over to a CPU that is running a task | |
1601 | * of lesser priority. | |
1602 | */ | |
697f0a48 | 1603 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1604 | { |
1605 | struct task_struct *next_task; | |
1606 | struct rq *lowest_rq; | |
311e800e | 1607 | int ret = 0; |
e8fa1362 | 1608 | |
a22d7fc1 GH |
1609 | if (!rq->rt.overloaded) |
1610 | return 0; | |
1611 | ||
917b627d | 1612 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1613 | if (!next_task) |
1614 | return 0; | |
1615 | ||
49246274 | 1616 | retry: |
697f0a48 | 1617 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1618 | WARN_ON(1); |
e8fa1362 | 1619 | return 0; |
f65eda4f | 1620 | } |
e8fa1362 SR |
1621 | |
1622 | /* | |
1623 | * It's possible that the next_task slipped in of | |
1624 | * higher priority than current. If that's the case | |
1625 | * just reschedule current. | |
1626 | */ | |
697f0a48 GH |
1627 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1628 | resched_task(rq->curr); | |
e8fa1362 SR |
1629 | return 0; |
1630 | } | |
1631 | ||
697f0a48 | 1632 | /* We might release rq lock */ |
e8fa1362 SR |
1633 | get_task_struct(next_task); |
1634 | ||
1635 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1636 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1637 | if (!lowest_rq) { |
1638 | struct task_struct *task; | |
1639 | /* | |
311e800e | 1640 | * find_lock_lowest_rq releases rq->lock |
1563513d GH |
1641 | * so it is possible that next_task has migrated. |
1642 | * | |
1643 | * We need to make sure that the task is still on the same | |
1644 | * run-queue and is also still the next task eligible for | |
1645 | * pushing. | |
e8fa1362 | 1646 | */ |
917b627d | 1647 | task = pick_next_pushable_task(rq); |
1563513d GH |
1648 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1649 | /* | |
311e800e HD |
1650 | * The task hasn't migrated, and is still the next |
1651 | * eligible task, but we failed to find a run-queue | |
1652 | * to push it to. Do not retry in this case, since | |
1653 | * other cpus will pull from us when ready. | |
1563513d | 1654 | */ |
1563513d | 1655 | goto out; |
e8fa1362 | 1656 | } |
917b627d | 1657 | |
1563513d GH |
1658 | if (!task) |
1659 | /* No more tasks, just exit */ | |
1660 | goto out; | |
1661 | ||
917b627d | 1662 | /* |
1563513d | 1663 | * Something has shifted, try again. |
917b627d | 1664 | */ |
1563513d GH |
1665 | put_task_struct(next_task); |
1666 | next_task = task; | |
1667 | goto retry; | |
e8fa1362 SR |
1668 | } |
1669 | ||
697f0a48 | 1670 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1671 | set_task_cpu(next_task, lowest_rq->cpu); |
1672 | activate_task(lowest_rq, next_task, 0); | |
311e800e | 1673 | ret = 1; |
e8fa1362 SR |
1674 | |
1675 | resched_task(lowest_rq->curr); | |
1676 | ||
1b12bbc7 | 1677 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1678 | |
e8fa1362 SR |
1679 | out: |
1680 | put_task_struct(next_task); | |
1681 | ||
311e800e | 1682 | return ret; |
e8fa1362 SR |
1683 | } |
1684 | ||
e8fa1362 SR |
1685 | static void push_rt_tasks(struct rq *rq) |
1686 | { | |
1687 | /* push_rt_task will return true if it moved an RT */ | |
1688 | while (push_rt_task(rq)) | |
1689 | ; | |
1690 | } | |
1691 | ||
f65eda4f SR |
1692 | static int pull_rt_task(struct rq *this_rq) |
1693 | { | |
80bf3171 | 1694 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
a8728944 | 1695 | struct task_struct *p; |
f65eda4f | 1696 | struct rq *src_rq; |
f65eda4f | 1697 | |
637f5085 | 1698 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
1699 | return 0; |
1700 | ||
7c3f2ab7 PZ |
1701 | /* |
1702 | * Match the barrier from rt_set_overloaded; this guarantees that if we | |
1703 | * see overloaded we must also see the rto_mask bit. | |
1704 | */ | |
1705 | smp_rmb(); | |
1706 | ||
c6c4927b | 1707 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1708 | if (this_cpu == cpu) |
1709 | continue; | |
1710 | ||
1711 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1712 | |
1713 | /* | |
1714 | * Don't bother taking the src_rq->lock if the next highest | |
1715 | * task is known to be lower-priority than our current task. | |
1716 | * This may look racy, but if this value is about to go | |
1717 | * logically higher, the src_rq will push this task away. | |
1718 | * And if its going logically lower, we do not care | |
1719 | */ | |
1720 | if (src_rq->rt.highest_prio.next >= | |
1721 | this_rq->rt.highest_prio.curr) | |
1722 | continue; | |
1723 | ||
f65eda4f SR |
1724 | /* |
1725 | * We can potentially drop this_rq's lock in | |
1726 | * double_lock_balance, and another CPU could | |
a8728944 | 1727 | * alter this_rq |
f65eda4f | 1728 | */ |
a8728944 | 1729 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
1730 | |
1731 | /* | |
e23ee747 KT |
1732 | * We can pull only a task, which is pushable |
1733 | * on its rq, and no others. | |
f65eda4f | 1734 | */ |
e23ee747 | 1735 | p = pick_highest_pushable_task(src_rq, this_cpu); |
f65eda4f SR |
1736 | |
1737 | /* | |
1738 | * Do we have an RT task that preempts | |
1739 | * the to-be-scheduled task? | |
1740 | */ | |
a8728944 | 1741 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f | 1742 | WARN_ON(p == src_rq->curr); |
fd2f4419 | 1743 | WARN_ON(!p->on_rq); |
f65eda4f SR |
1744 | |
1745 | /* | |
1746 | * There's a chance that p is higher in priority | |
1747 | * than what's currently running on its cpu. | |
1748 | * This is just that p is wakeing up and hasn't | |
1749 | * had a chance to schedule. We only pull | |
1750 | * p if it is lower in priority than the | |
a8728944 | 1751 | * current task on the run queue |
f65eda4f | 1752 | */ |
a8728944 | 1753 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 1754 | goto skip; |
f65eda4f SR |
1755 | |
1756 | ret = 1; | |
1757 | ||
1758 | deactivate_task(src_rq, p, 0); | |
1759 | set_task_cpu(p, this_cpu); | |
1760 | activate_task(this_rq, p, 0); | |
1761 | /* | |
1762 | * We continue with the search, just in | |
1763 | * case there's an even higher prio task | |
25985edc | 1764 | * in another runqueue. (low likelihood |
f65eda4f | 1765 | * but possible) |
f65eda4f | 1766 | */ |
f65eda4f | 1767 | } |
49246274 | 1768 | skip: |
1b12bbc7 | 1769 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
1770 | } |
1771 | ||
1772 | return ret; | |
1773 | } | |
1774 | ||
9a897c5a | 1775 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 | 1776 | { |
967fc046 | 1777 | push_rt_tasks(rq); |
e8fa1362 SR |
1778 | } |
1779 | ||
8ae121ac GH |
1780 | /* |
1781 | * If we are not running and we are not going to reschedule soon, we should | |
1782 | * try to push tasks away now | |
1783 | */ | |
efbbd05a | 1784 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1785 | { |
9a897c5a | 1786 | if (!task_running(rq, p) && |
8ae121ac | 1787 | !test_tsk_need_resched(rq->curr) && |
917b627d | 1788 | has_pushable_tasks(rq) && |
29baa747 | 1789 | p->nr_cpus_allowed > 1 && |
1baca4ce | 1790 | (dl_task(rq->curr) || rt_task(rq->curr)) && |
29baa747 | 1791 | (rq->curr->nr_cpus_allowed < 2 || |
3be209a8 | 1792 | rq->curr->prio <= p->prio)) |
4642dafd SR |
1793 | push_rt_tasks(rq); |
1794 | } | |
1795 | ||
cd8ba7cd | 1796 | static void set_cpus_allowed_rt(struct task_struct *p, |
96f874e2 | 1797 | const struct cpumask *new_mask) |
73fe6aae | 1798 | { |
8d3d5ada KT |
1799 | struct rq *rq; |
1800 | int weight; | |
73fe6aae GH |
1801 | |
1802 | BUG_ON(!rt_task(p)); | |
1803 | ||
8d3d5ada KT |
1804 | if (!p->on_rq) |
1805 | return; | |
917b627d | 1806 | |
8d3d5ada | 1807 | weight = cpumask_weight(new_mask); |
917b627d | 1808 | |
8d3d5ada KT |
1809 | /* |
1810 | * Only update if the process changes its state from whether it | |
1811 | * can migrate or not. | |
1812 | */ | |
29baa747 | 1813 | if ((p->nr_cpus_allowed > 1) == (weight > 1)) |
8d3d5ada | 1814 | return; |
917b627d | 1815 | |
8d3d5ada | 1816 | rq = task_rq(p); |
73fe6aae | 1817 | |
8d3d5ada KT |
1818 | /* |
1819 | * The process used to be able to migrate OR it can now migrate | |
1820 | */ | |
1821 | if (weight <= 1) { | |
1822 | if (!task_current(rq, p)) | |
1823 | dequeue_pushable_task(rq, p); | |
1824 | BUG_ON(!rq->rt.rt_nr_migratory); | |
1825 | rq->rt.rt_nr_migratory--; | |
1826 | } else { | |
1827 | if (!task_current(rq, p)) | |
1828 | enqueue_pushable_task(rq, p); | |
1829 | rq->rt.rt_nr_migratory++; | |
73fe6aae | 1830 | } |
8d3d5ada KT |
1831 | |
1832 | update_rt_migration(&rq->rt); | |
73fe6aae | 1833 | } |
deeeccd4 | 1834 | |
bdd7c81b | 1835 | /* Assumes rq->lock is held */ |
1f11eb6a | 1836 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
1837 | { |
1838 | if (rq->rt.overloaded) | |
1839 | rt_set_overload(rq); | |
6e0534f2 | 1840 | |
7def2be1 PZ |
1841 | __enable_runtime(rq); |
1842 | ||
e864c499 | 1843 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
1844 | } |
1845 | ||
1846 | /* Assumes rq->lock is held */ | |
1f11eb6a | 1847 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
1848 | { |
1849 | if (rq->rt.overloaded) | |
1850 | rt_clear_overload(rq); | |
6e0534f2 | 1851 | |
7def2be1 PZ |
1852 | __disable_runtime(rq); |
1853 | ||
6e0534f2 | 1854 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 1855 | } |
cb469845 SR |
1856 | |
1857 | /* | |
1858 | * When switch from the rt queue, we bring ourselves to a position | |
1859 | * that we might want to pull RT tasks from other runqueues. | |
1860 | */ | |
da7a735e | 1861 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1862 | { |
1863 | /* | |
1864 | * If there are other RT tasks then we will reschedule | |
1865 | * and the scheduling of the other RT tasks will handle | |
1866 | * the balancing. But if we are the last RT task | |
1867 | * we may need to handle the pulling of RT tasks | |
1868 | * now. | |
1869 | */ | |
1158ddb5 KT |
1870 | if (!p->on_rq || rq->rt.rt_nr_running) |
1871 | return; | |
1872 | ||
1873 | if (pull_rt_task(rq)) | |
1874 | resched_task(rq->curr); | |
cb469845 | 1875 | } |
3d8cbdf8 | 1876 | |
11c785b7 | 1877 | void __init init_sched_rt_class(void) |
3d8cbdf8 RR |
1878 | { |
1879 | unsigned int i; | |
1880 | ||
029632fb | 1881 | for_each_possible_cpu(i) { |
eaa95840 | 1882 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 1883 | GFP_KERNEL, cpu_to_node(i)); |
029632fb | 1884 | } |
3d8cbdf8 | 1885 | } |
cb469845 SR |
1886 | #endif /* CONFIG_SMP */ |
1887 | ||
1888 | /* | |
1889 | * When switching a task to RT, we may overload the runqueue | |
1890 | * with RT tasks. In this case we try to push them off to | |
1891 | * other runqueues. | |
1892 | */ | |
da7a735e | 1893 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1894 | { |
1895 | int check_resched = 1; | |
1896 | ||
1897 | /* | |
1898 | * If we are already running, then there's nothing | |
1899 | * that needs to be done. But if we are not running | |
1900 | * we may need to preempt the current running task. | |
1901 | * If that current running task is also an RT task | |
1902 | * then see if we can move to another run queue. | |
1903 | */ | |
fd2f4419 | 1904 | if (p->on_rq && rq->curr != p) { |
cb469845 | 1905 | #ifdef CONFIG_SMP |
10447917 | 1906 | if (p->nr_cpus_allowed > 1 && rq->rt.overloaded && |
cb469845 | 1907 | /* Don't resched if we changed runqueues */ |
10447917 | 1908 | push_rt_task(rq) && rq != task_rq(p)) |
cb469845 SR |
1909 | check_resched = 0; |
1910 | #endif /* CONFIG_SMP */ | |
1911 | if (check_resched && p->prio < rq->curr->prio) | |
1912 | resched_task(rq->curr); | |
1913 | } | |
1914 | } | |
1915 | ||
1916 | /* | |
1917 | * Priority of the task has changed. This may cause | |
1918 | * us to initiate a push or pull. | |
1919 | */ | |
da7a735e PZ |
1920 | static void |
1921 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) | |
cb469845 | 1922 | { |
fd2f4419 | 1923 | if (!p->on_rq) |
da7a735e PZ |
1924 | return; |
1925 | ||
1926 | if (rq->curr == p) { | |
cb469845 SR |
1927 | #ifdef CONFIG_SMP |
1928 | /* | |
1929 | * If our priority decreases while running, we | |
1930 | * may need to pull tasks to this runqueue. | |
1931 | */ | |
1932 | if (oldprio < p->prio) | |
1933 | pull_rt_task(rq); | |
1934 | /* | |
1935 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
1936 | * then reschedule. Note, the above pull_rt_task |
1937 | * can release the rq lock and p could migrate. | |
1938 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 1939 | */ |
e864c499 | 1940 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
cb469845 SR |
1941 | resched_task(p); |
1942 | #else | |
1943 | /* For UP simply resched on drop of prio */ | |
1944 | if (oldprio < p->prio) | |
1945 | resched_task(p); | |
e8fa1362 | 1946 | #endif /* CONFIG_SMP */ |
cb469845 SR |
1947 | } else { |
1948 | /* | |
1949 | * This task is not running, but if it is | |
1950 | * greater than the current running task | |
1951 | * then reschedule. | |
1952 | */ | |
1953 | if (p->prio < rq->curr->prio) | |
1954 | resched_task(rq->curr); | |
1955 | } | |
1956 | } | |
1957 | ||
78f2c7db PZ |
1958 | static void watchdog(struct rq *rq, struct task_struct *p) |
1959 | { | |
1960 | unsigned long soft, hard; | |
1961 | ||
78d7d407 JS |
1962 | /* max may change after cur was read, this will be fixed next tick */ |
1963 | soft = task_rlimit(p, RLIMIT_RTTIME); | |
1964 | hard = task_rlimit_max(p, RLIMIT_RTTIME); | |
78f2c7db PZ |
1965 | |
1966 | if (soft != RLIM_INFINITY) { | |
1967 | unsigned long next; | |
1968 | ||
57d2aa00 YX |
1969 | if (p->rt.watchdog_stamp != jiffies) { |
1970 | p->rt.timeout++; | |
1971 | p->rt.watchdog_stamp = jiffies; | |
1972 | } | |
1973 | ||
78f2c7db | 1974 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
5a52dd50 | 1975 | if (p->rt.timeout > next) |
f06febc9 | 1976 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
1977 | } |
1978 | } | |
bb44e5d1 | 1979 | |
8f4d37ec | 1980 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 1981 | { |
454c7999 CC |
1982 | struct sched_rt_entity *rt_se = &p->rt; |
1983 | ||
67e2be02 PZ |
1984 | update_curr_rt(rq); |
1985 | ||
78f2c7db PZ |
1986 | watchdog(rq, p); |
1987 | ||
bb44e5d1 IM |
1988 | /* |
1989 | * RR tasks need a special form of timeslice management. | |
1990 | * FIFO tasks have no timeslices. | |
1991 | */ | |
1992 | if (p->policy != SCHED_RR) | |
1993 | return; | |
1994 | ||
fa717060 | 1995 | if (--p->rt.time_slice) |
bb44e5d1 IM |
1996 | return; |
1997 | ||
ce0dbbbb | 1998 | p->rt.time_slice = sched_rr_timeslice; |
bb44e5d1 | 1999 | |
98fbc798 | 2000 | /* |
e9aa39bb LB |
2001 | * Requeue to the end of queue if we (and all of our ancestors) are not |
2002 | * the only element on the queue | |
98fbc798 | 2003 | */ |
454c7999 CC |
2004 | for_each_sched_rt_entity(rt_se) { |
2005 | if (rt_se->run_list.prev != rt_se->run_list.next) { | |
2006 | requeue_task_rt(rq, p, 0); | |
2007 | set_tsk_need_resched(p); | |
2008 | return; | |
2009 | } | |
98fbc798 | 2010 | } |
bb44e5d1 IM |
2011 | } |
2012 | ||
83b699ed SV |
2013 | static void set_curr_task_rt(struct rq *rq) |
2014 | { | |
2015 | struct task_struct *p = rq->curr; | |
2016 | ||
78becc27 | 2017 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
2018 | |
2019 | /* The running task is never eligible for pushing */ | |
2020 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
2021 | } |
2022 | ||
6d686f45 | 2023 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
0d721cea PW |
2024 | { |
2025 | /* | |
2026 | * Time slice is 0 for SCHED_FIFO tasks | |
2027 | */ | |
2028 | if (task->policy == SCHED_RR) | |
ce0dbbbb | 2029 | return sched_rr_timeslice; |
0d721cea PW |
2030 | else |
2031 | return 0; | |
2032 | } | |
2033 | ||
029632fb | 2034 | const struct sched_class rt_sched_class = { |
5522d5d5 | 2035 | .next = &fair_sched_class, |
bb44e5d1 IM |
2036 | .enqueue_task = enqueue_task_rt, |
2037 | .dequeue_task = dequeue_task_rt, | |
2038 | .yield_task = yield_task_rt, | |
2039 | ||
2040 | .check_preempt_curr = check_preempt_curr_rt, | |
2041 | ||
2042 | .pick_next_task = pick_next_task_rt, | |
2043 | .put_prev_task = put_prev_task_rt, | |
2044 | ||
681f3e68 | 2045 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
2046 | .select_task_rq = select_task_rq_rt, |
2047 | ||
73fe6aae | 2048 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
2049 | .rq_online = rq_online_rt, |
2050 | .rq_offline = rq_offline_rt, | |
9a897c5a | 2051 | .post_schedule = post_schedule_rt, |
efbbd05a | 2052 | .task_woken = task_woken_rt, |
cb469845 | 2053 | .switched_from = switched_from_rt, |
681f3e68 | 2054 | #endif |
bb44e5d1 | 2055 | |
83b699ed | 2056 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 2057 | .task_tick = task_tick_rt, |
cb469845 | 2058 | |
0d721cea PW |
2059 | .get_rr_interval = get_rr_interval_rt, |
2060 | ||
cb469845 SR |
2061 | .prio_changed = prio_changed_rt, |
2062 | .switched_to = switched_to_rt, | |
bb44e5d1 | 2063 | }; |
ada18de2 PZ |
2064 | |
2065 | #ifdef CONFIG_SCHED_DEBUG | |
2066 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2067 | ||
029632fb | 2068 | void print_rt_stats(struct seq_file *m, int cpu) |
ada18de2 | 2069 | { |
ec514c48 | 2070 | rt_rq_iter_t iter; |
ada18de2 PZ |
2071 | struct rt_rq *rt_rq; |
2072 | ||
2073 | rcu_read_lock(); | |
ec514c48 | 2074 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
ada18de2 PZ |
2075 | print_rt_rq(m, cpu, rt_rq); |
2076 | rcu_read_unlock(); | |
2077 | } | |
55e12e5e | 2078 | #endif /* CONFIG_SCHED_DEBUG */ |