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> | |
b6366f04 | 9 | #include <linux/irq_work.h> |
029632fb | 10 | |
ce0dbbbb CW |
11 | int sched_rr_timeslice = RR_TIMESLICE; |
12 | ||
029632fb PZ |
13 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); |
14 | ||
15 | struct rt_bandwidth def_rt_bandwidth; | |
16 | ||
17 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
18 | { | |
19 | struct rt_bandwidth *rt_b = | |
20 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
029632fb | 21 | int idle = 0; |
77a4d1a1 | 22 | int overrun; |
029632fb | 23 | |
77a4d1a1 | 24 | raw_spin_lock(&rt_b->rt_runtime_lock); |
029632fb | 25 | for (;;) { |
77a4d1a1 | 26 | overrun = hrtimer_forward_now(timer, rt_b->rt_period); |
029632fb PZ |
27 | if (!overrun) |
28 | break; | |
29 | ||
77a4d1a1 | 30 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
029632fb | 31 | idle = do_sched_rt_period_timer(rt_b, overrun); |
77a4d1a1 | 32 | raw_spin_lock(&rt_b->rt_runtime_lock); |
029632fb | 33 | } |
4cfafd30 PZ |
34 | if (idle) |
35 | rt_b->rt_period_active = 0; | |
77a4d1a1 | 36 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
029632fb PZ |
37 | |
38 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
39 | } | |
40 | ||
41 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
42 | { | |
43 | rt_b->rt_period = ns_to_ktime(period); | |
44 | rt_b->rt_runtime = runtime; | |
45 | ||
46 | raw_spin_lock_init(&rt_b->rt_runtime_lock); | |
47 | ||
48 | hrtimer_init(&rt_b->rt_period_timer, | |
49 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
50 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
51 | } | |
52 | ||
53 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
54 | { | |
55 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) | |
56 | return; | |
57 | ||
029632fb | 58 | raw_spin_lock(&rt_b->rt_runtime_lock); |
4cfafd30 PZ |
59 | if (!rt_b->rt_period_active) { |
60 | rt_b->rt_period_active = 1; | |
61 | hrtimer_forward_now(&rt_b->rt_period_timer, rt_b->rt_period); | |
62 | hrtimer_start_expires(&rt_b->rt_period_timer, HRTIMER_MODE_ABS_PINNED); | |
63 | } | |
029632fb PZ |
64 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
65 | } | |
66 | ||
b6366f04 SR |
67 | #ifdef CONFIG_SMP |
68 | static void push_irq_work_func(struct irq_work *work); | |
69 | #endif | |
70 | ||
07c54f7a | 71 | void init_rt_rq(struct rt_rq *rt_rq) |
029632fb PZ |
72 | { |
73 | struct rt_prio_array *array; | |
74 | int i; | |
75 | ||
76 | array = &rt_rq->active; | |
77 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
78 | INIT_LIST_HEAD(array->queue + i); | |
79 | __clear_bit(i, array->bitmap); | |
80 | } | |
81 | /* delimiter for bitsearch: */ | |
82 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
83 | ||
84 | #if defined CONFIG_SMP | |
85 | rt_rq->highest_prio.curr = MAX_RT_PRIO; | |
86 | rt_rq->highest_prio.next = MAX_RT_PRIO; | |
87 | rt_rq->rt_nr_migratory = 0; | |
88 | rt_rq->overloaded = 0; | |
89 | plist_head_init(&rt_rq->pushable_tasks); | |
b6366f04 SR |
90 | |
91 | #ifdef HAVE_RT_PUSH_IPI | |
92 | rt_rq->push_flags = 0; | |
93 | rt_rq->push_cpu = nr_cpu_ids; | |
94 | raw_spin_lock_init(&rt_rq->push_lock); | |
95 | init_irq_work(&rt_rq->push_work, push_irq_work_func); | |
029632fb | 96 | #endif |
b6366f04 | 97 | #endif /* CONFIG_SMP */ |
f4ebcbc0 KT |
98 | /* We start is dequeued state, because no RT tasks are queued */ |
99 | rt_rq->rt_queued = 0; | |
029632fb PZ |
100 | |
101 | rt_rq->rt_time = 0; | |
102 | rt_rq->rt_throttled = 0; | |
103 | rt_rq->rt_runtime = 0; | |
104 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); | |
105 | } | |
106 | ||
8f48894f | 107 | #ifdef CONFIG_RT_GROUP_SCHED |
029632fb PZ |
108 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) |
109 | { | |
110 | hrtimer_cancel(&rt_b->rt_period_timer); | |
111 | } | |
8f48894f PZ |
112 | |
113 | #define rt_entity_is_task(rt_se) (!(rt_se)->my_q) | |
114 | ||
398a153b GH |
115 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
116 | { | |
8f48894f PZ |
117 | #ifdef CONFIG_SCHED_DEBUG |
118 | WARN_ON_ONCE(!rt_entity_is_task(rt_se)); | |
119 | #endif | |
398a153b GH |
120 | return container_of(rt_se, struct task_struct, rt); |
121 | } | |
122 | ||
398a153b GH |
123 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
124 | { | |
125 | return rt_rq->rq; | |
126 | } | |
127 | ||
128 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
129 | { | |
130 | return rt_se->rt_rq; | |
131 | } | |
132 | ||
653d07a6 KT |
133 | static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se) |
134 | { | |
135 | struct rt_rq *rt_rq = rt_se->rt_rq; | |
136 | ||
137 | return rt_rq->rq; | |
138 | } | |
139 | ||
029632fb PZ |
140 | void free_rt_sched_group(struct task_group *tg) |
141 | { | |
142 | int i; | |
143 | ||
144 | if (tg->rt_se) | |
145 | destroy_rt_bandwidth(&tg->rt_bandwidth); | |
146 | ||
147 | for_each_possible_cpu(i) { | |
148 | if (tg->rt_rq) | |
149 | kfree(tg->rt_rq[i]); | |
150 | if (tg->rt_se) | |
151 | kfree(tg->rt_se[i]); | |
152 | } | |
153 | ||
154 | kfree(tg->rt_rq); | |
155 | kfree(tg->rt_se); | |
156 | } | |
157 | ||
158 | void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
159 | struct sched_rt_entity *rt_se, int cpu, | |
160 | struct sched_rt_entity *parent) | |
161 | { | |
162 | struct rq *rq = cpu_rq(cpu); | |
163 | ||
164 | rt_rq->highest_prio.curr = MAX_RT_PRIO; | |
165 | rt_rq->rt_nr_boosted = 0; | |
166 | rt_rq->rq = rq; | |
167 | rt_rq->tg = tg; | |
168 | ||
169 | tg->rt_rq[cpu] = rt_rq; | |
170 | tg->rt_se[cpu] = rt_se; | |
171 | ||
172 | if (!rt_se) | |
173 | return; | |
174 | ||
175 | if (!parent) | |
176 | rt_se->rt_rq = &rq->rt; | |
177 | else | |
178 | rt_se->rt_rq = parent->my_q; | |
179 | ||
180 | rt_se->my_q = rt_rq; | |
181 | rt_se->parent = parent; | |
182 | INIT_LIST_HEAD(&rt_se->run_list); | |
183 | } | |
184 | ||
185 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
186 | { | |
187 | struct rt_rq *rt_rq; | |
188 | struct sched_rt_entity *rt_se; | |
189 | int i; | |
190 | ||
191 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); | |
192 | if (!tg->rt_rq) | |
193 | goto err; | |
194 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); | |
195 | if (!tg->rt_se) | |
196 | goto err; | |
197 | ||
198 | init_rt_bandwidth(&tg->rt_bandwidth, | |
199 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
200 | ||
201 | for_each_possible_cpu(i) { | |
202 | rt_rq = kzalloc_node(sizeof(struct rt_rq), | |
203 | GFP_KERNEL, cpu_to_node(i)); | |
204 | if (!rt_rq) | |
205 | goto err; | |
206 | ||
207 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), | |
208 | GFP_KERNEL, cpu_to_node(i)); | |
209 | if (!rt_se) | |
210 | goto err_free_rq; | |
211 | ||
07c54f7a | 212 | init_rt_rq(rt_rq); |
029632fb PZ |
213 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
214 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); | |
215 | } | |
216 | ||
217 | return 1; | |
218 | ||
219 | err_free_rq: | |
220 | kfree(rt_rq); | |
221 | err: | |
222 | return 0; | |
223 | } | |
224 | ||
398a153b GH |
225 | #else /* CONFIG_RT_GROUP_SCHED */ |
226 | ||
a1ba4d8b PZ |
227 | #define rt_entity_is_task(rt_se) (1) |
228 | ||
8f48894f PZ |
229 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
230 | { | |
231 | return container_of(rt_se, struct task_struct, rt); | |
232 | } | |
233 | ||
398a153b GH |
234 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
235 | { | |
236 | return container_of(rt_rq, struct rq, rt); | |
237 | } | |
238 | ||
653d07a6 | 239 | static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se) |
398a153b GH |
240 | { |
241 | struct task_struct *p = rt_task_of(rt_se); | |
653d07a6 KT |
242 | |
243 | return task_rq(p); | |
244 | } | |
245 | ||
246 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
247 | { | |
248 | struct rq *rq = rq_of_rt_se(rt_se); | |
398a153b GH |
249 | |
250 | return &rq->rt; | |
251 | } | |
252 | ||
029632fb PZ |
253 | void free_rt_sched_group(struct task_group *tg) { } |
254 | ||
255 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
256 | { | |
257 | return 1; | |
258 | } | |
398a153b GH |
259 | #endif /* CONFIG_RT_GROUP_SCHED */ |
260 | ||
4fd29176 | 261 | #ifdef CONFIG_SMP |
84de4274 | 262 | |
8046d680 | 263 | static void pull_rt_task(struct rq *this_rq); |
38033c37 | 264 | |
dc877341 PZ |
265 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
266 | { | |
267 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
268 | return rq->rt.highest_prio.curr > prev->prio; | |
269 | } | |
270 | ||
637f5085 | 271 | static inline int rt_overloaded(struct rq *rq) |
4fd29176 | 272 | { |
637f5085 | 273 | return atomic_read(&rq->rd->rto_count); |
4fd29176 | 274 | } |
84de4274 | 275 | |
4fd29176 SR |
276 | static inline void rt_set_overload(struct rq *rq) |
277 | { | |
1f11eb6a GH |
278 | if (!rq->online) |
279 | return; | |
280 | ||
c6c4927b | 281 | cpumask_set_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 SR |
282 | /* |
283 | * Make sure the mask is visible before we set | |
284 | * the overload count. That is checked to determine | |
285 | * if we should look at the mask. It would be a shame | |
286 | * if we looked at the mask, but the mask was not | |
287 | * updated yet. | |
7c3f2ab7 PZ |
288 | * |
289 | * Matched by the barrier in pull_rt_task(). | |
4fd29176 | 290 | */ |
7c3f2ab7 | 291 | smp_wmb(); |
637f5085 | 292 | atomic_inc(&rq->rd->rto_count); |
4fd29176 | 293 | } |
84de4274 | 294 | |
4fd29176 SR |
295 | static inline void rt_clear_overload(struct rq *rq) |
296 | { | |
1f11eb6a GH |
297 | if (!rq->online) |
298 | return; | |
299 | ||
4fd29176 | 300 | /* the order here really doesn't matter */ |
637f5085 | 301 | atomic_dec(&rq->rd->rto_count); |
c6c4927b | 302 | cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 | 303 | } |
73fe6aae | 304 | |
398a153b | 305 | static void update_rt_migration(struct rt_rq *rt_rq) |
73fe6aae | 306 | { |
a1ba4d8b | 307 | if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) { |
398a153b GH |
308 | if (!rt_rq->overloaded) { |
309 | rt_set_overload(rq_of_rt_rq(rt_rq)); | |
310 | rt_rq->overloaded = 1; | |
cdc8eb98 | 311 | } |
398a153b GH |
312 | } else if (rt_rq->overloaded) { |
313 | rt_clear_overload(rq_of_rt_rq(rt_rq)); | |
314 | rt_rq->overloaded = 0; | |
637f5085 | 315 | } |
73fe6aae | 316 | } |
4fd29176 | 317 | |
398a153b GH |
318 | static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
319 | { | |
29baa747 PZ |
320 | struct task_struct *p; |
321 | ||
a1ba4d8b PZ |
322 | if (!rt_entity_is_task(rt_se)) |
323 | return; | |
324 | ||
29baa747 | 325 | p = rt_task_of(rt_se); |
a1ba4d8b PZ |
326 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
327 | ||
328 | rt_rq->rt_nr_total++; | |
29baa747 | 329 | if (p->nr_cpus_allowed > 1) |
398a153b GH |
330 | rt_rq->rt_nr_migratory++; |
331 | ||
332 | update_rt_migration(rt_rq); | |
333 | } | |
334 | ||
335 | static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
336 | { | |
29baa747 PZ |
337 | struct task_struct *p; |
338 | ||
a1ba4d8b PZ |
339 | if (!rt_entity_is_task(rt_se)) |
340 | return; | |
341 | ||
29baa747 | 342 | p = rt_task_of(rt_se); |
a1ba4d8b PZ |
343 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
344 | ||
345 | rt_rq->rt_nr_total--; | |
29baa747 | 346 | if (p->nr_cpus_allowed > 1) |
398a153b GH |
347 | rt_rq->rt_nr_migratory--; |
348 | ||
349 | update_rt_migration(rt_rq); | |
350 | } | |
351 | ||
5181f4a4 SR |
352 | static inline int has_pushable_tasks(struct rq *rq) |
353 | { | |
354 | return !plist_head_empty(&rq->rt.pushable_tasks); | |
355 | } | |
356 | ||
fd7a4bed PZ |
357 | static DEFINE_PER_CPU(struct callback_head, rt_push_head); |
358 | static DEFINE_PER_CPU(struct callback_head, rt_pull_head); | |
e3fca9e7 PZ |
359 | |
360 | static void push_rt_tasks(struct rq *); | |
fd7a4bed | 361 | static void pull_rt_task(struct rq *); |
e3fca9e7 PZ |
362 | |
363 | static inline void queue_push_tasks(struct rq *rq) | |
dc877341 | 364 | { |
e3fca9e7 PZ |
365 | if (!has_pushable_tasks(rq)) |
366 | return; | |
367 | ||
fd7a4bed PZ |
368 | queue_balance_callback(rq, &per_cpu(rt_push_head, rq->cpu), push_rt_tasks); |
369 | } | |
370 | ||
371 | static inline void queue_pull_task(struct rq *rq) | |
372 | { | |
373 | queue_balance_callback(rq, &per_cpu(rt_pull_head, rq->cpu), pull_rt_task); | |
dc877341 PZ |
374 | } |
375 | ||
917b627d GH |
376 | static void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
377 | { | |
378 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
379 | plist_node_init(&p->pushable_tasks, p->prio); | |
380 | plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
5181f4a4 SR |
381 | |
382 | /* Update the highest prio pushable task */ | |
383 | if (p->prio < rq->rt.highest_prio.next) | |
384 | rq->rt.highest_prio.next = p->prio; | |
917b627d GH |
385 | } |
386 | ||
387 | static void dequeue_pushable_task(struct rq *rq, struct task_struct *p) | |
388 | { | |
389 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
917b627d | 390 | |
5181f4a4 SR |
391 | /* Update the new highest prio pushable task */ |
392 | if (has_pushable_tasks(rq)) { | |
393 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
394 | struct task_struct, pushable_tasks); | |
395 | rq->rt.highest_prio.next = p->prio; | |
396 | } else | |
397 | rq->rt.highest_prio.next = MAX_RT_PRIO; | |
bcf08df3 IM |
398 | } |
399 | ||
917b627d GH |
400 | #else |
401 | ||
ceacc2c1 | 402 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
fa85ae24 | 403 | { |
6f505b16 PZ |
404 | } |
405 | ||
ceacc2c1 PZ |
406 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
407 | { | |
408 | } | |
409 | ||
b07430ac | 410 | static inline |
ceacc2c1 PZ |
411 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
412 | { | |
413 | } | |
414 | ||
398a153b | 415 | static inline |
ceacc2c1 PZ |
416 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
417 | { | |
418 | } | |
917b627d | 419 | |
dc877341 PZ |
420 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
421 | { | |
422 | return false; | |
423 | } | |
424 | ||
8046d680 | 425 | static inline void pull_rt_task(struct rq *this_rq) |
dc877341 | 426 | { |
dc877341 PZ |
427 | } |
428 | ||
e3fca9e7 | 429 | static inline void queue_push_tasks(struct rq *rq) |
dc877341 PZ |
430 | { |
431 | } | |
4fd29176 SR |
432 | #endif /* CONFIG_SMP */ |
433 | ||
f4ebcbc0 KT |
434 | static void enqueue_top_rt_rq(struct rt_rq *rt_rq); |
435 | static void dequeue_top_rt_rq(struct rt_rq *rt_rq); | |
436 | ||
6f505b16 PZ |
437 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
438 | { | |
439 | return !list_empty(&rt_se->run_list); | |
440 | } | |
441 | ||
052f1dc7 | 442 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 443 | |
9f0c1e56 | 444 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
445 | { |
446 | if (!rt_rq->tg) | |
9f0c1e56 | 447 | return RUNTIME_INF; |
6f505b16 | 448 | |
ac086bc2 PZ |
449 | return rt_rq->rt_runtime; |
450 | } | |
451 | ||
452 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
453 | { | |
454 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); | |
6f505b16 PZ |
455 | } |
456 | ||
ec514c48 CX |
457 | typedef struct task_group *rt_rq_iter_t; |
458 | ||
1c09ab0d YZ |
459 | static inline struct task_group *next_task_group(struct task_group *tg) |
460 | { | |
461 | do { | |
462 | tg = list_entry_rcu(tg->list.next, | |
463 | typeof(struct task_group), list); | |
464 | } while (&tg->list != &task_groups && task_group_is_autogroup(tg)); | |
465 | ||
466 | if (&tg->list == &task_groups) | |
467 | tg = NULL; | |
468 | ||
469 | return tg; | |
470 | } | |
471 | ||
472 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
473 | for (iter = container_of(&task_groups, typeof(*iter), list); \ | |
474 | (iter = next_task_group(iter)) && \ | |
475 | (rt_rq = iter->rt_rq[cpu_of(rq)]);) | |
ec514c48 | 476 | |
6f505b16 PZ |
477 | #define for_each_sched_rt_entity(rt_se) \ |
478 | for (; rt_se; rt_se = rt_se->parent) | |
479 | ||
480 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
481 | { | |
482 | return rt_se->my_q; | |
483 | } | |
484 | ||
37dad3fc | 485 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
6f505b16 PZ |
486 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
487 | ||
9f0c1e56 | 488 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 489 | { |
f6121f4f | 490 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
8875125e | 491 | struct rq *rq = rq_of_rt_rq(rt_rq); |
74b7eb58 YZ |
492 | struct sched_rt_entity *rt_se; |
493 | ||
8875125e | 494 | int cpu = cpu_of(rq); |
0c3b9168 BS |
495 | |
496 | rt_se = rt_rq->tg->rt_se[cpu]; | |
6f505b16 | 497 | |
f6121f4f | 498 | if (rt_rq->rt_nr_running) { |
f4ebcbc0 KT |
499 | if (!rt_se) |
500 | enqueue_top_rt_rq(rt_rq); | |
501 | else if (!on_rt_rq(rt_se)) | |
37dad3fc | 502 | enqueue_rt_entity(rt_se, false); |
f4ebcbc0 | 503 | |
e864c499 | 504 | if (rt_rq->highest_prio.curr < curr->prio) |
8875125e | 505 | resched_curr(rq); |
6f505b16 PZ |
506 | } |
507 | } | |
508 | ||
9f0c1e56 | 509 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 510 | { |
74b7eb58 | 511 | struct sched_rt_entity *rt_se; |
0c3b9168 | 512 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
74b7eb58 | 513 | |
0c3b9168 | 514 | rt_se = rt_rq->tg->rt_se[cpu]; |
6f505b16 | 515 | |
f4ebcbc0 KT |
516 | if (!rt_se) |
517 | dequeue_top_rt_rq(rt_rq); | |
518 | else if (on_rt_rq(rt_se)) | |
6f505b16 PZ |
519 | dequeue_rt_entity(rt_se); |
520 | } | |
521 | ||
46383648 KT |
522 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
523 | { | |
524 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
525 | } | |
526 | ||
23b0fdfc PZ |
527 | static int rt_se_boosted(struct sched_rt_entity *rt_se) |
528 | { | |
529 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
530 | struct task_struct *p; | |
531 | ||
532 | if (rt_rq) | |
533 | return !!rt_rq->rt_nr_boosted; | |
534 | ||
535 | p = rt_task_of(rt_se); | |
536 | return p->prio != p->normal_prio; | |
537 | } | |
538 | ||
d0b27fa7 | 539 | #ifdef CONFIG_SMP |
c6c4927b | 540 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 541 | { |
424c93fe | 542 | return this_rq()->rd->span; |
d0b27fa7 | 543 | } |
6f505b16 | 544 | #else |
c6c4927b | 545 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 546 | { |
c6c4927b | 547 | return cpu_online_mask; |
d0b27fa7 PZ |
548 | } |
549 | #endif | |
6f505b16 | 550 | |
d0b27fa7 PZ |
551 | static inline |
552 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 553 | { |
d0b27fa7 PZ |
554 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
555 | } | |
9f0c1e56 | 556 | |
ac086bc2 PZ |
557 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
558 | { | |
559 | return &rt_rq->tg->rt_bandwidth; | |
560 | } | |
561 | ||
55e12e5e | 562 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
563 | |
564 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
565 | { | |
ac086bc2 PZ |
566 | return rt_rq->rt_runtime; |
567 | } | |
568 | ||
569 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
570 | { | |
571 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
572 | } |
573 | ||
ec514c48 CX |
574 | typedef struct rt_rq *rt_rq_iter_t; |
575 | ||
576 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
577 | for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
578 | ||
6f505b16 PZ |
579 | #define for_each_sched_rt_entity(rt_se) \ |
580 | for (; rt_se; rt_se = NULL) | |
581 | ||
582 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
583 | { | |
584 | return NULL; | |
585 | } | |
586 | ||
9f0c1e56 | 587 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 588 | { |
f4ebcbc0 KT |
589 | struct rq *rq = rq_of_rt_rq(rt_rq); |
590 | ||
591 | if (!rt_rq->rt_nr_running) | |
592 | return; | |
593 | ||
594 | enqueue_top_rt_rq(rt_rq); | |
8875125e | 595 | resched_curr(rq); |
6f505b16 PZ |
596 | } |
597 | ||
9f0c1e56 | 598 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 599 | { |
f4ebcbc0 | 600 | dequeue_top_rt_rq(rt_rq); |
6f505b16 PZ |
601 | } |
602 | ||
46383648 KT |
603 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
604 | { | |
605 | return rt_rq->rt_throttled; | |
606 | } | |
607 | ||
c6c4927b | 608 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 609 | { |
c6c4927b | 610 | return cpu_online_mask; |
d0b27fa7 PZ |
611 | } |
612 | ||
613 | static inline | |
614 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
615 | { | |
616 | return &cpu_rq(cpu)->rt; | |
617 | } | |
618 | ||
ac086bc2 PZ |
619 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
620 | { | |
621 | return &def_rt_bandwidth; | |
622 | } | |
623 | ||
55e12e5e | 624 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 625 | |
faa59937 JL |
626 | bool sched_rt_bandwidth_account(struct rt_rq *rt_rq) |
627 | { | |
628 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
629 | ||
630 | return (hrtimer_active(&rt_b->rt_period_timer) || | |
631 | rt_rq->rt_time < rt_b->rt_runtime); | |
632 | } | |
633 | ||
ac086bc2 | 634 | #ifdef CONFIG_SMP |
78333cdd PZ |
635 | /* |
636 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
637 | */ | |
b79f3833 | 638 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
639 | { |
640 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
aa7f6730 | 641 | struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd; |
ac086bc2 PZ |
642 | int i, weight, more = 0; |
643 | u64 rt_period; | |
644 | ||
c6c4927b | 645 | weight = cpumask_weight(rd->span); |
ac086bc2 | 646 | |
0986b11b | 647 | raw_spin_lock(&rt_b->rt_runtime_lock); |
ac086bc2 | 648 | rt_period = ktime_to_ns(rt_b->rt_period); |
c6c4927b | 649 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
650 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
651 | s64 diff; | |
652 | ||
653 | if (iter == rt_rq) | |
654 | continue; | |
655 | ||
0986b11b | 656 | raw_spin_lock(&iter->rt_runtime_lock); |
78333cdd PZ |
657 | /* |
658 | * Either all rqs have inf runtime and there's nothing to steal | |
659 | * or __disable_runtime() below sets a specific rq to inf to | |
660 | * indicate its been disabled and disalow stealing. | |
661 | */ | |
7def2be1 PZ |
662 | if (iter->rt_runtime == RUNTIME_INF) |
663 | goto next; | |
664 | ||
78333cdd PZ |
665 | /* |
666 | * From runqueues with spare time, take 1/n part of their | |
667 | * spare time, but no more than our period. | |
668 | */ | |
ac086bc2 PZ |
669 | diff = iter->rt_runtime - iter->rt_time; |
670 | if (diff > 0) { | |
58838cf3 | 671 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
672 | if (rt_rq->rt_runtime + diff > rt_period) |
673 | diff = rt_period - rt_rq->rt_runtime; | |
674 | iter->rt_runtime -= diff; | |
675 | rt_rq->rt_runtime += diff; | |
676 | more = 1; | |
677 | if (rt_rq->rt_runtime == rt_period) { | |
0986b11b | 678 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 PZ |
679 | break; |
680 | } | |
681 | } | |
7def2be1 | 682 | next: |
0986b11b | 683 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 | 684 | } |
0986b11b | 685 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
ac086bc2 PZ |
686 | |
687 | return more; | |
688 | } | |
7def2be1 | 689 | |
78333cdd PZ |
690 | /* |
691 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
692 | */ | |
7def2be1 PZ |
693 | static void __disable_runtime(struct rq *rq) |
694 | { | |
695 | struct root_domain *rd = rq->rd; | |
ec514c48 | 696 | rt_rq_iter_t iter; |
7def2be1 PZ |
697 | struct rt_rq *rt_rq; |
698 | ||
699 | if (unlikely(!scheduler_running)) | |
700 | return; | |
701 | ||
ec514c48 | 702 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
703 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
704 | s64 want; | |
705 | int i; | |
706 | ||
0986b11b TG |
707 | raw_spin_lock(&rt_b->rt_runtime_lock); |
708 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
709 | /* |
710 | * Either we're all inf and nobody needs to borrow, or we're | |
711 | * already disabled and thus have nothing to do, or we have | |
712 | * exactly the right amount of runtime to take out. | |
713 | */ | |
7def2be1 PZ |
714 | if (rt_rq->rt_runtime == RUNTIME_INF || |
715 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
716 | goto balanced; | |
0986b11b | 717 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
7def2be1 | 718 | |
78333cdd PZ |
719 | /* |
720 | * Calculate the difference between what we started out with | |
721 | * and what we current have, that's the amount of runtime | |
722 | * we lend and now have to reclaim. | |
723 | */ | |
7def2be1 PZ |
724 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
725 | ||
78333cdd PZ |
726 | /* |
727 | * Greedy reclaim, take back as much as we can. | |
728 | */ | |
c6c4927b | 729 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
730 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
731 | s64 diff; | |
732 | ||
78333cdd PZ |
733 | /* |
734 | * Can't reclaim from ourselves or disabled runqueues. | |
735 | */ | |
f1679d08 | 736 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
737 | continue; |
738 | ||
0986b11b | 739 | raw_spin_lock(&iter->rt_runtime_lock); |
7def2be1 PZ |
740 | if (want > 0) { |
741 | diff = min_t(s64, iter->rt_runtime, want); | |
742 | iter->rt_runtime -= diff; | |
743 | want -= diff; | |
744 | } else { | |
745 | iter->rt_runtime -= want; | |
746 | want -= want; | |
747 | } | |
0986b11b | 748 | raw_spin_unlock(&iter->rt_runtime_lock); |
7def2be1 PZ |
749 | |
750 | if (!want) | |
751 | break; | |
752 | } | |
753 | ||
0986b11b | 754 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
78333cdd PZ |
755 | /* |
756 | * We cannot be left wanting - that would mean some runtime | |
757 | * leaked out of the system. | |
758 | */ | |
7def2be1 PZ |
759 | BUG_ON(want); |
760 | balanced: | |
78333cdd PZ |
761 | /* |
762 | * Disable all the borrow logic by pretending we have inf | |
763 | * runtime - in which case borrowing doesn't make sense. | |
764 | */ | |
7def2be1 | 765 | rt_rq->rt_runtime = RUNTIME_INF; |
a4c96ae3 | 766 | rt_rq->rt_throttled = 0; |
0986b11b TG |
767 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
768 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
99b62567 KT |
769 | |
770 | /* Make rt_rq available for pick_next_task() */ | |
771 | sched_rt_rq_enqueue(rt_rq); | |
7def2be1 PZ |
772 | } |
773 | } | |
774 | ||
7def2be1 PZ |
775 | static void __enable_runtime(struct rq *rq) |
776 | { | |
ec514c48 | 777 | rt_rq_iter_t iter; |
7def2be1 PZ |
778 | struct rt_rq *rt_rq; |
779 | ||
780 | if (unlikely(!scheduler_running)) | |
781 | return; | |
782 | ||
78333cdd PZ |
783 | /* |
784 | * Reset each runqueue's bandwidth settings | |
785 | */ | |
ec514c48 | 786 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
787 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
788 | ||
0986b11b TG |
789 | raw_spin_lock(&rt_b->rt_runtime_lock); |
790 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
7def2be1 PZ |
791 | rt_rq->rt_runtime = rt_b->rt_runtime; |
792 | rt_rq->rt_time = 0; | |
baf25731 | 793 | rt_rq->rt_throttled = 0; |
0986b11b TG |
794 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
795 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
796 | } |
797 | } | |
798 | ||
eff6549b PZ |
799 | static int balance_runtime(struct rt_rq *rt_rq) |
800 | { | |
801 | int more = 0; | |
802 | ||
4a6184ce PZ |
803 | if (!sched_feat(RT_RUNTIME_SHARE)) |
804 | return more; | |
805 | ||
eff6549b | 806 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
0986b11b | 807 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
eff6549b | 808 | more = do_balance_runtime(rt_rq); |
0986b11b | 809 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
810 | } |
811 | ||
812 | return more; | |
813 | } | |
55e12e5e | 814 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
815 | static inline int balance_runtime(struct rt_rq *rt_rq) |
816 | { | |
817 | return 0; | |
818 | } | |
55e12e5e | 819 | #endif /* CONFIG_SMP */ |
ac086bc2 | 820 | |
eff6549b PZ |
821 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
822 | { | |
42c62a58 | 823 | int i, idle = 1, throttled = 0; |
c6c4927b | 824 | const struct cpumask *span; |
eff6549b | 825 | |
eff6549b | 826 | span = sched_rt_period_mask(); |
e221d028 MG |
827 | #ifdef CONFIG_RT_GROUP_SCHED |
828 | /* | |
829 | * FIXME: isolated CPUs should really leave the root task group, | |
830 | * whether they are isolcpus or were isolated via cpusets, lest | |
831 | * the timer run on a CPU which does not service all runqueues, | |
832 | * potentially leaving other CPUs indefinitely throttled. If | |
833 | * isolation is really required, the user will turn the throttle | |
834 | * off to kill the perturbations it causes anyway. Meanwhile, | |
835 | * this maintains functionality for boot and/or troubleshooting. | |
836 | */ | |
837 | if (rt_b == &root_task_group.rt_bandwidth) | |
838 | span = cpu_online_mask; | |
839 | #endif | |
c6c4927b | 840 | for_each_cpu(i, span) { |
eff6549b PZ |
841 | int enqueue = 0; |
842 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
843 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
844 | ||
05fa785c | 845 | raw_spin_lock(&rq->lock); |
eff6549b PZ |
846 | if (rt_rq->rt_time) { |
847 | u64 runtime; | |
848 | ||
0986b11b | 849 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
850 | if (rt_rq->rt_throttled) |
851 | balance_runtime(rt_rq); | |
852 | runtime = rt_rq->rt_runtime; | |
853 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
854 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
855 | rt_rq->rt_throttled = 0; | |
856 | enqueue = 1; | |
61eadef6 MG |
857 | |
858 | /* | |
9edfbfed PZ |
859 | * When we're idle and a woken (rt) task is |
860 | * throttled check_preempt_curr() will set | |
861 | * skip_update and the time between the wakeup | |
862 | * and this unthrottle will get accounted as | |
863 | * 'runtime'. | |
61eadef6 MG |
864 | */ |
865 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) | |
9edfbfed | 866 | rq_clock_skip_update(rq, false); |
eff6549b PZ |
867 | } |
868 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
869 | idle = 0; | |
0986b11b | 870 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
0c3b9168 | 871 | } else if (rt_rq->rt_nr_running) { |
6c3df255 | 872 | idle = 0; |
0c3b9168 BS |
873 | if (!rt_rq_throttled(rt_rq)) |
874 | enqueue = 1; | |
875 | } | |
42c62a58 PZ |
876 | if (rt_rq->rt_throttled) |
877 | throttled = 1; | |
eff6549b PZ |
878 | |
879 | if (enqueue) | |
880 | sched_rt_rq_enqueue(rt_rq); | |
05fa785c | 881 | raw_spin_unlock(&rq->lock); |
eff6549b PZ |
882 | } |
883 | ||
42c62a58 PZ |
884 | if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)) |
885 | return 1; | |
886 | ||
eff6549b PZ |
887 | return idle; |
888 | } | |
ac086bc2 | 889 | |
6f505b16 PZ |
890 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
891 | { | |
052f1dc7 | 892 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
893 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
894 | ||
895 | if (rt_rq) | |
e864c499 | 896 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
897 | #endif |
898 | ||
899 | return rt_task_of(rt_se)->prio; | |
900 | } | |
901 | ||
9f0c1e56 | 902 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 903 | { |
9f0c1e56 | 904 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 905 | |
fa85ae24 | 906 | if (rt_rq->rt_throttled) |
23b0fdfc | 907 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 908 | |
5b680fd6 | 909 | if (runtime >= sched_rt_period(rt_rq)) |
ac086bc2 PZ |
910 | return 0; |
911 | ||
b79f3833 PZ |
912 | balance_runtime(rt_rq); |
913 | runtime = sched_rt_runtime(rt_rq); | |
914 | if (runtime == RUNTIME_INF) | |
915 | return 0; | |
ac086bc2 | 916 | |
9f0c1e56 | 917 | if (rt_rq->rt_time > runtime) { |
7abc63b1 PZ |
918 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
919 | ||
920 | /* | |
921 | * Don't actually throttle groups that have no runtime assigned | |
922 | * but accrue some time due to boosting. | |
923 | */ | |
924 | if (likely(rt_b->rt_runtime)) { | |
925 | rt_rq->rt_throttled = 1; | |
c224815d | 926 | printk_deferred_once("sched: RT throttling activated\n"); |
7abc63b1 PZ |
927 | } else { |
928 | /* | |
929 | * In case we did anyway, make it go away, | |
930 | * replenishment is a joke, since it will replenish us | |
931 | * with exactly 0 ns. | |
932 | */ | |
933 | rt_rq->rt_time = 0; | |
934 | } | |
935 | ||
23b0fdfc | 936 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 937 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
938 | return 1; |
939 | } | |
fa85ae24 PZ |
940 | } |
941 | ||
942 | return 0; | |
943 | } | |
944 | ||
bb44e5d1 IM |
945 | /* |
946 | * Update the current task's runtime statistics. Skip current tasks that | |
947 | * are not in our scheduling class. | |
948 | */ | |
a9957449 | 949 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
950 | { |
951 | struct task_struct *curr = rq->curr; | |
6f505b16 | 952 | struct sched_rt_entity *rt_se = &curr->rt; |
bb44e5d1 IM |
953 | u64 delta_exec; |
954 | ||
06c3bc65 | 955 | if (curr->sched_class != &rt_sched_class) |
bb44e5d1 IM |
956 | return; |
957 | ||
78becc27 | 958 | delta_exec = rq_clock_task(rq) - curr->se.exec_start; |
fc79e240 KT |
959 | if (unlikely((s64)delta_exec <= 0)) |
960 | return; | |
6cfb0d5d | 961 | |
42c62a58 PZ |
962 | schedstat_set(curr->se.statistics.exec_max, |
963 | max(curr->se.statistics.exec_max, delta_exec)); | |
bb44e5d1 IM |
964 | |
965 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
966 | account_group_exec_runtime(curr, delta_exec); |
967 | ||
78becc27 | 968 | curr->se.exec_start = rq_clock_task(rq); |
d842de87 | 969 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 970 | |
e9e9250b PZ |
971 | sched_rt_avg_update(rq, delta_exec); |
972 | ||
0b148fa0 PZ |
973 | if (!rt_bandwidth_enabled()) |
974 | return; | |
975 | ||
354d60c2 | 976 | for_each_sched_rt_entity(rt_se) { |
0b07939c | 977 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
354d60c2 | 978 | |
cc2991cf | 979 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
0986b11b | 980 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
981 | rt_rq->rt_time += delta_exec; |
982 | if (sched_rt_runtime_exceeded(rt_rq)) | |
8875125e | 983 | resched_curr(rq); |
0986b11b | 984 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 985 | } |
354d60c2 | 986 | } |
bb44e5d1 IM |
987 | } |
988 | ||
f4ebcbc0 KT |
989 | static void |
990 | dequeue_top_rt_rq(struct rt_rq *rt_rq) | |
991 | { | |
992 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
993 | ||
994 | BUG_ON(&rq->rt != rt_rq); | |
995 | ||
996 | if (!rt_rq->rt_queued) | |
997 | return; | |
998 | ||
999 | BUG_ON(!rq->nr_running); | |
1000 | ||
72465447 | 1001 | sub_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
1002 | rt_rq->rt_queued = 0; |
1003 | } | |
1004 | ||
1005 | static void | |
1006 | enqueue_top_rt_rq(struct rt_rq *rt_rq) | |
1007 | { | |
1008 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
1009 | ||
1010 | BUG_ON(&rq->rt != rt_rq); | |
1011 | ||
1012 | if (rt_rq->rt_queued) | |
1013 | return; | |
1014 | if (rt_rq_throttled(rt_rq) || !rt_rq->rt_nr_running) | |
1015 | return; | |
1016 | ||
72465447 | 1017 | add_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
1018 | rt_rq->rt_queued = 1; |
1019 | } | |
1020 | ||
398a153b | 1021 | #if defined CONFIG_SMP |
e864c499 | 1022 | |
398a153b GH |
1023 | static void |
1024 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 1025 | { |
4d984277 | 1026 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 1027 | |
757dfcaa KT |
1028 | #ifdef CONFIG_RT_GROUP_SCHED |
1029 | /* | |
1030 | * Change rq's cpupri only if rt_rq is the top queue. | |
1031 | */ | |
1032 | if (&rq->rt != rt_rq) | |
1033 | return; | |
1034 | #endif | |
5181f4a4 SR |
1035 | if (rq->online && prio < prev_prio) |
1036 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); | |
398a153b | 1037 | } |
73fe6aae | 1038 | |
398a153b GH |
1039 | static void |
1040 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
1041 | { | |
1042 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 1043 | |
757dfcaa KT |
1044 | #ifdef CONFIG_RT_GROUP_SCHED |
1045 | /* | |
1046 | * Change rq's cpupri only if rt_rq is the top queue. | |
1047 | */ | |
1048 | if (&rq->rt != rt_rq) | |
1049 | return; | |
1050 | #endif | |
398a153b GH |
1051 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
1052 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
1053 | } |
1054 | ||
398a153b GH |
1055 | #else /* CONFIG_SMP */ |
1056 | ||
6f505b16 | 1057 | static inline |
398a153b GH |
1058 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
1059 | static inline | |
1060 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
1061 | ||
1062 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 1063 | |
052f1dc7 | 1064 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
1065 | static void |
1066 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
1067 | { | |
1068 | int prev_prio = rt_rq->highest_prio.curr; | |
1069 | ||
1070 | if (prio < prev_prio) | |
1071 | rt_rq->highest_prio.curr = prio; | |
1072 | ||
1073 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
1074 | } | |
1075 | ||
1076 | static void | |
1077 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
1078 | { | |
1079 | int prev_prio = rt_rq->highest_prio.curr; | |
1080 | ||
6f505b16 | 1081 | if (rt_rq->rt_nr_running) { |
764a9d6f | 1082 | |
398a153b | 1083 | WARN_ON(prio < prev_prio); |
764a9d6f | 1084 | |
e864c499 | 1085 | /* |
398a153b GH |
1086 | * This may have been our highest task, and therefore |
1087 | * we may have some recomputation to do | |
e864c499 | 1088 | */ |
398a153b | 1089 | if (prio == prev_prio) { |
e864c499 GH |
1090 | struct rt_prio_array *array = &rt_rq->active; |
1091 | ||
1092 | rt_rq->highest_prio.curr = | |
764a9d6f | 1093 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
1094 | } |
1095 | ||
764a9d6f | 1096 | } else |
e864c499 | 1097 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 1098 | |
398a153b GH |
1099 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
1100 | } | |
1f11eb6a | 1101 | |
398a153b GH |
1102 | #else |
1103 | ||
1104 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1105 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1106 | ||
1107 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 1108 | |
052f1dc7 | 1109 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
1110 | |
1111 | static void | |
1112 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1113 | { | |
1114 | if (rt_se_boosted(rt_se)) | |
1115 | rt_rq->rt_nr_boosted++; | |
1116 | ||
1117 | if (rt_rq->tg) | |
1118 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
1119 | } | |
1120 | ||
1121 | static void | |
1122 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1123 | { | |
23b0fdfc PZ |
1124 | if (rt_se_boosted(rt_se)) |
1125 | rt_rq->rt_nr_boosted--; | |
1126 | ||
1127 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
1128 | } |
1129 | ||
1130 | #else /* CONFIG_RT_GROUP_SCHED */ | |
1131 | ||
1132 | static void | |
1133 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1134 | { | |
1135 | start_rt_bandwidth(&def_rt_bandwidth); | |
1136 | } | |
1137 | ||
1138 | static inline | |
1139 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
1140 | ||
1141 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
1142 | ||
22abdef3 KT |
1143 | static inline |
1144 | unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se) | |
1145 | { | |
1146 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
1147 | ||
1148 | if (group_rq) | |
1149 | return group_rq->rt_nr_running; | |
1150 | else | |
1151 | return 1; | |
1152 | } | |
1153 | ||
398a153b GH |
1154 | static inline |
1155 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1156 | { | |
1157 | int prio = rt_se_prio(rt_se); | |
1158 | ||
1159 | WARN_ON(!rt_prio(prio)); | |
22abdef3 | 1160 | rt_rq->rt_nr_running += rt_se_nr_running(rt_se); |
398a153b GH |
1161 | |
1162 | inc_rt_prio(rt_rq, prio); | |
1163 | inc_rt_migration(rt_se, rt_rq); | |
1164 | inc_rt_group(rt_se, rt_rq); | |
1165 | } | |
1166 | ||
1167 | static inline | |
1168 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1169 | { | |
1170 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
1171 | WARN_ON(!rt_rq->rt_nr_running); | |
22abdef3 | 1172 | rt_rq->rt_nr_running -= rt_se_nr_running(rt_se); |
398a153b GH |
1173 | |
1174 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
1175 | dec_rt_migration(rt_se, rt_rq); | |
1176 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
1177 | } |
1178 | ||
37dad3fc | 1179 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
bb44e5d1 | 1180 | { |
6f505b16 PZ |
1181 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
1182 | struct rt_prio_array *array = &rt_rq->active; | |
1183 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 1184 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 1185 | |
ad2a3f13 PZ |
1186 | /* |
1187 | * Don't enqueue the group if its throttled, or when empty. | |
1188 | * The latter is a consequence of the former when a child group | |
1189 | * get throttled and the current group doesn't have any other | |
1190 | * active members. | |
1191 | */ | |
1192 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 1193 | return; |
63489e45 | 1194 | |
37dad3fc TG |
1195 | if (head) |
1196 | list_add(&rt_se->run_list, queue); | |
1197 | else | |
1198 | list_add_tail(&rt_se->run_list, queue); | |
6f505b16 | 1199 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 1200 | |
6f505b16 PZ |
1201 | inc_rt_tasks(rt_se, rt_rq); |
1202 | } | |
1203 | ||
ad2a3f13 | 1204 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
1205 | { |
1206 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
1207 | struct rt_prio_array *array = &rt_rq->active; | |
1208 | ||
1209 | list_del_init(&rt_se->run_list); | |
1210 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
1211 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
1212 | ||
1213 | dec_rt_tasks(rt_se, rt_rq); | |
1214 | } | |
1215 | ||
1216 | /* | |
1217 | * Because the prio of an upper entry depends on the lower | |
1218 | * entries, we must remove entries top - down. | |
6f505b16 | 1219 | */ |
ad2a3f13 | 1220 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 1221 | { |
ad2a3f13 | 1222 | struct sched_rt_entity *back = NULL; |
6f505b16 | 1223 | |
58d6c2d7 PZ |
1224 | for_each_sched_rt_entity(rt_se) { |
1225 | rt_se->back = back; | |
1226 | back = rt_se; | |
1227 | } | |
1228 | ||
f4ebcbc0 KT |
1229 | dequeue_top_rt_rq(rt_rq_of_se(back)); |
1230 | ||
58d6c2d7 PZ |
1231 | for (rt_se = back; rt_se; rt_se = rt_se->back) { |
1232 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
1233 | __dequeue_rt_entity(rt_se); |
1234 | } | |
1235 | } | |
1236 | ||
37dad3fc | 1237 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
ad2a3f13 | 1238 | { |
f4ebcbc0 KT |
1239 | struct rq *rq = rq_of_rt_se(rt_se); |
1240 | ||
ad2a3f13 PZ |
1241 | dequeue_rt_stack(rt_se); |
1242 | for_each_sched_rt_entity(rt_se) | |
37dad3fc | 1243 | __enqueue_rt_entity(rt_se, head); |
f4ebcbc0 | 1244 | enqueue_top_rt_rq(&rq->rt); |
ad2a3f13 PZ |
1245 | } |
1246 | ||
1247 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
1248 | { | |
f4ebcbc0 KT |
1249 | struct rq *rq = rq_of_rt_se(rt_se); |
1250 | ||
ad2a3f13 PZ |
1251 | dequeue_rt_stack(rt_se); |
1252 | ||
1253 | for_each_sched_rt_entity(rt_se) { | |
1254 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
1255 | ||
1256 | if (rt_rq && rt_rq->rt_nr_running) | |
37dad3fc | 1257 | __enqueue_rt_entity(rt_se, false); |
58d6c2d7 | 1258 | } |
f4ebcbc0 | 1259 | enqueue_top_rt_rq(&rq->rt); |
bb44e5d1 IM |
1260 | } |
1261 | ||
1262 | /* | |
1263 | * Adding/removing a task to/from a priority array: | |
1264 | */ | |
ea87bb78 | 1265 | static void |
371fd7e7 | 1266 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
6f505b16 PZ |
1267 | { |
1268 | struct sched_rt_entity *rt_se = &p->rt; | |
1269 | ||
371fd7e7 | 1270 | if (flags & ENQUEUE_WAKEUP) |
6f505b16 PZ |
1271 | rt_se->timeout = 0; |
1272 | ||
371fd7e7 | 1273 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
c09595f6 | 1274 | |
29baa747 | 1275 | if (!task_current(rq, p) && p->nr_cpus_allowed > 1) |
917b627d | 1276 | enqueue_pushable_task(rq, p); |
6f505b16 PZ |
1277 | } |
1278 | ||
371fd7e7 | 1279 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1280 | { |
6f505b16 | 1281 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 1282 | |
f1e14ef6 | 1283 | update_curr_rt(rq); |
ad2a3f13 | 1284 | dequeue_rt_entity(rt_se); |
c09595f6 | 1285 | |
917b627d | 1286 | dequeue_pushable_task(rq, p); |
bb44e5d1 IM |
1287 | } |
1288 | ||
1289 | /* | |
60686317 RW |
1290 | * Put task to the head or the end of the run list without the overhead of |
1291 | * dequeue followed by enqueue. | |
bb44e5d1 | 1292 | */ |
7ebefa8c DA |
1293 | static void |
1294 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 1295 | { |
1cdad715 | 1296 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
1297 | struct rt_prio_array *array = &rt_rq->active; |
1298 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
1299 | ||
1300 | if (head) | |
1301 | list_move(&rt_se->run_list, queue); | |
1302 | else | |
1303 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 1304 | } |
6f505b16 PZ |
1305 | } |
1306 | ||
7ebefa8c | 1307 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 1308 | { |
6f505b16 PZ |
1309 | struct sched_rt_entity *rt_se = &p->rt; |
1310 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1311 | |
6f505b16 PZ |
1312 | for_each_sched_rt_entity(rt_se) { |
1313 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 1314 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 1315 | } |
bb44e5d1 IM |
1316 | } |
1317 | ||
6f505b16 | 1318 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 1319 | { |
7ebefa8c | 1320 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
1321 | } |
1322 | ||
e7693a36 | 1323 | #ifdef CONFIG_SMP |
318e0893 GH |
1324 | static int find_lowest_rq(struct task_struct *task); |
1325 | ||
0017d735 | 1326 | static int |
ac66f547 | 1327 | select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags) |
e7693a36 | 1328 | { |
7608dec2 PZ |
1329 | struct task_struct *curr; |
1330 | struct rq *rq; | |
c37495fd SR |
1331 | |
1332 | /* For anything but wake ups, just return the task_cpu */ | |
1333 | if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK) | |
1334 | goto out; | |
1335 | ||
7608dec2 PZ |
1336 | rq = cpu_rq(cpu); |
1337 | ||
1338 | rcu_read_lock(); | |
316c1608 | 1339 | curr = READ_ONCE(rq->curr); /* unlocked access */ |
7608dec2 | 1340 | |
318e0893 | 1341 | /* |
7608dec2 | 1342 | * If the current task on @p's runqueue is an RT task, then |
e1f47d89 SR |
1343 | * try to see if we can wake this RT task up on another |
1344 | * runqueue. Otherwise simply start this RT task | |
1345 | * on its current runqueue. | |
1346 | * | |
43fa5460 SR |
1347 | * We want to avoid overloading runqueues. If the woken |
1348 | * task is a higher priority, then it will stay on this CPU | |
1349 | * and the lower prio task should be moved to another CPU. | |
1350 | * Even though this will probably make the lower prio task | |
1351 | * lose its cache, we do not want to bounce a higher task | |
1352 | * around just because it gave up its CPU, perhaps for a | |
1353 | * lock? | |
1354 | * | |
1355 | * For equal prio tasks, we just let the scheduler sort it out. | |
7608dec2 PZ |
1356 | * |
1357 | * Otherwise, just let it ride on the affined RQ and the | |
1358 | * post-schedule router will push the preempted task away | |
1359 | * | |
1360 | * This test is optimistic, if we get it wrong the load-balancer | |
1361 | * will have to sort it out. | |
318e0893 | 1362 | */ |
7608dec2 | 1363 | if (curr && unlikely(rt_task(curr)) && |
29baa747 | 1364 | (curr->nr_cpus_allowed < 2 || |
6bfa687c | 1365 | curr->prio <= p->prio)) { |
7608dec2 | 1366 | int target = find_lowest_rq(p); |
318e0893 | 1367 | |
80e3d87b TC |
1368 | /* |
1369 | * Don't bother moving it if the destination CPU is | |
1370 | * not running a lower priority task. | |
1371 | */ | |
1372 | if (target != -1 && | |
1373 | p->prio < cpu_rq(target)->rt.highest_prio.curr) | |
7608dec2 | 1374 | cpu = target; |
318e0893 | 1375 | } |
7608dec2 | 1376 | rcu_read_unlock(); |
318e0893 | 1377 | |
c37495fd | 1378 | out: |
7608dec2 | 1379 | return cpu; |
e7693a36 | 1380 | } |
7ebefa8c DA |
1381 | |
1382 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
1383 | { | |
308a623a WL |
1384 | /* |
1385 | * Current can't be migrated, useless to reschedule, | |
1386 | * let's hope p can move out. | |
1387 | */ | |
1388 | if (rq->curr->nr_cpus_allowed == 1 || | |
1389 | !cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) | |
7ebefa8c DA |
1390 | return; |
1391 | ||
308a623a WL |
1392 | /* |
1393 | * p is migratable, so let's not schedule it and | |
1394 | * see if it is pushed or pulled somewhere else. | |
1395 | */ | |
29baa747 | 1396 | if (p->nr_cpus_allowed != 1 |
13b8bd0a RR |
1397 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1398 | return; | |
24600ce8 | 1399 | |
7ebefa8c DA |
1400 | /* |
1401 | * There appears to be other cpus that can accept | |
1402 | * current and none to run 'p', so lets reschedule | |
1403 | * to try and push current away: | |
1404 | */ | |
1405 | requeue_task_rt(rq, p, 1); | |
8875125e | 1406 | resched_curr(rq); |
7ebefa8c DA |
1407 | } |
1408 | ||
e7693a36 GH |
1409 | #endif /* CONFIG_SMP */ |
1410 | ||
bb44e5d1 IM |
1411 | /* |
1412 | * Preempt the current task with a newly woken task if needed: | |
1413 | */ | |
7d478721 | 1414 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1415 | { |
45c01e82 | 1416 | if (p->prio < rq->curr->prio) { |
8875125e | 1417 | resched_curr(rq); |
45c01e82 GH |
1418 | return; |
1419 | } | |
1420 | ||
1421 | #ifdef CONFIG_SMP | |
1422 | /* | |
1423 | * If: | |
1424 | * | |
1425 | * - the newly woken task is of equal priority to the current task | |
1426 | * - the newly woken task is non-migratable while current is migratable | |
1427 | * - current will be preempted on the next reschedule | |
1428 | * | |
1429 | * we should check to see if current can readily move to a different | |
1430 | * cpu. If so, we will reschedule to allow the push logic to try | |
1431 | * to move current somewhere else, making room for our non-migratable | |
1432 | * task. | |
1433 | */ | |
8dd0de8b | 1434 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
7ebefa8c | 1435 | check_preempt_equal_prio(rq, p); |
45c01e82 | 1436 | #endif |
bb44e5d1 IM |
1437 | } |
1438 | ||
6f505b16 PZ |
1439 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1440 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1441 | { |
6f505b16 PZ |
1442 | struct rt_prio_array *array = &rt_rq->active; |
1443 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1444 | struct list_head *queue; |
1445 | int idx; | |
1446 | ||
1447 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1448 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1449 | |
1450 | queue = array->queue + idx; | |
6f505b16 | 1451 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1452 | |
6f505b16 PZ |
1453 | return next; |
1454 | } | |
bb44e5d1 | 1455 | |
917b627d | 1456 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1457 | { |
1458 | struct sched_rt_entity *rt_se; | |
1459 | struct task_struct *p; | |
606dba2e | 1460 | struct rt_rq *rt_rq = &rq->rt; |
6f505b16 PZ |
1461 | |
1462 | do { | |
1463 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1464 | BUG_ON(!rt_se); |
6f505b16 PZ |
1465 | rt_rq = group_rt_rq(rt_se); |
1466 | } while (rt_rq); | |
1467 | ||
1468 | p = rt_task_of(rt_se); | |
78becc27 | 1469 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
1470 | |
1471 | return p; | |
1472 | } | |
1473 | ||
606dba2e PZ |
1474 | static struct task_struct * |
1475 | pick_next_task_rt(struct rq *rq, struct task_struct *prev) | |
917b627d | 1476 | { |
606dba2e PZ |
1477 | struct task_struct *p; |
1478 | struct rt_rq *rt_rq = &rq->rt; | |
1479 | ||
37e117c0 | 1480 | if (need_pull_rt_task(rq, prev)) { |
cbce1a68 PZ |
1481 | /* |
1482 | * This is OK, because current is on_cpu, which avoids it being | |
1483 | * picked for load-balance and preemption/IRQs are still | |
1484 | * disabled avoiding further scheduler activity on it and we're | |
1485 | * being very careful to re-start the picking loop. | |
1486 | */ | |
1487 | lockdep_unpin_lock(&rq->lock); | |
38033c37 | 1488 | pull_rt_task(rq); |
cbce1a68 | 1489 | lockdep_pin_lock(&rq->lock); |
37e117c0 PZ |
1490 | /* |
1491 | * pull_rt_task() can drop (and re-acquire) rq->lock; this | |
a1d9a323 KT |
1492 | * means a dl or stop task can slip in, in which case we need |
1493 | * to re-start task selection. | |
37e117c0 | 1494 | */ |
da0c1e65 | 1495 | if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) || |
a1d9a323 | 1496 | rq->dl.dl_nr_running)) |
37e117c0 PZ |
1497 | return RETRY_TASK; |
1498 | } | |
38033c37 | 1499 | |
734ff2a7 KT |
1500 | /* |
1501 | * We may dequeue prev's rt_rq in put_prev_task(). | |
1502 | * So, we update time before rt_nr_running check. | |
1503 | */ | |
1504 | if (prev->sched_class == &rt_sched_class) | |
1505 | update_curr_rt(rq); | |
1506 | ||
f4ebcbc0 | 1507 | if (!rt_rq->rt_queued) |
606dba2e PZ |
1508 | return NULL; |
1509 | ||
3f1d2a31 | 1510 | put_prev_task(rq, prev); |
606dba2e PZ |
1511 | |
1512 | p = _pick_next_task_rt(rq); | |
917b627d GH |
1513 | |
1514 | /* The running task is never eligible for pushing */ | |
f3f1768f | 1515 | dequeue_pushable_task(rq, p); |
917b627d | 1516 | |
e3fca9e7 | 1517 | queue_push_tasks(rq); |
3f029d3c | 1518 | |
6f505b16 | 1519 | return p; |
bb44e5d1 IM |
1520 | } |
1521 | ||
31ee529c | 1522 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1523 | { |
f1e14ef6 | 1524 | update_curr_rt(rq); |
917b627d GH |
1525 | |
1526 | /* | |
1527 | * The previous task needs to be made eligible for pushing | |
1528 | * if it is still active | |
1529 | */ | |
29baa747 | 1530 | if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1) |
917b627d | 1531 | enqueue_pushable_task(rq, p); |
bb44e5d1 IM |
1532 | } |
1533 | ||
681f3e68 | 1534 | #ifdef CONFIG_SMP |
6f505b16 | 1535 | |
e8fa1362 SR |
1536 | /* Only try algorithms three times */ |
1537 | #define RT_MAX_TRIES 3 | |
1538 | ||
f65eda4f SR |
1539 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1540 | { | |
1541 | if (!task_running(rq, p) && | |
60334caf | 1542 | cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) |
f65eda4f SR |
1543 | return 1; |
1544 | return 0; | |
1545 | } | |
1546 | ||
e23ee747 KT |
1547 | /* |
1548 | * Return the highest pushable rq's task, which is suitable to be executed | |
1549 | * on the cpu, NULL otherwise | |
1550 | */ | |
1551 | static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu) | |
e8fa1362 | 1552 | { |
e23ee747 KT |
1553 | struct plist_head *head = &rq->rt.pushable_tasks; |
1554 | struct task_struct *p; | |
3d07467b | 1555 | |
e23ee747 KT |
1556 | if (!has_pushable_tasks(rq)) |
1557 | return NULL; | |
3d07467b | 1558 | |
e23ee747 KT |
1559 | plist_for_each_entry(p, head, pushable_tasks) { |
1560 | if (pick_rt_task(rq, p, cpu)) | |
1561 | return p; | |
f65eda4f SR |
1562 | } |
1563 | ||
e23ee747 | 1564 | return NULL; |
e8fa1362 SR |
1565 | } |
1566 | ||
0e3900e6 | 1567 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1568 | |
6e1254d2 GH |
1569 | static int find_lowest_rq(struct task_struct *task) |
1570 | { | |
1571 | struct sched_domain *sd; | |
4ba29684 | 1572 | struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask); |
6e1254d2 GH |
1573 | int this_cpu = smp_processor_id(); |
1574 | int cpu = task_cpu(task); | |
06f90dbd | 1575 | |
0da938c4 SR |
1576 | /* Make sure the mask is initialized first */ |
1577 | if (unlikely(!lowest_mask)) | |
1578 | return -1; | |
1579 | ||
29baa747 | 1580 | if (task->nr_cpus_allowed == 1) |
6e0534f2 | 1581 | return -1; /* No other targets possible */ |
6e1254d2 | 1582 | |
6e0534f2 GH |
1583 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1584 | return -1; /* No targets found */ | |
6e1254d2 GH |
1585 | |
1586 | /* | |
1587 | * At this point we have built a mask of cpus representing the | |
1588 | * lowest priority tasks in the system. Now we want to elect | |
1589 | * the best one based on our affinity and topology. | |
1590 | * | |
1591 | * We prioritize the last cpu that the task executed on since | |
1592 | * it is most likely cache-hot in that location. | |
1593 | */ | |
96f874e2 | 1594 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1595 | return cpu; |
1596 | ||
1597 | /* | |
1598 | * Otherwise, we consult the sched_domains span maps to figure | |
1599 | * out which cpu is logically closest to our hot cache data. | |
1600 | */ | |
e2c88063 RR |
1601 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1602 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ | |
6e1254d2 | 1603 | |
cd4ae6ad | 1604 | rcu_read_lock(); |
e2c88063 RR |
1605 | for_each_domain(cpu, sd) { |
1606 | if (sd->flags & SD_WAKE_AFFINE) { | |
1607 | int best_cpu; | |
6e1254d2 | 1608 | |
e2c88063 RR |
1609 | /* |
1610 | * "this_cpu" is cheaper to preempt than a | |
1611 | * remote processor. | |
1612 | */ | |
1613 | if (this_cpu != -1 && | |
cd4ae6ad XF |
1614 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
1615 | rcu_read_unlock(); | |
e2c88063 | 1616 | return this_cpu; |
cd4ae6ad | 1617 | } |
e2c88063 RR |
1618 | |
1619 | best_cpu = cpumask_first_and(lowest_mask, | |
1620 | sched_domain_span(sd)); | |
cd4ae6ad XF |
1621 | if (best_cpu < nr_cpu_ids) { |
1622 | rcu_read_unlock(); | |
e2c88063 | 1623 | return best_cpu; |
cd4ae6ad | 1624 | } |
6e1254d2 GH |
1625 | } |
1626 | } | |
cd4ae6ad | 1627 | rcu_read_unlock(); |
6e1254d2 GH |
1628 | |
1629 | /* | |
1630 | * And finally, if there were no matches within the domains | |
1631 | * just give the caller *something* to work with from the compatible | |
1632 | * locations. | |
1633 | */ | |
e2c88063 RR |
1634 | if (this_cpu != -1) |
1635 | return this_cpu; | |
1636 | ||
1637 | cpu = cpumask_any(lowest_mask); | |
1638 | if (cpu < nr_cpu_ids) | |
1639 | return cpu; | |
1640 | return -1; | |
07b4032c GH |
1641 | } |
1642 | ||
1643 | /* Will lock the rq it finds */ | |
4df64c0b | 1644 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1645 | { |
1646 | struct rq *lowest_rq = NULL; | |
07b4032c | 1647 | int tries; |
4df64c0b | 1648 | int cpu; |
e8fa1362 | 1649 | |
07b4032c GH |
1650 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1651 | cpu = find_lowest_rq(task); | |
1652 | ||
2de0b463 | 1653 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1654 | break; |
1655 | ||
07b4032c GH |
1656 | lowest_rq = cpu_rq(cpu); |
1657 | ||
80e3d87b TC |
1658 | if (lowest_rq->rt.highest_prio.curr <= task->prio) { |
1659 | /* | |
1660 | * Target rq has tasks of equal or higher priority, | |
1661 | * retrying does not release any lock and is unlikely | |
1662 | * to yield a different result. | |
1663 | */ | |
1664 | lowest_rq = NULL; | |
1665 | break; | |
1666 | } | |
1667 | ||
e8fa1362 | 1668 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1669 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1670 | /* |
1671 | * We had to unlock the run queue. In | |
1672 | * the mean time, task could have | |
1673 | * migrated already or had its affinity changed. | |
1674 | * Also make sure that it wasn't scheduled on its rq. | |
1675 | */ | |
07b4032c | 1676 | if (unlikely(task_rq(task) != rq || |
96f874e2 | 1677 | !cpumask_test_cpu(lowest_rq->cpu, |
fa17b507 | 1678 | tsk_cpus_allowed(task)) || |
07b4032c | 1679 | task_running(rq, task) || |
da0c1e65 | 1680 | !task_on_rq_queued(task))) { |
4df64c0b | 1681 | |
7f1b4393 | 1682 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1683 | lowest_rq = NULL; |
1684 | break; | |
1685 | } | |
1686 | } | |
1687 | ||
1688 | /* If this rq is still suitable use it. */ | |
e864c499 | 1689 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1690 | break; |
1691 | ||
1692 | /* try again */ | |
1b12bbc7 | 1693 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1694 | lowest_rq = NULL; |
1695 | } | |
1696 | ||
1697 | return lowest_rq; | |
1698 | } | |
1699 | ||
917b627d GH |
1700 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1701 | { | |
1702 | struct task_struct *p; | |
1703 | ||
1704 | if (!has_pushable_tasks(rq)) | |
1705 | return NULL; | |
1706 | ||
1707 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1708 | struct task_struct, pushable_tasks); | |
1709 | ||
1710 | BUG_ON(rq->cpu != task_cpu(p)); | |
1711 | BUG_ON(task_current(rq, p)); | |
29baa747 | 1712 | BUG_ON(p->nr_cpus_allowed <= 1); |
917b627d | 1713 | |
da0c1e65 | 1714 | BUG_ON(!task_on_rq_queued(p)); |
917b627d GH |
1715 | BUG_ON(!rt_task(p)); |
1716 | ||
1717 | return p; | |
1718 | } | |
1719 | ||
e8fa1362 SR |
1720 | /* |
1721 | * If the current CPU has more than one RT task, see if the non | |
1722 | * running task can migrate over to a CPU that is running a task | |
1723 | * of lesser priority. | |
1724 | */ | |
697f0a48 | 1725 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1726 | { |
1727 | struct task_struct *next_task; | |
1728 | struct rq *lowest_rq; | |
311e800e | 1729 | int ret = 0; |
e8fa1362 | 1730 | |
a22d7fc1 GH |
1731 | if (!rq->rt.overloaded) |
1732 | return 0; | |
1733 | ||
917b627d | 1734 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1735 | if (!next_task) |
1736 | return 0; | |
1737 | ||
49246274 | 1738 | retry: |
697f0a48 | 1739 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1740 | WARN_ON(1); |
e8fa1362 | 1741 | return 0; |
f65eda4f | 1742 | } |
e8fa1362 SR |
1743 | |
1744 | /* | |
1745 | * It's possible that the next_task slipped in of | |
1746 | * higher priority than current. If that's the case | |
1747 | * just reschedule current. | |
1748 | */ | |
697f0a48 | 1749 | if (unlikely(next_task->prio < rq->curr->prio)) { |
8875125e | 1750 | resched_curr(rq); |
e8fa1362 SR |
1751 | return 0; |
1752 | } | |
1753 | ||
697f0a48 | 1754 | /* We might release rq lock */ |
e8fa1362 SR |
1755 | get_task_struct(next_task); |
1756 | ||
1757 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1758 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1759 | if (!lowest_rq) { |
1760 | struct task_struct *task; | |
1761 | /* | |
311e800e | 1762 | * find_lock_lowest_rq releases rq->lock |
1563513d GH |
1763 | * so it is possible that next_task has migrated. |
1764 | * | |
1765 | * We need to make sure that the task is still on the same | |
1766 | * run-queue and is also still the next task eligible for | |
1767 | * pushing. | |
e8fa1362 | 1768 | */ |
917b627d | 1769 | task = pick_next_pushable_task(rq); |
1563513d GH |
1770 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1771 | /* | |
311e800e HD |
1772 | * The task hasn't migrated, and is still the next |
1773 | * eligible task, but we failed to find a run-queue | |
1774 | * to push it to. Do not retry in this case, since | |
1775 | * other cpus will pull from us when ready. | |
1563513d | 1776 | */ |
1563513d | 1777 | goto out; |
e8fa1362 | 1778 | } |
917b627d | 1779 | |
1563513d GH |
1780 | if (!task) |
1781 | /* No more tasks, just exit */ | |
1782 | goto out; | |
1783 | ||
917b627d | 1784 | /* |
1563513d | 1785 | * Something has shifted, try again. |
917b627d | 1786 | */ |
1563513d GH |
1787 | put_task_struct(next_task); |
1788 | next_task = task; | |
1789 | goto retry; | |
e8fa1362 SR |
1790 | } |
1791 | ||
697f0a48 | 1792 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1793 | set_task_cpu(next_task, lowest_rq->cpu); |
1794 | activate_task(lowest_rq, next_task, 0); | |
311e800e | 1795 | ret = 1; |
e8fa1362 | 1796 | |
8875125e | 1797 | resched_curr(lowest_rq); |
e8fa1362 | 1798 | |
1b12bbc7 | 1799 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1800 | |
e8fa1362 SR |
1801 | out: |
1802 | put_task_struct(next_task); | |
1803 | ||
311e800e | 1804 | return ret; |
e8fa1362 SR |
1805 | } |
1806 | ||
e8fa1362 SR |
1807 | static void push_rt_tasks(struct rq *rq) |
1808 | { | |
1809 | /* push_rt_task will return true if it moved an RT */ | |
1810 | while (push_rt_task(rq)) | |
1811 | ; | |
1812 | } | |
1813 | ||
b6366f04 SR |
1814 | #ifdef HAVE_RT_PUSH_IPI |
1815 | /* | |
1816 | * The search for the next cpu always starts at rq->cpu and ends | |
1817 | * when we reach rq->cpu again. It will never return rq->cpu. | |
1818 | * This returns the next cpu to check, or nr_cpu_ids if the loop | |
1819 | * is complete. | |
1820 | * | |
1821 | * rq->rt.push_cpu holds the last cpu returned by this function, | |
1822 | * or if this is the first instance, it must hold rq->cpu. | |
1823 | */ | |
1824 | static int rto_next_cpu(struct rq *rq) | |
1825 | { | |
1826 | int prev_cpu = rq->rt.push_cpu; | |
1827 | int cpu; | |
1828 | ||
1829 | cpu = cpumask_next(prev_cpu, rq->rd->rto_mask); | |
1830 | ||
1831 | /* | |
1832 | * If the previous cpu is less than the rq's CPU, then it already | |
1833 | * passed the end of the mask, and has started from the beginning. | |
1834 | * We end if the next CPU is greater or equal to rq's CPU. | |
1835 | */ | |
1836 | if (prev_cpu < rq->cpu) { | |
1837 | if (cpu >= rq->cpu) | |
1838 | return nr_cpu_ids; | |
1839 | ||
1840 | } else if (cpu >= nr_cpu_ids) { | |
1841 | /* | |
1842 | * We passed the end of the mask, start at the beginning. | |
1843 | * If the result is greater or equal to the rq's CPU, then | |
1844 | * the loop is finished. | |
1845 | */ | |
1846 | cpu = cpumask_first(rq->rd->rto_mask); | |
1847 | if (cpu >= rq->cpu) | |
1848 | return nr_cpu_ids; | |
1849 | } | |
1850 | rq->rt.push_cpu = cpu; | |
1851 | ||
1852 | /* Return cpu to let the caller know if the loop is finished or not */ | |
1853 | return cpu; | |
1854 | } | |
1855 | ||
1856 | static int find_next_push_cpu(struct rq *rq) | |
1857 | { | |
1858 | struct rq *next_rq; | |
1859 | int cpu; | |
1860 | ||
1861 | while (1) { | |
1862 | cpu = rto_next_cpu(rq); | |
1863 | if (cpu >= nr_cpu_ids) | |
1864 | break; | |
1865 | next_rq = cpu_rq(cpu); | |
1866 | ||
1867 | /* Make sure the next rq can push to this rq */ | |
1868 | if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr) | |
1869 | break; | |
1870 | } | |
1871 | ||
1872 | return cpu; | |
1873 | } | |
1874 | ||
1875 | #define RT_PUSH_IPI_EXECUTING 1 | |
1876 | #define RT_PUSH_IPI_RESTART 2 | |
1877 | ||
1878 | static void tell_cpu_to_push(struct rq *rq) | |
1879 | { | |
1880 | int cpu; | |
1881 | ||
1882 | if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { | |
1883 | raw_spin_lock(&rq->rt.push_lock); | |
1884 | /* Make sure it's still executing */ | |
1885 | if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { | |
1886 | /* | |
1887 | * Tell the IPI to restart the loop as things have | |
1888 | * changed since it started. | |
1889 | */ | |
1890 | rq->rt.push_flags |= RT_PUSH_IPI_RESTART; | |
1891 | raw_spin_unlock(&rq->rt.push_lock); | |
1892 | return; | |
1893 | } | |
1894 | raw_spin_unlock(&rq->rt.push_lock); | |
1895 | } | |
1896 | ||
1897 | /* When here, there's no IPI going around */ | |
1898 | ||
1899 | rq->rt.push_cpu = rq->cpu; | |
1900 | cpu = find_next_push_cpu(rq); | |
1901 | if (cpu >= nr_cpu_ids) | |
1902 | return; | |
1903 | ||
1904 | rq->rt.push_flags = RT_PUSH_IPI_EXECUTING; | |
1905 | ||
1906 | irq_work_queue_on(&rq->rt.push_work, cpu); | |
1907 | } | |
1908 | ||
1909 | /* Called from hardirq context */ | |
1910 | static void try_to_push_tasks(void *arg) | |
1911 | { | |
1912 | struct rt_rq *rt_rq = arg; | |
1913 | struct rq *rq, *src_rq; | |
1914 | int this_cpu; | |
1915 | int cpu; | |
1916 | ||
1917 | this_cpu = rt_rq->push_cpu; | |
1918 | ||
1919 | /* Paranoid check */ | |
1920 | BUG_ON(this_cpu != smp_processor_id()); | |
1921 | ||
1922 | rq = cpu_rq(this_cpu); | |
1923 | src_rq = rq_of_rt_rq(rt_rq); | |
1924 | ||
1925 | again: | |
1926 | if (has_pushable_tasks(rq)) { | |
1927 | raw_spin_lock(&rq->lock); | |
1928 | push_rt_task(rq); | |
1929 | raw_spin_unlock(&rq->lock); | |
1930 | } | |
1931 | ||
1932 | /* Pass the IPI to the next rt overloaded queue */ | |
1933 | raw_spin_lock(&rt_rq->push_lock); | |
1934 | /* | |
1935 | * If the source queue changed since the IPI went out, | |
1936 | * we need to restart the search from that CPU again. | |
1937 | */ | |
1938 | if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) { | |
1939 | rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART; | |
1940 | rt_rq->push_cpu = src_rq->cpu; | |
1941 | } | |
1942 | ||
1943 | cpu = find_next_push_cpu(src_rq); | |
1944 | ||
1945 | if (cpu >= nr_cpu_ids) | |
1946 | rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING; | |
1947 | raw_spin_unlock(&rt_rq->push_lock); | |
1948 | ||
1949 | if (cpu >= nr_cpu_ids) | |
1950 | return; | |
1951 | ||
1952 | /* | |
1953 | * It is possible that a restart caused this CPU to be | |
1954 | * chosen again. Don't bother with an IPI, just see if we | |
1955 | * have more to push. | |
1956 | */ | |
1957 | if (unlikely(cpu == rq->cpu)) | |
1958 | goto again; | |
1959 | ||
1960 | /* Try the next RT overloaded CPU */ | |
1961 | irq_work_queue_on(&rt_rq->push_work, cpu); | |
1962 | } | |
1963 | ||
1964 | static void push_irq_work_func(struct irq_work *work) | |
1965 | { | |
1966 | struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work); | |
1967 | ||
1968 | try_to_push_tasks(rt_rq); | |
1969 | } | |
1970 | #endif /* HAVE_RT_PUSH_IPI */ | |
1971 | ||
8046d680 | 1972 | static void pull_rt_task(struct rq *this_rq) |
f65eda4f | 1973 | { |
8046d680 PZ |
1974 | int this_cpu = this_rq->cpu, cpu; |
1975 | bool resched = false; | |
a8728944 | 1976 | struct task_struct *p; |
f65eda4f | 1977 | struct rq *src_rq; |
f65eda4f | 1978 | |
637f5085 | 1979 | if (likely(!rt_overloaded(this_rq))) |
8046d680 | 1980 | return; |
f65eda4f | 1981 | |
7c3f2ab7 PZ |
1982 | /* |
1983 | * Match the barrier from rt_set_overloaded; this guarantees that if we | |
1984 | * see overloaded we must also see the rto_mask bit. | |
1985 | */ | |
1986 | smp_rmb(); | |
1987 | ||
b6366f04 SR |
1988 | #ifdef HAVE_RT_PUSH_IPI |
1989 | if (sched_feat(RT_PUSH_IPI)) { | |
1990 | tell_cpu_to_push(this_rq); | |
8046d680 | 1991 | return; |
b6366f04 SR |
1992 | } |
1993 | #endif | |
1994 | ||
c6c4927b | 1995 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1996 | if (this_cpu == cpu) |
1997 | continue; | |
1998 | ||
1999 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
2000 | |
2001 | /* | |
2002 | * Don't bother taking the src_rq->lock if the next highest | |
2003 | * task is known to be lower-priority than our current task. | |
2004 | * This may look racy, but if this value is about to go | |
2005 | * logically higher, the src_rq will push this task away. | |
2006 | * And if its going logically lower, we do not care | |
2007 | */ | |
2008 | if (src_rq->rt.highest_prio.next >= | |
2009 | this_rq->rt.highest_prio.curr) | |
2010 | continue; | |
2011 | ||
f65eda4f SR |
2012 | /* |
2013 | * We can potentially drop this_rq's lock in | |
2014 | * double_lock_balance, and another CPU could | |
a8728944 | 2015 | * alter this_rq |
f65eda4f | 2016 | */ |
a8728944 | 2017 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
2018 | |
2019 | /* | |
e23ee747 KT |
2020 | * We can pull only a task, which is pushable |
2021 | * on its rq, and no others. | |
f65eda4f | 2022 | */ |
e23ee747 | 2023 | p = pick_highest_pushable_task(src_rq, this_cpu); |
f65eda4f SR |
2024 | |
2025 | /* | |
2026 | * Do we have an RT task that preempts | |
2027 | * the to-be-scheduled task? | |
2028 | */ | |
a8728944 | 2029 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f | 2030 | WARN_ON(p == src_rq->curr); |
da0c1e65 | 2031 | WARN_ON(!task_on_rq_queued(p)); |
f65eda4f SR |
2032 | |
2033 | /* | |
2034 | * There's a chance that p is higher in priority | |
2035 | * than what's currently running on its cpu. | |
2036 | * This is just that p is wakeing up and hasn't | |
2037 | * had a chance to schedule. We only pull | |
2038 | * p if it is lower in priority than the | |
a8728944 | 2039 | * current task on the run queue |
f65eda4f | 2040 | */ |
a8728944 | 2041 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 2042 | goto skip; |
f65eda4f | 2043 | |
8046d680 | 2044 | resched = true; |
f65eda4f SR |
2045 | |
2046 | deactivate_task(src_rq, p, 0); | |
2047 | set_task_cpu(p, this_cpu); | |
2048 | activate_task(this_rq, p, 0); | |
2049 | /* | |
2050 | * We continue with the search, just in | |
2051 | * case there's an even higher prio task | |
25985edc | 2052 | * in another runqueue. (low likelihood |
f65eda4f | 2053 | * but possible) |
f65eda4f | 2054 | */ |
f65eda4f | 2055 | } |
49246274 | 2056 | skip: |
1b12bbc7 | 2057 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
2058 | } |
2059 | ||
8046d680 PZ |
2060 | if (resched) |
2061 | resched_curr(this_rq); | |
f65eda4f SR |
2062 | } |
2063 | ||
8ae121ac GH |
2064 | /* |
2065 | * If we are not running and we are not going to reschedule soon, we should | |
2066 | * try to push tasks away now | |
2067 | */ | |
efbbd05a | 2068 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 2069 | { |
9a897c5a | 2070 | if (!task_running(rq, p) && |
8ae121ac | 2071 | !test_tsk_need_resched(rq->curr) && |
29baa747 | 2072 | p->nr_cpus_allowed > 1 && |
1baca4ce | 2073 | (dl_task(rq->curr) || rt_task(rq->curr)) && |
29baa747 | 2074 | (rq->curr->nr_cpus_allowed < 2 || |
3be209a8 | 2075 | rq->curr->prio <= p->prio)) |
4642dafd SR |
2076 | push_rt_tasks(rq); |
2077 | } | |
2078 | ||
bdd7c81b | 2079 | /* Assumes rq->lock is held */ |
1f11eb6a | 2080 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
2081 | { |
2082 | if (rq->rt.overloaded) | |
2083 | rt_set_overload(rq); | |
6e0534f2 | 2084 | |
7def2be1 PZ |
2085 | __enable_runtime(rq); |
2086 | ||
e864c499 | 2087 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
2088 | } |
2089 | ||
2090 | /* Assumes rq->lock is held */ | |
1f11eb6a | 2091 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
2092 | { |
2093 | if (rq->rt.overloaded) | |
2094 | rt_clear_overload(rq); | |
6e0534f2 | 2095 | |
7def2be1 PZ |
2096 | __disable_runtime(rq); |
2097 | ||
6e0534f2 | 2098 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 2099 | } |
cb469845 SR |
2100 | |
2101 | /* | |
2102 | * When switch from the rt queue, we bring ourselves to a position | |
2103 | * that we might want to pull RT tasks from other runqueues. | |
2104 | */ | |
da7a735e | 2105 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
2106 | { |
2107 | /* | |
2108 | * If there are other RT tasks then we will reschedule | |
2109 | * and the scheduling of the other RT tasks will handle | |
2110 | * the balancing. But if we are the last RT task | |
2111 | * we may need to handle the pulling of RT tasks | |
2112 | * now. | |
2113 | */ | |
da0c1e65 | 2114 | if (!task_on_rq_queued(p) || rq->rt.rt_nr_running) |
1158ddb5 KT |
2115 | return; |
2116 | ||
fd7a4bed | 2117 | queue_pull_task(rq); |
cb469845 | 2118 | } |
3d8cbdf8 | 2119 | |
11c785b7 | 2120 | void __init init_sched_rt_class(void) |
3d8cbdf8 RR |
2121 | { |
2122 | unsigned int i; | |
2123 | ||
029632fb | 2124 | for_each_possible_cpu(i) { |
eaa95840 | 2125 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 2126 | GFP_KERNEL, cpu_to_node(i)); |
029632fb | 2127 | } |
3d8cbdf8 | 2128 | } |
cb469845 SR |
2129 | #endif /* CONFIG_SMP */ |
2130 | ||
2131 | /* | |
2132 | * When switching a task to RT, we may overload the runqueue | |
2133 | * with RT tasks. In this case we try to push them off to | |
2134 | * other runqueues. | |
2135 | */ | |
da7a735e | 2136 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
cb469845 | 2137 | { |
cb469845 SR |
2138 | /* |
2139 | * If we are already running, then there's nothing | |
2140 | * that needs to be done. But if we are not running | |
2141 | * we may need to preempt the current running task. | |
2142 | * If that current running task is also an RT task | |
2143 | * then see if we can move to another run queue. | |
2144 | */ | |
da0c1e65 | 2145 | if (task_on_rq_queued(p) && rq->curr != p) { |
cb469845 | 2146 | #ifdef CONFIG_SMP |
fd7a4bed PZ |
2147 | if (p->nr_cpus_allowed > 1 && rq->rt.overloaded) |
2148 | queue_push_tasks(rq); | |
2149 | #else | |
2150 | if (p->prio < rq->curr->prio) | |
8875125e | 2151 | resched_curr(rq); |
fd7a4bed | 2152 | #endif /* CONFIG_SMP */ |
cb469845 SR |
2153 | } |
2154 | } | |
2155 | ||
2156 | /* | |
2157 | * Priority of the task has changed. This may cause | |
2158 | * us to initiate a push or pull. | |
2159 | */ | |
da7a735e PZ |
2160 | static void |
2161 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) | |
cb469845 | 2162 | { |
da0c1e65 | 2163 | if (!task_on_rq_queued(p)) |
da7a735e PZ |
2164 | return; |
2165 | ||
2166 | if (rq->curr == p) { | |
cb469845 SR |
2167 | #ifdef CONFIG_SMP |
2168 | /* | |
2169 | * If our priority decreases while running, we | |
2170 | * may need to pull tasks to this runqueue. | |
2171 | */ | |
2172 | if (oldprio < p->prio) | |
fd7a4bed PZ |
2173 | queue_pull_task(rq); |
2174 | ||
cb469845 SR |
2175 | /* |
2176 | * If there's a higher priority task waiting to run | |
fd7a4bed | 2177 | * then reschedule. |
cb469845 | 2178 | */ |
fd7a4bed | 2179 | if (p->prio > rq->rt.highest_prio.curr) |
8875125e | 2180 | resched_curr(rq); |
cb469845 SR |
2181 | #else |
2182 | /* For UP simply resched on drop of prio */ | |
2183 | if (oldprio < p->prio) | |
8875125e | 2184 | resched_curr(rq); |
e8fa1362 | 2185 | #endif /* CONFIG_SMP */ |
cb469845 SR |
2186 | } else { |
2187 | /* | |
2188 | * This task is not running, but if it is | |
2189 | * greater than the current running task | |
2190 | * then reschedule. | |
2191 | */ | |
2192 | if (p->prio < rq->curr->prio) | |
8875125e | 2193 | resched_curr(rq); |
cb469845 SR |
2194 | } |
2195 | } | |
2196 | ||
78f2c7db PZ |
2197 | static void watchdog(struct rq *rq, struct task_struct *p) |
2198 | { | |
2199 | unsigned long soft, hard; | |
2200 | ||
78d7d407 JS |
2201 | /* max may change after cur was read, this will be fixed next tick */ |
2202 | soft = task_rlimit(p, RLIMIT_RTTIME); | |
2203 | hard = task_rlimit_max(p, RLIMIT_RTTIME); | |
78f2c7db PZ |
2204 | |
2205 | if (soft != RLIM_INFINITY) { | |
2206 | unsigned long next; | |
2207 | ||
57d2aa00 YX |
2208 | if (p->rt.watchdog_stamp != jiffies) { |
2209 | p->rt.timeout++; | |
2210 | p->rt.watchdog_stamp = jiffies; | |
2211 | } | |
2212 | ||
78f2c7db | 2213 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
5a52dd50 | 2214 | if (p->rt.timeout > next) |
f06febc9 | 2215 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
2216 | } |
2217 | } | |
bb44e5d1 | 2218 | |
8f4d37ec | 2219 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 2220 | { |
454c7999 CC |
2221 | struct sched_rt_entity *rt_se = &p->rt; |
2222 | ||
67e2be02 PZ |
2223 | update_curr_rt(rq); |
2224 | ||
78f2c7db PZ |
2225 | watchdog(rq, p); |
2226 | ||
bb44e5d1 IM |
2227 | /* |
2228 | * RR tasks need a special form of timeslice management. | |
2229 | * FIFO tasks have no timeslices. | |
2230 | */ | |
2231 | if (p->policy != SCHED_RR) | |
2232 | return; | |
2233 | ||
fa717060 | 2234 | if (--p->rt.time_slice) |
bb44e5d1 IM |
2235 | return; |
2236 | ||
ce0dbbbb | 2237 | p->rt.time_slice = sched_rr_timeslice; |
bb44e5d1 | 2238 | |
98fbc798 | 2239 | /* |
e9aa39bb LB |
2240 | * Requeue to the end of queue if we (and all of our ancestors) are not |
2241 | * the only element on the queue | |
98fbc798 | 2242 | */ |
454c7999 CC |
2243 | for_each_sched_rt_entity(rt_se) { |
2244 | if (rt_se->run_list.prev != rt_se->run_list.next) { | |
2245 | requeue_task_rt(rq, p, 0); | |
8aa6f0eb | 2246 | resched_curr(rq); |
454c7999 CC |
2247 | return; |
2248 | } | |
98fbc798 | 2249 | } |
bb44e5d1 IM |
2250 | } |
2251 | ||
83b699ed SV |
2252 | static void set_curr_task_rt(struct rq *rq) |
2253 | { | |
2254 | struct task_struct *p = rq->curr; | |
2255 | ||
78becc27 | 2256 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
2257 | |
2258 | /* The running task is never eligible for pushing */ | |
2259 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
2260 | } |
2261 | ||
6d686f45 | 2262 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
0d721cea PW |
2263 | { |
2264 | /* | |
2265 | * Time slice is 0 for SCHED_FIFO tasks | |
2266 | */ | |
2267 | if (task->policy == SCHED_RR) | |
ce0dbbbb | 2268 | return sched_rr_timeslice; |
0d721cea PW |
2269 | else |
2270 | return 0; | |
2271 | } | |
2272 | ||
029632fb | 2273 | const struct sched_class rt_sched_class = { |
5522d5d5 | 2274 | .next = &fair_sched_class, |
bb44e5d1 IM |
2275 | .enqueue_task = enqueue_task_rt, |
2276 | .dequeue_task = dequeue_task_rt, | |
2277 | .yield_task = yield_task_rt, | |
2278 | ||
2279 | .check_preempt_curr = check_preempt_curr_rt, | |
2280 | ||
2281 | .pick_next_task = pick_next_task_rt, | |
2282 | .put_prev_task = put_prev_task_rt, | |
2283 | ||
681f3e68 | 2284 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
2285 | .select_task_rq = select_task_rq_rt, |
2286 | ||
6c37067e | 2287 | .set_cpus_allowed = set_cpus_allowed_common, |
1f11eb6a GH |
2288 | .rq_online = rq_online_rt, |
2289 | .rq_offline = rq_offline_rt, | |
efbbd05a | 2290 | .task_woken = task_woken_rt, |
cb469845 | 2291 | .switched_from = switched_from_rt, |
681f3e68 | 2292 | #endif |
bb44e5d1 | 2293 | |
83b699ed | 2294 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 2295 | .task_tick = task_tick_rt, |
cb469845 | 2296 | |
0d721cea PW |
2297 | .get_rr_interval = get_rr_interval_rt, |
2298 | ||
cb469845 SR |
2299 | .prio_changed = prio_changed_rt, |
2300 | .switched_to = switched_to_rt, | |
6e998916 SG |
2301 | |
2302 | .update_curr = update_curr_rt, | |
bb44e5d1 | 2303 | }; |
ada18de2 PZ |
2304 | |
2305 | #ifdef CONFIG_SCHED_DEBUG | |
2306 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2307 | ||
029632fb | 2308 | void print_rt_stats(struct seq_file *m, int cpu) |
ada18de2 | 2309 | { |
ec514c48 | 2310 | rt_rq_iter_t iter; |
ada18de2 PZ |
2311 | struct rt_rq *rt_rq; |
2312 | ||
2313 | rcu_read_lock(); | |
ec514c48 | 2314 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
ada18de2 PZ |
2315 | print_rt_rq(m, cpu, rt_rq); |
2316 | rcu_read_unlock(); | |
2317 | } | |
55e12e5e | 2318 | #endif /* CONFIG_SCHED_DEBUG */ |