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 | */ | |
269b26a5 | 638 | static void 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; |
269b26a5 | 642 | int i, weight; |
ac086bc2 PZ |
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; | |
ac086bc2 | 676 | if (rt_rq->rt_runtime == rt_period) { |
0986b11b | 677 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 PZ |
678 | break; |
679 | } | |
680 | } | |
7def2be1 | 681 | next: |
0986b11b | 682 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 | 683 | } |
0986b11b | 684 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
ac086bc2 | 685 | } |
7def2be1 | 686 | |
78333cdd PZ |
687 | /* |
688 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
689 | */ | |
7def2be1 PZ |
690 | static void __disable_runtime(struct rq *rq) |
691 | { | |
692 | struct root_domain *rd = rq->rd; | |
ec514c48 | 693 | rt_rq_iter_t iter; |
7def2be1 PZ |
694 | struct rt_rq *rt_rq; |
695 | ||
696 | if (unlikely(!scheduler_running)) | |
697 | return; | |
698 | ||
ec514c48 | 699 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
700 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
701 | s64 want; | |
702 | int i; | |
703 | ||
0986b11b TG |
704 | raw_spin_lock(&rt_b->rt_runtime_lock); |
705 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
706 | /* |
707 | * Either we're all inf and nobody needs to borrow, or we're | |
708 | * already disabled and thus have nothing to do, or we have | |
709 | * exactly the right amount of runtime to take out. | |
710 | */ | |
7def2be1 PZ |
711 | if (rt_rq->rt_runtime == RUNTIME_INF || |
712 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
713 | goto balanced; | |
0986b11b | 714 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
7def2be1 | 715 | |
78333cdd PZ |
716 | /* |
717 | * Calculate the difference between what we started out with | |
718 | * and what we current have, that's the amount of runtime | |
719 | * we lend and now have to reclaim. | |
720 | */ | |
7def2be1 PZ |
721 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
722 | ||
78333cdd PZ |
723 | /* |
724 | * Greedy reclaim, take back as much as we can. | |
725 | */ | |
c6c4927b | 726 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
727 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
728 | s64 diff; | |
729 | ||
78333cdd PZ |
730 | /* |
731 | * Can't reclaim from ourselves or disabled runqueues. | |
732 | */ | |
f1679d08 | 733 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
734 | continue; |
735 | ||
0986b11b | 736 | raw_spin_lock(&iter->rt_runtime_lock); |
7def2be1 PZ |
737 | if (want > 0) { |
738 | diff = min_t(s64, iter->rt_runtime, want); | |
739 | iter->rt_runtime -= diff; | |
740 | want -= diff; | |
741 | } else { | |
742 | iter->rt_runtime -= want; | |
743 | want -= want; | |
744 | } | |
0986b11b | 745 | raw_spin_unlock(&iter->rt_runtime_lock); |
7def2be1 PZ |
746 | |
747 | if (!want) | |
748 | break; | |
749 | } | |
750 | ||
0986b11b | 751 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
78333cdd PZ |
752 | /* |
753 | * We cannot be left wanting - that would mean some runtime | |
754 | * leaked out of the system. | |
755 | */ | |
7def2be1 PZ |
756 | BUG_ON(want); |
757 | balanced: | |
78333cdd PZ |
758 | /* |
759 | * Disable all the borrow logic by pretending we have inf | |
760 | * runtime - in which case borrowing doesn't make sense. | |
761 | */ | |
7def2be1 | 762 | rt_rq->rt_runtime = RUNTIME_INF; |
a4c96ae3 | 763 | rt_rq->rt_throttled = 0; |
0986b11b TG |
764 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
765 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
99b62567 KT |
766 | |
767 | /* Make rt_rq available for pick_next_task() */ | |
768 | sched_rt_rq_enqueue(rt_rq); | |
7def2be1 PZ |
769 | } |
770 | } | |
771 | ||
7def2be1 PZ |
772 | static void __enable_runtime(struct rq *rq) |
773 | { | |
ec514c48 | 774 | rt_rq_iter_t iter; |
7def2be1 PZ |
775 | struct rt_rq *rt_rq; |
776 | ||
777 | if (unlikely(!scheduler_running)) | |
778 | return; | |
779 | ||
78333cdd PZ |
780 | /* |
781 | * Reset each runqueue's bandwidth settings | |
782 | */ | |
ec514c48 | 783 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
784 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
785 | ||
0986b11b TG |
786 | raw_spin_lock(&rt_b->rt_runtime_lock); |
787 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
7def2be1 PZ |
788 | rt_rq->rt_runtime = rt_b->rt_runtime; |
789 | rt_rq->rt_time = 0; | |
baf25731 | 790 | rt_rq->rt_throttled = 0; |
0986b11b TG |
791 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
792 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
793 | } |
794 | } | |
795 | ||
269b26a5 | 796 | static void balance_runtime(struct rt_rq *rt_rq) |
eff6549b | 797 | { |
4a6184ce | 798 | if (!sched_feat(RT_RUNTIME_SHARE)) |
269b26a5 | 799 | return; |
4a6184ce | 800 | |
eff6549b | 801 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
0986b11b | 802 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
269b26a5 | 803 | do_balance_runtime(rt_rq); |
0986b11b | 804 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b | 805 | } |
eff6549b | 806 | } |
55e12e5e | 807 | #else /* !CONFIG_SMP */ |
269b26a5 | 808 | static inline void balance_runtime(struct rt_rq *rt_rq) {} |
55e12e5e | 809 | #endif /* CONFIG_SMP */ |
ac086bc2 | 810 | |
eff6549b PZ |
811 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
812 | { | |
42c62a58 | 813 | int i, idle = 1, throttled = 0; |
c6c4927b | 814 | const struct cpumask *span; |
eff6549b | 815 | |
eff6549b | 816 | span = sched_rt_period_mask(); |
e221d028 MG |
817 | #ifdef CONFIG_RT_GROUP_SCHED |
818 | /* | |
819 | * FIXME: isolated CPUs should really leave the root task group, | |
820 | * whether they are isolcpus or were isolated via cpusets, lest | |
821 | * the timer run on a CPU which does not service all runqueues, | |
822 | * potentially leaving other CPUs indefinitely throttled. If | |
823 | * isolation is really required, the user will turn the throttle | |
824 | * off to kill the perturbations it causes anyway. Meanwhile, | |
825 | * this maintains functionality for boot and/or troubleshooting. | |
826 | */ | |
827 | if (rt_b == &root_task_group.rt_bandwidth) | |
828 | span = cpu_online_mask; | |
829 | #endif | |
c6c4927b | 830 | for_each_cpu(i, span) { |
eff6549b PZ |
831 | int enqueue = 0; |
832 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
833 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
834 | ||
05fa785c | 835 | raw_spin_lock(&rq->lock); |
eff6549b PZ |
836 | if (rt_rq->rt_time) { |
837 | u64 runtime; | |
838 | ||
0986b11b | 839 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
840 | if (rt_rq->rt_throttled) |
841 | balance_runtime(rt_rq); | |
842 | runtime = rt_rq->rt_runtime; | |
843 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
844 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
845 | rt_rq->rt_throttled = 0; | |
846 | enqueue = 1; | |
61eadef6 MG |
847 | |
848 | /* | |
9edfbfed PZ |
849 | * When we're idle and a woken (rt) task is |
850 | * throttled check_preempt_curr() will set | |
851 | * skip_update and the time between the wakeup | |
852 | * and this unthrottle will get accounted as | |
853 | * 'runtime'. | |
61eadef6 MG |
854 | */ |
855 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) | |
9edfbfed | 856 | rq_clock_skip_update(rq, false); |
eff6549b PZ |
857 | } |
858 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
859 | idle = 0; | |
0986b11b | 860 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
0c3b9168 | 861 | } else if (rt_rq->rt_nr_running) { |
6c3df255 | 862 | idle = 0; |
0c3b9168 BS |
863 | if (!rt_rq_throttled(rt_rq)) |
864 | enqueue = 1; | |
865 | } | |
42c62a58 PZ |
866 | if (rt_rq->rt_throttled) |
867 | throttled = 1; | |
eff6549b PZ |
868 | |
869 | if (enqueue) | |
870 | sched_rt_rq_enqueue(rt_rq); | |
05fa785c | 871 | raw_spin_unlock(&rq->lock); |
eff6549b PZ |
872 | } |
873 | ||
42c62a58 PZ |
874 | if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)) |
875 | return 1; | |
876 | ||
eff6549b PZ |
877 | return idle; |
878 | } | |
ac086bc2 | 879 | |
6f505b16 PZ |
880 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
881 | { | |
052f1dc7 | 882 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
883 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
884 | ||
885 | if (rt_rq) | |
e864c499 | 886 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
887 | #endif |
888 | ||
889 | return rt_task_of(rt_se)->prio; | |
890 | } | |
891 | ||
9f0c1e56 | 892 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 893 | { |
9f0c1e56 | 894 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 895 | |
fa85ae24 | 896 | if (rt_rq->rt_throttled) |
23b0fdfc | 897 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 898 | |
5b680fd6 | 899 | if (runtime >= sched_rt_period(rt_rq)) |
ac086bc2 PZ |
900 | return 0; |
901 | ||
b79f3833 PZ |
902 | balance_runtime(rt_rq); |
903 | runtime = sched_rt_runtime(rt_rq); | |
904 | if (runtime == RUNTIME_INF) | |
905 | return 0; | |
ac086bc2 | 906 | |
9f0c1e56 | 907 | if (rt_rq->rt_time > runtime) { |
7abc63b1 PZ |
908 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
909 | ||
910 | /* | |
911 | * Don't actually throttle groups that have no runtime assigned | |
912 | * but accrue some time due to boosting. | |
913 | */ | |
914 | if (likely(rt_b->rt_runtime)) { | |
915 | rt_rq->rt_throttled = 1; | |
c224815d | 916 | printk_deferred_once("sched: RT throttling activated\n"); |
7abc63b1 PZ |
917 | } else { |
918 | /* | |
919 | * In case we did anyway, make it go away, | |
920 | * replenishment is a joke, since it will replenish us | |
921 | * with exactly 0 ns. | |
922 | */ | |
923 | rt_rq->rt_time = 0; | |
924 | } | |
925 | ||
23b0fdfc | 926 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 927 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
928 | return 1; |
929 | } | |
fa85ae24 PZ |
930 | } |
931 | ||
932 | return 0; | |
933 | } | |
934 | ||
bb44e5d1 IM |
935 | /* |
936 | * Update the current task's runtime statistics. Skip current tasks that | |
937 | * are not in our scheduling class. | |
938 | */ | |
a9957449 | 939 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
940 | { |
941 | struct task_struct *curr = rq->curr; | |
6f505b16 | 942 | struct sched_rt_entity *rt_se = &curr->rt; |
bb44e5d1 IM |
943 | u64 delta_exec; |
944 | ||
06c3bc65 | 945 | if (curr->sched_class != &rt_sched_class) |
bb44e5d1 IM |
946 | return; |
947 | ||
78becc27 | 948 | delta_exec = rq_clock_task(rq) - curr->se.exec_start; |
fc79e240 KT |
949 | if (unlikely((s64)delta_exec <= 0)) |
950 | return; | |
6cfb0d5d | 951 | |
42c62a58 PZ |
952 | schedstat_set(curr->se.statistics.exec_max, |
953 | max(curr->se.statistics.exec_max, delta_exec)); | |
bb44e5d1 IM |
954 | |
955 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
956 | account_group_exec_runtime(curr, delta_exec); |
957 | ||
78becc27 | 958 | curr->se.exec_start = rq_clock_task(rq); |
d842de87 | 959 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 960 | |
e9e9250b PZ |
961 | sched_rt_avg_update(rq, delta_exec); |
962 | ||
0b148fa0 PZ |
963 | if (!rt_bandwidth_enabled()) |
964 | return; | |
965 | ||
354d60c2 | 966 | for_each_sched_rt_entity(rt_se) { |
0b07939c | 967 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
354d60c2 | 968 | |
cc2991cf | 969 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
0986b11b | 970 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
971 | rt_rq->rt_time += delta_exec; |
972 | if (sched_rt_runtime_exceeded(rt_rq)) | |
8875125e | 973 | resched_curr(rq); |
0986b11b | 974 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 975 | } |
354d60c2 | 976 | } |
bb44e5d1 IM |
977 | } |
978 | ||
f4ebcbc0 KT |
979 | static void |
980 | dequeue_top_rt_rq(struct rt_rq *rt_rq) | |
981 | { | |
982 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
983 | ||
984 | BUG_ON(&rq->rt != rt_rq); | |
985 | ||
986 | if (!rt_rq->rt_queued) | |
987 | return; | |
988 | ||
989 | BUG_ON(!rq->nr_running); | |
990 | ||
72465447 | 991 | sub_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
992 | rt_rq->rt_queued = 0; |
993 | } | |
994 | ||
995 | static void | |
996 | enqueue_top_rt_rq(struct rt_rq *rt_rq) | |
997 | { | |
998 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
999 | ||
1000 | BUG_ON(&rq->rt != rt_rq); | |
1001 | ||
1002 | if (rt_rq->rt_queued) | |
1003 | return; | |
1004 | if (rt_rq_throttled(rt_rq) || !rt_rq->rt_nr_running) | |
1005 | return; | |
1006 | ||
72465447 | 1007 | add_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
1008 | rt_rq->rt_queued = 1; |
1009 | } | |
1010 | ||
398a153b | 1011 | #if defined CONFIG_SMP |
e864c499 | 1012 | |
398a153b GH |
1013 | static void |
1014 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 1015 | { |
4d984277 | 1016 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 1017 | |
757dfcaa KT |
1018 | #ifdef CONFIG_RT_GROUP_SCHED |
1019 | /* | |
1020 | * Change rq's cpupri only if rt_rq is the top queue. | |
1021 | */ | |
1022 | if (&rq->rt != rt_rq) | |
1023 | return; | |
1024 | #endif | |
5181f4a4 SR |
1025 | if (rq->online && prio < prev_prio) |
1026 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); | |
398a153b | 1027 | } |
73fe6aae | 1028 | |
398a153b GH |
1029 | static void |
1030 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
1031 | { | |
1032 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 1033 | |
757dfcaa KT |
1034 | #ifdef CONFIG_RT_GROUP_SCHED |
1035 | /* | |
1036 | * Change rq's cpupri only if rt_rq is the top queue. | |
1037 | */ | |
1038 | if (&rq->rt != rt_rq) | |
1039 | return; | |
1040 | #endif | |
398a153b GH |
1041 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
1042 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
1043 | } |
1044 | ||
398a153b GH |
1045 | #else /* CONFIG_SMP */ |
1046 | ||
6f505b16 | 1047 | static inline |
398a153b GH |
1048 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
1049 | static inline | |
1050 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
1051 | ||
1052 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 1053 | |
052f1dc7 | 1054 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
1055 | static void |
1056 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
1057 | { | |
1058 | int prev_prio = rt_rq->highest_prio.curr; | |
1059 | ||
1060 | if (prio < prev_prio) | |
1061 | rt_rq->highest_prio.curr = prio; | |
1062 | ||
1063 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
1064 | } | |
1065 | ||
1066 | static void | |
1067 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
1068 | { | |
1069 | int prev_prio = rt_rq->highest_prio.curr; | |
1070 | ||
6f505b16 | 1071 | if (rt_rq->rt_nr_running) { |
764a9d6f | 1072 | |
398a153b | 1073 | WARN_ON(prio < prev_prio); |
764a9d6f | 1074 | |
e864c499 | 1075 | /* |
398a153b GH |
1076 | * This may have been our highest task, and therefore |
1077 | * we may have some recomputation to do | |
e864c499 | 1078 | */ |
398a153b | 1079 | if (prio == prev_prio) { |
e864c499 GH |
1080 | struct rt_prio_array *array = &rt_rq->active; |
1081 | ||
1082 | rt_rq->highest_prio.curr = | |
764a9d6f | 1083 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
1084 | } |
1085 | ||
764a9d6f | 1086 | } else |
e864c499 | 1087 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 1088 | |
398a153b GH |
1089 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
1090 | } | |
1f11eb6a | 1091 | |
398a153b GH |
1092 | #else |
1093 | ||
1094 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1095 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1096 | ||
1097 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 1098 | |
052f1dc7 | 1099 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
1100 | |
1101 | static void | |
1102 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1103 | { | |
1104 | if (rt_se_boosted(rt_se)) | |
1105 | rt_rq->rt_nr_boosted++; | |
1106 | ||
1107 | if (rt_rq->tg) | |
1108 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
1109 | } | |
1110 | ||
1111 | static void | |
1112 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1113 | { | |
23b0fdfc PZ |
1114 | if (rt_se_boosted(rt_se)) |
1115 | rt_rq->rt_nr_boosted--; | |
1116 | ||
1117 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
1118 | } |
1119 | ||
1120 | #else /* CONFIG_RT_GROUP_SCHED */ | |
1121 | ||
1122 | static void | |
1123 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1124 | { | |
1125 | start_rt_bandwidth(&def_rt_bandwidth); | |
1126 | } | |
1127 | ||
1128 | static inline | |
1129 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
1130 | ||
1131 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
1132 | ||
22abdef3 KT |
1133 | static inline |
1134 | unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se) | |
1135 | { | |
1136 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
1137 | ||
1138 | if (group_rq) | |
1139 | return group_rq->rt_nr_running; | |
1140 | else | |
1141 | return 1; | |
1142 | } | |
1143 | ||
398a153b GH |
1144 | static inline |
1145 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1146 | { | |
1147 | int prio = rt_se_prio(rt_se); | |
1148 | ||
1149 | WARN_ON(!rt_prio(prio)); | |
22abdef3 | 1150 | rt_rq->rt_nr_running += rt_se_nr_running(rt_se); |
398a153b GH |
1151 | |
1152 | inc_rt_prio(rt_rq, prio); | |
1153 | inc_rt_migration(rt_se, rt_rq); | |
1154 | inc_rt_group(rt_se, rt_rq); | |
1155 | } | |
1156 | ||
1157 | static inline | |
1158 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1159 | { | |
1160 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
1161 | WARN_ON(!rt_rq->rt_nr_running); | |
22abdef3 | 1162 | rt_rq->rt_nr_running -= rt_se_nr_running(rt_se); |
398a153b GH |
1163 | |
1164 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
1165 | dec_rt_migration(rt_se, rt_rq); | |
1166 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
1167 | } |
1168 | ||
37dad3fc | 1169 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
bb44e5d1 | 1170 | { |
6f505b16 PZ |
1171 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
1172 | struct rt_prio_array *array = &rt_rq->active; | |
1173 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 1174 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 1175 | |
ad2a3f13 PZ |
1176 | /* |
1177 | * Don't enqueue the group if its throttled, or when empty. | |
1178 | * The latter is a consequence of the former when a child group | |
1179 | * get throttled and the current group doesn't have any other | |
1180 | * active members. | |
1181 | */ | |
1182 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 1183 | return; |
63489e45 | 1184 | |
37dad3fc TG |
1185 | if (head) |
1186 | list_add(&rt_se->run_list, queue); | |
1187 | else | |
1188 | list_add_tail(&rt_se->run_list, queue); | |
6f505b16 | 1189 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 1190 | |
6f505b16 PZ |
1191 | inc_rt_tasks(rt_se, rt_rq); |
1192 | } | |
1193 | ||
ad2a3f13 | 1194 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
1195 | { |
1196 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
1197 | struct rt_prio_array *array = &rt_rq->active; | |
1198 | ||
1199 | list_del_init(&rt_se->run_list); | |
1200 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
1201 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
1202 | ||
1203 | dec_rt_tasks(rt_se, rt_rq); | |
1204 | } | |
1205 | ||
1206 | /* | |
1207 | * Because the prio of an upper entry depends on the lower | |
1208 | * entries, we must remove entries top - down. | |
6f505b16 | 1209 | */ |
ad2a3f13 | 1210 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 1211 | { |
ad2a3f13 | 1212 | struct sched_rt_entity *back = NULL; |
6f505b16 | 1213 | |
58d6c2d7 PZ |
1214 | for_each_sched_rt_entity(rt_se) { |
1215 | rt_se->back = back; | |
1216 | back = rt_se; | |
1217 | } | |
1218 | ||
f4ebcbc0 KT |
1219 | dequeue_top_rt_rq(rt_rq_of_se(back)); |
1220 | ||
58d6c2d7 PZ |
1221 | for (rt_se = back; rt_se; rt_se = rt_se->back) { |
1222 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
1223 | __dequeue_rt_entity(rt_se); |
1224 | } | |
1225 | } | |
1226 | ||
37dad3fc | 1227 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
ad2a3f13 | 1228 | { |
f4ebcbc0 KT |
1229 | struct rq *rq = rq_of_rt_se(rt_se); |
1230 | ||
ad2a3f13 PZ |
1231 | dequeue_rt_stack(rt_se); |
1232 | for_each_sched_rt_entity(rt_se) | |
37dad3fc | 1233 | __enqueue_rt_entity(rt_se, head); |
f4ebcbc0 | 1234 | enqueue_top_rt_rq(&rq->rt); |
ad2a3f13 PZ |
1235 | } |
1236 | ||
1237 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
1238 | { | |
f4ebcbc0 KT |
1239 | struct rq *rq = rq_of_rt_se(rt_se); |
1240 | ||
ad2a3f13 PZ |
1241 | dequeue_rt_stack(rt_se); |
1242 | ||
1243 | for_each_sched_rt_entity(rt_se) { | |
1244 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
1245 | ||
1246 | if (rt_rq && rt_rq->rt_nr_running) | |
37dad3fc | 1247 | __enqueue_rt_entity(rt_se, false); |
58d6c2d7 | 1248 | } |
f4ebcbc0 | 1249 | enqueue_top_rt_rq(&rq->rt); |
bb44e5d1 IM |
1250 | } |
1251 | ||
1252 | /* | |
1253 | * Adding/removing a task to/from a priority array: | |
1254 | */ | |
ea87bb78 | 1255 | static void |
371fd7e7 | 1256 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
6f505b16 PZ |
1257 | { |
1258 | struct sched_rt_entity *rt_se = &p->rt; | |
1259 | ||
371fd7e7 | 1260 | if (flags & ENQUEUE_WAKEUP) |
6f505b16 PZ |
1261 | rt_se->timeout = 0; |
1262 | ||
371fd7e7 | 1263 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
c09595f6 | 1264 | |
29baa747 | 1265 | if (!task_current(rq, p) && p->nr_cpus_allowed > 1) |
917b627d | 1266 | enqueue_pushable_task(rq, p); |
6f505b16 PZ |
1267 | } |
1268 | ||
371fd7e7 | 1269 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1270 | { |
6f505b16 | 1271 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 1272 | |
f1e14ef6 | 1273 | update_curr_rt(rq); |
ad2a3f13 | 1274 | dequeue_rt_entity(rt_se); |
c09595f6 | 1275 | |
917b627d | 1276 | dequeue_pushable_task(rq, p); |
bb44e5d1 IM |
1277 | } |
1278 | ||
1279 | /* | |
60686317 RW |
1280 | * Put task to the head or the end of the run list without the overhead of |
1281 | * dequeue followed by enqueue. | |
bb44e5d1 | 1282 | */ |
7ebefa8c DA |
1283 | static void |
1284 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 1285 | { |
1cdad715 | 1286 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
1287 | struct rt_prio_array *array = &rt_rq->active; |
1288 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
1289 | ||
1290 | if (head) | |
1291 | list_move(&rt_se->run_list, queue); | |
1292 | else | |
1293 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 1294 | } |
6f505b16 PZ |
1295 | } |
1296 | ||
7ebefa8c | 1297 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 1298 | { |
6f505b16 PZ |
1299 | struct sched_rt_entity *rt_se = &p->rt; |
1300 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1301 | |
6f505b16 PZ |
1302 | for_each_sched_rt_entity(rt_se) { |
1303 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 1304 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 1305 | } |
bb44e5d1 IM |
1306 | } |
1307 | ||
6f505b16 | 1308 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 1309 | { |
7ebefa8c | 1310 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
1311 | } |
1312 | ||
e7693a36 | 1313 | #ifdef CONFIG_SMP |
318e0893 GH |
1314 | static int find_lowest_rq(struct task_struct *task); |
1315 | ||
0017d735 | 1316 | static int |
ac66f547 | 1317 | select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags) |
e7693a36 | 1318 | { |
7608dec2 PZ |
1319 | struct task_struct *curr; |
1320 | struct rq *rq; | |
c37495fd SR |
1321 | |
1322 | /* For anything but wake ups, just return the task_cpu */ | |
1323 | if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK) | |
1324 | goto out; | |
1325 | ||
7608dec2 PZ |
1326 | rq = cpu_rq(cpu); |
1327 | ||
1328 | rcu_read_lock(); | |
316c1608 | 1329 | curr = READ_ONCE(rq->curr); /* unlocked access */ |
7608dec2 | 1330 | |
318e0893 | 1331 | /* |
7608dec2 | 1332 | * If the current task on @p's runqueue is an RT task, then |
e1f47d89 SR |
1333 | * try to see if we can wake this RT task up on another |
1334 | * runqueue. Otherwise simply start this RT task | |
1335 | * on its current runqueue. | |
1336 | * | |
43fa5460 SR |
1337 | * We want to avoid overloading runqueues. If the woken |
1338 | * task is a higher priority, then it will stay on this CPU | |
1339 | * and the lower prio task should be moved to another CPU. | |
1340 | * Even though this will probably make the lower prio task | |
1341 | * lose its cache, we do not want to bounce a higher task | |
1342 | * around just because it gave up its CPU, perhaps for a | |
1343 | * lock? | |
1344 | * | |
1345 | * For equal prio tasks, we just let the scheduler sort it out. | |
7608dec2 PZ |
1346 | * |
1347 | * Otherwise, just let it ride on the affined RQ and the | |
1348 | * post-schedule router will push the preempted task away | |
1349 | * | |
1350 | * This test is optimistic, if we get it wrong the load-balancer | |
1351 | * will have to sort it out. | |
318e0893 | 1352 | */ |
7608dec2 | 1353 | if (curr && unlikely(rt_task(curr)) && |
29baa747 | 1354 | (curr->nr_cpus_allowed < 2 || |
6bfa687c | 1355 | curr->prio <= p->prio)) { |
7608dec2 | 1356 | int target = find_lowest_rq(p); |
318e0893 | 1357 | |
80e3d87b TC |
1358 | /* |
1359 | * Don't bother moving it if the destination CPU is | |
1360 | * not running a lower priority task. | |
1361 | */ | |
1362 | if (target != -1 && | |
1363 | p->prio < cpu_rq(target)->rt.highest_prio.curr) | |
7608dec2 | 1364 | cpu = target; |
318e0893 | 1365 | } |
7608dec2 | 1366 | rcu_read_unlock(); |
318e0893 | 1367 | |
c37495fd | 1368 | out: |
7608dec2 | 1369 | return cpu; |
e7693a36 | 1370 | } |
7ebefa8c DA |
1371 | |
1372 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
1373 | { | |
308a623a WL |
1374 | /* |
1375 | * Current can't be migrated, useless to reschedule, | |
1376 | * let's hope p can move out. | |
1377 | */ | |
1378 | if (rq->curr->nr_cpus_allowed == 1 || | |
1379 | !cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) | |
7ebefa8c DA |
1380 | return; |
1381 | ||
308a623a WL |
1382 | /* |
1383 | * p is migratable, so let's not schedule it and | |
1384 | * see if it is pushed or pulled somewhere else. | |
1385 | */ | |
29baa747 | 1386 | if (p->nr_cpus_allowed != 1 |
13b8bd0a RR |
1387 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1388 | return; | |
24600ce8 | 1389 | |
7ebefa8c DA |
1390 | /* |
1391 | * There appears to be other cpus that can accept | |
1392 | * current and none to run 'p', so lets reschedule | |
1393 | * to try and push current away: | |
1394 | */ | |
1395 | requeue_task_rt(rq, p, 1); | |
8875125e | 1396 | resched_curr(rq); |
7ebefa8c DA |
1397 | } |
1398 | ||
e7693a36 GH |
1399 | #endif /* CONFIG_SMP */ |
1400 | ||
bb44e5d1 IM |
1401 | /* |
1402 | * Preempt the current task with a newly woken task if needed: | |
1403 | */ | |
7d478721 | 1404 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1405 | { |
45c01e82 | 1406 | if (p->prio < rq->curr->prio) { |
8875125e | 1407 | resched_curr(rq); |
45c01e82 GH |
1408 | return; |
1409 | } | |
1410 | ||
1411 | #ifdef CONFIG_SMP | |
1412 | /* | |
1413 | * If: | |
1414 | * | |
1415 | * - the newly woken task is of equal priority to the current task | |
1416 | * - the newly woken task is non-migratable while current is migratable | |
1417 | * - current will be preempted on the next reschedule | |
1418 | * | |
1419 | * we should check to see if current can readily move to a different | |
1420 | * cpu. If so, we will reschedule to allow the push logic to try | |
1421 | * to move current somewhere else, making room for our non-migratable | |
1422 | * task. | |
1423 | */ | |
8dd0de8b | 1424 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
7ebefa8c | 1425 | check_preempt_equal_prio(rq, p); |
45c01e82 | 1426 | #endif |
bb44e5d1 IM |
1427 | } |
1428 | ||
6f505b16 PZ |
1429 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1430 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1431 | { |
6f505b16 PZ |
1432 | struct rt_prio_array *array = &rt_rq->active; |
1433 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1434 | struct list_head *queue; |
1435 | int idx; | |
1436 | ||
1437 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1438 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1439 | |
1440 | queue = array->queue + idx; | |
6f505b16 | 1441 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1442 | |
6f505b16 PZ |
1443 | return next; |
1444 | } | |
bb44e5d1 | 1445 | |
917b627d | 1446 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1447 | { |
1448 | struct sched_rt_entity *rt_se; | |
1449 | struct task_struct *p; | |
606dba2e | 1450 | struct rt_rq *rt_rq = &rq->rt; |
6f505b16 PZ |
1451 | |
1452 | do { | |
1453 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1454 | BUG_ON(!rt_se); |
6f505b16 PZ |
1455 | rt_rq = group_rt_rq(rt_se); |
1456 | } while (rt_rq); | |
1457 | ||
1458 | p = rt_task_of(rt_se); | |
78becc27 | 1459 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
1460 | |
1461 | return p; | |
1462 | } | |
1463 | ||
606dba2e PZ |
1464 | static struct task_struct * |
1465 | pick_next_task_rt(struct rq *rq, struct task_struct *prev) | |
917b627d | 1466 | { |
606dba2e PZ |
1467 | struct task_struct *p; |
1468 | struct rt_rq *rt_rq = &rq->rt; | |
1469 | ||
37e117c0 | 1470 | if (need_pull_rt_task(rq, prev)) { |
cbce1a68 PZ |
1471 | /* |
1472 | * This is OK, because current is on_cpu, which avoids it being | |
1473 | * picked for load-balance and preemption/IRQs are still | |
1474 | * disabled avoiding further scheduler activity on it and we're | |
1475 | * being very careful to re-start the picking loop. | |
1476 | */ | |
1477 | lockdep_unpin_lock(&rq->lock); | |
38033c37 | 1478 | pull_rt_task(rq); |
cbce1a68 | 1479 | lockdep_pin_lock(&rq->lock); |
37e117c0 PZ |
1480 | /* |
1481 | * pull_rt_task() can drop (and re-acquire) rq->lock; this | |
a1d9a323 KT |
1482 | * means a dl or stop task can slip in, in which case we need |
1483 | * to re-start task selection. | |
37e117c0 | 1484 | */ |
da0c1e65 | 1485 | if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) || |
a1d9a323 | 1486 | rq->dl.dl_nr_running)) |
37e117c0 PZ |
1487 | return RETRY_TASK; |
1488 | } | |
38033c37 | 1489 | |
734ff2a7 KT |
1490 | /* |
1491 | * We may dequeue prev's rt_rq in put_prev_task(). | |
1492 | * So, we update time before rt_nr_running check. | |
1493 | */ | |
1494 | if (prev->sched_class == &rt_sched_class) | |
1495 | update_curr_rt(rq); | |
1496 | ||
f4ebcbc0 | 1497 | if (!rt_rq->rt_queued) |
606dba2e PZ |
1498 | return NULL; |
1499 | ||
3f1d2a31 | 1500 | put_prev_task(rq, prev); |
606dba2e PZ |
1501 | |
1502 | p = _pick_next_task_rt(rq); | |
917b627d GH |
1503 | |
1504 | /* The running task is never eligible for pushing */ | |
f3f1768f | 1505 | dequeue_pushable_task(rq, p); |
917b627d | 1506 | |
e3fca9e7 | 1507 | queue_push_tasks(rq); |
3f029d3c | 1508 | |
6f505b16 | 1509 | return p; |
bb44e5d1 IM |
1510 | } |
1511 | ||
31ee529c | 1512 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1513 | { |
f1e14ef6 | 1514 | update_curr_rt(rq); |
917b627d GH |
1515 | |
1516 | /* | |
1517 | * The previous task needs to be made eligible for pushing | |
1518 | * if it is still active | |
1519 | */ | |
29baa747 | 1520 | if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1) |
917b627d | 1521 | enqueue_pushable_task(rq, p); |
bb44e5d1 IM |
1522 | } |
1523 | ||
681f3e68 | 1524 | #ifdef CONFIG_SMP |
6f505b16 | 1525 | |
e8fa1362 SR |
1526 | /* Only try algorithms three times */ |
1527 | #define RT_MAX_TRIES 3 | |
1528 | ||
f65eda4f SR |
1529 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1530 | { | |
1531 | if (!task_running(rq, p) && | |
60334caf | 1532 | cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) |
f65eda4f SR |
1533 | return 1; |
1534 | return 0; | |
1535 | } | |
1536 | ||
e23ee747 KT |
1537 | /* |
1538 | * Return the highest pushable rq's task, which is suitable to be executed | |
1539 | * on the cpu, NULL otherwise | |
1540 | */ | |
1541 | static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu) | |
e8fa1362 | 1542 | { |
e23ee747 KT |
1543 | struct plist_head *head = &rq->rt.pushable_tasks; |
1544 | struct task_struct *p; | |
3d07467b | 1545 | |
e23ee747 KT |
1546 | if (!has_pushable_tasks(rq)) |
1547 | return NULL; | |
3d07467b | 1548 | |
e23ee747 KT |
1549 | plist_for_each_entry(p, head, pushable_tasks) { |
1550 | if (pick_rt_task(rq, p, cpu)) | |
1551 | return p; | |
f65eda4f SR |
1552 | } |
1553 | ||
e23ee747 | 1554 | return NULL; |
e8fa1362 SR |
1555 | } |
1556 | ||
0e3900e6 | 1557 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1558 | |
6e1254d2 GH |
1559 | static int find_lowest_rq(struct task_struct *task) |
1560 | { | |
1561 | struct sched_domain *sd; | |
4ba29684 | 1562 | struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask); |
6e1254d2 GH |
1563 | int this_cpu = smp_processor_id(); |
1564 | int cpu = task_cpu(task); | |
06f90dbd | 1565 | |
0da938c4 SR |
1566 | /* Make sure the mask is initialized first */ |
1567 | if (unlikely(!lowest_mask)) | |
1568 | return -1; | |
1569 | ||
29baa747 | 1570 | if (task->nr_cpus_allowed == 1) |
6e0534f2 | 1571 | return -1; /* No other targets possible */ |
6e1254d2 | 1572 | |
6e0534f2 GH |
1573 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1574 | return -1; /* No targets found */ | |
6e1254d2 GH |
1575 | |
1576 | /* | |
1577 | * At this point we have built a mask of cpus representing the | |
1578 | * lowest priority tasks in the system. Now we want to elect | |
1579 | * the best one based on our affinity and topology. | |
1580 | * | |
1581 | * We prioritize the last cpu that the task executed on since | |
1582 | * it is most likely cache-hot in that location. | |
1583 | */ | |
96f874e2 | 1584 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1585 | return cpu; |
1586 | ||
1587 | /* | |
1588 | * Otherwise, we consult the sched_domains span maps to figure | |
1589 | * out which cpu is logically closest to our hot cache data. | |
1590 | */ | |
e2c88063 RR |
1591 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1592 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ | |
6e1254d2 | 1593 | |
cd4ae6ad | 1594 | rcu_read_lock(); |
e2c88063 RR |
1595 | for_each_domain(cpu, sd) { |
1596 | if (sd->flags & SD_WAKE_AFFINE) { | |
1597 | int best_cpu; | |
6e1254d2 | 1598 | |
e2c88063 RR |
1599 | /* |
1600 | * "this_cpu" is cheaper to preempt than a | |
1601 | * remote processor. | |
1602 | */ | |
1603 | if (this_cpu != -1 && | |
cd4ae6ad XF |
1604 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
1605 | rcu_read_unlock(); | |
e2c88063 | 1606 | return this_cpu; |
cd4ae6ad | 1607 | } |
e2c88063 RR |
1608 | |
1609 | best_cpu = cpumask_first_and(lowest_mask, | |
1610 | sched_domain_span(sd)); | |
cd4ae6ad XF |
1611 | if (best_cpu < nr_cpu_ids) { |
1612 | rcu_read_unlock(); | |
e2c88063 | 1613 | return best_cpu; |
cd4ae6ad | 1614 | } |
6e1254d2 GH |
1615 | } |
1616 | } | |
cd4ae6ad | 1617 | rcu_read_unlock(); |
6e1254d2 GH |
1618 | |
1619 | /* | |
1620 | * And finally, if there were no matches within the domains | |
1621 | * just give the caller *something* to work with from the compatible | |
1622 | * locations. | |
1623 | */ | |
e2c88063 RR |
1624 | if (this_cpu != -1) |
1625 | return this_cpu; | |
1626 | ||
1627 | cpu = cpumask_any(lowest_mask); | |
1628 | if (cpu < nr_cpu_ids) | |
1629 | return cpu; | |
1630 | return -1; | |
07b4032c GH |
1631 | } |
1632 | ||
1633 | /* Will lock the rq it finds */ | |
4df64c0b | 1634 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1635 | { |
1636 | struct rq *lowest_rq = NULL; | |
07b4032c | 1637 | int tries; |
4df64c0b | 1638 | int cpu; |
e8fa1362 | 1639 | |
07b4032c GH |
1640 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1641 | cpu = find_lowest_rq(task); | |
1642 | ||
2de0b463 | 1643 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1644 | break; |
1645 | ||
07b4032c GH |
1646 | lowest_rq = cpu_rq(cpu); |
1647 | ||
80e3d87b TC |
1648 | if (lowest_rq->rt.highest_prio.curr <= task->prio) { |
1649 | /* | |
1650 | * Target rq has tasks of equal or higher priority, | |
1651 | * retrying does not release any lock and is unlikely | |
1652 | * to yield a different result. | |
1653 | */ | |
1654 | lowest_rq = NULL; | |
1655 | break; | |
1656 | } | |
1657 | ||
e8fa1362 | 1658 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1659 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1660 | /* |
1661 | * We had to unlock the run queue. In | |
1662 | * the mean time, task could have | |
1663 | * migrated already or had its affinity changed. | |
1664 | * Also make sure that it wasn't scheduled on its rq. | |
1665 | */ | |
07b4032c | 1666 | if (unlikely(task_rq(task) != rq || |
96f874e2 | 1667 | !cpumask_test_cpu(lowest_rq->cpu, |
fa17b507 | 1668 | tsk_cpus_allowed(task)) || |
07b4032c | 1669 | task_running(rq, task) || |
da0c1e65 | 1670 | !task_on_rq_queued(task))) { |
4df64c0b | 1671 | |
7f1b4393 | 1672 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1673 | lowest_rq = NULL; |
1674 | break; | |
1675 | } | |
1676 | } | |
1677 | ||
1678 | /* If this rq is still suitable use it. */ | |
e864c499 | 1679 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1680 | break; |
1681 | ||
1682 | /* try again */ | |
1b12bbc7 | 1683 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1684 | lowest_rq = NULL; |
1685 | } | |
1686 | ||
1687 | return lowest_rq; | |
1688 | } | |
1689 | ||
917b627d GH |
1690 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1691 | { | |
1692 | struct task_struct *p; | |
1693 | ||
1694 | if (!has_pushable_tasks(rq)) | |
1695 | return NULL; | |
1696 | ||
1697 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1698 | struct task_struct, pushable_tasks); | |
1699 | ||
1700 | BUG_ON(rq->cpu != task_cpu(p)); | |
1701 | BUG_ON(task_current(rq, p)); | |
29baa747 | 1702 | BUG_ON(p->nr_cpus_allowed <= 1); |
917b627d | 1703 | |
da0c1e65 | 1704 | BUG_ON(!task_on_rq_queued(p)); |
917b627d GH |
1705 | BUG_ON(!rt_task(p)); |
1706 | ||
1707 | return p; | |
1708 | } | |
1709 | ||
e8fa1362 SR |
1710 | /* |
1711 | * If the current CPU has more than one RT task, see if the non | |
1712 | * running task can migrate over to a CPU that is running a task | |
1713 | * of lesser priority. | |
1714 | */ | |
697f0a48 | 1715 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1716 | { |
1717 | struct task_struct *next_task; | |
1718 | struct rq *lowest_rq; | |
311e800e | 1719 | int ret = 0; |
e8fa1362 | 1720 | |
a22d7fc1 GH |
1721 | if (!rq->rt.overloaded) |
1722 | return 0; | |
1723 | ||
917b627d | 1724 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1725 | if (!next_task) |
1726 | return 0; | |
1727 | ||
49246274 | 1728 | retry: |
697f0a48 | 1729 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1730 | WARN_ON(1); |
e8fa1362 | 1731 | return 0; |
f65eda4f | 1732 | } |
e8fa1362 SR |
1733 | |
1734 | /* | |
1735 | * It's possible that the next_task slipped in of | |
1736 | * higher priority than current. If that's the case | |
1737 | * just reschedule current. | |
1738 | */ | |
697f0a48 | 1739 | if (unlikely(next_task->prio < rq->curr->prio)) { |
8875125e | 1740 | resched_curr(rq); |
e8fa1362 SR |
1741 | return 0; |
1742 | } | |
1743 | ||
697f0a48 | 1744 | /* We might release rq lock */ |
e8fa1362 SR |
1745 | get_task_struct(next_task); |
1746 | ||
1747 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1748 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1749 | if (!lowest_rq) { |
1750 | struct task_struct *task; | |
1751 | /* | |
311e800e | 1752 | * find_lock_lowest_rq releases rq->lock |
1563513d GH |
1753 | * so it is possible that next_task has migrated. |
1754 | * | |
1755 | * We need to make sure that the task is still on the same | |
1756 | * run-queue and is also still the next task eligible for | |
1757 | * pushing. | |
e8fa1362 | 1758 | */ |
917b627d | 1759 | task = pick_next_pushable_task(rq); |
1563513d GH |
1760 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1761 | /* | |
311e800e HD |
1762 | * The task hasn't migrated, and is still the next |
1763 | * eligible task, but we failed to find a run-queue | |
1764 | * to push it to. Do not retry in this case, since | |
1765 | * other cpus will pull from us when ready. | |
1563513d | 1766 | */ |
1563513d | 1767 | goto out; |
e8fa1362 | 1768 | } |
917b627d | 1769 | |
1563513d GH |
1770 | if (!task) |
1771 | /* No more tasks, just exit */ | |
1772 | goto out; | |
1773 | ||
917b627d | 1774 | /* |
1563513d | 1775 | * Something has shifted, try again. |
917b627d | 1776 | */ |
1563513d GH |
1777 | put_task_struct(next_task); |
1778 | next_task = task; | |
1779 | goto retry; | |
e8fa1362 SR |
1780 | } |
1781 | ||
697f0a48 | 1782 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1783 | set_task_cpu(next_task, lowest_rq->cpu); |
1784 | activate_task(lowest_rq, next_task, 0); | |
311e800e | 1785 | ret = 1; |
e8fa1362 | 1786 | |
8875125e | 1787 | resched_curr(lowest_rq); |
e8fa1362 | 1788 | |
1b12bbc7 | 1789 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1790 | |
e8fa1362 SR |
1791 | out: |
1792 | put_task_struct(next_task); | |
1793 | ||
311e800e | 1794 | return ret; |
e8fa1362 SR |
1795 | } |
1796 | ||
e8fa1362 SR |
1797 | static void push_rt_tasks(struct rq *rq) |
1798 | { | |
1799 | /* push_rt_task will return true if it moved an RT */ | |
1800 | while (push_rt_task(rq)) | |
1801 | ; | |
1802 | } | |
1803 | ||
b6366f04 SR |
1804 | #ifdef HAVE_RT_PUSH_IPI |
1805 | /* | |
1806 | * The search for the next cpu always starts at rq->cpu and ends | |
1807 | * when we reach rq->cpu again. It will never return rq->cpu. | |
1808 | * This returns the next cpu to check, or nr_cpu_ids if the loop | |
1809 | * is complete. | |
1810 | * | |
1811 | * rq->rt.push_cpu holds the last cpu returned by this function, | |
1812 | * or if this is the first instance, it must hold rq->cpu. | |
1813 | */ | |
1814 | static int rto_next_cpu(struct rq *rq) | |
1815 | { | |
1816 | int prev_cpu = rq->rt.push_cpu; | |
1817 | int cpu; | |
1818 | ||
1819 | cpu = cpumask_next(prev_cpu, rq->rd->rto_mask); | |
1820 | ||
1821 | /* | |
1822 | * If the previous cpu is less than the rq's CPU, then it already | |
1823 | * passed the end of the mask, and has started from the beginning. | |
1824 | * We end if the next CPU is greater or equal to rq's CPU. | |
1825 | */ | |
1826 | if (prev_cpu < rq->cpu) { | |
1827 | if (cpu >= rq->cpu) | |
1828 | return nr_cpu_ids; | |
1829 | ||
1830 | } else if (cpu >= nr_cpu_ids) { | |
1831 | /* | |
1832 | * We passed the end of the mask, start at the beginning. | |
1833 | * If the result is greater or equal to the rq's CPU, then | |
1834 | * the loop is finished. | |
1835 | */ | |
1836 | cpu = cpumask_first(rq->rd->rto_mask); | |
1837 | if (cpu >= rq->cpu) | |
1838 | return nr_cpu_ids; | |
1839 | } | |
1840 | rq->rt.push_cpu = cpu; | |
1841 | ||
1842 | /* Return cpu to let the caller know if the loop is finished or not */ | |
1843 | return cpu; | |
1844 | } | |
1845 | ||
1846 | static int find_next_push_cpu(struct rq *rq) | |
1847 | { | |
1848 | struct rq *next_rq; | |
1849 | int cpu; | |
1850 | ||
1851 | while (1) { | |
1852 | cpu = rto_next_cpu(rq); | |
1853 | if (cpu >= nr_cpu_ids) | |
1854 | break; | |
1855 | next_rq = cpu_rq(cpu); | |
1856 | ||
1857 | /* Make sure the next rq can push to this rq */ | |
1858 | if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr) | |
1859 | break; | |
1860 | } | |
1861 | ||
1862 | return cpu; | |
1863 | } | |
1864 | ||
1865 | #define RT_PUSH_IPI_EXECUTING 1 | |
1866 | #define RT_PUSH_IPI_RESTART 2 | |
1867 | ||
1868 | static void tell_cpu_to_push(struct rq *rq) | |
1869 | { | |
1870 | int cpu; | |
1871 | ||
1872 | if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { | |
1873 | raw_spin_lock(&rq->rt.push_lock); | |
1874 | /* Make sure it's still executing */ | |
1875 | if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { | |
1876 | /* | |
1877 | * Tell the IPI to restart the loop as things have | |
1878 | * changed since it started. | |
1879 | */ | |
1880 | rq->rt.push_flags |= RT_PUSH_IPI_RESTART; | |
1881 | raw_spin_unlock(&rq->rt.push_lock); | |
1882 | return; | |
1883 | } | |
1884 | raw_spin_unlock(&rq->rt.push_lock); | |
1885 | } | |
1886 | ||
1887 | /* When here, there's no IPI going around */ | |
1888 | ||
1889 | rq->rt.push_cpu = rq->cpu; | |
1890 | cpu = find_next_push_cpu(rq); | |
1891 | if (cpu >= nr_cpu_ids) | |
1892 | return; | |
1893 | ||
1894 | rq->rt.push_flags = RT_PUSH_IPI_EXECUTING; | |
1895 | ||
1896 | irq_work_queue_on(&rq->rt.push_work, cpu); | |
1897 | } | |
1898 | ||
1899 | /* Called from hardirq context */ | |
1900 | static void try_to_push_tasks(void *arg) | |
1901 | { | |
1902 | struct rt_rq *rt_rq = arg; | |
1903 | struct rq *rq, *src_rq; | |
1904 | int this_cpu; | |
1905 | int cpu; | |
1906 | ||
1907 | this_cpu = rt_rq->push_cpu; | |
1908 | ||
1909 | /* Paranoid check */ | |
1910 | BUG_ON(this_cpu != smp_processor_id()); | |
1911 | ||
1912 | rq = cpu_rq(this_cpu); | |
1913 | src_rq = rq_of_rt_rq(rt_rq); | |
1914 | ||
1915 | again: | |
1916 | if (has_pushable_tasks(rq)) { | |
1917 | raw_spin_lock(&rq->lock); | |
1918 | push_rt_task(rq); | |
1919 | raw_spin_unlock(&rq->lock); | |
1920 | } | |
1921 | ||
1922 | /* Pass the IPI to the next rt overloaded queue */ | |
1923 | raw_spin_lock(&rt_rq->push_lock); | |
1924 | /* | |
1925 | * If the source queue changed since the IPI went out, | |
1926 | * we need to restart the search from that CPU again. | |
1927 | */ | |
1928 | if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) { | |
1929 | rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART; | |
1930 | rt_rq->push_cpu = src_rq->cpu; | |
1931 | } | |
1932 | ||
1933 | cpu = find_next_push_cpu(src_rq); | |
1934 | ||
1935 | if (cpu >= nr_cpu_ids) | |
1936 | rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING; | |
1937 | raw_spin_unlock(&rt_rq->push_lock); | |
1938 | ||
1939 | if (cpu >= nr_cpu_ids) | |
1940 | return; | |
1941 | ||
1942 | /* | |
1943 | * It is possible that a restart caused this CPU to be | |
1944 | * chosen again. Don't bother with an IPI, just see if we | |
1945 | * have more to push. | |
1946 | */ | |
1947 | if (unlikely(cpu == rq->cpu)) | |
1948 | goto again; | |
1949 | ||
1950 | /* Try the next RT overloaded CPU */ | |
1951 | irq_work_queue_on(&rt_rq->push_work, cpu); | |
1952 | } | |
1953 | ||
1954 | static void push_irq_work_func(struct irq_work *work) | |
1955 | { | |
1956 | struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work); | |
1957 | ||
1958 | try_to_push_tasks(rt_rq); | |
1959 | } | |
1960 | #endif /* HAVE_RT_PUSH_IPI */ | |
1961 | ||
8046d680 | 1962 | static void pull_rt_task(struct rq *this_rq) |
f65eda4f | 1963 | { |
8046d680 PZ |
1964 | int this_cpu = this_rq->cpu, cpu; |
1965 | bool resched = false; | |
a8728944 | 1966 | struct task_struct *p; |
f65eda4f | 1967 | struct rq *src_rq; |
f65eda4f | 1968 | |
637f5085 | 1969 | if (likely(!rt_overloaded(this_rq))) |
8046d680 | 1970 | return; |
f65eda4f | 1971 | |
7c3f2ab7 PZ |
1972 | /* |
1973 | * Match the barrier from rt_set_overloaded; this guarantees that if we | |
1974 | * see overloaded we must also see the rto_mask bit. | |
1975 | */ | |
1976 | smp_rmb(); | |
1977 | ||
b6366f04 SR |
1978 | #ifdef HAVE_RT_PUSH_IPI |
1979 | if (sched_feat(RT_PUSH_IPI)) { | |
1980 | tell_cpu_to_push(this_rq); | |
8046d680 | 1981 | return; |
b6366f04 SR |
1982 | } |
1983 | #endif | |
1984 | ||
c6c4927b | 1985 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1986 | if (this_cpu == cpu) |
1987 | continue; | |
1988 | ||
1989 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1990 | |
1991 | /* | |
1992 | * Don't bother taking the src_rq->lock if the next highest | |
1993 | * task is known to be lower-priority than our current task. | |
1994 | * This may look racy, but if this value is about to go | |
1995 | * logically higher, the src_rq will push this task away. | |
1996 | * And if its going logically lower, we do not care | |
1997 | */ | |
1998 | if (src_rq->rt.highest_prio.next >= | |
1999 | this_rq->rt.highest_prio.curr) | |
2000 | continue; | |
2001 | ||
f65eda4f SR |
2002 | /* |
2003 | * We can potentially drop this_rq's lock in | |
2004 | * double_lock_balance, and another CPU could | |
a8728944 | 2005 | * alter this_rq |
f65eda4f | 2006 | */ |
a8728944 | 2007 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
2008 | |
2009 | /* | |
e23ee747 KT |
2010 | * We can pull only a task, which is pushable |
2011 | * on its rq, and no others. | |
f65eda4f | 2012 | */ |
e23ee747 | 2013 | p = pick_highest_pushable_task(src_rq, this_cpu); |
f65eda4f SR |
2014 | |
2015 | /* | |
2016 | * Do we have an RT task that preempts | |
2017 | * the to-be-scheduled task? | |
2018 | */ | |
a8728944 | 2019 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f | 2020 | WARN_ON(p == src_rq->curr); |
da0c1e65 | 2021 | WARN_ON(!task_on_rq_queued(p)); |
f65eda4f SR |
2022 | |
2023 | /* | |
2024 | * There's a chance that p is higher in priority | |
2025 | * than what's currently running on its cpu. | |
2026 | * This is just that p is wakeing up and hasn't | |
2027 | * had a chance to schedule. We only pull | |
2028 | * p if it is lower in priority than the | |
a8728944 | 2029 | * current task on the run queue |
f65eda4f | 2030 | */ |
a8728944 | 2031 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 2032 | goto skip; |
f65eda4f | 2033 | |
8046d680 | 2034 | resched = true; |
f65eda4f SR |
2035 | |
2036 | deactivate_task(src_rq, p, 0); | |
2037 | set_task_cpu(p, this_cpu); | |
2038 | activate_task(this_rq, p, 0); | |
2039 | /* | |
2040 | * We continue with the search, just in | |
2041 | * case there's an even higher prio task | |
25985edc | 2042 | * in another runqueue. (low likelihood |
f65eda4f | 2043 | * but possible) |
f65eda4f | 2044 | */ |
f65eda4f | 2045 | } |
49246274 | 2046 | skip: |
1b12bbc7 | 2047 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
2048 | } |
2049 | ||
8046d680 PZ |
2050 | if (resched) |
2051 | resched_curr(this_rq); | |
f65eda4f SR |
2052 | } |
2053 | ||
8ae121ac GH |
2054 | /* |
2055 | * If we are not running and we are not going to reschedule soon, we should | |
2056 | * try to push tasks away now | |
2057 | */ | |
efbbd05a | 2058 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 2059 | { |
9a897c5a | 2060 | if (!task_running(rq, p) && |
8ae121ac | 2061 | !test_tsk_need_resched(rq->curr) && |
29baa747 | 2062 | p->nr_cpus_allowed > 1 && |
1baca4ce | 2063 | (dl_task(rq->curr) || rt_task(rq->curr)) && |
29baa747 | 2064 | (rq->curr->nr_cpus_allowed < 2 || |
3be209a8 | 2065 | rq->curr->prio <= p->prio)) |
4642dafd SR |
2066 | push_rt_tasks(rq); |
2067 | } | |
2068 | ||
bdd7c81b | 2069 | /* Assumes rq->lock is held */ |
1f11eb6a | 2070 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
2071 | { |
2072 | if (rq->rt.overloaded) | |
2073 | rt_set_overload(rq); | |
6e0534f2 | 2074 | |
7def2be1 PZ |
2075 | __enable_runtime(rq); |
2076 | ||
e864c499 | 2077 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
2078 | } |
2079 | ||
2080 | /* Assumes rq->lock is held */ | |
1f11eb6a | 2081 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
2082 | { |
2083 | if (rq->rt.overloaded) | |
2084 | rt_clear_overload(rq); | |
6e0534f2 | 2085 | |
7def2be1 PZ |
2086 | __disable_runtime(rq); |
2087 | ||
6e0534f2 | 2088 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 2089 | } |
cb469845 SR |
2090 | |
2091 | /* | |
2092 | * When switch from the rt queue, we bring ourselves to a position | |
2093 | * that we might want to pull RT tasks from other runqueues. | |
2094 | */ | |
da7a735e | 2095 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
2096 | { |
2097 | /* | |
2098 | * If there are other RT tasks then we will reschedule | |
2099 | * and the scheduling of the other RT tasks will handle | |
2100 | * the balancing. But if we are the last RT task | |
2101 | * we may need to handle the pulling of RT tasks | |
2102 | * now. | |
2103 | */ | |
da0c1e65 | 2104 | if (!task_on_rq_queued(p) || rq->rt.rt_nr_running) |
1158ddb5 KT |
2105 | return; |
2106 | ||
fd7a4bed | 2107 | queue_pull_task(rq); |
cb469845 | 2108 | } |
3d8cbdf8 | 2109 | |
11c785b7 | 2110 | void __init init_sched_rt_class(void) |
3d8cbdf8 RR |
2111 | { |
2112 | unsigned int i; | |
2113 | ||
029632fb | 2114 | for_each_possible_cpu(i) { |
eaa95840 | 2115 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 2116 | GFP_KERNEL, cpu_to_node(i)); |
029632fb | 2117 | } |
3d8cbdf8 | 2118 | } |
cb469845 SR |
2119 | #endif /* CONFIG_SMP */ |
2120 | ||
2121 | /* | |
2122 | * When switching a task to RT, we may overload the runqueue | |
2123 | * with RT tasks. In this case we try to push them off to | |
2124 | * other runqueues. | |
2125 | */ | |
da7a735e | 2126 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
cb469845 | 2127 | { |
cb469845 SR |
2128 | /* |
2129 | * If we are already running, then there's nothing | |
2130 | * that needs to be done. But if we are not running | |
2131 | * we may need to preempt the current running task. | |
2132 | * If that current running task is also an RT task | |
2133 | * then see if we can move to another run queue. | |
2134 | */ | |
da0c1e65 | 2135 | if (task_on_rq_queued(p) && rq->curr != p) { |
cb469845 | 2136 | #ifdef CONFIG_SMP |
fd7a4bed PZ |
2137 | if (p->nr_cpus_allowed > 1 && rq->rt.overloaded) |
2138 | queue_push_tasks(rq); | |
2139 | #else | |
2140 | if (p->prio < rq->curr->prio) | |
8875125e | 2141 | resched_curr(rq); |
fd7a4bed | 2142 | #endif /* CONFIG_SMP */ |
cb469845 SR |
2143 | } |
2144 | } | |
2145 | ||
2146 | /* | |
2147 | * Priority of the task has changed. This may cause | |
2148 | * us to initiate a push or pull. | |
2149 | */ | |
da7a735e PZ |
2150 | static void |
2151 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) | |
cb469845 | 2152 | { |
da0c1e65 | 2153 | if (!task_on_rq_queued(p)) |
da7a735e PZ |
2154 | return; |
2155 | ||
2156 | if (rq->curr == p) { | |
cb469845 SR |
2157 | #ifdef CONFIG_SMP |
2158 | /* | |
2159 | * If our priority decreases while running, we | |
2160 | * may need to pull tasks to this runqueue. | |
2161 | */ | |
2162 | if (oldprio < p->prio) | |
fd7a4bed PZ |
2163 | queue_pull_task(rq); |
2164 | ||
cb469845 SR |
2165 | /* |
2166 | * If there's a higher priority task waiting to run | |
fd7a4bed | 2167 | * then reschedule. |
cb469845 | 2168 | */ |
fd7a4bed | 2169 | if (p->prio > rq->rt.highest_prio.curr) |
8875125e | 2170 | resched_curr(rq); |
cb469845 SR |
2171 | #else |
2172 | /* For UP simply resched on drop of prio */ | |
2173 | if (oldprio < p->prio) | |
8875125e | 2174 | resched_curr(rq); |
e8fa1362 | 2175 | #endif /* CONFIG_SMP */ |
cb469845 SR |
2176 | } else { |
2177 | /* | |
2178 | * This task is not running, but if it is | |
2179 | * greater than the current running task | |
2180 | * then reschedule. | |
2181 | */ | |
2182 | if (p->prio < rq->curr->prio) | |
8875125e | 2183 | resched_curr(rq); |
cb469845 SR |
2184 | } |
2185 | } | |
2186 | ||
78f2c7db PZ |
2187 | static void watchdog(struct rq *rq, struct task_struct *p) |
2188 | { | |
2189 | unsigned long soft, hard; | |
2190 | ||
78d7d407 JS |
2191 | /* max may change after cur was read, this will be fixed next tick */ |
2192 | soft = task_rlimit(p, RLIMIT_RTTIME); | |
2193 | hard = task_rlimit_max(p, RLIMIT_RTTIME); | |
78f2c7db PZ |
2194 | |
2195 | if (soft != RLIM_INFINITY) { | |
2196 | unsigned long next; | |
2197 | ||
57d2aa00 YX |
2198 | if (p->rt.watchdog_stamp != jiffies) { |
2199 | p->rt.timeout++; | |
2200 | p->rt.watchdog_stamp = jiffies; | |
2201 | } | |
2202 | ||
78f2c7db | 2203 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
5a52dd50 | 2204 | if (p->rt.timeout > next) |
f06febc9 | 2205 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
2206 | } |
2207 | } | |
bb44e5d1 | 2208 | |
8f4d37ec | 2209 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 2210 | { |
454c7999 CC |
2211 | struct sched_rt_entity *rt_se = &p->rt; |
2212 | ||
67e2be02 PZ |
2213 | update_curr_rt(rq); |
2214 | ||
78f2c7db PZ |
2215 | watchdog(rq, p); |
2216 | ||
bb44e5d1 IM |
2217 | /* |
2218 | * RR tasks need a special form of timeslice management. | |
2219 | * FIFO tasks have no timeslices. | |
2220 | */ | |
2221 | if (p->policy != SCHED_RR) | |
2222 | return; | |
2223 | ||
fa717060 | 2224 | if (--p->rt.time_slice) |
bb44e5d1 IM |
2225 | return; |
2226 | ||
ce0dbbbb | 2227 | p->rt.time_slice = sched_rr_timeslice; |
bb44e5d1 | 2228 | |
98fbc798 | 2229 | /* |
e9aa39bb LB |
2230 | * Requeue to the end of queue if we (and all of our ancestors) are not |
2231 | * the only element on the queue | |
98fbc798 | 2232 | */ |
454c7999 CC |
2233 | for_each_sched_rt_entity(rt_se) { |
2234 | if (rt_se->run_list.prev != rt_se->run_list.next) { | |
2235 | requeue_task_rt(rq, p, 0); | |
8aa6f0eb | 2236 | resched_curr(rq); |
454c7999 CC |
2237 | return; |
2238 | } | |
98fbc798 | 2239 | } |
bb44e5d1 IM |
2240 | } |
2241 | ||
83b699ed SV |
2242 | static void set_curr_task_rt(struct rq *rq) |
2243 | { | |
2244 | struct task_struct *p = rq->curr; | |
2245 | ||
78becc27 | 2246 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
2247 | |
2248 | /* The running task is never eligible for pushing */ | |
2249 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
2250 | } |
2251 | ||
6d686f45 | 2252 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
0d721cea PW |
2253 | { |
2254 | /* | |
2255 | * Time slice is 0 for SCHED_FIFO tasks | |
2256 | */ | |
2257 | if (task->policy == SCHED_RR) | |
ce0dbbbb | 2258 | return sched_rr_timeslice; |
0d721cea PW |
2259 | else |
2260 | return 0; | |
2261 | } | |
2262 | ||
029632fb | 2263 | const struct sched_class rt_sched_class = { |
5522d5d5 | 2264 | .next = &fair_sched_class, |
bb44e5d1 IM |
2265 | .enqueue_task = enqueue_task_rt, |
2266 | .dequeue_task = dequeue_task_rt, | |
2267 | .yield_task = yield_task_rt, | |
2268 | ||
2269 | .check_preempt_curr = check_preempt_curr_rt, | |
2270 | ||
2271 | .pick_next_task = pick_next_task_rt, | |
2272 | .put_prev_task = put_prev_task_rt, | |
2273 | ||
681f3e68 | 2274 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
2275 | .select_task_rq = select_task_rq_rt, |
2276 | ||
6c37067e | 2277 | .set_cpus_allowed = set_cpus_allowed_common, |
1f11eb6a GH |
2278 | .rq_online = rq_online_rt, |
2279 | .rq_offline = rq_offline_rt, | |
efbbd05a | 2280 | .task_woken = task_woken_rt, |
cb469845 | 2281 | .switched_from = switched_from_rt, |
681f3e68 | 2282 | #endif |
bb44e5d1 | 2283 | |
83b699ed | 2284 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 2285 | .task_tick = task_tick_rt, |
cb469845 | 2286 | |
0d721cea PW |
2287 | .get_rr_interval = get_rr_interval_rt, |
2288 | ||
cb469845 SR |
2289 | .prio_changed = prio_changed_rt, |
2290 | .switched_to = switched_to_rt, | |
6e998916 SG |
2291 | |
2292 | .update_curr = update_curr_rt, | |
bb44e5d1 | 2293 | }; |
ada18de2 PZ |
2294 | |
2295 | #ifdef CONFIG_SCHED_DEBUG | |
2296 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2297 | ||
029632fb | 2298 | void print_rt_stats(struct seq_file *m, int cpu) |
ada18de2 | 2299 | { |
ec514c48 | 2300 | rt_rq_iter_t iter; |
ada18de2 PZ |
2301 | struct rt_rq *rt_rq; |
2302 | ||
2303 | rcu_read_lock(); | |
ec514c48 | 2304 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
ada18de2 PZ |
2305 | print_rt_rq(m, cpu, rt_rq); |
2306 | rcu_read_unlock(); | |
2307 | } | |
55e12e5e | 2308 | #endif /* CONFIG_SCHED_DEBUG */ |