]>
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 | ||
e3fca9e7 PZ |
357 | static DEFINE_PER_CPU(struct callback_head, rt_balance_head); |
358 | ||
359 | static void push_rt_tasks(struct rq *); | |
360 | ||
361 | static inline void queue_push_tasks(struct rq *rq) | |
dc877341 | 362 | { |
e3fca9e7 PZ |
363 | if (!has_pushable_tasks(rq)) |
364 | return; | |
365 | ||
366 | queue_balance_callback(rq, &per_cpu(rt_balance_head, rq->cpu), push_rt_tasks); | |
dc877341 PZ |
367 | } |
368 | ||
917b627d GH |
369 | static void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
370 | { | |
371 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
372 | plist_node_init(&p->pushable_tasks, p->prio); | |
373 | plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
5181f4a4 SR |
374 | |
375 | /* Update the highest prio pushable task */ | |
376 | if (p->prio < rq->rt.highest_prio.next) | |
377 | rq->rt.highest_prio.next = p->prio; | |
917b627d GH |
378 | } |
379 | ||
380 | static void dequeue_pushable_task(struct rq *rq, struct task_struct *p) | |
381 | { | |
382 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
917b627d | 383 | |
5181f4a4 SR |
384 | /* Update the new highest prio pushable task */ |
385 | if (has_pushable_tasks(rq)) { | |
386 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
387 | struct task_struct, pushable_tasks); | |
388 | rq->rt.highest_prio.next = p->prio; | |
389 | } else | |
390 | rq->rt.highest_prio.next = MAX_RT_PRIO; | |
bcf08df3 IM |
391 | } |
392 | ||
917b627d GH |
393 | #else |
394 | ||
ceacc2c1 | 395 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
fa85ae24 | 396 | { |
6f505b16 PZ |
397 | } |
398 | ||
ceacc2c1 PZ |
399 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
400 | { | |
401 | } | |
402 | ||
b07430ac | 403 | static inline |
ceacc2c1 PZ |
404 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
405 | { | |
406 | } | |
407 | ||
398a153b | 408 | static inline |
ceacc2c1 PZ |
409 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
410 | { | |
411 | } | |
917b627d | 412 | |
dc877341 PZ |
413 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
414 | { | |
415 | return false; | |
416 | } | |
417 | ||
8046d680 | 418 | static inline void pull_rt_task(struct rq *this_rq) |
dc877341 | 419 | { |
dc877341 PZ |
420 | } |
421 | ||
e3fca9e7 | 422 | static inline void queue_push_tasks(struct rq *rq) |
dc877341 PZ |
423 | { |
424 | } | |
4fd29176 SR |
425 | #endif /* CONFIG_SMP */ |
426 | ||
f4ebcbc0 KT |
427 | static void enqueue_top_rt_rq(struct rt_rq *rt_rq); |
428 | static void dequeue_top_rt_rq(struct rt_rq *rt_rq); | |
429 | ||
6f505b16 PZ |
430 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
431 | { | |
432 | return !list_empty(&rt_se->run_list); | |
433 | } | |
434 | ||
052f1dc7 | 435 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 436 | |
9f0c1e56 | 437 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
438 | { |
439 | if (!rt_rq->tg) | |
9f0c1e56 | 440 | return RUNTIME_INF; |
6f505b16 | 441 | |
ac086bc2 PZ |
442 | return rt_rq->rt_runtime; |
443 | } | |
444 | ||
445 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
446 | { | |
447 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); | |
6f505b16 PZ |
448 | } |
449 | ||
ec514c48 CX |
450 | typedef struct task_group *rt_rq_iter_t; |
451 | ||
1c09ab0d YZ |
452 | static inline struct task_group *next_task_group(struct task_group *tg) |
453 | { | |
454 | do { | |
455 | tg = list_entry_rcu(tg->list.next, | |
456 | typeof(struct task_group), list); | |
457 | } while (&tg->list != &task_groups && task_group_is_autogroup(tg)); | |
458 | ||
459 | if (&tg->list == &task_groups) | |
460 | tg = NULL; | |
461 | ||
462 | return tg; | |
463 | } | |
464 | ||
465 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
466 | for (iter = container_of(&task_groups, typeof(*iter), list); \ | |
467 | (iter = next_task_group(iter)) && \ | |
468 | (rt_rq = iter->rt_rq[cpu_of(rq)]);) | |
ec514c48 | 469 | |
6f505b16 PZ |
470 | #define for_each_sched_rt_entity(rt_se) \ |
471 | for (; rt_se; rt_se = rt_se->parent) | |
472 | ||
473 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
474 | { | |
475 | return rt_se->my_q; | |
476 | } | |
477 | ||
37dad3fc | 478 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
6f505b16 PZ |
479 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
480 | ||
9f0c1e56 | 481 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 482 | { |
f6121f4f | 483 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
8875125e | 484 | struct rq *rq = rq_of_rt_rq(rt_rq); |
74b7eb58 YZ |
485 | struct sched_rt_entity *rt_se; |
486 | ||
8875125e | 487 | int cpu = cpu_of(rq); |
0c3b9168 BS |
488 | |
489 | rt_se = rt_rq->tg->rt_se[cpu]; | |
6f505b16 | 490 | |
f6121f4f | 491 | if (rt_rq->rt_nr_running) { |
f4ebcbc0 KT |
492 | if (!rt_se) |
493 | enqueue_top_rt_rq(rt_rq); | |
494 | else if (!on_rt_rq(rt_se)) | |
37dad3fc | 495 | enqueue_rt_entity(rt_se, false); |
f4ebcbc0 | 496 | |
e864c499 | 497 | if (rt_rq->highest_prio.curr < curr->prio) |
8875125e | 498 | resched_curr(rq); |
6f505b16 PZ |
499 | } |
500 | } | |
501 | ||
9f0c1e56 | 502 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 503 | { |
74b7eb58 | 504 | struct sched_rt_entity *rt_se; |
0c3b9168 | 505 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
74b7eb58 | 506 | |
0c3b9168 | 507 | rt_se = rt_rq->tg->rt_se[cpu]; |
6f505b16 | 508 | |
f4ebcbc0 KT |
509 | if (!rt_se) |
510 | dequeue_top_rt_rq(rt_rq); | |
511 | else if (on_rt_rq(rt_se)) | |
6f505b16 PZ |
512 | dequeue_rt_entity(rt_se); |
513 | } | |
514 | ||
46383648 KT |
515 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
516 | { | |
517 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
518 | } | |
519 | ||
23b0fdfc PZ |
520 | static int rt_se_boosted(struct sched_rt_entity *rt_se) |
521 | { | |
522 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
523 | struct task_struct *p; | |
524 | ||
525 | if (rt_rq) | |
526 | return !!rt_rq->rt_nr_boosted; | |
527 | ||
528 | p = rt_task_of(rt_se); | |
529 | return p->prio != p->normal_prio; | |
530 | } | |
531 | ||
d0b27fa7 | 532 | #ifdef CONFIG_SMP |
c6c4927b | 533 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 534 | { |
424c93fe | 535 | return this_rq()->rd->span; |
d0b27fa7 | 536 | } |
6f505b16 | 537 | #else |
c6c4927b | 538 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 539 | { |
c6c4927b | 540 | return cpu_online_mask; |
d0b27fa7 PZ |
541 | } |
542 | #endif | |
6f505b16 | 543 | |
d0b27fa7 PZ |
544 | static inline |
545 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 546 | { |
d0b27fa7 PZ |
547 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
548 | } | |
9f0c1e56 | 549 | |
ac086bc2 PZ |
550 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
551 | { | |
552 | return &rt_rq->tg->rt_bandwidth; | |
553 | } | |
554 | ||
55e12e5e | 555 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
556 | |
557 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
558 | { | |
ac086bc2 PZ |
559 | return rt_rq->rt_runtime; |
560 | } | |
561 | ||
562 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
563 | { | |
564 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
565 | } |
566 | ||
ec514c48 CX |
567 | typedef struct rt_rq *rt_rq_iter_t; |
568 | ||
569 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
570 | for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
571 | ||
6f505b16 PZ |
572 | #define for_each_sched_rt_entity(rt_se) \ |
573 | for (; rt_se; rt_se = NULL) | |
574 | ||
575 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
576 | { | |
577 | return NULL; | |
578 | } | |
579 | ||
9f0c1e56 | 580 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 581 | { |
f4ebcbc0 KT |
582 | struct rq *rq = rq_of_rt_rq(rt_rq); |
583 | ||
584 | if (!rt_rq->rt_nr_running) | |
585 | return; | |
586 | ||
587 | enqueue_top_rt_rq(rt_rq); | |
8875125e | 588 | resched_curr(rq); |
6f505b16 PZ |
589 | } |
590 | ||
9f0c1e56 | 591 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 592 | { |
f4ebcbc0 | 593 | dequeue_top_rt_rq(rt_rq); |
6f505b16 PZ |
594 | } |
595 | ||
46383648 KT |
596 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
597 | { | |
598 | return rt_rq->rt_throttled; | |
599 | } | |
600 | ||
c6c4927b | 601 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 602 | { |
c6c4927b | 603 | return cpu_online_mask; |
d0b27fa7 PZ |
604 | } |
605 | ||
606 | static inline | |
607 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
608 | { | |
609 | return &cpu_rq(cpu)->rt; | |
610 | } | |
611 | ||
ac086bc2 PZ |
612 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
613 | { | |
614 | return &def_rt_bandwidth; | |
615 | } | |
616 | ||
55e12e5e | 617 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 618 | |
faa59937 JL |
619 | bool sched_rt_bandwidth_account(struct rt_rq *rt_rq) |
620 | { | |
621 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
622 | ||
623 | return (hrtimer_active(&rt_b->rt_period_timer) || | |
624 | rt_rq->rt_time < rt_b->rt_runtime); | |
625 | } | |
626 | ||
ac086bc2 | 627 | #ifdef CONFIG_SMP |
78333cdd PZ |
628 | /* |
629 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
630 | */ | |
b79f3833 | 631 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
632 | { |
633 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
aa7f6730 | 634 | struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd; |
ac086bc2 PZ |
635 | int i, weight, more = 0; |
636 | u64 rt_period; | |
637 | ||
c6c4927b | 638 | weight = cpumask_weight(rd->span); |
ac086bc2 | 639 | |
0986b11b | 640 | raw_spin_lock(&rt_b->rt_runtime_lock); |
ac086bc2 | 641 | rt_period = ktime_to_ns(rt_b->rt_period); |
c6c4927b | 642 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
643 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
644 | s64 diff; | |
645 | ||
646 | if (iter == rt_rq) | |
647 | continue; | |
648 | ||
0986b11b | 649 | raw_spin_lock(&iter->rt_runtime_lock); |
78333cdd PZ |
650 | /* |
651 | * Either all rqs have inf runtime and there's nothing to steal | |
652 | * or __disable_runtime() below sets a specific rq to inf to | |
653 | * indicate its been disabled and disalow stealing. | |
654 | */ | |
7def2be1 PZ |
655 | if (iter->rt_runtime == RUNTIME_INF) |
656 | goto next; | |
657 | ||
78333cdd PZ |
658 | /* |
659 | * From runqueues with spare time, take 1/n part of their | |
660 | * spare time, but no more than our period. | |
661 | */ | |
ac086bc2 PZ |
662 | diff = iter->rt_runtime - iter->rt_time; |
663 | if (diff > 0) { | |
58838cf3 | 664 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
665 | if (rt_rq->rt_runtime + diff > rt_period) |
666 | diff = rt_period - rt_rq->rt_runtime; | |
667 | iter->rt_runtime -= diff; | |
668 | rt_rq->rt_runtime += diff; | |
669 | more = 1; | |
670 | if (rt_rq->rt_runtime == rt_period) { | |
0986b11b | 671 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 PZ |
672 | break; |
673 | } | |
674 | } | |
7def2be1 | 675 | next: |
0986b11b | 676 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 | 677 | } |
0986b11b | 678 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
ac086bc2 PZ |
679 | |
680 | return more; | |
681 | } | |
7def2be1 | 682 | |
78333cdd PZ |
683 | /* |
684 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
685 | */ | |
7def2be1 PZ |
686 | static void __disable_runtime(struct rq *rq) |
687 | { | |
688 | struct root_domain *rd = rq->rd; | |
ec514c48 | 689 | rt_rq_iter_t iter; |
7def2be1 PZ |
690 | struct rt_rq *rt_rq; |
691 | ||
692 | if (unlikely(!scheduler_running)) | |
693 | return; | |
694 | ||
ec514c48 | 695 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
696 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
697 | s64 want; | |
698 | int i; | |
699 | ||
0986b11b TG |
700 | raw_spin_lock(&rt_b->rt_runtime_lock); |
701 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
702 | /* |
703 | * Either we're all inf and nobody needs to borrow, or we're | |
704 | * already disabled and thus have nothing to do, or we have | |
705 | * exactly the right amount of runtime to take out. | |
706 | */ | |
7def2be1 PZ |
707 | if (rt_rq->rt_runtime == RUNTIME_INF || |
708 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
709 | goto balanced; | |
0986b11b | 710 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
7def2be1 | 711 | |
78333cdd PZ |
712 | /* |
713 | * Calculate the difference between what we started out with | |
714 | * and what we current have, that's the amount of runtime | |
715 | * we lend and now have to reclaim. | |
716 | */ | |
7def2be1 PZ |
717 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
718 | ||
78333cdd PZ |
719 | /* |
720 | * Greedy reclaim, take back as much as we can. | |
721 | */ | |
c6c4927b | 722 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
723 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
724 | s64 diff; | |
725 | ||
78333cdd PZ |
726 | /* |
727 | * Can't reclaim from ourselves or disabled runqueues. | |
728 | */ | |
f1679d08 | 729 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
730 | continue; |
731 | ||
0986b11b | 732 | raw_spin_lock(&iter->rt_runtime_lock); |
7def2be1 PZ |
733 | if (want > 0) { |
734 | diff = min_t(s64, iter->rt_runtime, want); | |
735 | iter->rt_runtime -= diff; | |
736 | want -= diff; | |
737 | } else { | |
738 | iter->rt_runtime -= want; | |
739 | want -= want; | |
740 | } | |
0986b11b | 741 | raw_spin_unlock(&iter->rt_runtime_lock); |
7def2be1 PZ |
742 | |
743 | if (!want) | |
744 | break; | |
745 | } | |
746 | ||
0986b11b | 747 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
78333cdd PZ |
748 | /* |
749 | * We cannot be left wanting - that would mean some runtime | |
750 | * leaked out of the system. | |
751 | */ | |
7def2be1 PZ |
752 | BUG_ON(want); |
753 | balanced: | |
78333cdd PZ |
754 | /* |
755 | * Disable all the borrow logic by pretending we have inf | |
756 | * runtime - in which case borrowing doesn't make sense. | |
757 | */ | |
7def2be1 | 758 | rt_rq->rt_runtime = RUNTIME_INF; |
a4c96ae3 | 759 | rt_rq->rt_throttled = 0; |
0986b11b TG |
760 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
761 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
99b62567 KT |
762 | |
763 | /* Make rt_rq available for pick_next_task() */ | |
764 | sched_rt_rq_enqueue(rt_rq); | |
7def2be1 PZ |
765 | } |
766 | } | |
767 | ||
7def2be1 PZ |
768 | static void __enable_runtime(struct rq *rq) |
769 | { | |
ec514c48 | 770 | rt_rq_iter_t iter; |
7def2be1 PZ |
771 | struct rt_rq *rt_rq; |
772 | ||
773 | if (unlikely(!scheduler_running)) | |
774 | return; | |
775 | ||
78333cdd PZ |
776 | /* |
777 | * Reset each runqueue's bandwidth settings | |
778 | */ | |
ec514c48 | 779 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
780 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
781 | ||
0986b11b TG |
782 | raw_spin_lock(&rt_b->rt_runtime_lock); |
783 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
7def2be1 PZ |
784 | rt_rq->rt_runtime = rt_b->rt_runtime; |
785 | rt_rq->rt_time = 0; | |
baf25731 | 786 | rt_rq->rt_throttled = 0; |
0986b11b TG |
787 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
788 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
789 | } |
790 | } | |
791 | ||
eff6549b PZ |
792 | static int balance_runtime(struct rt_rq *rt_rq) |
793 | { | |
794 | int more = 0; | |
795 | ||
4a6184ce PZ |
796 | if (!sched_feat(RT_RUNTIME_SHARE)) |
797 | return more; | |
798 | ||
eff6549b | 799 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
0986b11b | 800 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
eff6549b | 801 | more = do_balance_runtime(rt_rq); |
0986b11b | 802 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
803 | } |
804 | ||
805 | return more; | |
806 | } | |
55e12e5e | 807 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
808 | static inline int balance_runtime(struct rt_rq *rt_rq) |
809 | { | |
810 | return 0; | |
811 | } | |
55e12e5e | 812 | #endif /* CONFIG_SMP */ |
ac086bc2 | 813 | |
eff6549b PZ |
814 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
815 | { | |
42c62a58 | 816 | int i, idle = 1, throttled = 0; |
c6c4927b | 817 | const struct cpumask *span; |
eff6549b | 818 | |
eff6549b | 819 | span = sched_rt_period_mask(); |
e221d028 MG |
820 | #ifdef CONFIG_RT_GROUP_SCHED |
821 | /* | |
822 | * FIXME: isolated CPUs should really leave the root task group, | |
823 | * whether they are isolcpus or were isolated via cpusets, lest | |
824 | * the timer run on a CPU which does not service all runqueues, | |
825 | * potentially leaving other CPUs indefinitely throttled. If | |
826 | * isolation is really required, the user will turn the throttle | |
827 | * off to kill the perturbations it causes anyway. Meanwhile, | |
828 | * this maintains functionality for boot and/or troubleshooting. | |
829 | */ | |
830 | if (rt_b == &root_task_group.rt_bandwidth) | |
831 | span = cpu_online_mask; | |
832 | #endif | |
c6c4927b | 833 | for_each_cpu(i, span) { |
eff6549b PZ |
834 | int enqueue = 0; |
835 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
836 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
837 | ||
05fa785c | 838 | raw_spin_lock(&rq->lock); |
eff6549b PZ |
839 | if (rt_rq->rt_time) { |
840 | u64 runtime; | |
841 | ||
0986b11b | 842 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
843 | if (rt_rq->rt_throttled) |
844 | balance_runtime(rt_rq); | |
845 | runtime = rt_rq->rt_runtime; | |
846 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
847 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
848 | rt_rq->rt_throttled = 0; | |
849 | enqueue = 1; | |
61eadef6 MG |
850 | |
851 | /* | |
9edfbfed PZ |
852 | * When we're idle and a woken (rt) task is |
853 | * throttled check_preempt_curr() will set | |
854 | * skip_update and the time between the wakeup | |
855 | * and this unthrottle will get accounted as | |
856 | * 'runtime'. | |
61eadef6 MG |
857 | */ |
858 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) | |
9edfbfed | 859 | rq_clock_skip_update(rq, false); |
eff6549b PZ |
860 | } |
861 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
862 | idle = 0; | |
0986b11b | 863 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
0c3b9168 | 864 | } else if (rt_rq->rt_nr_running) { |
6c3df255 | 865 | idle = 0; |
0c3b9168 BS |
866 | if (!rt_rq_throttled(rt_rq)) |
867 | enqueue = 1; | |
868 | } | |
42c62a58 PZ |
869 | if (rt_rq->rt_throttled) |
870 | throttled = 1; | |
eff6549b PZ |
871 | |
872 | if (enqueue) | |
873 | sched_rt_rq_enqueue(rt_rq); | |
05fa785c | 874 | raw_spin_unlock(&rq->lock); |
eff6549b PZ |
875 | } |
876 | ||
42c62a58 PZ |
877 | if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)) |
878 | return 1; | |
879 | ||
eff6549b PZ |
880 | return idle; |
881 | } | |
ac086bc2 | 882 | |
6f505b16 PZ |
883 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
884 | { | |
052f1dc7 | 885 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
886 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
887 | ||
888 | if (rt_rq) | |
e864c499 | 889 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
890 | #endif |
891 | ||
892 | return rt_task_of(rt_se)->prio; | |
893 | } | |
894 | ||
9f0c1e56 | 895 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 896 | { |
9f0c1e56 | 897 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 898 | |
fa85ae24 | 899 | if (rt_rq->rt_throttled) |
23b0fdfc | 900 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 901 | |
5b680fd6 | 902 | if (runtime >= sched_rt_period(rt_rq)) |
ac086bc2 PZ |
903 | return 0; |
904 | ||
b79f3833 PZ |
905 | balance_runtime(rt_rq); |
906 | runtime = sched_rt_runtime(rt_rq); | |
907 | if (runtime == RUNTIME_INF) | |
908 | return 0; | |
ac086bc2 | 909 | |
9f0c1e56 | 910 | if (rt_rq->rt_time > runtime) { |
7abc63b1 PZ |
911 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
912 | ||
913 | /* | |
914 | * Don't actually throttle groups that have no runtime assigned | |
915 | * but accrue some time due to boosting. | |
916 | */ | |
917 | if (likely(rt_b->rt_runtime)) { | |
918 | rt_rq->rt_throttled = 1; | |
c224815d | 919 | printk_deferred_once("sched: RT throttling activated\n"); |
7abc63b1 PZ |
920 | } else { |
921 | /* | |
922 | * In case we did anyway, make it go away, | |
923 | * replenishment is a joke, since it will replenish us | |
924 | * with exactly 0 ns. | |
925 | */ | |
926 | rt_rq->rt_time = 0; | |
927 | } | |
928 | ||
23b0fdfc | 929 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 930 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
931 | return 1; |
932 | } | |
fa85ae24 PZ |
933 | } |
934 | ||
935 | return 0; | |
936 | } | |
937 | ||
bb44e5d1 IM |
938 | /* |
939 | * Update the current task's runtime statistics. Skip current tasks that | |
940 | * are not in our scheduling class. | |
941 | */ | |
a9957449 | 942 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
943 | { |
944 | struct task_struct *curr = rq->curr; | |
6f505b16 | 945 | struct sched_rt_entity *rt_se = &curr->rt; |
bb44e5d1 IM |
946 | u64 delta_exec; |
947 | ||
06c3bc65 | 948 | if (curr->sched_class != &rt_sched_class) |
bb44e5d1 IM |
949 | return; |
950 | ||
78becc27 | 951 | delta_exec = rq_clock_task(rq) - curr->se.exec_start; |
fc79e240 KT |
952 | if (unlikely((s64)delta_exec <= 0)) |
953 | return; | |
6cfb0d5d | 954 | |
42c62a58 PZ |
955 | schedstat_set(curr->se.statistics.exec_max, |
956 | max(curr->se.statistics.exec_max, delta_exec)); | |
bb44e5d1 IM |
957 | |
958 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
959 | account_group_exec_runtime(curr, delta_exec); |
960 | ||
78becc27 | 961 | curr->se.exec_start = rq_clock_task(rq); |
d842de87 | 962 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 963 | |
e9e9250b PZ |
964 | sched_rt_avg_update(rq, delta_exec); |
965 | ||
0b148fa0 PZ |
966 | if (!rt_bandwidth_enabled()) |
967 | return; | |
968 | ||
354d60c2 | 969 | for_each_sched_rt_entity(rt_se) { |
0b07939c | 970 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
354d60c2 | 971 | |
cc2991cf | 972 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
0986b11b | 973 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
974 | rt_rq->rt_time += delta_exec; |
975 | if (sched_rt_runtime_exceeded(rt_rq)) | |
8875125e | 976 | resched_curr(rq); |
0986b11b | 977 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 978 | } |
354d60c2 | 979 | } |
bb44e5d1 IM |
980 | } |
981 | ||
f4ebcbc0 KT |
982 | static void |
983 | dequeue_top_rt_rq(struct rt_rq *rt_rq) | |
984 | { | |
985 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
986 | ||
987 | BUG_ON(&rq->rt != rt_rq); | |
988 | ||
989 | if (!rt_rq->rt_queued) | |
990 | return; | |
991 | ||
992 | BUG_ON(!rq->nr_running); | |
993 | ||
72465447 | 994 | sub_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
995 | rt_rq->rt_queued = 0; |
996 | } | |
997 | ||
998 | static void | |
999 | enqueue_top_rt_rq(struct rt_rq *rt_rq) | |
1000 | { | |
1001 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
1002 | ||
1003 | BUG_ON(&rq->rt != rt_rq); | |
1004 | ||
1005 | if (rt_rq->rt_queued) | |
1006 | return; | |
1007 | if (rt_rq_throttled(rt_rq) || !rt_rq->rt_nr_running) | |
1008 | return; | |
1009 | ||
72465447 | 1010 | add_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
1011 | rt_rq->rt_queued = 1; |
1012 | } | |
1013 | ||
398a153b | 1014 | #if defined CONFIG_SMP |
e864c499 | 1015 | |
398a153b GH |
1016 | static void |
1017 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 1018 | { |
4d984277 | 1019 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 1020 | |
757dfcaa KT |
1021 | #ifdef CONFIG_RT_GROUP_SCHED |
1022 | /* | |
1023 | * Change rq's cpupri only if rt_rq is the top queue. | |
1024 | */ | |
1025 | if (&rq->rt != rt_rq) | |
1026 | return; | |
1027 | #endif | |
5181f4a4 SR |
1028 | if (rq->online && prio < prev_prio) |
1029 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); | |
398a153b | 1030 | } |
73fe6aae | 1031 | |
398a153b GH |
1032 | static void |
1033 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
1034 | { | |
1035 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 1036 | |
757dfcaa KT |
1037 | #ifdef CONFIG_RT_GROUP_SCHED |
1038 | /* | |
1039 | * Change rq's cpupri only if rt_rq is the top queue. | |
1040 | */ | |
1041 | if (&rq->rt != rt_rq) | |
1042 | return; | |
1043 | #endif | |
398a153b GH |
1044 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
1045 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
1046 | } |
1047 | ||
398a153b GH |
1048 | #else /* CONFIG_SMP */ |
1049 | ||
6f505b16 | 1050 | static inline |
398a153b GH |
1051 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
1052 | static inline | |
1053 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
1054 | ||
1055 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 1056 | |
052f1dc7 | 1057 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
1058 | static void |
1059 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
1060 | { | |
1061 | int prev_prio = rt_rq->highest_prio.curr; | |
1062 | ||
1063 | if (prio < prev_prio) | |
1064 | rt_rq->highest_prio.curr = prio; | |
1065 | ||
1066 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
1067 | } | |
1068 | ||
1069 | static void | |
1070 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
1071 | { | |
1072 | int prev_prio = rt_rq->highest_prio.curr; | |
1073 | ||
6f505b16 | 1074 | if (rt_rq->rt_nr_running) { |
764a9d6f | 1075 | |
398a153b | 1076 | WARN_ON(prio < prev_prio); |
764a9d6f | 1077 | |
e864c499 | 1078 | /* |
398a153b GH |
1079 | * This may have been our highest task, and therefore |
1080 | * we may have some recomputation to do | |
e864c499 | 1081 | */ |
398a153b | 1082 | if (prio == prev_prio) { |
e864c499 GH |
1083 | struct rt_prio_array *array = &rt_rq->active; |
1084 | ||
1085 | rt_rq->highest_prio.curr = | |
764a9d6f | 1086 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
1087 | } |
1088 | ||
764a9d6f | 1089 | } else |
e864c499 | 1090 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 1091 | |
398a153b GH |
1092 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
1093 | } | |
1f11eb6a | 1094 | |
398a153b GH |
1095 | #else |
1096 | ||
1097 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1098 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1099 | ||
1100 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 1101 | |
052f1dc7 | 1102 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
1103 | |
1104 | static void | |
1105 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1106 | { | |
1107 | if (rt_se_boosted(rt_se)) | |
1108 | rt_rq->rt_nr_boosted++; | |
1109 | ||
1110 | if (rt_rq->tg) | |
1111 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
1112 | } | |
1113 | ||
1114 | static void | |
1115 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1116 | { | |
23b0fdfc PZ |
1117 | if (rt_se_boosted(rt_se)) |
1118 | rt_rq->rt_nr_boosted--; | |
1119 | ||
1120 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
1121 | } |
1122 | ||
1123 | #else /* CONFIG_RT_GROUP_SCHED */ | |
1124 | ||
1125 | static void | |
1126 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1127 | { | |
1128 | start_rt_bandwidth(&def_rt_bandwidth); | |
1129 | } | |
1130 | ||
1131 | static inline | |
1132 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
1133 | ||
1134 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
1135 | ||
22abdef3 KT |
1136 | static inline |
1137 | unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se) | |
1138 | { | |
1139 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
1140 | ||
1141 | if (group_rq) | |
1142 | return group_rq->rt_nr_running; | |
1143 | else | |
1144 | return 1; | |
1145 | } | |
1146 | ||
398a153b GH |
1147 | static inline |
1148 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1149 | { | |
1150 | int prio = rt_se_prio(rt_se); | |
1151 | ||
1152 | WARN_ON(!rt_prio(prio)); | |
22abdef3 | 1153 | rt_rq->rt_nr_running += rt_se_nr_running(rt_se); |
398a153b GH |
1154 | |
1155 | inc_rt_prio(rt_rq, prio); | |
1156 | inc_rt_migration(rt_se, rt_rq); | |
1157 | inc_rt_group(rt_se, rt_rq); | |
1158 | } | |
1159 | ||
1160 | static inline | |
1161 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1162 | { | |
1163 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
1164 | WARN_ON(!rt_rq->rt_nr_running); | |
22abdef3 | 1165 | rt_rq->rt_nr_running -= rt_se_nr_running(rt_se); |
398a153b GH |
1166 | |
1167 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
1168 | dec_rt_migration(rt_se, rt_rq); | |
1169 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
1170 | } |
1171 | ||
37dad3fc | 1172 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
bb44e5d1 | 1173 | { |
6f505b16 PZ |
1174 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
1175 | struct rt_prio_array *array = &rt_rq->active; | |
1176 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 1177 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 1178 | |
ad2a3f13 PZ |
1179 | /* |
1180 | * Don't enqueue the group if its throttled, or when empty. | |
1181 | * The latter is a consequence of the former when a child group | |
1182 | * get throttled and the current group doesn't have any other | |
1183 | * active members. | |
1184 | */ | |
1185 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 1186 | return; |
63489e45 | 1187 | |
37dad3fc TG |
1188 | if (head) |
1189 | list_add(&rt_se->run_list, queue); | |
1190 | else | |
1191 | list_add_tail(&rt_se->run_list, queue); | |
6f505b16 | 1192 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 1193 | |
6f505b16 PZ |
1194 | inc_rt_tasks(rt_se, rt_rq); |
1195 | } | |
1196 | ||
ad2a3f13 | 1197 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
1198 | { |
1199 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
1200 | struct rt_prio_array *array = &rt_rq->active; | |
1201 | ||
1202 | list_del_init(&rt_se->run_list); | |
1203 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
1204 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
1205 | ||
1206 | dec_rt_tasks(rt_se, rt_rq); | |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * Because the prio of an upper entry depends on the lower | |
1211 | * entries, we must remove entries top - down. | |
6f505b16 | 1212 | */ |
ad2a3f13 | 1213 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 1214 | { |
ad2a3f13 | 1215 | struct sched_rt_entity *back = NULL; |
6f505b16 | 1216 | |
58d6c2d7 PZ |
1217 | for_each_sched_rt_entity(rt_se) { |
1218 | rt_se->back = back; | |
1219 | back = rt_se; | |
1220 | } | |
1221 | ||
f4ebcbc0 KT |
1222 | dequeue_top_rt_rq(rt_rq_of_se(back)); |
1223 | ||
58d6c2d7 PZ |
1224 | for (rt_se = back; rt_se; rt_se = rt_se->back) { |
1225 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
1226 | __dequeue_rt_entity(rt_se); |
1227 | } | |
1228 | } | |
1229 | ||
37dad3fc | 1230 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
ad2a3f13 | 1231 | { |
f4ebcbc0 KT |
1232 | struct rq *rq = rq_of_rt_se(rt_se); |
1233 | ||
ad2a3f13 PZ |
1234 | dequeue_rt_stack(rt_se); |
1235 | for_each_sched_rt_entity(rt_se) | |
37dad3fc | 1236 | __enqueue_rt_entity(rt_se, head); |
f4ebcbc0 | 1237 | enqueue_top_rt_rq(&rq->rt); |
ad2a3f13 PZ |
1238 | } |
1239 | ||
1240 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
1241 | { | |
f4ebcbc0 KT |
1242 | struct rq *rq = rq_of_rt_se(rt_se); |
1243 | ||
ad2a3f13 PZ |
1244 | dequeue_rt_stack(rt_se); |
1245 | ||
1246 | for_each_sched_rt_entity(rt_se) { | |
1247 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
1248 | ||
1249 | if (rt_rq && rt_rq->rt_nr_running) | |
37dad3fc | 1250 | __enqueue_rt_entity(rt_se, false); |
58d6c2d7 | 1251 | } |
f4ebcbc0 | 1252 | enqueue_top_rt_rq(&rq->rt); |
bb44e5d1 IM |
1253 | } |
1254 | ||
1255 | /* | |
1256 | * Adding/removing a task to/from a priority array: | |
1257 | */ | |
ea87bb78 | 1258 | static void |
371fd7e7 | 1259 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
6f505b16 PZ |
1260 | { |
1261 | struct sched_rt_entity *rt_se = &p->rt; | |
1262 | ||
371fd7e7 | 1263 | if (flags & ENQUEUE_WAKEUP) |
6f505b16 PZ |
1264 | rt_se->timeout = 0; |
1265 | ||
371fd7e7 | 1266 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
c09595f6 | 1267 | |
29baa747 | 1268 | if (!task_current(rq, p) && p->nr_cpus_allowed > 1) |
917b627d | 1269 | enqueue_pushable_task(rq, p); |
6f505b16 PZ |
1270 | } |
1271 | ||
371fd7e7 | 1272 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1273 | { |
6f505b16 | 1274 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 1275 | |
f1e14ef6 | 1276 | update_curr_rt(rq); |
ad2a3f13 | 1277 | dequeue_rt_entity(rt_se); |
c09595f6 | 1278 | |
917b627d | 1279 | dequeue_pushable_task(rq, p); |
bb44e5d1 IM |
1280 | } |
1281 | ||
1282 | /* | |
60686317 RW |
1283 | * Put task to the head or the end of the run list without the overhead of |
1284 | * dequeue followed by enqueue. | |
bb44e5d1 | 1285 | */ |
7ebefa8c DA |
1286 | static void |
1287 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 1288 | { |
1cdad715 | 1289 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
1290 | struct rt_prio_array *array = &rt_rq->active; |
1291 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
1292 | ||
1293 | if (head) | |
1294 | list_move(&rt_se->run_list, queue); | |
1295 | else | |
1296 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 1297 | } |
6f505b16 PZ |
1298 | } |
1299 | ||
7ebefa8c | 1300 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 1301 | { |
6f505b16 PZ |
1302 | struct sched_rt_entity *rt_se = &p->rt; |
1303 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1304 | |
6f505b16 PZ |
1305 | for_each_sched_rt_entity(rt_se) { |
1306 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 1307 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 1308 | } |
bb44e5d1 IM |
1309 | } |
1310 | ||
6f505b16 | 1311 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 1312 | { |
7ebefa8c | 1313 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
1314 | } |
1315 | ||
e7693a36 | 1316 | #ifdef CONFIG_SMP |
318e0893 GH |
1317 | static int find_lowest_rq(struct task_struct *task); |
1318 | ||
0017d735 | 1319 | static int |
ac66f547 | 1320 | select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags) |
e7693a36 | 1321 | { |
7608dec2 PZ |
1322 | struct task_struct *curr; |
1323 | struct rq *rq; | |
c37495fd SR |
1324 | |
1325 | /* For anything but wake ups, just return the task_cpu */ | |
1326 | if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK) | |
1327 | goto out; | |
1328 | ||
7608dec2 PZ |
1329 | rq = cpu_rq(cpu); |
1330 | ||
1331 | rcu_read_lock(); | |
316c1608 | 1332 | curr = READ_ONCE(rq->curr); /* unlocked access */ |
7608dec2 | 1333 | |
318e0893 | 1334 | /* |
7608dec2 | 1335 | * If the current task on @p's runqueue is an RT task, then |
e1f47d89 SR |
1336 | * try to see if we can wake this RT task up on another |
1337 | * runqueue. Otherwise simply start this RT task | |
1338 | * on its current runqueue. | |
1339 | * | |
43fa5460 SR |
1340 | * We want to avoid overloading runqueues. If the woken |
1341 | * task is a higher priority, then it will stay on this CPU | |
1342 | * and the lower prio task should be moved to another CPU. | |
1343 | * Even though this will probably make the lower prio task | |
1344 | * lose its cache, we do not want to bounce a higher task | |
1345 | * around just because it gave up its CPU, perhaps for a | |
1346 | * lock? | |
1347 | * | |
1348 | * For equal prio tasks, we just let the scheduler sort it out. | |
7608dec2 PZ |
1349 | * |
1350 | * Otherwise, just let it ride on the affined RQ and the | |
1351 | * post-schedule router will push the preempted task away | |
1352 | * | |
1353 | * This test is optimistic, if we get it wrong the load-balancer | |
1354 | * will have to sort it out. | |
318e0893 | 1355 | */ |
7608dec2 | 1356 | if (curr && unlikely(rt_task(curr)) && |
29baa747 | 1357 | (curr->nr_cpus_allowed < 2 || |
6bfa687c | 1358 | curr->prio <= p->prio)) { |
7608dec2 | 1359 | int target = find_lowest_rq(p); |
318e0893 | 1360 | |
80e3d87b TC |
1361 | /* |
1362 | * Don't bother moving it if the destination CPU is | |
1363 | * not running a lower priority task. | |
1364 | */ | |
1365 | if (target != -1 && | |
1366 | p->prio < cpu_rq(target)->rt.highest_prio.curr) | |
7608dec2 | 1367 | cpu = target; |
318e0893 | 1368 | } |
7608dec2 | 1369 | rcu_read_unlock(); |
318e0893 | 1370 | |
c37495fd | 1371 | out: |
7608dec2 | 1372 | return cpu; |
e7693a36 | 1373 | } |
7ebefa8c DA |
1374 | |
1375 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
1376 | { | |
308a623a WL |
1377 | /* |
1378 | * Current can't be migrated, useless to reschedule, | |
1379 | * let's hope p can move out. | |
1380 | */ | |
1381 | if (rq->curr->nr_cpus_allowed == 1 || | |
1382 | !cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) | |
7ebefa8c DA |
1383 | return; |
1384 | ||
308a623a WL |
1385 | /* |
1386 | * p is migratable, so let's not schedule it and | |
1387 | * see if it is pushed or pulled somewhere else. | |
1388 | */ | |
29baa747 | 1389 | if (p->nr_cpus_allowed != 1 |
13b8bd0a RR |
1390 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1391 | return; | |
24600ce8 | 1392 | |
7ebefa8c DA |
1393 | /* |
1394 | * There appears to be other cpus that can accept | |
1395 | * current and none to run 'p', so lets reschedule | |
1396 | * to try and push current away: | |
1397 | */ | |
1398 | requeue_task_rt(rq, p, 1); | |
8875125e | 1399 | resched_curr(rq); |
7ebefa8c DA |
1400 | } |
1401 | ||
e7693a36 GH |
1402 | #endif /* CONFIG_SMP */ |
1403 | ||
bb44e5d1 IM |
1404 | /* |
1405 | * Preempt the current task with a newly woken task if needed: | |
1406 | */ | |
7d478721 | 1407 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1408 | { |
45c01e82 | 1409 | if (p->prio < rq->curr->prio) { |
8875125e | 1410 | resched_curr(rq); |
45c01e82 GH |
1411 | return; |
1412 | } | |
1413 | ||
1414 | #ifdef CONFIG_SMP | |
1415 | /* | |
1416 | * If: | |
1417 | * | |
1418 | * - the newly woken task is of equal priority to the current task | |
1419 | * - the newly woken task is non-migratable while current is migratable | |
1420 | * - current will be preempted on the next reschedule | |
1421 | * | |
1422 | * we should check to see if current can readily move to a different | |
1423 | * cpu. If so, we will reschedule to allow the push logic to try | |
1424 | * to move current somewhere else, making room for our non-migratable | |
1425 | * task. | |
1426 | */ | |
8dd0de8b | 1427 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
7ebefa8c | 1428 | check_preempt_equal_prio(rq, p); |
45c01e82 | 1429 | #endif |
bb44e5d1 IM |
1430 | } |
1431 | ||
6f505b16 PZ |
1432 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1433 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1434 | { |
6f505b16 PZ |
1435 | struct rt_prio_array *array = &rt_rq->active; |
1436 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1437 | struct list_head *queue; |
1438 | int idx; | |
1439 | ||
1440 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1441 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1442 | |
1443 | queue = array->queue + idx; | |
6f505b16 | 1444 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1445 | |
6f505b16 PZ |
1446 | return next; |
1447 | } | |
bb44e5d1 | 1448 | |
917b627d | 1449 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1450 | { |
1451 | struct sched_rt_entity *rt_se; | |
1452 | struct task_struct *p; | |
606dba2e | 1453 | struct rt_rq *rt_rq = &rq->rt; |
6f505b16 PZ |
1454 | |
1455 | do { | |
1456 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1457 | BUG_ON(!rt_se); |
6f505b16 PZ |
1458 | rt_rq = group_rt_rq(rt_se); |
1459 | } while (rt_rq); | |
1460 | ||
1461 | p = rt_task_of(rt_se); | |
78becc27 | 1462 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
1463 | |
1464 | return p; | |
1465 | } | |
1466 | ||
606dba2e PZ |
1467 | static struct task_struct * |
1468 | pick_next_task_rt(struct rq *rq, struct task_struct *prev) | |
917b627d | 1469 | { |
606dba2e PZ |
1470 | struct task_struct *p; |
1471 | struct rt_rq *rt_rq = &rq->rt; | |
1472 | ||
37e117c0 | 1473 | if (need_pull_rt_task(rq, prev)) { |
38033c37 | 1474 | pull_rt_task(rq); |
37e117c0 PZ |
1475 | /* |
1476 | * pull_rt_task() can drop (and re-acquire) rq->lock; this | |
a1d9a323 KT |
1477 | * means a dl or stop task can slip in, in which case we need |
1478 | * to re-start task selection. | |
37e117c0 | 1479 | */ |
da0c1e65 | 1480 | if (unlikely((rq->stop && task_on_rq_queued(rq->stop)) || |
a1d9a323 | 1481 | rq->dl.dl_nr_running)) |
37e117c0 PZ |
1482 | return RETRY_TASK; |
1483 | } | |
38033c37 | 1484 | |
734ff2a7 KT |
1485 | /* |
1486 | * We may dequeue prev's rt_rq in put_prev_task(). | |
1487 | * So, we update time before rt_nr_running check. | |
1488 | */ | |
1489 | if (prev->sched_class == &rt_sched_class) | |
1490 | update_curr_rt(rq); | |
1491 | ||
f4ebcbc0 | 1492 | if (!rt_rq->rt_queued) |
606dba2e PZ |
1493 | return NULL; |
1494 | ||
3f1d2a31 | 1495 | put_prev_task(rq, prev); |
606dba2e PZ |
1496 | |
1497 | p = _pick_next_task_rt(rq); | |
917b627d GH |
1498 | |
1499 | /* The running task is never eligible for pushing */ | |
f3f1768f | 1500 | dequeue_pushable_task(rq, p); |
917b627d | 1501 | |
e3fca9e7 | 1502 | queue_push_tasks(rq); |
3f029d3c | 1503 | |
6f505b16 | 1504 | return p; |
bb44e5d1 IM |
1505 | } |
1506 | ||
31ee529c | 1507 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1508 | { |
f1e14ef6 | 1509 | update_curr_rt(rq); |
917b627d GH |
1510 | |
1511 | /* | |
1512 | * The previous task needs to be made eligible for pushing | |
1513 | * if it is still active | |
1514 | */ | |
29baa747 | 1515 | if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1) |
917b627d | 1516 | enqueue_pushable_task(rq, p); |
bb44e5d1 IM |
1517 | } |
1518 | ||
681f3e68 | 1519 | #ifdef CONFIG_SMP |
6f505b16 | 1520 | |
e8fa1362 SR |
1521 | /* Only try algorithms three times */ |
1522 | #define RT_MAX_TRIES 3 | |
1523 | ||
f65eda4f SR |
1524 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1525 | { | |
1526 | if (!task_running(rq, p) && | |
60334caf | 1527 | cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) |
f65eda4f SR |
1528 | return 1; |
1529 | return 0; | |
1530 | } | |
1531 | ||
e23ee747 KT |
1532 | /* |
1533 | * Return the highest pushable rq's task, which is suitable to be executed | |
1534 | * on the cpu, NULL otherwise | |
1535 | */ | |
1536 | static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu) | |
e8fa1362 | 1537 | { |
e23ee747 KT |
1538 | struct plist_head *head = &rq->rt.pushable_tasks; |
1539 | struct task_struct *p; | |
3d07467b | 1540 | |
e23ee747 KT |
1541 | if (!has_pushable_tasks(rq)) |
1542 | return NULL; | |
3d07467b | 1543 | |
e23ee747 KT |
1544 | plist_for_each_entry(p, head, pushable_tasks) { |
1545 | if (pick_rt_task(rq, p, cpu)) | |
1546 | return p; | |
f65eda4f SR |
1547 | } |
1548 | ||
e23ee747 | 1549 | return NULL; |
e8fa1362 SR |
1550 | } |
1551 | ||
0e3900e6 | 1552 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1553 | |
6e1254d2 GH |
1554 | static int find_lowest_rq(struct task_struct *task) |
1555 | { | |
1556 | struct sched_domain *sd; | |
4ba29684 | 1557 | struct cpumask *lowest_mask = this_cpu_cpumask_var_ptr(local_cpu_mask); |
6e1254d2 GH |
1558 | int this_cpu = smp_processor_id(); |
1559 | int cpu = task_cpu(task); | |
06f90dbd | 1560 | |
0da938c4 SR |
1561 | /* Make sure the mask is initialized first */ |
1562 | if (unlikely(!lowest_mask)) | |
1563 | return -1; | |
1564 | ||
29baa747 | 1565 | if (task->nr_cpus_allowed == 1) |
6e0534f2 | 1566 | return -1; /* No other targets possible */ |
6e1254d2 | 1567 | |
6e0534f2 GH |
1568 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1569 | return -1; /* No targets found */ | |
6e1254d2 GH |
1570 | |
1571 | /* | |
1572 | * At this point we have built a mask of cpus representing the | |
1573 | * lowest priority tasks in the system. Now we want to elect | |
1574 | * the best one based on our affinity and topology. | |
1575 | * | |
1576 | * We prioritize the last cpu that the task executed on since | |
1577 | * it is most likely cache-hot in that location. | |
1578 | */ | |
96f874e2 | 1579 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1580 | return cpu; |
1581 | ||
1582 | /* | |
1583 | * Otherwise, we consult the sched_domains span maps to figure | |
1584 | * out which cpu is logically closest to our hot cache data. | |
1585 | */ | |
e2c88063 RR |
1586 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1587 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ | |
6e1254d2 | 1588 | |
cd4ae6ad | 1589 | rcu_read_lock(); |
e2c88063 RR |
1590 | for_each_domain(cpu, sd) { |
1591 | if (sd->flags & SD_WAKE_AFFINE) { | |
1592 | int best_cpu; | |
6e1254d2 | 1593 | |
e2c88063 RR |
1594 | /* |
1595 | * "this_cpu" is cheaper to preempt than a | |
1596 | * remote processor. | |
1597 | */ | |
1598 | if (this_cpu != -1 && | |
cd4ae6ad XF |
1599 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
1600 | rcu_read_unlock(); | |
e2c88063 | 1601 | return this_cpu; |
cd4ae6ad | 1602 | } |
e2c88063 RR |
1603 | |
1604 | best_cpu = cpumask_first_and(lowest_mask, | |
1605 | sched_domain_span(sd)); | |
cd4ae6ad XF |
1606 | if (best_cpu < nr_cpu_ids) { |
1607 | rcu_read_unlock(); | |
e2c88063 | 1608 | return best_cpu; |
cd4ae6ad | 1609 | } |
6e1254d2 GH |
1610 | } |
1611 | } | |
cd4ae6ad | 1612 | rcu_read_unlock(); |
6e1254d2 GH |
1613 | |
1614 | /* | |
1615 | * And finally, if there were no matches within the domains | |
1616 | * just give the caller *something* to work with from the compatible | |
1617 | * locations. | |
1618 | */ | |
e2c88063 RR |
1619 | if (this_cpu != -1) |
1620 | return this_cpu; | |
1621 | ||
1622 | cpu = cpumask_any(lowest_mask); | |
1623 | if (cpu < nr_cpu_ids) | |
1624 | return cpu; | |
1625 | return -1; | |
07b4032c GH |
1626 | } |
1627 | ||
1628 | /* Will lock the rq it finds */ | |
4df64c0b | 1629 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1630 | { |
1631 | struct rq *lowest_rq = NULL; | |
07b4032c | 1632 | int tries; |
4df64c0b | 1633 | int cpu; |
e8fa1362 | 1634 | |
07b4032c GH |
1635 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1636 | cpu = find_lowest_rq(task); | |
1637 | ||
2de0b463 | 1638 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1639 | break; |
1640 | ||
07b4032c GH |
1641 | lowest_rq = cpu_rq(cpu); |
1642 | ||
80e3d87b TC |
1643 | if (lowest_rq->rt.highest_prio.curr <= task->prio) { |
1644 | /* | |
1645 | * Target rq has tasks of equal or higher priority, | |
1646 | * retrying does not release any lock and is unlikely | |
1647 | * to yield a different result. | |
1648 | */ | |
1649 | lowest_rq = NULL; | |
1650 | break; | |
1651 | } | |
1652 | ||
e8fa1362 | 1653 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1654 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1655 | /* |
1656 | * We had to unlock the run queue. In | |
1657 | * the mean time, task could have | |
1658 | * migrated already or had its affinity changed. | |
1659 | * Also make sure that it wasn't scheduled on its rq. | |
1660 | */ | |
07b4032c | 1661 | if (unlikely(task_rq(task) != rq || |
96f874e2 | 1662 | !cpumask_test_cpu(lowest_rq->cpu, |
fa17b507 | 1663 | tsk_cpus_allowed(task)) || |
07b4032c | 1664 | task_running(rq, task) || |
da0c1e65 | 1665 | !task_on_rq_queued(task))) { |
4df64c0b | 1666 | |
7f1b4393 | 1667 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1668 | lowest_rq = NULL; |
1669 | break; | |
1670 | } | |
1671 | } | |
1672 | ||
1673 | /* If this rq is still suitable use it. */ | |
e864c499 | 1674 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1675 | break; |
1676 | ||
1677 | /* try again */ | |
1b12bbc7 | 1678 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1679 | lowest_rq = NULL; |
1680 | } | |
1681 | ||
1682 | return lowest_rq; | |
1683 | } | |
1684 | ||
917b627d GH |
1685 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1686 | { | |
1687 | struct task_struct *p; | |
1688 | ||
1689 | if (!has_pushable_tasks(rq)) | |
1690 | return NULL; | |
1691 | ||
1692 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1693 | struct task_struct, pushable_tasks); | |
1694 | ||
1695 | BUG_ON(rq->cpu != task_cpu(p)); | |
1696 | BUG_ON(task_current(rq, p)); | |
29baa747 | 1697 | BUG_ON(p->nr_cpus_allowed <= 1); |
917b627d | 1698 | |
da0c1e65 | 1699 | BUG_ON(!task_on_rq_queued(p)); |
917b627d GH |
1700 | BUG_ON(!rt_task(p)); |
1701 | ||
1702 | return p; | |
1703 | } | |
1704 | ||
e8fa1362 SR |
1705 | /* |
1706 | * If the current CPU has more than one RT task, see if the non | |
1707 | * running task can migrate over to a CPU that is running a task | |
1708 | * of lesser priority. | |
1709 | */ | |
697f0a48 | 1710 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1711 | { |
1712 | struct task_struct *next_task; | |
1713 | struct rq *lowest_rq; | |
311e800e | 1714 | int ret = 0; |
e8fa1362 | 1715 | |
a22d7fc1 GH |
1716 | if (!rq->rt.overloaded) |
1717 | return 0; | |
1718 | ||
917b627d | 1719 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1720 | if (!next_task) |
1721 | return 0; | |
1722 | ||
49246274 | 1723 | retry: |
697f0a48 | 1724 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1725 | WARN_ON(1); |
e8fa1362 | 1726 | return 0; |
f65eda4f | 1727 | } |
e8fa1362 SR |
1728 | |
1729 | /* | |
1730 | * It's possible that the next_task slipped in of | |
1731 | * higher priority than current. If that's the case | |
1732 | * just reschedule current. | |
1733 | */ | |
697f0a48 | 1734 | if (unlikely(next_task->prio < rq->curr->prio)) { |
8875125e | 1735 | resched_curr(rq); |
e8fa1362 SR |
1736 | return 0; |
1737 | } | |
1738 | ||
697f0a48 | 1739 | /* We might release rq lock */ |
e8fa1362 SR |
1740 | get_task_struct(next_task); |
1741 | ||
1742 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1743 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1744 | if (!lowest_rq) { |
1745 | struct task_struct *task; | |
1746 | /* | |
311e800e | 1747 | * find_lock_lowest_rq releases rq->lock |
1563513d GH |
1748 | * so it is possible that next_task has migrated. |
1749 | * | |
1750 | * We need to make sure that the task is still on the same | |
1751 | * run-queue and is also still the next task eligible for | |
1752 | * pushing. | |
e8fa1362 | 1753 | */ |
917b627d | 1754 | task = pick_next_pushable_task(rq); |
1563513d GH |
1755 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1756 | /* | |
311e800e HD |
1757 | * The task hasn't migrated, and is still the next |
1758 | * eligible task, but we failed to find a run-queue | |
1759 | * to push it to. Do not retry in this case, since | |
1760 | * other cpus will pull from us when ready. | |
1563513d | 1761 | */ |
1563513d | 1762 | goto out; |
e8fa1362 | 1763 | } |
917b627d | 1764 | |
1563513d GH |
1765 | if (!task) |
1766 | /* No more tasks, just exit */ | |
1767 | goto out; | |
1768 | ||
917b627d | 1769 | /* |
1563513d | 1770 | * Something has shifted, try again. |
917b627d | 1771 | */ |
1563513d GH |
1772 | put_task_struct(next_task); |
1773 | next_task = task; | |
1774 | goto retry; | |
e8fa1362 SR |
1775 | } |
1776 | ||
697f0a48 | 1777 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1778 | set_task_cpu(next_task, lowest_rq->cpu); |
1779 | activate_task(lowest_rq, next_task, 0); | |
311e800e | 1780 | ret = 1; |
e8fa1362 | 1781 | |
8875125e | 1782 | resched_curr(lowest_rq); |
e8fa1362 | 1783 | |
1b12bbc7 | 1784 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1785 | |
e8fa1362 SR |
1786 | out: |
1787 | put_task_struct(next_task); | |
1788 | ||
311e800e | 1789 | return ret; |
e8fa1362 SR |
1790 | } |
1791 | ||
e8fa1362 SR |
1792 | static void push_rt_tasks(struct rq *rq) |
1793 | { | |
1794 | /* push_rt_task will return true if it moved an RT */ | |
1795 | while (push_rt_task(rq)) | |
1796 | ; | |
1797 | } | |
1798 | ||
b6366f04 SR |
1799 | #ifdef HAVE_RT_PUSH_IPI |
1800 | /* | |
1801 | * The search for the next cpu always starts at rq->cpu and ends | |
1802 | * when we reach rq->cpu again. It will never return rq->cpu. | |
1803 | * This returns the next cpu to check, or nr_cpu_ids if the loop | |
1804 | * is complete. | |
1805 | * | |
1806 | * rq->rt.push_cpu holds the last cpu returned by this function, | |
1807 | * or if this is the first instance, it must hold rq->cpu. | |
1808 | */ | |
1809 | static int rto_next_cpu(struct rq *rq) | |
1810 | { | |
1811 | int prev_cpu = rq->rt.push_cpu; | |
1812 | int cpu; | |
1813 | ||
1814 | cpu = cpumask_next(prev_cpu, rq->rd->rto_mask); | |
1815 | ||
1816 | /* | |
1817 | * If the previous cpu is less than the rq's CPU, then it already | |
1818 | * passed the end of the mask, and has started from the beginning. | |
1819 | * We end if the next CPU is greater or equal to rq's CPU. | |
1820 | */ | |
1821 | if (prev_cpu < rq->cpu) { | |
1822 | if (cpu >= rq->cpu) | |
1823 | return nr_cpu_ids; | |
1824 | ||
1825 | } else if (cpu >= nr_cpu_ids) { | |
1826 | /* | |
1827 | * We passed the end of the mask, start at the beginning. | |
1828 | * If the result is greater or equal to the rq's CPU, then | |
1829 | * the loop is finished. | |
1830 | */ | |
1831 | cpu = cpumask_first(rq->rd->rto_mask); | |
1832 | if (cpu >= rq->cpu) | |
1833 | return nr_cpu_ids; | |
1834 | } | |
1835 | rq->rt.push_cpu = cpu; | |
1836 | ||
1837 | /* Return cpu to let the caller know if the loop is finished or not */ | |
1838 | return cpu; | |
1839 | } | |
1840 | ||
1841 | static int find_next_push_cpu(struct rq *rq) | |
1842 | { | |
1843 | struct rq *next_rq; | |
1844 | int cpu; | |
1845 | ||
1846 | while (1) { | |
1847 | cpu = rto_next_cpu(rq); | |
1848 | if (cpu >= nr_cpu_ids) | |
1849 | break; | |
1850 | next_rq = cpu_rq(cpu); | |
1851 | ||
1852 | /* Make sure the next rq can push to this rq */ | |
1853 | if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr) | |
1854 | break; | |
1855 | } | |
1856 | ||
1857 | return cpu; | |
1858 | } | |
1859 | ||
1860 | #define RT_PUSH_IPI_EXECUTING 1 | |
1861 | #define RT_PUSH_IPI_RESTART 2 | |
1862 | ||
1863 | static void tell_cpu_to_push(struct rq *rq) | |
1864 | { | |
1865 | int cpu; | |
1866 | ||
1867 | if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { | |
1868 | raw_spin_lock(&rq->rt.push_lock); | |
1869 | /* Make sure it's still executing */ | |
1870 | if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { | |
1871 | /* | |
1872 | * Tell the IPI to restart the loop as things have | |
1873 | * changed since it started. | |
1874 | */ | |
1875 | rq->rt.push_flags |= RT_PUSH_IPI_RESTART; | |
1876 | raw_spin_unlock(&rq->rt.push_lock); | |
1877 | return; | |
1878 | } | |
1879 | raw_spin_unlock(&rq->rt.push_lock); | |
1880 | } | |
1881 | ||
1882 | /* When here, there's no IPI going around */ | |
1883 | ||
1884 | rq->rt.push_cpu = rq->cpu; | |
1885 | cpu = find_next_push_cpu(rq); | |
1886 | if (cpu >= nr_cpu_ids) | |
1887 | return; | |
1888 | ||
1889 | rq->rt.push_flags = RT_PUSH_IPI_EXECUTING; | |
1890 | ||
1891 | irq_work_queue_on(&rq->rt.push_work, cpu); | |
1892 | } | |
1893 | ||
1894 | /* Called from hardirq context */ | |
1895 | static void try_to_push_tasks(void *arg) | |
1896 | { | |
1897 | struct rt_rq *rt_rq = arg; | |
1898 | struct rq *rq, *src_rq; | |
1899 | int this_cpu; | |
1900 | int cpu; | |
1901 | ||
1902 | this_cpu = rt_rq->push_cpu; | |
1903 | ||
1904 | /* Paranoid check */ | |
1905 | BUG_ON(this_cpu != smp_processor_id()); | |
1906 | ||
1907 | rq = cpu_rq(this_cpu); | |
1908 | src_rq = rq_of_rt_rq(rt_rq); | |
1909 | ||
1910 | again: | |
1911 | if (has_pushable_tasks(rq)) { | |
1912 | raw_spin_lock(&rq->lock); | |
1913 | push_rt_task(rq); | |
1914 | raw_spin_unlock(&rq->lock); | |
1915 | } | |
1916 | ||
1917 | /* Pass the IPI to the next rt overloaded queue */ | |
1918 | raw_spin_lock(&rt_rq->push_lock); | |
1919 | /* | |
1920 | * If the source queue changed since the IPI went out, | |
1921 | * we need to restart the search from that CPU again. | |
1922 | */ | |
1923 | if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) { | |
1924 | rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART; | |
1925 | rt_rq->push_cpu = src_rq->cpu; | |
1926 | } | |
1927 | ||
1928 | cpu = find_next_push_cpu(src_rq); | |
1929 | ||
1930 | if (cpu >= nr_cpu_ids) | |
1931 | rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING; | |
1932 | raw_spin_unlock(&rt_rq->push_lock); | |
1933 | ||
1934 | if (cpu >= nr_cpu_ids) | |
1935 | return; | |
1936 | ||
1937 | /* | |
1938 | * It is possible that a restart caused this CPU to be | |
1939 | * chosen again. Don't bother with an IPI, just see if we | |
1940 | * have more to push. | |
1941 | */ | |
1942 | if (unlikely(cpu == rq->cpu)) | |
1943 | goto again; | |
1944 | ||
1945 | /* Try the next RT overloaded CPU */ | |
1946 | irq_work_queue_on(&rt_rq->push_work, cpu); | |
1947 | } | |
1948 | ||
1949 | static void push_irq_work_func(struct irq_work *work) | |
1950 | { | |
1951 | struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work); | |
1952 | ||
1953 | try_to_push_tasks(rt_rq); | |
1954 | } | |
1955 | #endif /* HAVE_RT_PUSH_IPI */ | |
1956 | ||
8046d680 | 1957 | static void pull_rt_task(struct rq *this_rq) |
f65eda4f | 1958 | { |
8046d680 PZ |
1959 | int this_cpu = this_rq->cpu, cpu; |
1960 | bool resched = false; | |
a8728944 | 1961 | struct task_struct *p; |
f65eda4f | 1962 | struct rq *src_rq; |
f65eda4f | 1963 | |
637f5085 | 1964 | if (likely(!rt_overloaded(this_rq))) |
8046d680 | 1965 | return; |
f65eda4f | 1966 | |
7c3f2ab7 PZ |
1967 | /* |
1968 | * Match the barrier from rt_set_overloaded; this guarantees that if we | |
1969 | * see overloaded we must also see the rto_mask bit. | |
1970 | */ | |
1971 | smp_rmb(); | |
1972 | ||
b6366f04 SR |
1973 | #ifdef HAVE_RT_PUSH_IPI |
1974 | if (sched_feat(RT_PUSH_IPI)) { | |
1975 | tell_cpu_to_push(this_rq); | |
8046d680 | 1976 | return; |
b6366f04 SR |
1977 | } |
1978 | #endif | |
1979 | ||
c6c4927b | 1980 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1981 | if (this_cpu == cpu) |
1982 | continue; | |
1983 | ||
1984 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1985 | |
1986 | /* | |
1987 | * Don't bother taking the src_rq->lock if the next highest | |
1988 | * task is known to be lower-priority than our current task. | |
1989 | * This may look racy, but if this value is about to go | |
1990 | * logically higher, the src_rq will push this task away. | |
1991 | * And if its going logically lower, we do not care | |
1992 | */ | |
1993 | if (src_rq->rt.highest_prio.next >= | |
1994 | this_rq->rt.highest_prio.curr) | |
1995 | continue; | |
1996 | ||
f65eda4f SR |
1997 | /* |
1998 | * We can potentially drop this_rq's lock in | |
1999 | * double_lock_balance, and another CPU could | |
a8728944 | 2000 | * alter this_rq |
f65eda4f | 2001 | */ |
a8728944 | 2002 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
2003 | |
2004 | /* | |
e23ee747 KT |
2005 | * We can pull only a task, which is pushable |
2006 | * on its rq, and no others. | |
f65eda4f | 2007 | */ |
e23ee747 | 2008 | p = pick_highest_pushable_task(src_rq, this_cpu); |
f65eda4f SR |
2009 | |
2010 | /* | |
2011 | * Do we have an RT task that preempts | |
2012 | * the to-be-scheduled task? | |
2013 | */ | |
a8728944 | 2014 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f | 2015 | WARN_ON(p == src_rq->curr); |
da0c1e65 | 2016 | WARN_ON(!task_on_rq_queued(p)); |
f65eda4f SR |
2017 | |
2018 | /* | |
2019 | * There's a chance that p is higher in priority | |
2020 | * than what's currently running on its cpu. | |
2021 | * This is just that p is wakeing up and hasn't | |
2022 | * had a chance to schedule. We only pull | |
2023 | * p if it is lower in priority than the | |
a8728944 | 2024 | * current task on the run queue |
f65eda4f | 2025 | */ |
a8728944 | 2026 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 2027 | goto skip; |
f65eda4f | 2028 | |
8046d680 | 2029 | resched = true; |
f65eda4f SR |
2030 | |
2031 | deactivate_task(src_rq, p, 0); | |
2032 | set_task_cpu(p, this_cpu); | |
2033 | activate_task(this_rq, p, 0); | |
2034 | /* | |
2035 | * We continue with the search, just in | |
2036 | * case there's an even higher prio task | |
25985edc | 2037 | * in another runqueue. (low likelihood |
f65eda4f | 2038 | * but possible) |
f65eda4f | 2039 | */ |
f65eda4f | 2040 | } |
49246274 | 2041 | skip: |
1b12bbc7 | 2042 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
2043 | } |
2044 | ||
8046d680 PZ |
2045 | if (resched) |
2046 | resched_curr(this_rq); | |
f65eda4f SR |
2047 | } |
2048 | ||
8ae121ac GH |
2049 | /* |
2050 | * If we are not running and we are not going to reschedule soon, we should | |
2051 | * try to push tasks away now | |
2052 | */ | |
efbbd05a | 2053 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 2054 | { |
9a897c5a | 2055 | if (!task_running(rq, p) && |
8ae121ac | 2056 | !test_tsk_need_resched(rq->curr) && |
917b627d | 2057 | has_pushable_tasks(rq) && |
29baa747 | 2058 | p->nr_cpus_allowed > 1 && |
1baca4ce | 2059 | (dl_task(rq->curr) || rt_task(rq->curr)) && |
29baa747 | 2060 | (rq->curr->nr_cpus_allowed < 2 || |
3be209a8 | 2061 | rq->curr->prio <= p->prio)) |
4642dafd SR |
2062 | push_rt_tasks(rq); |
2063 | } | |
2064 | ||
cd8ba7cd | 2065 | static void set_cpus_allowed_rt(struct task_struct *p, |
96f874e2 | 2066 | const struct cpumask *new_mask) |
73fe6aae | 2067 | { |
8d3d5ada KT |
2068 | struct rq *rq; |
2069 | int weight; | |
73fe6aae GH |
2070 | |
2071 | BUG_ON(!rt_task(p)); | |
2072 | ||
da0c1e65 | 2073 | if (!task_on_rq_queued(p)) |
8d3d5ada | 2074 | return; |
917b627d | 2075 | |
8d3d5ada | 2076 | weight = cpumask_weight(new_mask); |
917b627d | 2077 | |
8d3d5ada KT |
2078 | /* |
2079 | * Only update if the process changes its state from whether it | |
2080 | * can migrate or not. | |
2081 | */ | |
29baa747 | 2082 | if ((p->nr_cpus_allowed > 1) == (weight > 1)) |
8d3d5ada | 2083 | return; |
917b627d | 2084 | |
8d3d5ada | 2085 | rq = task_rq(p); |
73fe6aae | 2086 | |
8d3d5ada KT |
2087 | /* |
2088 | * The process used to be able to migrate OR it can now migrate | |
2089 | */ | |
2090 | if (weight <= 1) { | |
2091 | if (!task_current(rq, p)) | |
2092 | dequeue_pushable_task(rq, p); | |
2093 | BUG_ON(!rq->rt.rt_nr_migratory); | |
2094 | rq->rt.rt_nr_migratory--; | |
2095 | } else { | |
2096 | if (!task_current(rq, p)) | |
2097 | enqueue_pushable_task(rq, p); | |
2098 | rq->rt.rt_nr_migratory++; | |
73fe6aae | 2099 | } |
8d3d5ada KT |
2100 | |
2101 | update_rt_migration(&rq->rt); | |
73fe6aae | 2102 | } |
deeeccd4 | 2103 | |
bdd7c81b | 2104 | /* Assumes rq->lock is held */ |
1f11eb6a | 2105 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
2106 | { |
2107 | if (rq->rt.overloaded) | |
2108 | rt_set_overload(rq); | |
6e0534f2 | 2109 | |
7def2be1 PZ |
2110 | __enable_runtime(rq); |
2111 | ||
e864c499 | 2112 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
2113 | } |
2114 | ||
2115 | /* Assumes rq->lock is held */ | |
1f11eb6a | 2116 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
2117 | { |
2118 | if (rq->rt.overloaded) | |
2119 | rt_clear_overload(rq); | |
6e0534f2 | 2120 | |
7def2be1 PZ |
2121 | __disable_runtime(rq); |
2122 | ||
6e0534f2 | 2123 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 2124 | } |
cb469845 SR |
2125 | |
2126 | /* | |
2127 | * When switch from the rt queue, we bring ourselves to a position | |
2128 | * that we might want to pull RT tasks from other runqueues. | |
2129 | */ | |
da7a735e | 2130 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
2131 | { |
2132 | /* | |
2133 | * If there are other RT tasks then we will reschedule | |
2134 | * and the scheduling of the other RT tasks will handle | |
2135 | * the balancing. But if we are the last RT task | |
2136 | * we may need to handle the pulling of RT tasks | |
2137 | * now. | |
2138 | */ | |
da0c1e65 | 2139 | if (!task_on_rq_queued(p) || rq->rt.rt_nr_running) |
1158ddb5 KT |
2140 | return; |
2141 | ||
8046d680 | 2142 | pull_rt_task(rq); |
cb469845 | 2143 | } |
3d8cbdf8 | 2144 | |
11c785b7 | 2145 | void __init init_sched_rt_class(void) |
3d8cbdf8 RR |
2146 | { |
2147 | unsigned int i; | |
2148 | ||
029632fb | 2149 | for_each_possible_cpu(i) { |
eaa95840 | 2150 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 2151 | GFP_KERNEL, cpu_to_node(i)); |
029632fb | 2152 | } |
3d8cbdf8 | 2153 | } |
cb469845 SR |
2154 | #endif /* CONFIG_SMP */ |
2155 | ||
2156 | /* | |
2157 | * When switching a task to RT, we may overload the runqueue | |
2158 | * with RT tasks. In this case we try to push them off to | |
2159 | * other runqueues. | |
2160 | */ | |
da7a735e | 2161 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
2162 | { |
2163 | int check_resched = 1; | |
2164 | ||
2165 | /* | |
2166 | * If we are already running, then there's nothing | |
2167 | * that needs to be done. But if we are not running | |
2168 | * we may need to preempt the current running task. | |
2169 | * If that current running task is also an RT task | |
2170 | * then see if we can move to another run queue. | |
2171 | */ | |
da0c1e65 | 2172 | if (task_on_rq_queued(p) && rq->curr != p) { |
cb469845 | 2173 | #ifdef CONFIG_SMP |
10447917 | 2174 | if (p->nr_cpus_allowed > 1 && rq->rt.overloaded && |
cb469845 | 2175 | /* Don't resched if we changed runqueues */ |
10447917 | 2176 | push_rt_task(rq) && rq != task_rq(p)) |
cb469845 SR |
2177 | check_resched = 0; |
2178 | #endif /* CONFIG_SMP */ | |
2179 | if (check_resched && p->prio < rq->curr->prio) | |
8875125e | 2180 | resched_curr(rq); |
cb469845 SR |
2181 | } |
2182 | } | |
2183 | ||
2184 | /* | |
2185 | * Priority of the task has changed. This may cause | |
2186 | * us to initiate a push or pull. | |
2187 | */ | |
da7a735e PZ |
2188 | static void |
2189 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) | |
cb469845 | 2190 | { |
da0c1e65 | 2191 | if (!task_on_rq_queued(p)) |
da7a735e PZ |
2192 | return; |
2193 | ||
2194 | if (rq->curr == p) { | |
cb469845 SR |
2195 | #ifdef CONFIG_SMP |
2196 | /* | |
2197 | * If our priority decreases while running, we | |
2198 | * may need to pull tasks to this runqueue. | |
2199 | */ | |
2200 | if (oldprio < p->prio) | |
2201 | pull_rt_task(rq); | |
2202 | /* | |
2203 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
2204 | * then reschedule. Note, the above pull_rt_task |
2205 | * can release the rq lock and p could migrate. | |
2206 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 2207 | */ |
e864c499 | 2208 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
8875125e | 2209 | resched_curr(rq); |
cb469845 SR |
2210 | #else |
2211 | /* For UP simply resched on drop of prio */ | |
2212 | if (oldprio < p->prio) | |
8875125e | 2213 | resched_curr(rq); |
e8fa1362 | 2214 | #endif /* CONFIG_SMP */ |
cb469845 SR |
2215 | } else { |
2216 | /* | |
2217 | * This task is not running, but if it is | |
2218 | * greater than the current running task | |
2219 | * then reschedule. | |
2220 | */ | |
2221 | if (p->prio < rq->curr->prio) | |
8875125e | 2222 | resched_curr(rq); |
cb469845 SR |
2223 | } |
2224 | } | |
2225 | ||
78f2c7db PZ |
2226 | static void watchdog(struct rq *rq, struct task_struct *p) |
2227 | { | |
2228 | unsigned long soft, hard; | |
2229 | ||
78d7d407 JS |
2230 | /* max may change after cur was read, this will be fixed next tick */ |
2231 | soft = task_rlimit(p, RLIMIT_RTTIME); | |
2232 | hard = task_rlimit_max(p, RLIMIT_RTTIME); | |
78f2c7db PZ |
2233 | |
2234 | if (soft != RLIM_INFINITY) { | |
2235 | unsigned long next; | |
2236 | ||
57d2aa00 YX |
2237 | if (p->rt.watchdog_stamp != jiffies) { |
2238 | p->rt.timeout++; | |
2239 | p->rt.watchdog_stamp = jiffies; | |
2240 | } | |
2241 | ||
78f2c7db | 2242 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
5a52dd50 | 2243 | if (p->rt.timeout > next) |
f06febc9 | 2244 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
2245 | } |
2246 | } | |
bb44e5d1 | 2247 | |
8f4d37ec | 2248 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 2249 | { |
454c7999 CC |
2250 | struct sched_rt_entity *rt_se = &p->rt; |
2251 | ||
67e2be02 PZ |
2252 | update_curr_rt(rq); |
2253 | ||
78f2c7db PZ |
2254 | watchdog(rq, p); |
2255 | ||
bb44e5d1 IM |
2256 | /* |
2257 | * RR tasks need a special form of timeslice management. | |
2258 | * FIFO tasks have no timeslices. | |
2259 | */ | |
2260 | if (p->policy != SCHED_RR) | |
2261 | return; | |
2262 | ||
fa717060 | 2263 | if (--p->rt.time_slice) |
bb44e5d1 IM |
2264 | return; |
2265 | ||
ce0dbbbb | 2266 | p->rt.time_slice = sched_rr_timeslice; |
bb44e5d1 | 2267 | |
98fbc798 | 2268 | /* |
e9aa39bb LB |
2269 | * Requeue to the end of queue if we (and all of our ancestors) are not |
2270 | * the only element on the queue | |
98fbc798 | 2271 | */ |
454c7999 CC |
2272 | for_each_sched_rt_entity(rt_se) { |
2273 | if (rt_se->run_list.prev != rt_se->run_list.next) { | |
2274 | requeue_task_rt(rq, p, 0); | |
8aa6f0eb | 2275 | resched_curr(rq); |
454c7999 CC |
2276 | return; |
2277 | } | |
98fbc798 | 2278 | } |
bb44e5d1 IM |
2279 | } |
2280 | ||
83b699ed SV |
2281 | static void set_curr_task_rt(struct rq *rq) |
2282 | { | |
2283 | struct task_struct *p = rq->curr; | |
2284 | ||
78becc27 | 2285 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
2286 | |
2287 | /* The running task is never eligible for pushing */ | |
2288 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
2289 | } |
2290 | ||
6d686f45 | 2291 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
0d721cea PW |
2292 | { |
2293 | /* | |
2294 | * Time slice is 0 for SCHED_FIFO tasks | |
2295 | */ | |
2296 | if (task->policy == SCHED_RR) | |
ce0dbbbb | 2297 | return sched_rr_timeslice; |
0d721cea PW |
2298 | else |
2299 | return 0; | |
2300 | } | |
2301 | ||
029632fb | 2302 | const struct sched_class rt_sched_class = { |
5522d5d5 | 2303 | .next = &fair_sched_class, |
bb44e5d1 IM |
2304 | .enqueue_task = enqueue_task_rt, |
2305 | .dequeue_task = dequeue_task_rt, | |
2306 | .yield_task = yield_task_rt, | |
2307 | ||
2308 | .check_preempt_curr = check_preempt_curr_rt, | |
2309 | ||
2310 | .pick_next_task = pick_next_task_rt, | |
2311 | .put_prev_task = put_prev_task_rt, | |
2312 | ||
681f3e68 | 2313 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
2314 | .select_task_rq = select_task_rq_rt, |
2315 | ||
73fe6aae | 2316 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
2317 | .rq_online = rq_online_rt, |
2318 | .rq_offline = rq_offline_rt, | |
efbbd05a | 2319 | .task_woken = task_woken_rt, |
cb469845 | 2320 | .switched_from = switched_from_rt, |
681f3e68 | 2321 | #endif |
bb44e5d1 | 2322 | |
83b699ed | 2323 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 2324 | .task_tick = task_tick_rt, |
cb469845 | 2325 | |
0d721cea PW |
2326 | .get_rr_interval = get_rr_interval_rt, |
2327 | ||
cb469845 SR |
2328 | .prio_changed = prio_changed_rt, |
2329 | .switched_to = switched_to_rt, | |
6e998916 SG |
2330 | |
2331 | .update_curr = update_curr_rt, | |
bb44e5d1 | 2332 | }; |
ada18de2 PZ |
2333 | |
2334 | #ifdef CONFIG_SCHED_DEBUG | |
2335 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2336 | ||
029632fb | 2337 | void print_rt_stats(struct seq_file *m, int cpu) |
ada18de2 | 2338 | { |
ec514c48 | 2339 | rt_rq_iter_t iter; |
ada18de2 PZ |
2340 | struct rt_rq *rt_rq; |
2341 | ||
2342 | rcu_read_lock(); | |
ec514c48 | 2343 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
ada18de2 PZ |
2344 | print_rt_rq(m, cpu, rt_rq); |
2345 | rcu_read_unlock(); | |
2346 | } | |
55e12e5e | 2347 | #endif /* CONFIG_SCHED_DEBUG */ |