]>
Commit | Line | Data |
---|---|---|
bb44e5d1 IM |
1 | /* |
2 | * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR | |
3 | * policies) | |
4 | */ | |
5 | ||
029632fb PZ |
6 | #include "sched.h" |
7 | ||
8 | #include <linux/slab.h> | |
9 | ||
ce0dbbbb CW |
10 | int sched_rr_timeslice = RR_TIMESLICE; |
11 | ||
029632fb PZ |
12 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); |
13 | ||
14 | struct rt_bandwidth def_rt_bandwidth; | |
15 | ||
16 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
17 | { | |
18 | struct rt_bandwidth *rt_b = | |
19 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
20 | ktime_t now; | |
21 | int overrun; | |
22 | int idle = 0; | |
23 | ||
24 | for (;;) { | |
25 | now = hrtimer_cb_get_time(timer); | |
26 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
27 | ||
28 | if (!overrun) | |
29 | break; | |
30 | ||
31 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
32 | } | |
33 | ||
34 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
35 | } | |
36 | ||
37 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
38 | { | |
39 | rt_b->rt_period = ns_to_ktime(period); | |
40 | rt_b->rt_runtime = runtime; | |
41 | ||
42 | raw_spin_lock_init(&rt_b->rt_runtime_lock); | |
43 | ||
44 | hrtimer_init(&rt_b->rt_period_timer, | |
45 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
46 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
47 | } | |
48 | ||
49 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
50 | { | |
51 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) | |
52 | return; | |
53 | ||
54 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
55 | return; | |
56 | ||
57 | raw_spin_lock(&rt_b->rt_runtime_lock); | |
58 | start_bandwidth_timer(&rt_b->rt_period_timer, rt_b->rt_period); | |
59 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
60 | } | |
61 | ||
62 | void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) | |
63 | { | |
64 | struct rt_prio_array *array; | |
65 | int i; | |
66 | ||
67 | array = &rt_rq->active; | |
68 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
69 | INIT_LIST_HEAD(array->queue + i); | |
70 | __clear_bit(i, array->bitmap); | |
71 | } | |
72 | /* delimiter for bitsearch: */ | |
73 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
74 | ||
75 | #if defined CONFIG_SMP | |
76 | rt_rq->highest_prio.curr = MAX_RT_PRIO; | |
77 | rt_rq->highest_prio.next = MAX_RT_PRIO; | |
78 | rt_rq->rt_nr_migratory = 0; | |
79 | rt_rq->overloaded = 0; | |
80 | plist_head_init(&rt_rq->pushable_tasks); | |
81 | #endif | |
f4ebcbc0 KT |
82 | /* We start is dequeued state, because no RT tasks are queued */ |
83 | rt_rq->rt_queued = 0; | |
029632fb PZ |
84 | |
85 | rt_rq->rt_time = 0; | |
86 | rt_rq->rt_throttled = 0; | |
87 | rt_rq->rt_runtime = 0; | |
88 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); | |
89 | } | |
90 | ||
8f48894f | 91 | #ifdef CONFIG_RT_GROUP_SCHED |
029632fb PZ |
92 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) |
93 | { | |
94 | hrtimer_cancel(&rt_b->rt_period_timer); | |
95 | } | |
8f48894f PZ |
96 | |
97 | #define rt_entity_is_task(rt_se) (!(rt_se)->my_q) | |
98 | ||
398a153b GH |
99 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
100 | { | |
8f48894f PZ |
101 | #ifdef CONFIG_SCHED_DEBUG |
102 | WARN_ON_ONCE(!rt_entity_is_task(rt_se)); | |
103 | #endif | |
398a153b GH |
104 | return container_of(rt_se, struct task_struct, rt); |
105 | } | |
106 | ||
398a153b GH |
107 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
108 | { | |
109 | return rt_rq->rq; | |
110 | } | |
111 | ||
112 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
113 | { | |
114 | return rt_se->rt_rq; | |
115 | } | |
116 | ||
653d07a6 KT |
117 | static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se) |
118 | { | |
119 | struct rt_rq *rt_rq = rt_se->rt_rq; | |
120 | ||
121 | return rt_rq->rq; | |
122 | } | |
123 | ||
029632fb PZ |
124 | void free_rt_sched_group(struct task_group *tg) |
125 | { | |
126 | int i; | |
127 | ||
128 | if (tg->rt_se) | |
129 | destroy_rt_bandwidth(&tg->rt_bandwidth); | |
130 | ||
131 | for_each_possible_cpu(i) { | |
132 | if (tg->rt_rq) | |
133 | kfree(tg->rt_rq[i]); | |
134 | if (tg->rt_se) | |
135 | kfree(tg->rt_se[i]); | |
136 | } | |
137 | ||
138 | kfree(tg->rt_rq); | |
139 | kfree(tg->rt_se); | |
140 | } | |
141 | ||
142 | void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
143 | struct sched_rt_entity *rt_se, int cpu, | |
144 | struct sched_rt_entity *parent) | |
145 | { | |
146 | struct rq *rq = cpu_rq(cpu); | |
147 | ||
148 | rt_rq->highest_prio.curr = MAX_RT_PRIO; | |
149 | rt_rq->rt_nr_boosted = 0; | |
150 | rt_rq->rq = rq; | |
151 | rt_rq->tg = tg; | |
152 | ||
153 | tg->rt_rq[cpu] = rt_rq; | |
154 | tg->rt_se[cpu] = rt_se; | |
155 | ||
156 | if (!rt_se) | |
157 | return; | |
158 | ||
159 | if (!parent) | |
160 | rt_se->rt_rq = &rq->rt; | |
161 | else | |
162 | rt_se->rt_rq = parent->my_q; | |
163 | ||
164 | rt_se->my_q = rt_rq; | |
165 | rt_se->parent = parent; | |
166 | INIT_LIST_HEAD(&rt_se->run_list); | |
167 | } | |
168 | ||
169 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
170 | { | |
171 | struct rt_rq *rt_rq; | |
172 | struct sched_rt_entity *rt_se; | |
173 | int i; | |
174 | ||
175 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); | |
176 | if (!tg->rt_rq) | |
177 | goto err; | |
178 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); | |
179 | if (!tg->rt_se) | |
180 | goto err; | |
181 | ||
182 | init_rt_bandwidth(&tg->rt_bandwidth, | |
183 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
184 | ||
185 | for_each_possible_cpu(i) { | |
186 | rt_rq = kzalloc_node(sizeof(struct rt_rq), | |
187 | GFP_KERNEL, cpu_to_node(i)); | |
188 | if (!rt_rq) | |
189 | goto err; | |
190 | ||
191 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), | |
192 | GFP_KERNEL, cpu_to_node(i)); | |
193 | if (!rt_se) | |
194 | goto err_free_rq; | |
195 | ||
196 | init_rt_rq(rt_rq, cpu_rq(i)); | |
197 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; | |
198 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); | |
199 | } | |
200 | ||
201 | return 1; | |
202 | ||
203 | err_free_rq: | |
204 | kfree(rt_rq); | |
205 | err: | |
206 | return 0; | |
207 | } | |
208 | ||
398a153b GH |
209 | #else /* CONFIG_RT_GROUP_SCHED */ |
210 | ||
a1ba4d8b PZ |
211 | #define rt_entity_is_task(rt_se) (1) |
212 | ||
8f48894f PZ |
213 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
214 | { | |
215 | return container_of(rt_se, struct task_struct, rt); | |
216 | } | |
217 | ||
398a153b GH |
218 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) |
219 | { | |
220 | return container_of(rt_rq, struct rq, rt); | |
221 | } | |
222 | ||
653d07a6 | 223 | static inline struct rq *rq_of_rt_se(struct sched_rt_entity *rt_se) |
398a153b GH |
224 | { |
225 | struct task_struct *p = rt_task_of(rt_se); | |
653d07a6 KT |
226 | |
227 | return task_rq(p); | |
228 | } | |
229 | ||
230 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
231 | { | |
232 | struct rq *rq = rq_of_rt_se(rt_se); | |
398a153b GH |
233 | |
234 | return &rq->rt; | |
235 | } | |
236 | ||
029632fb PZ |
237 | void free_rt_sched_group(struct task_group *tg) { } |
238 | ||
239 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
240 | { | |
241 | return 1; | |
242 | } | |
398a153b GH |
243 | #endif /* CONFIG_RT_GROUP_SCHED */ |
244 | ||
4fd29176 | 245 | #ifdef CONFIG_SMP |
84de4274 | 246 | |
38033c37 PZ |
247 | static int pull_rt_task(struct rq *this_rq); |
248 | ||
dc877341 PZ |
249 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
250 | { | |
251 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
252 | return rq->rt.highest_prio.curr > prev->prio; | |
253 | } | |
254 | ||
637f5085 | 255 | static inline int rt_overloaded(struct rq *rq) |
4fd29176 | 256 | { |
637f5085 | 257 | return atomic_read(&rq->rd->rto_count); |
4fd29176 | 258 | } |
84de4274 | 259 | |
4fd29176 SR |
260 | static inline void rt_set_overload(struct rq *rq) |
261 | { | |
1f11eb6a GH |
262 | if (!rq->online) |
263 | return; | |
264 | ||
c6c4927b | 265 | cpumask_set_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 SR |
266 | /* |
267 | * Make sure the mask is visible before we set | |
268 | * the overload count. That is checked to determine | |
269 | * if we should look at the mask. It would be a shame | |
270 | * if we looked at the mask, but the mask was not | |
271 | * updated yet. | |
7c3f2ab7 PZ |
272 | * |
273 | * Matched by the barrier in pull_rt_task(). | |
4fd29176 | 274 | */ |
7c3f2ab7 | 275 | smp_wmb(); |
637f5085 | 276 | atomic_inc(&rq->rd->rto_count); |
4fd29176 | 277 | } |
84de4274 | 278 | |
4fd29176 SR |
279 | static inline void rt_clear_overload(struct rq *rq) |
280 | { | |
1f11eb6a GH |
281 | if (!rq->online) |
282 | return; | |
283 | ||
4fd29176 | 284 | /* the order here really doesn't matter */ |
637f5085 | 285 | atomic_dec(&rq->rd->rto_count); |
c6c4927b | 286 | cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); |
4fd29176 | 287 | } |
73fe6aae | 288 | |
398a153b | 289 | static void update_rt_migration(struct rt_rq *rt_rq) |
73fe6aae | 290 | { |
a1ba4d8b | 291 | if (rt_rq->rt_nr_migratory && rt_rq->rt_nr_total > 1) { |
398a153b GH |
292 | if (!rt_rq->overloaded) { |
293 | rt_set_overload(rq_of_rt_rq(rt_rq)); | |
294 | rt_rq->overloaded = 1; | |
cdc8eb98 | 295 | } |
398a153b GH |
296 | } else if (rt_rq->overloaded) { |
297 | rt_clear_overload(rq_of_rt_rq(rt_rq)); | |
298 | rt_rq->overloaded = 0; | |
637f5085 | 299 | } |
73fe6aae | 300 | } |
4fd29176 | 301 | |
398a153b GH |
302 | static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
303 | { | |
29baa747 PZ |
304 | struct task_struct *p; |
305 | ||
a1ba4d8b PZ |
306 | if (!rt_entity_is_task(rt_se)) |
307 | return; | |
308 | ||
29baa747 | 309 | p = rt_task_of(rt_se); |
a1ba4d8b PZ |
310 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
311 | ||
312 | rt_rq->rt_nr_total++; | |
29baa747 | 313 | if (p->nr_cpus_allowed > 1) |
398a153b GH |
314 | rt_rq->rt_nr_migratory++; |
315 | ||
316 | update_rt_migration(rt_rq); | |
317 | } | |
318 | ||
319 | static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
320 | { | |
29baa747 PZ |
321 | struct task_struct *p; |
322 | ||
a1ba4d8b PZ |
323 | if (!rt_entity_is_task(rt_se)) |
324 | return; | |
325 | ||
29baa747 | 326 | p = rt_task_of(rt_se); |
a1ba4d8b PZ |
327 | rt_rq = &rq_of_rt_rq(rt_rq)->rt; |
328 | ||
329 | rt_rq->rt_nr_total--; | |
29baa747 | 330 | if (p->nr_cpus_allowed > 1) |
398a153b GH |
331 | rt_rq->rt_nr_migratory--; |
332 | ||
333 | update_rt_migration(rt_rq); | |
334 | } | |
335 | ||
5181f4a4 SR |
336 | static inline int has_pushable_tasks(struct rq *rq) |
337 | { | |
338 | return !plist_head_empty(&rq->rt.pushable_tasks); | |
339 | } | |
340 | ||
dc877341 PZ |
341 | static inline void set_post_schedule(struct rq *rq) |
342 | { | |
343 | /* | |
344 | * We detect this state here so that we can avoid taking the RQ | |
345 | * lock again later if there is no need to push | |
346 | */ | |
347 | rq->post_schedule = has_pushable_tasks(rq); | |
348 | } | |
349 | ||
917b627d GH |
350 | static void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
351 | { | |
352 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
353 | plist_node_init(&p->pushable_tasks, p->prio); | |
354 | plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
5181f4a4 SR |
355 | |
356 | /* Update the highest prio pushable task */ | |
357 | if (p->prio < rq->rt.highest_prio.next) | |
358 | rq->rt.highest_prio.next = p->prio; | |
917b627d GH |
359 | } |
360 | ||
361 | static void dequeue_pushable_task(struct rq *rq, struct task_struct *p) | |
362 | { | |
363 | plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks); | |
917b627d | 364 | |
5181f4a4 SR |
365 | /* Update the new highest prio pushable task */ |
366 | if (has_pushable_tasks(rq)) { | |
367 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
368 | struct task_struct, pushable_tasks); | |
369 | rq->rt.highest_prio.next = p->prio; | |
370 | } else | |
371 | rq->rt.highest_prio.next = MAX_RT_PRIO; | |
bcf08df3 IM |
372 | } |
373 | ||
917b627d GH |
374 | #else |
375 | ||
ceacc2c1 | 376 | static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p) |
fa85ae24 | 377 | { |
6f505b16 PZ |
378 | } |
379 | ||
ceacc2c1 PZ |
380 | static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p) |
381 | { | |
382 | } | |
383 | ||
b07430ac | 384 | static inline |
ceacc2c1 PZ |
385 | void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
386 | { | |
387 | } | |
388 | ||
398a153b | 389 | static inline |
ceacc2c1 PZ |
390 | void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) |
391 | { | |
392 | } | |
917b627d | 393 | |
dc877341 PZ |
394 | static inline bool need_pull_rt_task(struct rq *rq, struct task_struct *prev) |
395 | { | |
396 | return false; | |
397 | } | |
398 | ||
399 | static inline int pull_rt_task(struct rq *this_rq) | |
400 | { | |
401 | return 0; | |
402 | } | |
403 | ||
404 | static inline void set_post_schedule(struct rq *rq) | |
405 | { | |
406 | } | |
4fd29176 SR |
407 | #endif /* CONFIG_SMP */ |
408 | ||
f4ebcbc0 KT |
409 | static void enqueue_top_rt_rq(struct rt_rq *rt_rq); |
410 | static void dequeue_top_rt_rq(struct rt_rq *rt_rq); | |
411 | ||
6f505b16 PZ |
412 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) |
413 | { | |
414 | return !list_empty(&rt_se->run_list); | |
415 | } | |
416 | ||
052f1dc7 | 417 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 418 | |
9f0c1e56 | 419 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
420 | { |
421 | if (!rt_rq->tg) | |
9f0c1e56 | 422 | return RUNTIME_INF; |
6f505b16 | 423 | |
ac086bc2 PZ |
424 | return rt_rq->rt_runtime; |
425 | } | |
426 | ||
427 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
428 | { | |
429 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); | |
6f505b16 PZ |
430 | } |
431 | ||
ec514c48 CX |
432 | typedef struct task_group *rt_rq_iter_t; |
433 | ||
1c09ab0d YZ |
434 | static inline struct task_group *next_task_group(struct task_group *tg) |
435 | { | |
436 | do { | |
437 | tg = list_entry_rcu(tg->list.next, | |
438 | typeof(struct task_group), list); | |
439 | } while (&tg->list != &task_groups && task_group_is_autogroup(tg)); | |
440 | ||
441 | if (&tg->list == &task_groups) | |
442 | tg = NULL; | |
443 | ||
444 | return tg; | |
445 | } | |
446 | ||
447 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
448 | for (iter = container_of(&task_groups, typeof(*iter), list); \ | |
449 | (iter = next_task_group(iter)) && \ | |
450 | (rt_rq = iter->rt_rq[cpu_of(rq)]);) | |
ec514c48 | 451 | |
6f505b16 PZ |
452 | #define for_each_sched_rt_entity(rt_se) \ |
453 | for (; rt_se; rt_se = rt_se->parent) | |
454 | ||
455 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
456 | { | |
457 | return rt_se->my_q; | |
458 | } | |
459 | ||
37dad3fc | 460 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head); |
6f505b16 PZ |
461 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); |
462 | ||
9f0c1e56 | 463 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 464 | { |
f6121f4f | 465 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
74b7eb58 YZ |
466 | struct sched_rt_entity *rt_se; |
467 | ||
0c3b9168 BS |
468 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
469 | ||
470 | rt_se = rt_rq->tg->rt_se[cpu]; | |
6f505b16 | 471 | |
f6121f4f | 472 | if (rt_rq->rt_nr_running) { |
f4ebcbc0 KT |
473 | if (!rt_se) |
474 | enqueue_top_rt_rq(rt_rq); | |
475 | else if (!on_rt_rq(rt_se)) | |
37dad3fc | 476 | enqueue_rt_entity(rt_se, false); |
f4ebcbc0 | 477 | |
e864c499 | 478 | if (rt_rq->highest_prio.curr < curr->prio) |
1020387f | 479 | resched_task(curr); |
6f505b16 PZ |
480 | } |
481 | } | |
482 | ||
9f0c1e56 | 483 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 484 | { |
74b7eb58 | 485 | struct sched_rt_entity *rt_se; |
0c3b9168 | 486 | int cpu = cpu_of(rq_of_rt_rq(rt_rq)); |
74b7eb58 | 487 | |
0c3b9168 | 488 | rt_se = rt_rq->tg->rt_se[cpu]; |
6f505b16 | 489 | |
f4ebcbc0 KT |
490 | if (!rt_se) |
491 | dequeue_top_rt_rq(rt_rq); | |
492 | else if (on_rt_rq(rt_se)) | |
6f505b16 PZ |
493 | dequeue_rt_entity(rt_se); |
494 | } | |
495 | ||
46383648 KT |
496 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
497 | { | |
498 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
499 | } | |
500 | ||
23b0fdfc PZ |
501 | static int rt_se_boosted(struct sched_rt_entity *rt_se) |
502 | { | |
503 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
504 | struct task_struct *p; | |
505 | ||
506 | if (rt_rq) | |
507 | return !!rt_rq->rt_nr_boosted; | |
508 | ||
509 | p = rt_task_of(rt_se); | |
510 | return p->prio != p->normal_prio; | |
511 | } | |
512 | ||
d0b27fa7 | 513 | #ifdef CONFIG_SMP |
c6c4927b | 514 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 515 | { |
424c93fe | 516 | return this_rq()->rd->span; |
d0b27fa7 | 517 | } |
6f505b16 | 518 | #else |
c6c4927b | 519 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 520 | { |
c6c4927b | 521 | return cpu_online_mask; |
d0b27fa7 PZ |
522 | } |
523 | #endif | |
6f505b16 | 524 | |
d0b27fa7 PZ |
525 | static inline |
526 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 527 | { |
d0b27fa7 PZ |
528 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
529 | } | |
9f0c1e56 | 530 | |
ac086bc2 PZ |
531 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
532 | { | |
533 | return &rt_rq->tg->rt_bandwidth; | |
534 | } | |
535 | ||
55e12e5e | 536 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
537 | |
538 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
539 | { | |
ac086bc2 PZ |
540 | return rt_rq->rt_runtime; |
541 | } | |
542 | ||
543 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
544 | { | |
545 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
546 | } |
547 | ||
ec514c48 CX |
548 | typedef struct rt_rq *rt_rq_iter_t; |
549 | ||
550 | #define for_each_rt_rq(rt_rq, iter, rq) \ | |
551 | for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
552 | ||
6f505b16 PZ |
553 | #define for_each_sched_rt_entity(rt_se) \ |
554 | for (; rt_se; rt_se = NULL) | |
555 | ||
556 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
557 | { | |
558 | return NULL; | |
559 | } | |
560 | ||
9f0c1e56 | 561 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 | 562 | { |
f4ebcbc0 KT |
563 | struct rq *rq = rq_of_rt_rq(rt_rq); |
564 | ||
565 | if (!rt_rq->rt_nr_running) | |
566 | return; | |
567 | ||
568 | enqueue_top_rt_rq(rt_rq); | |
569 | resched_task(rq->curr); | |
6f505b16 PZ |
570 | } |
571 | ||
9f0c1e56 | 572 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 | 573 | { |
f4ebcbc0 | 574 | dequeue_top_rt_rq(rt_rq); |
6f505b16 PZ |
575 | } |
576 | ||
46383648 KT |
577 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
578 | { | |
579 | return rt_rq->rt_throttled; | |
580 | } | |
581 | ||
c6c4927b | 582 | static inline const struct cpumask *sched_rt_period_mask(void) |
d0b27fa7 | 583 | { |
c6c4927b | 584 | return cpu_online_mask; |
d0b27fa7 PZ |
585 | } |
586 | ||
587 | static inline | |
588 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
589 | { | |
590 | return &cpu_rq(cpu)->rt; | |
591 | } | |
592 | ||
ac086bc2 PZ |
593 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
594 | { | |
595 | return &def_rt_bandwidth; | |
596 | } | |
597 | ||
55e12e5e | 598 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 599 | |
faa59937 JL |
600 | bool sched_rt_bandwidth_account(struct rt_rq *rt_rq) |
601 | { | |
602 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
603 | ||
604 | return (hrtimer_active(&rt_b->rt_period_timer) || | |
605 | rt_rq->rt_time < rt_b->rt_runtime); | |
606 | } | |
607 | ||
ac086bc2 | 608 | #ifdef CONFIG_SMP |
78333cdd PZ |
609 | /* |
610 | * We ran out of runtime, see if we can borrow some from our neighbours. | |
611 | */ | |
b79f3833 | 612 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
613 | { |
614 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
aa7f6730 | 615 | struct root_domain *rd = rq_of_rt_rq(rt_rq)->rd; |
ac086bc2 PZ |
616 | int i, weight, more = 0; |
617 | u64 rt_period; | |
618 | ||
c6c4927b | 619 | weight = cpumask_weight(rd->span); |
ac086bc2 | 620 | |
0986b11b | 621 | raw_spin_lock(&rt_b->rt_runtime_lock); |
ac086bc2 | 622 | rt_period = ktime_to_ns(rt_b->rt_period); |
c6c4927b | 623 | for_each_cpu(i, rd->span) { |
ac086bc2 PZ |
624 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
625 | s64 diff; | |
626 | ||
627 | if (iter == rt_rq) | |
628 | continue; | |
629 | ||
0986b11b | 630 | raw_spin_lock(&iter->rt_runtime_lock); |
78333cdd PZ |
631 | /* |
632 | * Either all rqs have inf runtime and there's nothing to steal | |
633 | * or __disable_runtime() below sets a specific rq to inf to | |
634 | * indicate its been disabled and disalow stealing. | |
635 | */ | |
7def2be1 PZ |
636 | if (iter->rt_runtime == RUNTIME_INF) |
637 | goto next; | |
638 | ||
78333cdd PZ |
639 | /* |
640 | * From runqueues with spare time, take 1/n part of their | |
641 | * spare time, but no more than our period. | |
642 | */ | |
ac086bc2 PZ |
643 | diff = iter->rt_runtime - iter->rt_time; |
644 | if (diff > 0) { | |
58838cf3 | 645 | diff = div_u64((u64)diff, weight); |
ac086bc2 PZ |
646 | if (rt_rq->rt_runtime + diff > rt_period) |
647 | diff = rt_period - rt_rq->rt_runtime; | |
648 | iter->rt_runtime -= diff; | |
649 | rt_rq->rt_runtime += diff; | |
650 | more = 1; | |
651 | if (rt_rq->rt_runtime == rt_period) { | |
0986b11b | 652 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 PZ |
653 | break; |
654 | } | |
655 | } | |
7def2be1 | 656 | next: |
0986b11b | 657 | raw_spin_unlock(&iter->rt_runtime_lock); |
ac086bc2 | 658 | } |
0986b11b | 659 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
ac086bc2 PZ |
660 | |
661 | return more; | |
662 | } | |
7def2be1 | 663 | |
78333cdd PZ |
664 | /* |
665 | * Ensure this RQ takes back all the runtime it lend to its neighbours. | |
666 | */ | |
7def2be1 PZ |
667 | static void __disable_runtime(struct rq *rq) |
668 | { | |
669 | struct root_domain *rd = rq->rd; | |
ec514c48 | 670 | rt_rq_iter_t iter; |
7def2be1 PZ |
671 | struct rt_rq *rt_rq; |
672 | ||
673 | if (unlikely(!scheduler_running)) | |
674 | return; | |
675 | ||
ec514c48 | 676 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
677 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
678 | s64 want; | |
679 | int i; | |
680 | ||
0986b11b TG |
681 | raw_spin_lock(&rt_b->rt_runtime_lock); |
682 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
78333cdd PZ |
683 | /* |
684 | * Either we're all inf and nobody needs to borrow, or we're | |
685 | * already disabled and thus have nothing to do, or we have | |
686 | * exactly the right amount of runtime to take out. | |
687 | */ | |
7def2be1 PZ |
688 | if (rt_rq->rt_runtime == RUNTIME_INF || |
689 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
690 | goto balanced; | |
0986b11b | 691 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
7def2be1 | 692 | |
78333cdd PZ |
693 | /* |
694 | * Calculate the difference between what we started out with | |
695 | * and what we current have, that's the amount of runtime | |
696 | * we lend and now have to reclaim. | |
697 | */ | |
7def2be1 PZ |
698 | want = rt_b->rt_runtime - rt_rq->rt_runtime; |
699 | ||
78333cdd PZ |
700 | /* |
701 | * Greedy reclaim, take back as much as we can. | |
702 | */ | |
c6c4927b | 703 | for_each_cpu(i, rd->span) { |
7def2be1 PZ |
704 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); |
705 | s64 diff; | |
706 | ||
78333cdd PZ |
707 | /* |
708 | * Can't reclaim from ourselves or disabled runqueues. | |
709 | */ | |
f1679d08 | 710 | if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF) |
7def2be1 PZ |
711 | continue; |
712 | ||
0986b11b | 713 | raw_spin_lock(&iter->rt_runtime_lock); |
7def2be1 PZ |
714 | if (want > 0) { |
715 | diff = min_t(s64, iter->rt_runtime, want); | |
716 | iter->rt_runtime -= diff; | |
717 | want -= diff; | |
718 | } else { | |
719 | iter->rt_runtime -= want; | |
720 | want -= want; | |
721 | } | |
0986b11b | 722 | raw_spin_unlock(&iter->rt_runtime_lock); |
7def2be1 PZ |
723 | |
724 | if (!want) | |
725 | break; | |
726 | } | |
727 | ||
0986b11b | 728 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
78333cdd PZ |
729 | /* |
730 | * We cannot be left wanting - that would mean some runtime | |
731 | * leaked out of the system. | |
732 | */ | |
7def2be1 PZ |
733 | BUG_ON(want); |
734 | balanced: | |
78333cdd PZ |
735 | /* |
736 | * Disable all the borrow logic by pretending we have inf | |
737 | * runtime - in which case borrowing doesn't make sense. | |
738 | */ | |
7def2be1 | 739 | rt_rq->rt_runtime = RUNTIME_INF; |
a4c96ae3 | 740 | rt_rq->rt_throttled = 0; |
0986b11b TG |
741 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
742 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
743 | } |
744 | } | |
745 | ||
7def2be1 PZ |
746 | static void __enable_runtime(struct rq *rq) |
747 | { | |
ec514c48 | 748 | rt_rq_iter_t iter; |
7def2be1 PZ |
749 | struct rt_rq *rt_rq; |
750 | ||
751 | if (unlikely(!scheduler_running)) | |
752 | return; | |
753 | ||
78333cdd PZ |
754 | /* |
755 | * Reset each runqueue's bandwidth settings | |
756 | */ | |
ec514c48 | 757 | for_each_rt_rq(rt_rq, iter, rq) { |
7def2be1 PZ |
758 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
759 | ||
0986b11b TG |
760 | raw_spin_lock(&rt_b->rt_runtime_lock); |
761 | raw_spin_lock(&rt_rq->rt_runtime_lock); | |
7def2be1 PZ |
762 | rt_rq->rt_runtime = rt_b->rt_runtime; |
763 | rt_rq->rt_time = 0; | |
baf25731 | 764 | rt_rq->rt_throttled = 0; |
0986b11b TG |
765 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
766 | raw_spin_unlock(&rt_b->rt_runtime_lock); | |
7def2be1 PZ |
767 | } |
768 | } | |
769 | ||
eff6549b PZ |
770 | static int balance_runtime(struct rt_rq *rt_rq) |
771 | { | |
772 | int more = 0; | |
773 | ||
4a6184ce PZ |
774 | if (!sched_feat(RT_RUNTIME_SHARE)) |
775 | return more; | |
776 | ||
eff6549b | 777 | if (rt_rq->rt_time > rt_rq->rt_runtime) { |
0986b11b | 778 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
eff6549b | 779 | more = do_balance_runtime(rt_rq); |
0986b11b | 780 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
781 | } |
782 | ||
783 | return more; | |
784 | } | |
55e12e5e | 785 | #else /* !CONFIG_SMP */ |
eff6549b PZ |
786 | static inline int balance_runtime(struct rt_rq *rt_rq) |
787 | { | |
788 | return 0; | |
789 | } | |
55e12e5e | 790 | #endif /* CONFIG_SMP */ |
ac086bc2 | 791 | |
eff6549b PZ |
792 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
793 | { | |
42c62a58 | 794 | int i, idle = 1, throttled = 0; |
c6c4927b | 795 | const struct cpumask *span; |
eff6549b | 796 | |
eff6549b | 797 | span = sched_rt_period_mask(); |
e221d028 MG |
798 | #ifdef CONFIG_RT_GROUP_SCHED |
799 | /* | |
800 | * FIXME: isolated CPUs should really leave the root task group, | |
801 | * whether they are isolcpus or were isolated via cpusets, lest | |
802 | * the timer run on a CPU which does not service all runqueues, | |
803 | * potentially leaving other CPUs indefinitely throttled. If | |
804 | * isolation is really required, the user will turn the throttle | |
805 | * off to kill the perturbations it causes anyway. Meanwhile, | |
806 | * this maintains functionality for boot and/or troubleshooting. | |
807 | */ | |
808 | if (rt_b == &root_task_group.rt_bandwidth) | |
809 | span = cpu_online_mask; | |
810 | #endif | |
c6c4927b | 811 | for_each_cpu(i, span) { |
eff6549b PZ |
812 | int enqueue = 0; |
813 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
814 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
815 | ||
05fa785c | 816 | raw_spin_lock(&rq->lock); |
eff6549b PZ |
817 | if (rt_rq->rt_time) { |
818 | u64 runtime; | |
819 | ||
0986b11b | 820 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
eff6549b PZ |
821 | if (rt_rq->rt_throttled) |
822 | balance_runtime(rt_rq); | |
823 | runtime = rt_rq->rt_runtime; | |
824 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
825 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
826 | rt_rq->rt_throttled = 0; | |
827 | enqueue = 1; | |
61eadef6 MG |
828 | |
829 | /* | |
830 | * Force a clock update if the CPU was idle, | |
831 | * lest wakeup -> unthrottle time accumulate. | |
832 | */ | |
833 | if (rt_rq->rt_nr_running && rq->curr == rq->idle) | |
834 | rq->skip_clock_update = -1; | |
eff6549b PZ |
835 | } |
836 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
837 | idle = 0; | |
0986b11b | 838 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
0c3b9168 | 839 | } else if (rt_rq->rt_nr_running) { |
6c3df255 | 840 | idle = 0; |
0c3b9168 BS |
841 | if (!rt_rq_throttled(rt_rq)) |
842 | enqueue = 1; | |
843 | } | |
42c62a58 PZ |
844 | if (rt_rq->rt_throttled) |
845 | throttled = 1; | |
eff6549b PZ |
846 | |
847 | if (enqueue) | |
848 | sched_rt_rq_enqueue(rt_rq); | |
05fa785c | 849 | raw_spin_unlock(&rq->lock); |
eff6549b PZ |
850 | } |
851 | ||
42c62a58 PZ |
852 | if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)) |
853 | return 1; | |
854 | ||
eff6549b PZ |
855 | return idle; |
856 | } | |
ac086bc2 | 857 | |
6f505b16 PZ |
858 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
859 | { | |
052f1dc7 | 860 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
861 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
862 | ||
863 | if (rt_rq) | |
e864c499 | 864 | return rt_rq->highest_prio.curr; |
6f505b16 PZ |
865 | #endif |
866 | ||
867 | return rt_task_of(rt_se)->prio; | |
868 | } | |
869 | ||
9f0c1e56 | 870 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 871 | { |
9f0c1e56 | 872 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 873 | |
fa85ae24 | 874 | if (rt_rq->rt_throttled) |
23b0fdfc | 875 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 876 | |
5b680fd6 | 877 | if (runtime >= sched_rt_period(rt_rq)) |
ac086bc2 PZ |
878 | return 0; |
879 | ||
b79f3833 PZ |
880 | balance_runtime(rt_rq); |
881 | runtime = sched_rt_runtime(rt_rq); | |
882 | if (runtime == RUNTIME_INF) | |
883 | return 0; | |
ac086bc2 | 884 | |
9f0c1e56 | 885 | if (rt_rq->rt_time > runtime) { |
7abc63b1 PZ |
886 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); |
887 | ||
888 | /* | |
889 | * Don't actually throttle groups that have no runtime assigned | |
890 | * but accrue some time due to boosting. | |
891 | */ | |
892 | if (likely(rt_b->rt_runtime)) { | |
893 | rt_rq->rt_throttled = 1; | |
c224815d | 894 | printk_deferred_once("sched: RT throttling activated\n"); |
7abc63b1 PZ |
895 | } else { |
896 | /* | |
897 | * In case we did anyway, make it go away, | |
898 | * replenishment is a joke, since it will replenish us | |
899 | * with exactly 0 ns. | |
900 | */ | |
901 | rt_rq->rt_time = 0; | |
902 | } | |
903 | ||
23b0fdfc | 904 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 905 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
906 | return 1; |
907 | } | |
fa85ae24 PZ |
908 | } |
909 | ||
910 | return 0; | |
911 | } | |
912 | ||
bb44e5d1 IM |
913 | /* |
914 | * Update the current task's runtime statistics. Skip current tasks that | |
915 | * are not in our scheduling class. | |
916 | */ | |
a9957449 | 917 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
918 | { |
919 | struct task_struct *curr = rq->curr; | |
6f505b16 | 920 | struct sched_rt_entity *rt_se = &curr->rt; |
bb44e5d1 IM |
921 | u64 delta_exec; |
922 | ||
06c3bc65 | 923 | if (curr->sched_class != &rt_sched_class) |
bb44e5d1 IM |
924 | return; |
925 | ||
78becc27 | 926 | delta_exec = rq_clock_task(rq) - curr->se.exec_start; |
fc79e240 KT |
927 | if (unlikely((s64)delta_exec <= 0)) |
928 | return; | |
6cfb0d5d | 929 | |
42c62a58 PZ |
930 | schedstat_set(curr->se.statistics.exec_max, |
931 | max(curr->se.statistics.exec_max, delta_exec)); | |
bb44e5d1 IM |
932 | |
933 | curr->se.sum_exec_runtime += delta_exec; | |
f06febc9 FM |
934 | account_group_exec_runtime(curr, delta_exec); |
935 | ||
78becc27 | 936 | curr->se.exec_start = rq_clock_task(rq); |
d842de87 | 937 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 938 | |
e9e9250b PZ |
939 | sched_rt_avg_update(rq, delta_exec); |
940 | ||
0b148fa0 PZ |
941 | if (!rt_bandwidth_enabled()) |
942 | return; | |
943 | ||
354d60c2 | 944 | for_each_sched_rt_entity(rt_se) { |
0b07939c | 945 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
354d60c2 | 946 | |
cc2991cf | 947 | if (sched_rt_runtime(rt_rq) != RUNTIME_INF) { |
0986b11b | 948 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
cc2991cf PZ |
949 | rt_rq->rt_time += delta_exec; |
950 | if (sched_rt_runtime_exceeded(rt_rq)) | |
951 | resched_task(curr); | |
0986b11b | 952 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
cc2991cf | 953 | } |
354d60c2 | 954 | } |
bb44e5d1 IM |
955 | } |
956 | ||
f4ebcbc0 KT |
957 | static void |
958 | dequeue_top_rt_rq(struct rt_rq *rt_rq) | |
959 | { | |
960 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
961 | ||
962 | BUG_ON(&rq->rt != rt_rq); | |
963 | ||
964 | if (!rt_rq->rt_queued) | |
965 | return; | |
966 | ||
967 | BUG_ON(!rq->nr_running); | |
968 | ||
72465447 | 969 | sub_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
970 | rt_rq->rt_queued = 0; |
971 | } | |
972 | ||
973 | static void | |
974 | enqueue_top_rt_rq(struct rt_rq *rt_rq) | |
975 | { | |
976 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
977 | ||
978 | BUG_ON(&rq->rt != rt_rq); | |
979 | ||
980 | if (rt_rq->rt_queued) | |
981 | return; | |
982 | if (rt_rq_throttled(rt_rq) || !rt_rq->rt_nr_running) | |
983 | return; | |
984 | ||
72465447 | 985 | add_nr_running(rq, rt_rq->rt_nr_running); |
f4ebcbc0 KT |
986 | rt_rq->rt_queued = 1; |
987 | } | |
988 | ||
398a153b | 989 | #if defined CONFIG_SMP |
e864c499 | 990 | |
398a153b GH |
991 | static void |
992 | inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
63489e45 | 993 | { |
4d984277 | 994 | struct rq *rq = rq_of_rt_rq(rt_rq); |
1f11eb6a | 995 | |
757dfcaa KT |
996 | #ifdef CONFIG_RT_GROUP_SCHED |
997 | /* | |
998 | * Change rq's cpupri only if rt_rq is the top queue. | |
999 | */ | |
1000 | if (&rq->rt != rt_rq) | |
1001 | return; | |
1002 | #endif | |
5181f4a4 SR |
1003 | if (rq->online && prio < prev_prio) |
1004 | cpupri_set(&rq->rd->cpupri, rq->cpu, prio); | |
398a153b | 1005 | } |
73fe6aae | 1006 | |
398a153b GH |
1007 | static void |
1008 | dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) | |
1009 | { | |
1010 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
d0b27fa7 | 1011 | |
757dfcaa KT |
1012 | #ifdef CONFIG_RT_GROUP_SCHED |
1013 | /* | |
1014 | * Change rq's cpupri only if rt_rq is the top queue. | |
1015 | */ | |
1016 | if (&rq->rt != rt_rq) | |
1017 | return; | |
1018 | #endif | |
398a153b GH |
1019 | if (rq->online && rt_rq->highest_prio.curr != prev_prio) |
1020 | cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr); | |
63489e45 SR |
1021 | } |
1022 | ||
398a153b GH |
1023 | #else /* CONFIG_SMP */ |
1024 | ||
6f505b16 | 1025 | static inline |
398a153b GH |
1026 | void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} |
1027 | static inline | |
1028 | void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {} | |
1029 | ||
1030 | #endif /* CONFIG_SMP */ | |
6e0534f2 | 1031 | |
052f1dc7 | 1032 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
398a153b GH |
1033 | static void |
1034 | inc_rt_prio(struct rt_rq *rt_rq, int prio) | |
1035 | { | |
1036 | int prev_prio = rt_rq->highest_prio.curr; | |
1037 | ||
1038 | if (prio < prev_prio) | |
1039 | rt_rq->highest_prio.curr = prio; | |
1040 | ||
1041 | inc_rt_prio_smp(rt_rq, prio, prev_prio); | |
1042 | } | |
1043 | ||
1044 | static void | |
1045 | dec_rt_prio(struct rt_rq *rt_rq, int prio) | |
1046 | { | |
1047 | int prev_prio = rt_rq->highest_prio.curr; | |
1048 | ||
6f505b16 | 1049 | if (rt_rq->rt_nr_running) { |
764a9d6f | 1050 | |
398a153b | 1051 | WARN_ON(prio < prev_prio); |
764a9d6f | 1052 | |
e864c499 | 1053 | /* |
398a153b GH |
1054 | * This may have been our highest task, and therefore |
1055 | * we may have some recomputation to do | |
e864c499 | 1056 | */ |
398a153b | 1057 | if (prio == prev_prio) { |
e864c499 GH |
1058 | struct rt_prio_array *array = &rt_rq->active; |
1059 | ||
1060 | rt_rq->highest_prio.curr = | |
764a9d6f | 1061 | sched_find_first_bit(array->bitmap); |
e864c499 GH |
1062 | } |
1063 | ||
764a9d6f | 1064 | } else |
e864c499 | 1065 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
73fe6aae | 1066 | |
398a153b GH |
1067 | dec_rt_prio_smp(rt_rq, prio, prev_prio); |
1068 | } | |
1f11eb6a | 1069 | |
398a153b GH |
1070 | #else |
1071 | ||
1072 | static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1073 | static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {} | |
1074 | ||
1075 | #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */ | |
6e0534f2 | 1076 | |
052f1dc7 | 1077 | #ifdef CONFIG_RT_GROUP_SCHED |
398a153b GH |
1078 | |
1079 | static void | |
1080 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1081 | { | |
1082 | if (rt_se_boosted(rt_se)) | |
1083 | rt_rq->rt_nr_boosted++; | |
1084 | ||
1085 | if (rt_rq->tg) | |
1086 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
1087 | } | |
1088 | ||
1089 | static void | |
1090 | dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1091 | { | |
23b0fdfc PZ |
1092 | if (rt_se_boosted(rt_se)) |
1093 | rt_rq->rt_nr_boosted--; | |
1094 | ||
1095 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
398a153b GH |
1096 | } |
1097 | ||
1098 | #else /* CONFIG_RT_GROUP_SCHED */ | |
1099 | ||
1100 | static void | |
1101 | inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1102 | { | |
1103 | start_rt_bandwidth(&def_rt_bandwidth); | |
1104 | } | |
1105 | ||
1106 | static inline | |
1107 | void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {} | |
1108 | ||
1109 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
1110 | ||
22abdef3 KT |
1111 | static inline |
1112 | unsigned int rt_se_nr_running(struct sched_rt_entity *rt_se) | |
1113 | { | |
1114 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
1115 | ||
1116 | if (group_rq) | |
1117 | return group_rq->rt_nr_running; | |
1118 | else | |
1119 | return 1; | |
1120 | } | |
1121 | ||
398a153b GH |
1122 | static inline |
1123 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1124 | { | |
1125 | int prio = rt_se_prio(rt_se); | |
1126 | ||
1127 | WARN_ON(!rt_prio(prio)); | |
22abdef3 | 1128 | rt_rq->rt_nr_running += rt_se_nr_running(rt_se); |
398a153b GH |
1129 | |
1130 | inc_rt_prio(rt_rq, prio); | |
1131 | inc_rt_migration(rt_se, rt_rq); | |
1132 | inc_rt_group(rt_se, rt_rq); | |
1133 | } | |
1134 | ||
1135 | static inline | |
1136 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
1137 | { | |
1138 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); | |
1139 | WARN_ON(!rt_rq->rt_nr_running); | |
22abdef3 | 1140 | rt_rq->rt_nr_running -= rt_se_nr_running(rt_se); |
398a153b GH |
1141 | |
1142 | dec_rt_prio(rt_rq, rt_se_prio(rt_se)); | |
1143 | dec_rt_migration(rt_se, rt_rq); | |
1144 | dec_rt_group(rt_se, rt_rq); | |
63489e45 SR |
1145 | } |
1146 | ||
37dad3fc | 1147 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
bb44e5d1 | 1148 | { |
6f505b16 PZ |
1149 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
1150 | struct rt_prio_array *array = &rt_rq->active; | |
1151 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 1152 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 1153 | |
ad2a3f13 PZ |
1154 | /* |
1155 | * Don't enqueue the group if its throttled, or when empty. | |
1156 | * The latter is a consequence of the former when a child group | |
1157 | * get throttled and the current group doesn't have any other | |
1158 | * active members. | |
1159 | */ | |
1160 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 1161 | return; |
63489e45 | 1162 | |
37dad3fc TG |
1163 | if (head) |
1164 | list_add(&rt_se->run_list, queue); | |
1165 | else | |
1166 | list_add_tail(&rt_se->run_list, queue); | |
6f505b16 | 1167 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 1168 | |
6f505b16 PZ |
1169 | inc_rt_tasks(rt_se, rt_rq); |
1170 | } | |
1171 | ||
ad2a3f13 | 1172 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
1173 | { |
1174 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
1175 | struct rt_prio_array *array = &rt_rq->active; | |
1176 | ||
1177 | list_del_init(&rt_se->run_list); | |
1178 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
1179 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
1180 | ||
1181 | dec_rt_tasks(rt_se, rt_rq); | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * Because the prio of an upper entry depends on the lower | |
1186 | * entries, we must remove entries top - down. | |
6f505b16 | 1187 | */ |
ad2a3f13 | 1188 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 1189 | { |
ad2a3f13 | 1190 | struct sched_rt_entity *back = NULL; |
6f505b16 | 1191 | |
58d6c2d7 PZ |
1192 | for_each_sched_rt_entity(rt_se) { |
1193 | rt_se->back = back; | |
1194 | back = rt_se; | |
1195 | } | |
1196 | ||
f4ebcbc0 KT |
1197 | dequeue_top_rt_rq(rt_rq_of_se(back)); |
1198 | ||
58d6c2d7 PZ |
1199 | for (rt_se = back; rt_se; rt_se = rt_se->back) { |
1200 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
1201 | __dequeue_rt_entity(rt_se); |
1202 | } | |
1203 | } | |
1204 | ||
37dad3fc | 1205 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se, bool head) |
ad2a3f13 | 1206 | { |
f4ebcbc0 KT |
1207 | struct rq *rq = rq_of_rt_se(rt_se); |
1208 | ||
ad2a3f13 PZ |
1209 | dequeue_rt_stack(rt_se); |
1210 | for_each_sched_rt_entity(rt_se) | |
37dad3fc | 1211 | __enqueue_rt_entity(rt_se, head); |
f4ebcbc0 | 1212 | enqueue_top_rt_rq(&rq->rt); |
ad2a3f13 PZ |
1213 | } |
1214 | ||
1215 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
1216 | { | |
f4ebcbc0 KT |
1217 | struct rq *rq = rq_of_rt_se(rt_se); |
1218 | ||
ad2a3f13 PZ |
1219 | dequeue_rt_stack(rt_se); |
1220 | ||
1221 | for_each_sched_rt_entity(rt_se) { | |
1222 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
1223 | ||
1224 | if (rt_rq && rt_rq->rt_nr_running) | |
37dad3fc | 1225 | __enqueue_rt_entity(rt_se, false); |
58d6c2d7 | 1226 | } |
f4ebcbc0 | 1227 | enqueue_top_rt_rq(&rq->rt); |
bb44e5d1 IM |
1228 | } |
1229 | ||
1230 | /* | |
1231 | * Adding/removing a task to/from a priority array: | |
1232 | */ | |
ea87bb78 | 1233 | static void |
371fd7e7 | 1234 | enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
6f505b16 PZ |
1235 | { |
1236 | struct sched_rt_entity *rt_se = &p->rt; | |
1237 | ||
371fd7e7 | 1238 | if (flags & ENQUEUE_WAKEUP) |
6f505b16 PZ |
1239 | rt_se->timeout = 0; |
1240 | ||
371fd7e7 | 1241 | enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD); |
c09595f6 | 1242 | |
29baa747 | 1243 | if (!task_current(rq, p) && p->nr_cpus_allowed > 1) |
917b627d | 1244 | enqueue_pushable_task(rq, p); |
6f505b16 PZ |
1245 | } |
1246 | ||
371fd7e7 | 1247 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1248 | { |
6f505b16 | 1249 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 1250 | |
f1e14ef6 | 1251 | update_curr_rt(rq); |
ad2a3f13 | 1252 | dequeue_rt_entity(rt_se); |
c09595f6 | 1253 | |
917b627d | 1254 | dequeue_pushable_task(rq, p); |
bb44e5d1 IM |
1255 | } |
1256 | ||
1257 | /* | |
60686317 RW |
1258 | * Put task to the head or the end of the run list without the overhead of |
1259 | * dequeue followed by enqueue. | |
bb44e5d1 | 1260 | */ |
7ebefa8c DA |
1261 | static void |
1262 | requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head) | |
6f505b16 | 1263 | { |
1cdad715 | 1264 | if (on_rt_rq(rt_se)) { |
7ebefa8c DA |
1265 | struct rt_prio_array *array = &rt_rq->active; |
1266 | struct list_head *queue = array->queue + rt_se_prio(rt_se); | |
1267 | ||
1268 | if (head) | |
1269 | list_move(&rt_se->run_list, queue); | |
1270 | else | |
1271 | list_move_tail(&rt_se->run_list, queue); | |
1cdad715 | 1272 | } |
6f505b16 PZ |
1273 | } |
1274 | ||
7ebefa8c | 1275 | static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head) |
bb44e5d1 | 1276 | { |
6f505b16 PZ |
1277 | struct sched_rt_entity *rt_se = &p->rt; |
1278 | struct rt_rq *rt_rq; | |
bb44e5d1 | 1279 | |
6f505b16 PZ |
1280 | for_each_sched_rt_entity(rt_se) { |
1281 | rt_rq = rt_rq_of_se(rt_se); | |
7ebefa8c | 1282 | requeue_rt_entity(rt_rq, rt_se, head); |
6f505b16 | 1283 | } |
bb44e5d1 IM |
1284 | } |
1285 | ||
6f505b16 | 1286 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 1287 | { |
7ebefa8c | 1288 | requeue_task_rt(rq, rq->curr, 0); |
bb44e5d1 IM |
1289 | } |
1290 | ||
e7693a36 | 1291 | #ifdef CONFIG_SMP |
318e0893 GH |
1292 | static int find_lowest_rq(struct task_struct *task); |
1293 | ||
0017d735 | 1294 | static int |
ac66f547 | 1295 | select_task_rq_rt(struct task_struct *p, int cpu, int sd_flag, int flags) |
e7693a36 | 1296 | { |
7608dec2 PZ |
1297 | struct task_struct *curr; |
1298 | struct rq *rq; | |
c37495fd | 1299 | |
29baa747 | 1300 | if (p->nr_cpus_allowed == 1) |
76854c7e MG |
1301 | goto out; |
1302 | ||
c37495fd SR |
1303 | /* For anything but wake ups, just return the task_cpu */ |
1304 | if (sd_flag != SD_BALANCE_WAKE && sd_flag != SD_BALANCE_FORK) | |
1305 | goto out; | |
1306 | ||
7608dec2 PZ |
1307 | rq = cpu_rq(cpu); |
1308 | ||
1309 | rcu_read_lock(); | |
1310 | curr = ACCESS_ONCE(rq->curr); /* unlocked access */ | |
1311 | ||
318e0893 | 1312 | /* |
7608dec2 | 1313 | * If the current task on @p's runqueue is an RT task, then |
e1f47d89 SR |
1314 | * try to see if we can wake this RT task up on another |
1315 | * runqueue. Otherwise simply start this RT task | |
1316 | * on its current runqueue. | |
1317 | * | |
43fa5460 SR |
1318 | * We want to avoid overloading runqueues. If the woken |
1319 | * task is a higher priority, then it will stay on this CPU | |
1320 | * and the lower prio task should be moved to another CPU. | |
1321 | * Even though this will probably make the lower prio task | |
1322 | * lose its cache, we do not want to bounce a higher task | |
1323 | * around just because it gave up its CPU, perhaps for a | |
1324 | * lock? | |
1325 | * | |
1326 | * For equal prio tasks, we just let the scheduler sort it out. | |
7608dec2 PZ |
1327 | * |
1328 | * Otherwise, just let it ride on the affined RQ and the | |
1329 | * post-schedule router will push the preempted task away | |
1330 | * | |
1331 | * This test is optimistic, if we get it wrong the load-balancer | |
1332 | * will have to sort it out. | |
318e0893 | 1333 | */ |
7608dec2 | 1334 | if (curr && unlikely(rt_task(curr)) && |
29baa747 | 1335 | (curr->nr_cpus_allowed < 2 || |
6bfa687c | 1336 | curr->prio <= p->prio)) { |
7608dec2 | 1337 | int target = find_lowest_rq(p); |
318e0893 | 1338 | |
7608dec2 PZ |
1339 | if (target != -1) |
1340 | cpu = target; | |
318e0893 | 1341 | } |
7608dec2 | 1342 | rcu_read_unlock(); |
318e0893 | 1343 | |
c37495fd | 1344 | out: |
7608dec2 | 1345 | return cpu; |
e7693a36 | 1346 | } |
7ebefa8c DA |
1347 | |
1348 | static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p) | |
1349 | { | |
29baa747 | 1350 | if (rq->curr->nr_cpus_allowed == 1) |
7ebefa8c DA |
1351 | return; |
1352 | ||
29baa747 | 1353 | if (p->nr_cpus_allowed != 1 |
13b8bd0a RR |
1354 | && cpupri_find(&rq->rd->cpupri, p, NULL)) |
1355 | return; | |
24600ce8 | 1356 | |
13b8bd0a RR |
1357 | if (!cpupri_find(&rq->rd->cpupri, rq->curr, NULL)) |
1358 | return; | |
7ebefa8c DA |
1359 | |
1360 | /* | |
1361 | * There appears to be other cpus that can accept | |
1362 | * current and none to run 'p', so lets reschedule | |
1363 | * to try and push current away: | |
1364 | */ | |
1365 | requeue_task_rt(rq, p, 1); | |
1366 | resched_task(rq->curr); | |
1367 | } | |
1368 | ||
e7693a36 GH |
1369 | #endif /* CONFIG_SMP */ |
1370 | ||
bb44e5d1 IM |
1371 | /* |
1372 | * Preempt the current task with a newly woken task if needed: | |
1373 | */ | |
7d478721 | 1374 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p, int flags) |
bb44e5d1 | 1375 | { |
45c01e82 | 1376 | if (p->prio < rq->curr->prio) { |
bb44e5d1 | 1377 | resched_task(rq->curr); |
45c01e82 GH |
1378 | return; |
1379 | } | |
1380 | ||
1381 | #ifdef CONFIG_SMP | |
1382 | /* | |
1383 | * If: | |
1384 | * | |
1385 | * - the newly woken task is of equal priority to the current task | |
1386 | * - the newly woken task is non-migratable while current is migratable | |
1387 | * - current will be preempted on the next reschedule | |
1388 | * | |
1389 | * we should check to see if current can readily move to a different | |
1390 | * cpu. If so, we will reschedule to allow the push logic to try | |
1391 | * to move current somewhere else, making room for our non-migratable | |
1392 | * task. | |
1393 | */ | |
8dd0de8b | 1394 | if (p->prio == rq->curr->prio && !test_tsk_need_resched(rq->curr)) |
7ebefa8c | 1395 | check_preempt_equal_prio(rq, p); |
45c01e82 | 1396 | #endif |
bb44e5d1 IM |
1397 | } |
1398 | ||
6f505b16 PZ |
1399 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
1400 | struct rt_rq *rt_rq) | |
bb44e5d1 | 1401 | { |
6f505b16 PZ |
1402 | struct rt_prio_array *array = &rt_rq->active; |
1403 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
1404 | struct list_head *queue; |
1405 | int idx; | |
1406 | ||
1407 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 1408 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
1409 | |
1410 | queue = array->queue + idx; | |
6f505b16 | 1411 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 1412 | |
6f505b16 PZ |
1413 | return next; |
1414 | } | |
bb44e5d1 | 1415 | |
917b627d | 1416 | static struct task_struct *_pick_next_task_rt(struct rq *rq) |
6f505b16 PZ |
1417 | { |
1418 | struct sched_rt_entity *rt_se; | |
1419 | struct task_struct *p; | |
606dba2e | 1420 | struct rt_rq *rt_rq = &rq->rt; |
6f505b16 PZ |
1421 | |
1422 | do { | |
1423 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 1424 | BUG_ON(!rt_se); |
6f505b16 PZ |
1425 | rt_rq = group_rt_rq(rt_se); |
1426 | } while (rt_rq); | |
1427 | ||
1428 | p = rt_task_of(rt_se); | |
78becc27 | 1429 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
1430 | |
1431 | return p; | |
1432 | } | |
1433 | ||
606dba2e PZ |
1434 | static struct task_struct * |
1435 | pick_next_task_rt(struct rq *rq, struct task_struct *prev) | |
917b627d | 1436 | { |
606dba2e PZ |
1437 | struct task_struct *p; |
1438 | struct rt_rq *rt_rq = &rq->rt; | |
1439 | ||
37e117c0 | 1440 | if (need_pull_rt_task(rq, prev)) { |
38033c37 | 1441 | pull_rt_task(rq); |
37e117c0 PZ |
1442 | /* |
1443 | * pull_rt_task() can drop (and re-acquire) rq->lock; this | |
a1d9a323 KT |
1444 | * means a dl or stop task can slip in, in which case we need |
1445 | * to re-start task selection. | |
37e117c0 | 1446 | */ |
a1d9a323 KT |
1447 | if (unlikely((rq->stop && rq->stop->on_rq) || |
1448 | rq->dl.dl_nr_running)) | |
37e117c0 PZ |
1449 | return RETRY_TASK; |
1450 | } | |
38033c37 | 1451 | |
734ff2a7 KT |
1452 | /* |
1453 | * We may dequeue prev's rt_rq in put_prev_task(). | |
1454 | * So, we update time before rt_nr_running check. | |
1455 | */ | |
1456 | if (prev->sched_class == &rt_sched_class) | |
1457 | update_curr_rt(rq); | |
1458 | ||
f4ebcbc0 | 1459 | if (!rt_rq->rt_queued) |
606dba2e PZ |
1460 | return NULL; |
1461 | ||
3f1d2a31 | 1462 | put_prev_task(rq, prev); |
606dba2e PZ |
1463 | |
1464 | p = _pick_next_task_rt(rq); | |
917b627d GH |
1465 | |
1466 | /* The running task is never eligible for pushing */ | |
1467 | if (p) | |
1468 | dequeue_pushable_task(rq, p); | |
1469 | ||
dc877341 | 1470 | set_post_schedule(rq); |
3f029d3c | 1471 | |
6f505b16 | 1472 | return p; |
bb44e5d1 IM |
1473 | } |
1474 | ||
31ee529c | 1475 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 1476 | { |
f1e14ef6 | 1477 | update_curr_rt(rq); |
917b627d GH |
1478 | |
1479 | /* | |
1480 | * The previous task needs to be made eligible for pushing | |
1481 | * if it is still active | |
1482 | */ | |
29baa747 | 1483 | if (on_rt_rq(&p->rt) && p->nr_cpus_allowed > 1) |
917b627d | 1484 | enqueue_pushable_task(rq, p); |
bb44e5d1 IM |
1485 | } |
1486 | ||
681f3e68 | 1487 | #ifdef CONFIG_SMP |
6f505b16 | 1488 | |
e8fa1362 SR |
1489 | /* Only try algorithms three times */ |
1490 | #define RT_MAX_TRIES 3 | |
1491 | ||
f65eda4f SR |
1492 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
1493 | { | |
1494 | if (!task_running(rq, p) && | |
60334caf | 1495 | cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) |
f65eda4f SR |
1496 | return 1; |
1497 | return 0; | |
1498 | } | |
1499 | ||
e23ee747 KT |
1500 | /* |
1501 | * Return the highest pushable rq's task, which is suitable to be executed | |
1502 | * on the cpu, NULL otherwise | |
1503 | */ | |
1504 | static struct task_struct *pick_highest_pushable_task(struct rq *rq, int cpu) | |
e8fa1362 | 1505 | { |
e23ee747 KT |
1506 | struct plist_head *head = &rq->rt.pushable_tasks; |
1507 | struct task_struct *p; | |
3d07467b | 1508 | |
e23ee747 KT |
1509 | if (!has_pushable_tasks(rq)) |
1510 | return NULL; | |
3d07467b | 1511 | |
e23ee747 KT |
1512 | plist_for_each_entry(p, head, pushable_tasks) { |
1513 | if (pick_rt_task(rq, p, cpu)) | |
1514 | return p; | |
f65eda4f SR |
1515 | } |
1516 | ||
e23ee747 | 1517 | return NULL; |
e8fa1362 SR |
1518 | } |
1519 | ||
0e3900e6 | 1520 | static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask); |
e8fa1362 | 1521 | |
6e1254d2 GH |
1522 | static int find_lowest_rq(struct task_struct *task) |
1523 | { | |
1524 | struct sched_domain *sd; | |
96f874e2 | 1525 | struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask); |
6e1254d2 GH |
1526 | int this_cpu = smp_processor_id(); |
1527 | int cpu = task_cpu(task); | |
06f90dbd | 1528 | |
0da938c4 SR |
1529 | /* Make sure the mask is initialized first */ |
1530 | if (unlikely(!lowest_mask)) | |
1531 | return -1; | |
1532 | ||
29baa747 | 1533 | if (task->nr_cpus_allowed == 1) |
6e0534f2 | 1534 | return -1; /* No other targets possible */ |
6e1254d2 | 1535 | |
6e0534f2 GH |
1536 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
1537 | return -1; /* No targets found */ | |
6e1254d2 GH |
1538 | |
1539 | /* | |
1540 | * At this point we have built a mask of cpus representing the | |
1541 | * lowest priority tasks in the system. Now we want to elect | |
1542 | * the best one based on our affinity and topology. | |
1543 | * | |
1544 | * We prioritize the last cpu that the task executed on since | |
1545 | * it is most likely cache-hot in that location. | |
1546 | */ | |
96f874e2 | 1547 | if (cpumask_test_cpu(cpu, lowest_mask)) |
6e1254d2 GH |
1548 | return cpu; |
1549 | ||
1550 | /* | |
1551 | * Otherwise, we consult the sched_domains span maps to figure | |
1552 | * out which cpu is logically closest to our hot cache data. | |
1553 | */ | |
e2c88063 RR |
1554 | if (!cpumask_test_cpu(this_cpu, lowest_mask)) |
1555 | this_cpu = -1; /* Skip this_cpu opt if not among lowest */ | |
6e1254d2 | 1556 | |
cd4ae6ad | 1557 | rcu_read_lock(); |
e2c88063 RR |
1558 | for_each_domain(cpu, sd) { |
1559 | if (sd->flags & SD_WAKE_AFFINE) { | |
1560 | int best_cpu; | |
6e1254d2 | 1561 | |
e2c88063 RR |
1562 | /* |
1563 | * "this_cpu" is cheaper to preempt than a | |
1564 | * remote processor. | |
1565 | */ | |
1566 | if (this_cpu != -1 && | |
cd4ae6ad XF |
1567 | cpumask_test_cpu(this_cpu, sched_domain_span(sd))) { |
1568 | rcu_read_unlock(); | |
e2c88063 | 1569 | return this_cpu; |
cd4ae6ad | 1570 | } |
e2c88063 RR |
1571 | |
1572 | best_cpu = cpumask_first_and(lowest_mask, | |
1573 | sched_domain_span(sd)); | |
cd4ae6ad XF |
1574 | if (best_cpu < nr_cpu_ids) { |
1575 | rcu_read_unlock(); | |
e2c88063 | 1576 | return best_cpu; |
cd4ae6ad | 1577 | } |
6e1254d2 GH |
1578 | } |
1579 | } | |
cd4ae6ad | 1580 | rcu_read_unlock(); |
6e1254d2 GH |
1581 | |
1582 | /* | |
1583 | * And finally, if there were no matches within the domains | |
1584 | * just give the caller *something* to work with from the compatible | |
1585 | * locations. | |
1586 | */ | |
e2c88063 RR |
1587 | if (this_cpu != -1) |
1588 | return this_cpu; | |
1589 | ||
1590 | cpu = cpumask_any(lowest_mask); | |
1591 | if (cpu < nr_cpu_ids) | |
1592 | return cpu; | |
1593 | return -1; | |
07b4032c GH |
1594 | } |
1595 | ||
1596 | /* Will lock the rq it finds */ | |
4df64c0b | 1597 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
1598 | { |
1599 | struct rq *lowest_rq = NULL; | |
07b4032c | 1600 | int tries; |
4df64c0b | 1601 | int cpu; |
e8fa1362 | 1602 | |
07b4032c GH |
1603 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
1604 | cpu = find_lowest_rq(task); | |
1605 | ||
2de0b463 | 1606 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
1607 | break; |
1608 | ||
07b4032c GH |
1609 | lowest_rq = cpu_rq(cpu); |
1610 | ||
e8fa1362 | 1611 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 1612 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
1613 | /* |
1614 | * We had to unlock the run queue. In | |
1615 | * the mean time, task could have | |
1616 | * migrated already or had its affinity changed. | |
1617 | * Also make sure that it wasn't scheduled on its rq. | |
1618 | */ | |
07b4032c | 1619 | if (unlikely(task_rq(task) != rq || |
96f874e2 | 1620 | !cpumask_test_cpu(lowest_rq->cpu, |
fa17b507 | 1621 | tsk_cpus_allowed(task)) || |
07b4032c | 1622 | task_running(rq, task) || |
fd2f4419 | 1623 | !task->on_rq)) { |
4df64c0b | 1624 | |
7f1b4393 | 1625 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1626 | lowest_rq = NULL; |
1627 | break; | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | /* If this rq is still suitable use it. */ | |
e864c499 | 1632 | if (lowest_rq->rt.highest_prio.curr > task->prio) |
e8fa1362 SR |
1633 | break; |
1634 | ||
1635 | /* try again */ | |
1b12bbc7 | 1636 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 SR |
1637 | lowest_rq = NULL; |
1638 | } | |
1639 | ||
1640 | return lowest_rq; | |
1641 | } | |
1642 | ||
917b627d GH |
1643 | static struct task_struct *pick_next_pushable_task(struct rq *rq) |
1644 | { | |
1645 | struct task_struct *p; | |
1646 | ||
1647 | if (!has_pushable_tasks(rq)) | |
1648 | return NULL; | |
1649 | ||
1650 | p = plist_first_entry(&rq->rt.pushable_tasks, | |
1651 | struct task_struct, pushable_tasks); | |
1652 | ||
1653 | BUG_ON(rq->cpu != task_cpu(p)); | |
1654 | BUG_ON(task_current(rq, p)); | |
29baa747 | 1655 | BUG_ON(p->nr_cpus_allowed <= 1); |
917b627d | 1656 | |
fd2f4419 | 1657 | BUG_ON(!p->on_rq); |
917b627d GH |
1658 | BUG_ON(!rt_task(p)); |
1659 | ||
1660 | return p; | |
1661 | } | |
1662 | ||
e8fa1362 SR |
1663 | /* |
1664 | * If the current CPU has more than one RT task, see if the non | |
1665 | * running task can migrate over to a CPU that is running a task | |
1666 | * of lesser priority. | |
1667 | */ | |
697f0a48 | 1668 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1669 | { |
1670 | struct task_struct *next_task; | |
1671 | struct rq *lowest_rq; | |
311e800e | 1672 | int ret = 0; |
e8fa1362 | 1673 | |
a22d7fc1 GH |
1674 | if (!rq->rt.overloaded) |
1675 | return 0; | |
1676 | ||
917b627d | 1677 | next_task = pick_next_pushable_task(rq); |
e8fa1362 SR |
1678 | if (!next_task) |
1679 | return 0; | |
1680 | ||
49246274 | 1681 | retry: |
697f0a48 | 1682 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1683 | WARN_ON(1); |
e8fa1362 | 1684 | return 0; |
f65eda4f | 1685 | } |
e8fa1362 SR |
1686 | |
1687 | /* | |
1688 | * It's possible that the next_task slipped in of | |
1689 | * higher priority than current. If that's the case | |
1690 | * just reschedule current. | |
1691 | */ | |
697f0a48 GH |
1692 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1693 | resched_task(rq->curr); | |
e8fa1362 SR |
1694 | return 0; |
1695 | } | |
1696 | ||
697f0a48 | 1697 | /* We might release rq lock */ |
e8fa1362 SR |
1698 | get_task_struct(next_task); |
1699 | ||
1700 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1701 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1702 | if (!lowest_rq) { |
1703 | struct task_struct *task; | |
1704 | /* | |
311e800e | 1705 | * find_lock_lowest_rq releases rq->lock |
1563513d GH |
1706 | * so it is possible that next_task has migrated. |
1707 | * | |
1708 | * We need to make sure that the task is still on the same | |
1709 | * run-queue and is also still the next task eligible for | |
1710 | * pushing. | |
e8fa1362 | 1711 | */ |
917b627d | 1712 | task = pick_next_pushable_task(rq); |
1563513d GH |
1713 | if (task_cpu(next_task) == rq->cpu && task == next_task) { |
1714 | /* | |
311e800e HD |
1715 | * The task hasn't migrated, and is still the next |
1716 | * eligible task, but we failed to find a run-queue | |
1717 | * to push it to. Do not retry in this case, since | |
1718 | * other cpus will pull from us when ready. | |
1563513d | 1719 | */ |
1563513d | 1720 | goto out; |
e8fa1362 | 1721 | } |
917b627d | 1722 | |
1563513d GH |
1723 | if (!task) |
1724 | /* No more tasks, just exit */ | |
1725 | goto out; | |
1726 | ||
917b627d | 1727 | /* |
1563513d | 1728 | * Something has shifted, try again. |
917b627d | 1729 | */ |
1563513d GH |
1730 | put_task_struct(next_task); |
1731 | next_task = task; | |
1732 | goto retry; | |
e8fa1362 SR |
1733 | } |
1734 | ||
697f0a48 | 1735 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1736 | set_task_cpu(next_task, lowest_rq->cpu); |
1737 | activate_task(lowest_rq, next_task, 0); | |
311e800e | 1738 | ret = 1; |
e8fa1362 SR |
1739 | |
1740 | resched_task(lowest_rq->curr); | |
1741 | ||
1b12bbc7 | 1742 | double_unlock_balance(rq, lowest_rq); |
e8fa1362 | 1743 | |
e8fa1362 SR |
1744 | out: |
1745 | put_task_struct(next_task); | |
1746 | ||
311e800e | 1747 | return ret; |
e8fa1362 SR |
1748 | } |
1749 | ||
e8fa1362 SR |
1750 | static void push_rt_tasks(struct rq *rq) |
1751 | { | |
1752 | /* push_rt_task will return true if it moved an RT */ | |
1753 | while (push_rt_task(rq)) | |
1754 | ; | |
1755 | } | |
1756 | ||
f65eda4f SR |
1757 | static int pull_rt_task(struct rq *this_rq) |
1758 | { | |
80bf3171 | 1759 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
a8728944 | 1760 | struct task_struct *p; |
f65eda4f | 1761 | struct rq *src_rq; |
f65eda4f | 1762 | |
637f5085 | 1763 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
1764 | return 0; |
1765 | ||
7c3f2ab7 PZ |
1766 | /* |
1767 | * Match the barrier from rt_set_overloaded; this guarantees that if we | |
1768 | * see overloaded we must also see the rto_mask bit. | |
1769 | */ | |
1770 | smp_rmb(); | |
1771 | ||
c6c4927b | 1772 | for_each_cpu(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1773 | if (this_cpu == cpu) |
1774 | continue; | |
1775 | ||
1776 | src_rq = cpu_rq(cpu); | |
74ab8e4f GH |
1777 | |
1778 | /* | |
1779 | * Don't bother taking the src_rq->lock if the next highest | |
1780 | * task is known to be lower-priority than our current task. | |
1781 | * This may look racy, but if this value is about to go | |
1782 | * logically higher, the src_rq will push this task away. | |
1783 | * And if its going logically lower, we do not care | |
1784 | */ | |
1785 | if (src_rq->rt.highest_prio.next >= | |
1786 | this_rq->rt.highest_prio.curr) | |
1787 | continue; | |
1788 | ||
f65eda4f SR |
1789 | /* |
1790 | * We can potentially drop this_rq's lock in | |
1791 | * double_lock_balance, and another CPU could | |
a8728944 | 1792 | * alter this_rq |
f65eda4f | 1793 | */ |
a8728944 | 1794 | double_lock_balance(this_rq, src_rq); |
f65eda4f SR |
1795 | |
1796 | /* | |
e23ee747 KT |
1797 | * We can pull only a task, which is pushable |
1798 | * on its rq, and no others. | |
f65eda4f | 1799 | */ |
e23ee747 | 1800 | p = pick_highest_pushable_task(src_rq, this_cpu); |
f65eda4f SR |
1801 | |
1802 | /* | |
1803 | * Do we have an RT task that preempts | |
1804 | * the to-be-scheduled task? | |
1805 | */ | |
a8728944 | 1806 | if (p && (p->prio < this_rq->rt.highest_prio.curr)) { |
f65eda4f | 1807 | WARN_ON(p == src_rq->curr); |
fd2f4419 | 1808 | WARN_ON(!p->on_rq); |
f65eda4f SR |
1809 | |
1810 | /* | |
1811 | * There's a chance that p is higher in priority | |
1812 | * than what's currently running on its cpu. | |
1813 | * This is just that p is wakeing up and hasn't | |
1814 | * had a chance to schedule. We only pull | |
1815 | * p if it is lower in priority than the | |
a8728944 | 1816 | * current task on the run queue |
f65eda4f | 1817 | */ |
a8728944 | 1818 | if (p->prio < src_rq->curr->prio) |
614ee1f6 | 1819 | goto skip; |
f65eda4f SR |
1820 | |
1821 | ret = 1; | |
1822 | ||
1823 | deactivate_task(src_rq, p, 0); | |
1824 | set_task_cpu(p, this_cpu); | |
1825 | activate_task(this_rq, p, 0); | |
1826 | /* | |
1827 | * We continue with the search, just in | |
1828 | * case there's an even higher prio task | |
25985edc | 1829 | * in another runqueue. (low likelihood |
f65eda4f | 1830 | * but possible) |
f65eda4f | 1831 | */ |
f65eda4f | 1832 | } |
49246274 | 1833 | skip: |
1b12bbc7 | 1834 | double_unlock_balance(this_rq, src_rq); |
f65eda4f SR |
1835 | } |
1836 | ||
1837 | return ret; | |
1838 | } | |
1839 | ||
9a897c5a | 1840 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 | 1841 | { |
967fc046 | 1842 | push_rt_tasks(rq); |
e8fa1362 SR |
1843 | } |
1844 | ||
8ae121ac GH |
1845 | /* |
1846 | * If we are not running and we are not going to reschedule soon, we should | |
1847 | * try to push tasks away now | |
1848 | */ | |
efbbd05a | 1849 | static void task_woken_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1850 | { |
9a897c5a | 1851 | if (!task_running(rq, p) && |
8ae121ac | 1852 | !test_tsk_need_resched(rq->curr) && |
917b627d | 1853 | has_pushable_tasks(rq) && |
29baa747 | 1854 | p->nr_cpus_allowed > 1 && |
1baca4ce | 1855 | (dl_task(rq->curr) || rt_task(rq->curr)) && |
29baa747 | 1856 | (rq->curr->nr_cpus_allowed < 2 || |
3be209a8 | 1857 | rq->curr->prio <= p->prio)) |
4642dafd SR |
1858 | push_rt_tasks(rq); |
1859 | } | |
1860 | ||
cd8ba7cd | 1861 | static void set_cpus_allowed_rt(struct task_struct *p, |
96f874e2 | 1862 | const struct cpumask *new_mask) |
73fe6aae | 1863 | { |
8d3d5ada KT |
1864 | struct rq *rq; |
1865 | int weight; | |
73fe6aae GH |
1866 | |
1867 | BUG_ON(!rt_task(p)); | |
1868 | ||
8d3d5ada KT |
1869 | if (!p->on_rq) |
1870 | return; | |
917b627d | 1871 | |
8d3d5ada | 1872 | weight = cpumask_weight(new_mask); |
917b627d | 1873 | |
8d3d5ada KT |
1874 | /* |
1875 | * Only update if the process changes its state from whether it | |
1876 | * can migrate or not. | |
1877 | */ | |
29baa747 | 1878 | if ((p->nr_cpus_allowed > 1) == (weight > 1)) |
8d3d5ada | 1879 | return; |
917b627d | 1880 | |
8d3d5ada | 1881 | rq = task_rq(p); |
73fe6aae | 1882 | |
8d3d5ada KT |
1883 | /* |
1884 | * The process used to be able to migrate OR it can now migrate | |
1885 | */ | |
1886 | if (weight <= 1) { | |
1887 | if (!task_current(rq, p)) | |
1888 | dequeue_pushable_task(rq, p); | |
1889 | BUG_ON(!rq->rt.rt_nr_migratory); | |
1890 | rq->rt.rt_nr_migratory--; | |
1891 | } else { | |
1892 | if (!task_current(rq, p)) | |
1893 | enqueue_pushable_task(rq, p); | |
1894 | rq->rt.rt_nr_migratory++; | |
73fe6aae | 1895 | } |
8d3d5ada KT |
1896 | |
1897 | update_rt_migration(&rq->rt); | |
73fe6aae | 1898 | } |
deeeccd4 | 1899 | |
bdd7c81b | 1900 | /* Assumes rq->lock is held */ |
1f11eb6a | 1901 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
1902 | { |
1903 | if (rq->rt.overloaded) | |
1904 | rt_set_overload(rq); | |
6e0534f2 | 1905 | |
7def2be1 PZ |
1906 | __enable_runtime(rq); |
1907 | ||
e864c499 | 1908 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr); |
bdd7c81b IM |
1909 | } |
1910 | ||
1911 | /* Assumes rq->lock is held */ | |
1f11eb6a | 1912 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
1913 | { |
1914 | if (rq->rt.overloaded) | |
1915 | rt_clear_overload(rq); | |
6e0534f2 | 1916 | |
7def2be1 PZ |
1917 | __disable_runtime(rq); |
1918 | ||
6e0534f2 | 1919 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 1920 | } |
cb469845 SR |
1921 | |
1922 | /* | |
1923 | * When switch from the rt queue, we bring ourselves to a position | |
1924 | * that we might want to pull RT tasks from other runqueues. | |
1925 | */ | |
da7a735e | 1926 | static void switched_from_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1927 | { |
1928 | /* | |
1929 | * If there are other RT tasks then we will reschedule | |
1930 | * and the scheduling of the other RT tasks will handle | |
1931 | * the balancing. But if we are the last RT task | |
1932 | * we may need to handle the pulling of RT tasks | |
1933 | * now. | |
1934 | */ | |
1158ddb5 KT |
1935 | if (!p->on_rq || rq->rt.rt_nr_running) |
1936 | return; | |
1937 | ||
1938 | if (pull_rt_task(rq)) | |
1939 | resched_task(rq->curr); | |
cb469845 | 1940 | } |
3d8cbdf8 | 1941 | |
11c785b7 | 1942 | void __init init_sched_rt_class(void) |
3d8cbdf8 RR |
1943 | { |
1944 | unsigned int i; | |
1945 | ||
029632fb | 1946 | for_each_possible_cpu(i) { |
eaa95840 | 1947 | zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), |
6ca09dfc | 1948 | GFP_KERNEL, cpu_to_node(i)); |
029632fb | 1949 | } |
3d8cbdf8 | 1950 | } |
cb469845 SR |
1951 | #endif /* CONFIG_SMP */ |
1952 | ||
1953 | /* | |
1954 | * When switching a task to RT, we may overload the runqueue | |
1955 | * with RT tasks. In this case we try to push them off to | |
1956 | * other runqueues. | |
1957 | */ | |
da7a735e | 1958 | static void switched_to_rt(struct rq *rq, struct task_struct *p) |
cb469845 SR |
1959 | { |
1960 | int check_resched = 1; | |
1961 | ||
1962 | /* | |
1963 | * If we are already running, then there's nothing | |
1964 | * that needs to be done. But if we are not running | |
1965 | * we may need to preempt the current running task. | |
1966 | * If that current running task is also an RT task | |
1967 | * then see if we can move to another run queue. | |
1968 | */ | |
fd2f4419 | 1969 | if (p->on_rq && rq->curr != p) { |
cb469845 | 1970 | #ifdef CONFIG_SMP |
10447917 | 1971 | if (p->nr_cpus_allowed > 1 && rq->rt.overloaded && |
cb469845 | 1972 | /* Don't resched if we changed runqueues */ |
10447917 | 1973 | push_rt_task(rq) && rq != task_rq(p)) |
cb469845 SR |
1974 | check_resched = 0; |
1975 | #endif /* CONFIG_SMP */ | |
1976 | if (check_resched && p->prio < rq->curr->prio) | |
1977 | resched_task(rq->curr); | |
1978 | } | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * Priority of the task has changed. This may cause | |
1983 | * us to initiate a push or pull. | |
1984 | */ | |
da7a735e PZ |
1985 | static void |
1986 | prio_changed_rt(struct rq *rq, struct task_struct *p, int oldprio) | |
cb469845 | 1987 | { |
fd2f4419 | 1988 | if (!p->on_rq) |
da7a735e PZ |
1989 | return; |
1990 | ||
1991 | if (rq->curr == p) { | |
cb469845 SR |
1992 | #ifdef CONFIG_SMP |
1993 | /* | |
1994 | * If our priority decreases while running, we | |
1995 | * may need to pull tasks to this runqueue. | |
1996 | */ | |
1997 | if (oldprio < p->prio) | |
1998 | pull_rt_task(rq); | |
1999 | /* | |
2000 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
2001 | * then reschedule. Note, the above pull_rt_task |
2002 | * can release the rq lock and p could migrate. | |
2003 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 2004 | */ |
e864c499 | 2005 | if (p->prio > rq->rt.highest_prio.curr && rq->curr == p) |
cb469845 SR |
2006 | resched_task(p); |
2007 | #else | |
2008 | /* For UP simply resched on drop of prio */ | |
2009 | if (oldprio < p->prio) | |
2010 | resched_task(p); | |
e8fa1362 | 2011 | #endif /* CONFIG_SMP */ |
cb469845 SR |
2012 | } else { |
2013 | /* | |
2014 | * This task is not running, but if it is | |
2015 | * greater than the current running task | |
2016 | * then reschedule. | |
2017 | */ | |
2018 | if (p->prio < rq->curr->prio) | |
2019 | resched_task(rq->curr); | |
2020 | } | |
2021 | } | |
2022 | ||
78f2c7db PZ |
2023 | static void watchdog(struct rq *rq, struct task_struct *p) |
2024 | { | |
2025 | unsigned long soft, hard; | |
2026 | ||
78d7d407 JS |
2027 | /* max may change after cur was read, this will be fixed next tick */ |
2028 | soft = task_rlimit(p, RLIMIT_RTTIME); | |
2029 | hard = task_rlimit_max(p, RLIMIT_RTTIME); | |
78f2c7db PZ |
2030 | |
2031 | if (soft != RLIM_INFINITY) { | |
2032 | unsigned long next; | |
2033 | ||
57d2aa00 YX |
2034 | if (p->rt.watchdog_stamp != jiffies) { |
2035 | p->rt.timeout++; | |
2036 | p->rt.watchdog_stamp = jiffies; | |
2037 | } | |
2038 | ||
78f2c7db | 2039 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); |
5a52dd50 | 2040 | if (p->rt.timeout > next) |
f06febc9 | 2041 | p->cputime_expires.sched_exp = p->se.sum_exec_runtime; |
78f2c7db PZ |
2042 | } |
2043 | } | |
bb44e5d1 | 2044 | |
8f4d37ec | 2045 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 2046 | { |
454c7999 CC |
2047 | struct sched_rt_entity *rt_se = &p->rt; |
2048 | ||
67e2be02 PZ |
2049 | update_curr_rt(rq); |
2050 | ||
78f2c7db PZ |
2051 | watchdog(rq, p); |
2052 | ||
bb44e5d1 IM |
2053 | /* |
2054 | * RR tasks need a special form of timeslice management. | |
2055 | * FIFO tasks have no timeslices. | |
2056 | */ | |
2057 | if (p->policy != SCHED_RR) | |
2058 | return; | |
2059 | ||
fa717060 | 2060 | if (--p->rt.time_slice) |
bb44e5d1 IM |
2061 | return; |
2062 | ||
ce0dbbbb | 2063 | p->rt.time_slice = sched_rr_timeslice; |
bb44e5d1 | 2064 | |
98fbc798 | 2065 | /* |
e9aa39bb LB |
2066 | * Requeue to the end of queue if we (and all of our ancestors) are not |
2067 | * the only element on the queue | |
98fbc798 | 2068 | */ |
454c7999 CC |
2069 | for_each_sched_rt_entity(rt_se) { |
2070 | if (rt_se->run_list.prev != rt_se->run_list.next) { | |
2071 | requeue_task_rt(rq, p, 0); | |
2072 | set_tsk_need_resched(p); | |
2073 | return; | |
2074 | } | |
98fbc798 | 2075 | } |
bb44e5d1 IM |
2076 | } |
2077 | ||
83b699ed SV |
2078 | static void set_curr_task_rt(struct rq *rq) |
2079 | { | |
2080 | struct task_struct *p = rq->curr; | |
2081 | ||
78becc27 | 2082 | p->se.exec_start = rq_clock_task(rq); |
917b627d GH |
2083 | |
2084 | /* The running task is never eligible for pushing */ | |
2085 | dequeue_pushable_task(rq, p); | |
83b699ed SV |
2086 | } |
2087 | ||
6d686f45 | 2088 | static unsigned int get_rr_interval_rt(struct rq *rq, struct task_struct *task) |
0d721cea PW |
2089 | { |
2090 | /* | |
2091 | * Time slice is 0 for SCHED_FIFO tasks | |
2092 | */ | |
2093 | if (task->policy == SCHED_RR) | |
ce0dbbbb | 2094 | return sched_rr_timeslice; |
0d721cea PW |
2095 | else |
2096 | return 0; | |
2097 | } | |
2098 | ||
029632fb | 2099 | const struct sched_class rt_sched_class = { |
5522d5d5 | 2100 | .next = &fair_sched_class, |
bb44e5d1 IM |
2101 | .enqueue_task = enqueue_task_rt, |
2102 | .dequeue_task = dequeue_task_rt, | |
2103 | .yield_task = yield_task_rt, | |
2104 | ||
2105 | .check_preempt_curr = check_preempt_curr_rt, | |
2106 | ||
2107 | .pick_next_task = pick_next_task_rt, | |
2108 | .put_prev_task = put_prev_task_rt, | |
2109 | ||
681f3e68 | 2110 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
2111 | .select_task_rq = select_task_rq_rt, |
2112 | ||
73fe6aae | 2113 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
2114 | .rq_online = rq_online_rt, |
2115 | .rq_offline = rq_offline_rt, | |
9a897c5a | 2116 | .post_schedule = post_schedule_rt, |
efbbd05a | 2117 | .task_woken = task_woken_rt, |
cb469845 | 2118 | .switched_from = switched_from_rt, |
681f3e68 | 2119 | #endif |
bb44e5d1 | 2120 | |
83b699ed | 2121 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 2122 | .task_tick = task_tick_rt, |
cb469845 | 2123 | |
0d721cea PW |
2124 | .get_rr_interval = get_rr_interval_rt, |
2125 | ||
cb469845 SR |
2126 | .prio_changed = prio_changed_rt, |
2127 | .switched_to = switched_to_rt, | |
bb44e5d1 | 2128 | }; |
ada18de2 PZ |
2129 | |
2130 | #ifdef CONFIG_SCHED_DEBUG | |
2131 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2132 | ||
029632fb | 2133 | void print_rt_stats(struct seq_file *m, int cpu) |
ada18de2 | 2134 | { |
ec514c48 | 2135 | rt_rq_iter_t iter; |
ada18de2 PZ |
2136 | struct rt_rq *rt_rq; |
2137 | ||
2138 | rcu_read_lock(); | |
ec514c48 | 2139 | for_each_rt_rq(rt_rq, iter, cpu_rq(cpu)) |
ada18de2 PZ |
2140 | print_rt_rq(m, cpu, rt_rq); |
2141 | rcu_read_unlock(); | |
2142 | } | |
55e12e5e | 2143 | #endif /* CONFIG_SCHED_DEBUG */ |