Commit | Line | Data |
---|---|---|
bb44e5d1 IM |
1 | /* |
2 | * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR | |
3 | * policies) | |
4 | */ | |
5 | ||
4fd29176 | 6 | #ifdef CONFIG_SMP |
84de4274 | 7 | |
637f5085 | 8 | static inline int rt_overloaded(struct rq *rq) |
4fd29176 | 9 | { |
637f5085 | 10 | return atomic_read(&rq->rd->rto_count); |
4fd29176 | 11 | } |
84de4274 | 12 | |
4fd29176 SR |
13 | static inline void rt_set_overload(struct rq *rq) |
14 | { | |
1f11eb6a GH |
15 | if (!rq->online) |
16 | return; | |
17 | ||
637f5085 | 18 | cpu_set(rq->cpu, rq->rd->rto_mask); |
4fd29176 SR |
19 | /* |
20 | * Make sure the mask is visible before we set | |
21 | * the overload count. That is checked to determine | |
22 | * if we should look at the mask. It would be a shame | |
23 | * if we looked at the mask, but the mask was not | |
24 | * updated yet. | |
25 | */ | |
26 | wmb(); | |
637f5085 | 27 | atomic_inc(&rq->rd->rto_count); |
4fd29176 | 28 | } |
84de4274 | 29 | |
4fd29176 SR |
30 | static inline void rt_clear_overload(struct rq *rq) |
31 | { | |
1f11eb6a GH |
32 | if (!rq->online) |
33 | return; | |
34 | ||
4fd29176 | 35 | /* the order here really doesn't matter */ |
637f5085 GH |
36 | atomic_dec(&rq->rd->rto_count); |
37 | cpu_clear(rq->cpu, rq->rd->rto_mask); | |
4fd29176 | 38 | } |
73fe6aae GH |
39 | |
40 | static void update_rt_migration(struct rq *rq) | |
41 | { | |
637f5085 | 42 | if (rq->rt.rt_nr_migratory && (rq->rt.rt_nr_running > 1)) { |
cdc8eb98 GH |
43 | if (!rq->rt.overloaded) { |
44 | rt_set_overload(rq); | |
45 | rq->rt.overloaded = 1; | |
46 | } | |
47 | } else if (rq->rt.overloaded) { | |
73fe6aae | 48 | rt_clear_overload(rq); |
637f5085 GH |
49 | rq->rt.overloaded = 0; |
50 | } | |
73fe6aae | 51 | } |
4fd29176 SR |
52 | #endif /* CONFIG_SMP */ |
53 | ||
6f505b16 | 54 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
fa85ae24 | 55 | { |
6f505b16 PZ |
56 | return container_of(rt_se, struct task_struct, rt); |
57 | } | |
58 | ||
59 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) | |
60 | { | |
61 | return !list_empty(&rt_se->run_list); | |
62 | } | |
63 | ||
052f1dc7 | 64 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 65 | |
9f0c1e56 | 66 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
67 | { |
68 | if (!rt_rq->tg) | |
9f0c1e56 | 69 | return RUNTIME_INF; |
6f505b16 | 70 | |
ac086bc2 PZ |
71 | return rt_rq->rt_runtime; |
72 | } | |
73 | ||
74 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
75 | { | |
76 | return ktime_to_ns(rt_rq->tg->rt_bandwidth.rt_period); | |
6f505b16 PZ |
77 | } |
78 | ||
79 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
80 | list_for_each_entry(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) | |
81 | ||
82 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) | |
83 | { | |
84 | return rt_rq->rq; | |
85 | } | |
86 | ||
87 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
88 | { | |
89 | return rt_se->rt_rq; | |
90 | } | |
91 | ||
92 | #define for_each_sched_rt_entity(rt_se) \ | |
93 | for (; rt_se; rt_se = rt_se->parent) | |
94 | ||
95 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
96 | { | |
97 | return rt_se->my_q; | |
98 | } | |
99 | ||
100 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se); | |
101 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); | |
102 | ||
9f0c1e56 | 103 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 PZ |
104 | { |
105 | struct sched_rt_entity *rt_se = rt_rq->rt_se; | |
106 | ||
107 | if (rt_se && !on_rt_rq(rt_se) && rt_rq->rt_nr_running) { | |
1020387f PZ |
108 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
109 | ||
6f505b16 | 110 | enqueue_rt_entity(rt_se); |
1020387f PZ |
111 | if (rt_rq->highest_prio < curr->prio) |
112 | resched_task(curr); | |
6f505b16 PZ |
113 | } |
114 | } | |
115 | ||
9f0c1e56 | 116 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
117 | { |
118 | struct sched_rt_entity *rt_se = rt_rq->rt_se; | |
119 | ||
120 | if (rt_se && on_rt_rq(rt_se)) | |
121 | dequeue_rt_entity(rt_se); | |
122 | } | |
123 | ||
23b0fdfc PZ |
124 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
125 | { | |
126 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
127 | } | |
128 | ||
129 | static int rt_se_boosted(struct sched_rt_entity *rt_se) | |
130 | { | |
131 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
132 | struct task_struct *p; | |
133 | ||
134 | if (rt_rq) | |
135 | return !!rt_rq->rt_nr_boosted; | |
136 | ||
137 | p = rt_task_of(rt_se); | |
138 | return p->prio != p->normal_prio; | |
139 | } | |
140 | ||
d0b27fa7 PZ |
141 | #ifdef CONFIG_SMP |
142 | static inline cpumask_t sched_rt_period_mask(void) | |
143 | { | |
144 | return cpu_rq(smp_processor_id())->rd->span; | |
145 | } | |
6f505b16 | 146 | #else |
d0b27fa7 PZ |
147 | static inline cpumask_t sched_rt_period_mask(void) |
148 | { | |
149 | return cpu_online_map; | |
150 | } | |
151 | #endif | |
6f505b16 | 152 | |
d0b27fa7 PZ |
153 | static inline |
154 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 155 | { |
d0b27fa7 PZ |
156 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
157 | } | |
9f0c1e56 | 158 | |
ac086bc2 PZ |
159 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
160 | { | |
161 | return &rt_rq->tg->rt_bandwidth; | |
162 | } | |
163 | ||
d0b27fa7 PZ |
164 | #else |
165 | ||
166 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
167 | { | |
ac086bc2 PZ |
168 | return rt_rq->rt_runtime; |
169 | } | |
170 | ||
171 | static inline u64 sched_rt_period(struct rt_rq *rt_rq) | |
172 | { | |
173 | return ktime_to_ns(def_rt_bandwidth.rt_period); | |
6f505b16 PZ |
174 | } |
175 | ||
176 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
177 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
178 | ||
179 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) | |
180 | { | |
181 | return container_of(rt_rq, struct rq, rt); | |
182 | } | |
183 | ||
184 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
185 | { | |
186 | struct task_struct *p = rt_task_of(rt_se); | |
187 | struct rq *rq = task_rq(p); | |
188 | ||
189 | return &rq->rt; | |
190 | } | |
191 | ||
192 | #define for_each_sched_rt_entity(rt_se) \ | |
193 | for (; rt_se; rt_se = NULL) | |
194 | ||
195 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
196 | { | |
197 | return NULL; | |
198 | } | |
199 | ||
9f0c1e56 | 200 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 PZ |
201 | { |
202 | } | |
203 | ||
9f0c1e56 | 204 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
205 | { |
206 | } | |
207 | ||
23b0fdfc PZ |
208 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
209 | { | |
210 | return rt_rq->rt_throttled; | |
211 | } | |
d0b27fa7 PZ |
212 | |
213 | static inline cpumask_t sched_rt_period_mask(void) | |
214 | { | |
215 | return cpu_online_map; | |
216 | } | |
217 | ||
218 | static inline | |
219 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
220 | { | |
221 | return &cpu_rq(cpu)->rt; | |
222 | } | |
223 | ||
ac086bc2 PZ |
224 | static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq) |
225 | { | |
226 | return &def_rt_bandwidth; | |
227 | } | |
228 | ||
6f505b16 PZ |
229 | #endif |
230 | ||
b79f3833 PZ |
231 | #ifdef CONFIG_SMP |
232 | static int do_balance_runtime(struct rt_rq *rt_rq); | |
233 | ||
234 | static int balance_runtime(struct rt_rq *rt_rq) | |
235 | { | |
236 | int more = 0; | |
237 | ||
238 | if (rt_rq->rt_time > rt_rq->rt_runtime) { | |
239 | spin_unlock(&rt_rq->rt_runtime_lock); | |
240 | more = do_balance_runtime(rt_rq); | |
241 | spin_lock(&rt_rq->rt_runtime_lock); | |
242 | } | |
243 | ||
244 | return more; | |
245 | } | |
246 | #else | |
247 | static inline int balance_runtime(struct rt_rq *rt_rq) | |
248 | { | |
249 | return 0; | |
250 | } | |
251 | #endif | |
252 | ||
d0b27fa7 PZ |
253 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
254 | { | |
255 | int i, idle = 1; | |
256 | cpumask_t span; | |
257 | ||
258 | if (rt_b->rt_runtime == RUNTIME_INF) | |
259 | return 1; | |
260 | ||
261 | span = sched_rt_period_mask(); | |
262 | for_each_cpu_mask(i, span) { | |
263 | int enqueue = 0; | |
264 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
265 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
266 | ||
267 | spin_lock(&rq->lock); | |
268 | if (rt_rq->rt_time) { | |
ac086bc2 | 269 | u64 runtime; |
d0b27fa7 | 270 | |
ac086bc2 | 271 | spin_lock(&rt_rq->rt_runtime_lock); |
b79f3833 PZ |
272 | if (rt_rq->rt_throttled) |
273 | balance_runtime(rt_rq); | |
ac086bc2 | 274 | runtime = rt_rq->rt_runtime; |
d0b27fa7 PZ |
275 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); |
276 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
277 | rt_rq->rt_throttled = 0; | |
278 | enqueue = 1; | |
279 | } | |
280 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
281 | idle = 0; | |
ac086bc2 | 282 | spin_unlock(&rt_rq->rt_runtime_lock); |
d0b27fa7 PZ |
283 | } |
284 | ||
285 | if (enqueue) | |
286 | sched_rt_rq_enqueue(rt_rq); | |
287 | spin_unlock(&rq->lock); | |
288 | } | |
289 | ||
290 | return idle; | |
291 | } | |
292 | ||
ac086bc2 | 293 | #ifdef CONFIG_SMP |
b79f3833 | 294 | static int do_balance_runtime(struct rt_rq *rt_rq) |
ac086bc2 PZ |
295 | { |
296 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
297 | struct root_domain *rd = cpu_rq(smp_processor_id())->rd; | |
298 | int i, weight, more = 0; | |
299 | u64 rt_period; | |
300 | ||
301 | weight = cpus_weight(rd->span); | |
302 | ||
303 | spin_lock(&rt_b->rt_runtime_lock); | |
304 | rt_period = ktime_to_ns(rt_b->rt_period); | |
305 | for_each_cpu_mask(i, rd->span) { | |
306 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); | |
307 | s64 diff; | |
308 | ||
309 | if (iter == rt_rq) | |
310 | continue; | |
311 | ||
312 | spin_lock(&iter->rt_runtime_lock); | |
7def2be1 PZ |
313 | if (iter->rt_runtime == RUNTIME_INF) |
314 | goto next; | |
315 | ||
ac086bc2 PZ |
316 | diff = iter->rt_runtime - iter->rt_time; |
317 | if (diff > 0) { | |
318 | do_div(diff, weight); | |
319 | if (rt_rq->rt_runtime + diff > rt_period) | |
320 | diff = rt_period - rt_rq->rt_runtime; | |
321 | iter->rt_runtime -= diff; | |
322 | rt_rq->rt_runtime += diff; | |
323 | more = 1; | |
324 | if (rt_rq->rt_runtime == rt_period) { | |
325 | spin_unlock(&iter->rt_runtime_lock); | |
326 | break; | |
327 | } | |
328 | } | |
7def2be1 | 329 | next: |
ac086bc2 PZ |
330 | spin_unlock(&iter->rt_runtime_lock); |
331 | } | |
332 | spin_unlock(&rt_b->rt_runtime_lock); | |
333 | ||
334 | return more; | |
335 | } | |
7def2be1 PZ |
336 | |
337 | static void __disable_runtime(struct rq *rq) | |
338 | { | |
339 | struct root_domain *rd = rq->rd; | |
340 | struct rt_rq *rt_rq; | |
341 | ||
342 | if (unlikely(!scheduler_running)) | |
343 | return; | |
344 | ||
345 | for_each_leaf_rt_rq(rt_rq, rq) { | |
346 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
347 | s64 want; | |
348 | int i; | |
349 | ||
350 | spin_lock(&rt_b->rt_runtime_lock); | |
351 | spin_lock(&rt_rq->rt_runtime_lock); | |
352 | if (rt_rq->rt_runtime == RUNTIME_INF || | |
353 | rt_rq->rt_runtime == rt_b->rt_runtime) | |
354 | goto balanced; | |
355 | spin_unlock(&rt_rq->rt_runtime_lock); | |
356 | ||
357 | want = rt_b->rt_runtime - rt_rq->rt_runtime; | |
358 | ||
359 | for_each_cpu_mask(i, rd->span) { | |
360 | struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i); | |
361 | s64 diff; | |
362 | ||
363 | if (iter == rt_rq) | |
364 | continue; | |
365 | ||
366 | spin_lock(&iter->rt_runtime_lock); | |
367 | if (want > 0) { | |
368 | diff = min_t(s64, iter->rt_runtime, want); | |
369 | iter->rt_runtime -= diff; | |
370 | want -= diff; | |
371 | } else { | |
372 | iter->rt_runtime -= want; | |
373 | want -= want; | |
374 | } | |
375 | spin_unlock(&iter->rt_runtime_lock); | |
376 | ||
377 | if (!want) | |
378 | break; | |
379 | } | |
380 | ||
381 | spin_lock(&rt_rq->rt_runtime_lock); | |
382 | BUG_ON(want); | |
383 | balanced: | |
384 | rt_rq->rt_runtime = RUNTIME_INF; | |
385 | spin_unlock(&rt_rq->rt_runtime_lock); | |
386 | spin_unlock(&rt_b->rt_runtime_lock); | |
387 | } | |
388 | } | |
389 | ||
390 | static void disable_runtime(struct rq *rq) | |
391 | { | |
392 | unsigned long flags; | |
393 | ||
394 | spin_lock_irqsave(&rq->lock, flags); | |
395 | __disable_runtime(rq); | |
396 | spin_unlock_irqrestore(&rq->lock, flags); | |
397 | } | |
398 | ||
399 | static void __enable_runtime(struct rq *rq) | |
400 | { | |
401 | struct root_domain *rd = rq->rd; | |
402 | struct rt_rq *rt_rq; | |
403 | ||
404 | if (unlikely(!scheduler_running)) | |
405 | return; | |
406 | ||
407 | for_each_leaf_rt_rq(rt_rq, rq) { | |
408 | struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq); | |
409 | ||
410 | spin_lock(&rt_b->rt_runtime_lock); | |
411 | spin_lock(&rt_rq->rt_runtime_lock); | |
412 | rt_rq->rt_runtime = rt_b->rt_runtime; | |
413 | rt_rq->rt_time = 0; | |
414 | spin_unlock(&rt_rq->rt_runtime_lock); | |
415 | spin_unlock(&rt_b->rt_runtime_lock); | |
416 | } | |
417 | } | |
418 | ||
419 | static void enable_runtime(struct rq *rq) | |
420 | { | |
421 | unsigned long flags; | |
422 | ||
423 | spin_lock_irqsave(&rq->lock, flags); | |
424 | __enable_runtime(rq); | |
425 | spin_unlock_irqrestore(&rq->lock, flags); | |
426 | } | |
427 | ||
ac086bc2 PZ |
428 | #endif |
429 | ||
6f505b16 PZ |
430 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
431 | { | |
052f1dc7 | 432 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
433 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
434 | ||
435 | if (rt_rq) | |
436 | return rt_rq->highest_prio; | |
437 | #endif | |
438 | ||
439 | return rt_task_of(rt_se)->prio; | |
440 | } | |
441 | ||
9f0c1e56 | 442 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 443 | { |
9f0c1e56 | 444 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 445 | |
9f0c1e56 | 446 | if (runtime == RUNTIME_INF) |
fa85ae24 PZ |
447 | return 0; |
448 | ||
449 | if (rt_rq->rt_throttled) | |
23b0fdfc | 450 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 451 | |
ac086bc2 PZ |
452 | if (sched_rt_runtime(rt_rq) >= sched_rt_period(rt_rq)) |
453 | return 0; | |
454 | ||
b79f3833 PZ |
455 | balance_runtime(rt_rq); |
456 | runtime = sched_rt_runtime(rt_rq); | |
457 | if (runtime == RUNTIME_INF) | |
458 | return 0; | |
ac086bc2 | 459 | |
9f0c1e56 | 460 | if (rt_rq->rt_time > runtime) { |
6f505b16 | 461 | rt_rq->rt_throttled = 1; |
23b0fdfc | 462 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 463 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
464 | return 1; |
465 | } | |
fa85ae24 PZ |
466 | } |
467 | ||
468 | return 0; | |
469 | } | |
470 | ||
bb44e5d1 IM |
471 | /* |
472 | * Update the current task's runtime statistics. Skip current tasks that | |
473 | * are not in our scheduling class. | |
474 | */ | |
a9957449 | 475 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
476 | { |
477 | struct task_struct *curr = rq->curr; | |
6f505b16 PZ |
478 | struct sched_rt_entity *rt_se = &curr->rt; |
479 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
bb44e5d1 IM |
480 | u64 delta_exec; |
481 | ||
482 | if (!task_has_rt_policy(curr)) | |
483 | return; | |
484 | ||
d281918d | 485 | delta_exec = rq->clock - curr->se.exec_start; |
bb44e5d1 IM |
486 | if (unlikely((s64)delta_exec < 0)) |
487 | delta_exec = 0; | |
6cfb0d5d IM |
488 | |
489 | schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec)); | |
bb44e5d1 IM |
490 | |
491 | curr->se.sum_exec_runtime += delta_exec; | |
d281918d | 492 | curr->se.exec_start = rq->clock; |
d842de87 | 493 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 494 | |
354d60c2 DG |
495 | for_each_sched_rt_entity(rt_se) { |
496 | rt_rq = rt_rq_of_se(rt_se); | |
497 | ||
498 | spin_lock(&rt_rq->rt_runtime_lock); | |
499 | rt_rq->rt_time += delta_exec; | |
500 | if (sched_rt_runtime_exceeded(rt_rq)) | |
501 | resched_task(curr); | |
502 | spin_unlock(&rt_rq->rt_runtime_lock); | |
503 | } | |
bb44e5d1 IM |
504 | } |
505 | ||
6f505b16 PZ |
506 | static inline |
507 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
63489e45 | 508 | { |
6f505b16 PZ |
509 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
510 | rt_rq->rt_nr_running++; | |
052f1dc7 | 511 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6e0534f2 GH |
512 | if (rt_se_prio(rt_se) < rt_rq->highest_prio) { |
513 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
1f11eb6a | 514 | |
1100ac91 IM |
515 | rt_rq->highest_prio = rt_se_prio(rt_se); |
516 | #ifdef CONFIG_SMP | |
1f11eb6a GH |
517 | if (rq->online) |
518 | cpupri_set(&rq->rd->cpupri, rq->cpu, | |
519 | rt_se_prio(rt_se)); | |
1100ac91 | 520 | #endif |
6e0534f2 | 521 | } |
6f505b16 | 522 | #endif |
764a9d6f | 523 | #ifdef CONFIG_SMP |
6f505b16 PZ |
524 | if (rt_se->nr_cpus_allowed > 1) { |
525 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
1100ac91 | 526 | |
73fe6aae | 527 | rq->rt.rt_nr_migratory++; |
6f505b16 | 528 | } |
73fe6aae | 529 | |
6f505b16 PZ |
530 | update_rt_migration(rq_of_rt_rq(rt_rq)); |
531 | #endif | |
052f1dc7 | 532 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
533 | if (rt_se_boosted(rt_se)) |
534 | rt_rq->rt_nr_boosted++; | |
d0b27fa7 PZ |
535 | |
536 | if (rt_rq->tg) | |
537 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
538 | #else | |
539 | start_rt_bandwidth(&def_rt_bandwidth); | |
23b0fdfc | 540 | #endif |
63489e45 SR |
541 | } |
542 | ||
6f505b16 PZ |
543 | static inline |
544 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
63489e45 | 545 | { |
6e0534f2 GH |
546 | #ifdef CONFIG_SMP |
547 | int highest_prio = rt_rq->highest_prio; | |
548 | #endif | |
549 | ||
6f505b16 PZ |
550 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
551 | WARN_ON(!rt_rq->rt_nr_running); | |
552 | rt_rq->rt_nr_running--; | |
052f1dc7 | 553 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 | 554 | if (rt_rq->rt_nr_running) { |
764a9d6f SR |
555 | struct rt_prio_array *array; |
556 | ||
6f505b16 PZ |
557 | WARN_ON(rt_se_prio(rt_se) < rt_rq->highest_prio); |
558 | if (rt_se_prio(rt_se) == rt_rq->highest_prio) { | |
764a9d6f | 559 | /* recalculate */ |
6f505b16 PZ |
560 | array = &rt_rq->active; |
561 | rt_rq->highest_prio = | |
764a9d6f SR |
562 | sched_find_first_bit(array->bitmap); |
563 | } /* otherwise leave rq->highest prio alone */ | |
564 | } else | |
6f505b16 PZ |
565 | rt_rq->highest_prio = MAX_RT_PRIO; |
566 | #endif | |
567 | #ifdef CONFIG_SMP | |
568 | if (rt_se->nr_cpus_allowed > 1) { | |
569 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
73fe6aae | 570 | rq->rt.rt_nr_migratory--; |
6f505b16 | 571 | } |
73fe6aae | 572 | |
6e0534f2 GH |
573 | if (rt_rq->highest_prio != highest_prio) { |
574 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
1f11eb6a GH |
575 | |
576 | if (rq->online) | |
577 | cpupri_set(&rq->rd->cpupri, rq->cpu, | |
578 | rt_rq->highest_prio); | |
6e0534f2 GH |
579 | } |
580 | ||
6f505b16 | 581 | update_rt_migration(rq_of_rt_rq(rt_rq)); |
764a9d6f | 582 | #endif /* CONFIG_SMP */ |
052f1dc7 | 583 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
584 | if (rt_se_boosted(rt_se)) |
585 | rt_rq->rt_nr_boosted--; | |
586 | ||
587 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
588 | #endif | |
63489e45 SR |
589 | } |
590 | ||
ad2a3f13 | 591 | static void __enqueue_rt_entity(struct sched_rt_entity *rt_se) |
bb44e5d1 | 592 | { |
6f505b16 PZ |
593 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
594 | struct rt_prio_array *array = &rt_rq->active; | |
595 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
20b6331b | 596 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
bb44e5d1 | 597 | |
ad2a3f13 PZ |
598 | /* |
599 | * Don't enqueue the group if its throttled, or when empty. | |
600 | * The latter is a consequence of the former when a child group | |
601 | * get throttled and the current group doesn't have any other | |
602 | * active members. | |
603 | */ | |
604 | if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running)) | |
6f505b16 | 605 | return; |
63489e45 | 606 | |
45c01e82 | 607 | if (rt_se->nr_cpus_allowed == 1) |
20b6331b | 608 | list_add(&rt_se->run_list, queue); |
45c01e82 | 609 | else |
20b6331b | 610 | list_add_tail(&rt_se->run_list, queue); |
45c01e82 | 611 | |
6f505b16 | 612 | __set_bit(rt_se_prio(rt_se), array->bitmap); |
78f2c7db | 613 | |
6f505b16 PZ |
614 | inc_rt_tasks(rt_se, rt_rq); |
615 | } | |
616 | ||
ad2a3f13 | 617 | static void __dequeue_rt_entity(struct sched_rt_entity *rt_se) |
6f505b16 PZ |
618 | { |
619 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
620 | struct rt_prio_array *array = &rt_rq->active; | |
621 | ||
622 | list_del_init(&rt_se->run_list); | |
20b6331b | 623 | if (list_empty(array->queue + rt_se_prio(rt_se))) |
6f505b16 PZ |
624 | __clear_bit(rt_se_prio(rt_se), array->bitmap); |
625 | ||
626 | dec_rt_tasks(rt_se, rt_rq); | |
627 | } | |
628 | ||
629 | /* | |
630 | * Because the prio of an upper entry depends on the lower | |
631 | * entries, we must remove entries top - down. | |
6f505b16 | 632 | */ |
ad2a3f13 | 633 | static void dequeue_rt_stack(struct sched_rt_entity *rt_se) |
6f505b16 | 634 | { |
ad2a3f13 | 635 | struct sched_rt_entity *back = NULL; |
6f505b16 | 636 | |
58d6c2d7 PZ |
637 | for_each_sched_rt_entity(rt_se) { |
638 | rt_se->back = back; | |
639 | back = rt_se; | |
640 | } | |
641 | ||
642 | for (rt_se = back; rt_se; rt_se = rt_se->back) { | |
643 | if (on_rt_rq(rt_se)) | |
ad2a3f13 PZ |
644 | __dequeue_rt_entity(rt_se); |
645 | } | |
646 | } | |
647 | ||
648 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se) | |
649 | { | |
650 | dequeue_rt_stack(rt_se); | |
651 | for_each_sched_rt_entity(rt_se) | |
652 | __enqueue_rt_entity(rt_se); | |
653 | } | |
654 | ||
655 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
656 | { | |
657 | dequeue_rt_stack(rt_se); | |
658 | ||
659 | for_each_sched_rt_entity(rt_se) { | |
660 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
661 | ||
662 | if (rt_rq && rt_rq->rt_nr_running) | |
663 | __enqueue_rt_entity(rt_se); | |
58d6c2d7 | 664 | } |
bb44e5d1 IM |
665 | } |
666 | ||
667 | /* | |
668 | * Adding/removing a task to/from a priority array: | |
669 | */ | |
6f505b16 PZ |
670 | static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup) |
671 | { | |
672 | struct sched_rt_entity *rt_se = &p->rt; | |
673 | ||
674 | if (wakeup) | |
675 | rt_se->timeout = 0; | |
676 | ||
ad2a3f13 | 677 | enqueue_rt_entity(rt_se); |
6f505b16 PZ |
678 | } |
679 | ||
f02231e5 | 680 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep) |
bb44e5d1 | 681 | { |
6f505b16 | 682 | struct sched_rt_entity *rt_se = &p->rt; |
bb44e5d1 | 683 | |
f1e14ef6 | 684 | update_curr_rt(rq); |
ad2a3f13 | 685 | dequeue_rt_entity(rt_se); |
bb44e5d1 IM |
686 | } |
687 | ||
688 | /* | |
689 | * Put task to the end of the run list without the overhead of dequeue | |
690 | * followed by enqueue. | |
691 | */ | |
6f505b16 PZ |
692 | static |
693 | void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se) | |
694 | { | |
695 | struct rt_prio_array *array = &rt_rq->active; | |
15a8641e | 696 | struct list_head *queue = array->queue + rt_se_prio(rt_se); |
6f505b16 | 697 | |
1cdad715 IM |
698 | if (on_rt_rq(rt_se)) { |
699 | list_del_init(&rt_se->run_list); | |
700 | list_add_tail(&rt_se->run_list, | |
701 | array->queue + rt_se_prio(rt_se)); | |
702 | } | |
6f505b16 PZ |
703 | } |
704 | ||
bb44e5d1 IM |
705 | static void requeue_task_rt(struct rq *rq, struct task_struct *p) |
706 | { | |
6f505b16 PZ |
707 | struct sched_rt_entity *rt_se = &p->rt; |
708 | struct rt_rq *rt_rq; | |
bb44e5d1 | 709 | |
6f505b16 PZ |
710 | for_each_sched_rt_entity(rt_se) { |
711 | rt_rq = rt_rq_of_se(rt_se); | |
712 | requeue_rt_entity(rt_rq, rt_se); | |
713 | } | |
bb44e5d1 IM |
714 | } |
715 | ||
6f505b16 | 716 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 717 | { |
4530d7ab | 718 | requeue_task_rt(rq, rq->curr); |
bb44e5d1 IM |
719 | } |
720 | ||
e7693a36 | 721 | #ifdef CONFIG_SMP |
318e0893 GH |
722 | static int find_lowest_rq(struct task_struct *task); |
723 | ||
e7693a36 GH |
724 | static int select_task_rq_rt(struct task_struct *p, int sync) |
725 | { | |
318e0893 GH |
726 | struct rq *rq = task_rq(p); |
727 | ||
728 | /* | |
e1f47d89 SR |
729 | * If the current task is an RT task, then |
730 | * try to see if we can wake this RT task up on another | |
731 | * runqueue. Otherwise simply start this RT task | |
732 | * on its current runqueue. | |
733 | * | |
734 | * We want to avoid overloading runqueues. Even if | |
735 | * the RT task is of higher priority than the current RT task. | |
736 | * RT tasks behave differently than other tasks. If | |
737 | * one gets preempted, we try to push it off to another queue. | |
738 | * So trying to keep a preempting RT task on the same | |
739 | * cache hot CPU will force the running RT task to | |
740 | * a cold CPU. So we waste all the cache for the lower | |
741 | * RT task in hopes of saving some of a RT task | |
742 | * that is just being woken and probably will have | |
743 | * cold cache anyway. | |
318e0893 | 744 | */ |
17b3279b | 745 | if (unlikely(rt_task(rq->curr)) && |
6f505b16 | 746 | (p->rt.nr_cpus_allowed > 1)) { |
318e0893 GH |
747 | int cpu = find_lowest_rq(p); |
748 | ||
749 | return (cpu == -1) ? task_cpu(p) : cpu; | |
750 | } | |
751 | ||
752 | /* | |
753 | * Otherwise, just let it ride on the affined RQ and the | |
754 | * post-schedule router will push the preempted task away | |
755 | */ | |
e7693a36 GH |
756 | return task_cpu(p); |
757 | } | |
758 | #endif /* CONFIG_SMP */ | |
759 | ||
bb44e5d1 IM |
760 | /* |
761 | * Preempt the current task with a newly woken task if needed: | |
762 | */ | |
763 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p) | |
764 | { | |
45c01e82 | 765 | if (p->prio < rq->curr->prio) { |
bb44e5d1 | 766 | resched_task(rq->curr); |
45c01e82 GH |
767 | return; |
768 | } | |
769 | ||
770 | #ifdef CONFIG_SMP | |
771 | /* | |
772 | * If: | |
773 | * | |
774 | * - the newly woken task is of equal priority to the current task | |
775 | * - the newly woken task is non-migratable while current is migratable | |
776 | * - current will be preempted on the next reschedule | |
777 | * | |
778 | * we should check to see if current can readily move to a different | |
779 | * cpu. If so, we will reschedule to allow the push logic to try | |
780 | * to move current somewhere else, making room for our non-migratable | |
781 | * task. | |
782 | */ | |
783 | if((p->prio == rq->curr->prio) | |
784 | && p->rt.nr_cpus_allowed == 1 | |
20b6331b | 785 | && rq->curr->rt.nr_cpus_allowed != 1) { |
45c01e82 GH |
786 | cpumask_t mask; |
787 | ||
788 | if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask)) | |
789 | /* | |
790 | * There appears to be other cpus that can accept | |
791 | * current, so lets reschedule to try and push it away | |
792 | */ | |
793 | resched_task(rq->curr); | |
794 | } | |
795 | #endif | |
bb44e5d1 IM |
796 | } |
797 | ||
6f505b16 PZ |
798 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
799 | struct rt_rq *rt_rq) | |
bb44e5d1 | 800 | { |
6f505b16 PZ |
801 | struct rt_prio_array *array = &rt_rq->active; |
802 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
803 | struct list_head *queue; |
804 | int idx; | |
805 | ||
806 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 807 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 | 808 | |
20b6331b DA |
809 | queue = array->queue + idx; |
810 | next = list_entry(queue->next, struct sched_rt_entity, run_list); | |
326587b8 | 811 | |
6f505b16 PZ |
812 | return next; |
813 | } | |
bb44e5d1 | 814 | |
6f505b16 PZ |
815 | static struct task_struct *pick_next_task_rt(struct rq *rq) |
816 | { | |
817 | struct sched_rt_entity *rt_se; | |
818 | struct task_struct *p; | |
819 | struct rt_rq *rt_rq; | |
bb44e5d1 | 820 | |
6f505b16 PZ |
821 | rt_rq = &rq->rt; |
822 | ||
823 | if (unlikely(!rt_rq->rt_nr_running)) | |
824 | return NULL; | |
825 | ||
23b0fdfc | 826 | if (rt_rq_throttled(rt_rq)) |
6f505b16 PZ |
827 | return NULL; |
828 | ||
829 | do { | |
830 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 831 | BUG_ON(!rt_se); |
6f505b16 PZ |
832 | rt_rq = group_rt_rq(rt_se); |
833 | } while (rt_rq); | |
834 | ||
835 | p = rt_task_of(rt_se); | |
836 | p->se.exec_start = rq->clock; | |
837 | return p; | |
bb44e5d1 IM |
838 | } |
839 | ||
31ee529c | 840 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 841 | { |
f1e14ef6 | 842 | update_curr_rt(rq); |
bb44e5d1 IM |
843 | p->se.exec_start = 0; |
844 | } | |
845 | ||
681f3e68 | 846 | #ifdef CONFIG_SMP |
6f505b16 | 847 | |
e8fa1362 SR |
848 | /* Only try algorithms three times */ |
849 | #define RT_MAX_TRIES 3 | |
850 | ||
851 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest); | |
852 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); | |
853 | ||
f65eda4f SR |
854 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
855 | { | |
856 | if (!task_running(rq, p) && | |
73fe6aae | 857 | (cpu < 0 || cpu_isset(cpu, p->cpus_allowed)) && |
6f505b16 | 858 | (p->rt.nr_cpus_allowed > 1)) |
f65eda4f SR |
859 | return 1; |
860 | return 0; | |
861 | } | |
862 | ||
e8fa1362 | 863 | /* Return the second highest RT task, NULL otherwise */ |
79064fbf | 864 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
e8fa1362 | 865 | { |
6f505b16 PZ |
866 | struct task_struct *next = NULL; |
867 | struct sched_rt_entity *rt_se; | |
868 | struct rt_prio_array *array; | |
869 | struct rt_rq *rt_rq; | |
e8fa1362 SR |
870 | int idx; |
871 | ||
6f505b16 PZ |
872 | for_each_leaf_rt_rq(rt_rq, rq) { |
873 | array = &rt_rq->active; | |
874 | idx = sched_find_first_bit(array->bitmap); | |
875 | next_idx: | |
876 | if (idx >= MAX_RT_PRIO) | |
877 | continue; | |
878 | if (next && next->prio < idx) | |
879 | continue; | |
20b6331b | 880 | list_for_each_entry(rt_se, array->queue + idx, run_list) { |
6f505b16 PZ |
881 | struct task_struct *p = rt_task_of(rt_se); |
882 | if (pick_rt_task(rq, p, cpu)) { | |
883 | next = p; | |
884 | break; | |
885 | } | |
886 | } | |
887 | if (!next) { | |
888 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); | |
889 | goto next_idx; | |
890 | } | |
f65eda4f SR |
891 | } |
892 | ||
e8fa1362 SR |
893 | return next; |
894 | } | |
895 | ||
896 | static DEFINE_PER_CPU(cpumask_t, local_cpu_mask); | |
897 | ||
6e1254d2 GH |
898 | static inline int pick_optimal_cpu(int this_cpu, cpumask_t *mask) |
899 | { | |
900 | int first; | |
901 | ||
902 | /* "this_cpu" is cheaper to preempt than a remote processor */ | |
903 | if ((this_cpu != -1) && cpu_isset(this_cpu, *mask)) | |
904 | return this_cpu; | |
905 | ||
906 | first = first_cpu(*mask); | |
907 | if (first != NR_CPUS) | |
908 | return first; | |
909 | ||
910 | return -1; | |
911 | } | |
912 | ||
913 | static int find_lowest_rq(struct task_struct *task) | |
914 | { | |
915 | struct sched_domain *sd; | |
916 | cpumask_t *lowest_mask = &__get_cpu_var(local_cpu_mask); | |
917 | int this_cpu = smp_processor_id(); | |
918 | int cpu = task_cpu(task); | |
06f90dbd | 919 | |
6e0534f2 GH |
920 | if (task->rt.nr_cpus_allowed == 1) |
921 | return -1; /* No other targets possible */ | |
6e1254d2 | 922 | |
6e0534f2 GH |
923 | if (!cpupri_find(&task_rq(task)->rd->cpupri, task, lowest_mask)) |
924 | return -1; /* No targets found */ | |
6e1254d2 GH |
925 | |
926 | /* | |
927 | * At this point we have built a mask of cpus representing the | |
928 | * lowest priority tasks in the system. Now we want to elect | |
929 | * the best one based on our affinity and topology. | |
930 | * | |
931 | * We prioritize the last cpu that the task executed on since | |
932 | * it is most likely cache-hot in that location. | |
933 | */ | |
934 | if (cpu_isset(cpu, *lowest_mask)) | |
935 | return cpu; | |
936 | ||
937 | /* | |
938 | * Otherwise, we consult the sched_domains span maps to figure | |
939 | * out which cpu is logically closest to our hot cache data. | |
940 | */ | |
941 | if (this_cpu == cpu) | |
942 | this_cpu = -1; /* Skip this_cpu opt if the same */ | |
943 | ||
944 | for_each_domain(cpu, sd) { | |
945 | if (sd->flags & SD_WAKE_AFFINE) { | |
946 | cpumask_t domain_mask; | |
947 | int best_cpu; | |
948 | ||
949 | cpus_and(domain_mask, sd->span, *lowest_mask); | |
950 | ||
951 | best_cpu = pick_optimal_cpu(this_cpu, | |
952 | &domain_mask); | |
953 | if (best_cpu != -1) | |
954 | return best_cpu; | |
955 | } | |
956 | } | |
957 | ||
958 | /* | |
959 | * And finally, if there were no matches within the domains | |
960 | * just give the caller *something* to work with from the compatible | |
961 | * locations. | |
962 | */ | |
963 | return pick_optimal_cpu(this_cpu, lowest_mask); | |
07b4032c GH |
964 | } |
965 | ||
966 | /* Will lock the rq it finds */ | |
4df64c0b | 967 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
968 | { |
969 | struct rq *lowest_rq = NULL; | |
07b4032c | 970 | int tries; |
4df64c0b | 971 | int cpu; |
e8fa1362 | 972 | |
07b4032c GH |
973 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
974 | cpu = find_lowest_rq(task); | |
975 | ||
2de0b463 | 976 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
977 | break; |
978 | ||
07b4032c GH |
979 | lowest_rq = cpu_rq(cpu); |
980 | ||
e8fa1362 | 981 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 982 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
983 | /* |
984 | * We had to unlock the run queue. In | |
985 | * the mean time, task could have | |
986 | * migrated already or had its affinity changed. | |
987 | * Also make sure that it wasn't scheduled on its rq. | |
988 | */ | |
07b4032c | 989 | if (unlikely(task_rq(task) != rq || |
4df64c0b IM |
990 | !cpu_isset(lowest_rq->cpu, |
991 | task->cpus_allowed) || | |
07b4032c | 992 | task_running(rq, task) || |
e8fa1362 | 993 | !task->se.on_rq)) { |
4df64c0b | 994 | |
e8fa1362 SR |
995 | spin_unlock(&lowest_rq->lock); |
996 | lowest_rq = NULL; | |
997 | break; | |
998 | } | |
999 | } | |
1000 | ||
1001 | /* If this rq is still suitable use it. */ | |
1002 | if (lowest_rq->rt.highest_prio > task->prio) | |
1003 | break; | |
1004 | ||
1005 | /* try again */ | |
1006 | spin_unlock(&lowest_rq->lock); | |
1007 | lowest_rq = NULL; | |
1008 | } | |
1009 | ||
1010 | return lowest_rq; | |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * If the current CPU has more than one RT task, see if the non | |
1015 | * running task can migrate over to a CPU that is running a task | |
1016 | * of lesser priority. | |
1017 | */ | |
697f0a48 | 1018 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
1019 | { |
1020 | struct task_struct *next_task; | |
1021 | struct rq *lowest_rq; | |
1022 | int ret = 0; | |
1023 | int paranoid = RT_MAX_TRIES; | |
1024 | ||
a22d7fc1 GH |
1025 | if (!rq->rt.overloaded) |
1026 | return 0; | |
1027 | ||
697f0a48 | 1028 | next_task = pick_next_highest_task_rt(rq, -1); |
e8fa1362 SR |
1029 | if (!next_task) |
1030 | return 0; | |
1031 | ||
1032 | retry: | |
697f0a48 | 1033 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 1034 | WARN_ON(1); |
e8fa1362 | 1035 | return 0; |
f65eda4f | 1036 | } |
e8fa1362 SR |
1037 | |
1038 | /* | |
1039 | * It's possible that the next_task slipped in of | |
1040 | * higher priority than current. If that's the case | |
1041 | * just reschedule current. | |
1042 | */ | |
697f0a48 GH |
1043 | if (unlikely(next_task->prio < rq->curr->prio)) { |
1044 | resched_task(rq->curr); | |
e8fa1362 SR |
1045 | return 0; |
1046 | } | |
1047 | ||
697f0a48 | 1048 | /* We might release rq lock */ |
e8fa1362 SR |
1049 | get_task_struct(next_task); |
1050 | ||
1051 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 1052 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
1053 | if (!lowest_rq) { |
1054 | struct task_struct *task; | |
1055 | /* | |
697f0a48 | 1056 | * find lock_lowest_rq releases rq->lock |
e8fa1362 SR |
1057 | * so it is possible that next_task has changed. |
1058 | * If it has, then try again. | |
1059 | */ | |
697f0a48 | 1060 | task = pick_next_highest_task_rt(rq, -1); |
e8fa1362 SR |
1061 | if (unlikely(task != next_task) && task && paranoid--) { |
1062 | put_task_struct(next_task); | |
1063 | next_task = task; | |
1064 | goto retry; | |
1065 | } | |
1066 | goto out; | |
1067 | } | |
1068 | ||
697f0a48 | 1069 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
1070 | set_task_cpu(next_task, lowest_rq->cpu); |
1071 | activate_task(lowest_rq, next_task, 0); | |
1072 | ||
1073 | resched_task(lowest_rq->curr); | |
1074 | ||
1075 | spin_unlock(&lowest_rq->lock); | |
1076 | ||
1077 | ret = 1; | |
1078 | out: | |
1079 | put_task_struct(next_task); | |
1080 | ||
1081 | return ret; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * TODO: Currently we just use the second highest prio task on | |
1086 | * the queue, and stop when it can't migrate (or there's | |
1087 | * no more RT tasks). There may be a case where a lower | |
1088 | * priority RT task has a different affinity than the | |
1089 | * higher RT task. In this case the lower RT task could | |
1090 | * possibly be able to migrate where as the higher priority | |
1091 | * RT task could not. We currently ignore this issue. | |
1092 | * Enhancements are welcome! | |
1093 | */ | |
1094 | static void push_rt_tasks(struct rq *rq) | |
1095 | { | |
1096 | /* push_rt_task will return true if it moved an RT */ | |
1097 | while (push_rt_task(rq)) | |
1098 | ; | |
1099 | } | |
1100 | ||
f65eda4f SR |
1101 | static int pull_rt_task(struct rq *this_rq) |
1102 | { | |
80bf3171 IM |
1103 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
1104 | struct task_struct *p, *next; | |
f65eda4f | 1105 | struct rq *src_rq; |
f65eda4f | 1106 | |
637f5085 | 1107 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
1108 | return 0; |
1109 | ||
1110 | next = pick_next_task_rt(this_rq); | |
1111 | ||
637f5085 | 1112 | for_each_cpu_mask(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
1113 | if (this_cpu == cpu) |
1114 | continue; | |
1115 | ||
1116 | src_rq = cpu_rq(cpu); | |
f65eda4f SR |
1117 | /* |
1118 | * We can potentially drop this_rq's lock in | |
1119 | * double_lock_balance, and another CPU could | |
1120 | * steal our next task - hence we must cause | |
1121 | * the caller to recalculate the next task | |
1122 | * in that case: | |
1123 | */ | |
1124 | if (double_lock_balance(this_rq, src_rq)) { | |
1125 | struct task_struct *old_next = next; | |
80bf3171 | 1126 | |
f65eda4f SR |
1127 | next = pick_next_task_rt(this_rq); |
1128 | if (next != old_next) | |
1129 | ret = 1; | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * Are there still pullable RT tasks? | |
1134 | */ | |
614ee1f6 MG |
1135 | if (src_rq->rt.rt_nr_running <= 1) |
1136 | goto skip; | |
f65eda4f | 1137 | |
f65eda4f SR |
1138 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
1139 | ||
1140 | /* | |
1141 | * Do we have an RT task that preempts | |
1142 | * the to-be-scheduled task? | |
1143 | */ | |
1144 | if (p && (!next || (p->prio < next->prio))) { | |
1145 | WARN_ON(p == src_rq->curr); | |
1146 | WARN_ON(!p->se.on_rq); | |
1147 | ||
1148 | /* | |
1149 | * There's a chance that p is higher in priority | |
1150 | * than what's currently running on its cpu. | |
1151 | * This is just that p is wakeing up and hasn't | |
1152 | * had a chance to schedule. We only pull | |
1153 | * p if it is lower in priority than the | |
1154 | * current task on the run queue or | |
1155 | * this_rq next task is lower in prio than | |
1156 | * the current task on that rq. | |
1157 | */ | |
1158 | if (p->prio < src_rq->curr->prio || | |
1159 | (next && next->prio < src_rq->curr->prio)) | |
614ee1f6 | 1160 | goto skip; |
f65eda4f SR |
1161 | |
1162 | ret = 1; | |
1163 | ||
1164 | deactivate_task(src_rq, p, 0); | |
1165 | set_task_cpu(p, this_cpu); | |
1166 | activate_task(this_rq, p, 0); | |
1167 | /* | |
1168 | * We continue with the search, just in | |
1169 | * case there's an even higher prio task | |
1170 | * in another runqueue. (low likelyhood | |
1171 | * but possible) | |
80bf3171 | 1172 | * |
f65eda4f SR |
1173 | * Update next so that we won't pick a task |
1174 | * on another cpu with a priority lower (or equal) | |
1175 | * than the one we just picked. | |
1176 | */ | |
1177 | next = p; | |
1178 | ||
1179 | } | |
614ee1f6 | 1180 | skip: |
f65eda4f SR |
1181 | spin_unlock(&src_rq->lock); |
1182 | } | |
1183 | ||
1184 | return ret; | |
1185 | } | |
1186 | ||
9a897c5a | 1187 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
f65eda4f SR |
1188 | { |
1189 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
7f51f298 | 1190 | if (unlikely(rt_task(prev)) && rq->rt.highest_prio > prev->prio) |
f65eda4f SR |
1191 | pull_rt_task(rq); |
1192 | } | |
1193 | ||
9a897c5a | 1194 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 SR |
1195 | { |
1196 | /* | |
1197 | * If we have more than one rt_task queued, then | |
1198 | * see if we can push the other rt_tasks off to other CPUS. | |
1199 | * Note we may release the rq lock, and since | |
1200 | * the lock was owned by prev, we need to release it | |
1201 | * first via finish_lock_switch and then reaquire it here. | |
1202 | */ | |
a22d7fc1 | 1203 | if (unlikely(rq->rt.overloaded)) { |
e8fa1362 SR |
1204 | spin_lock_irq(&rq->lock); |
1205 | push_rt_tasks(rq); | |
1206 | spin_unlock_irq(&rq->lock); | |
1207 | } | |
1208 | } | |
1209 | ||
8ae121ac GH |
1210 | /* |
1211 | * If we are not running and we are not going to reschedule soon, we should | |
1212 | * try to push tasks away now | |
1213 | */ | |
9a897c5a | 1214 | static void task_wake_up_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1215 | { |
9a897c5a | 1216 | if (!task_running(rq, p) && |
8ae121ac | 1217 | !test_tsk_need_resched(rq->curr) && |
a22d7fc1 | 1218 | rq->rt.overloaded) |
4642dafd SR |
1219 | push_rt_tasks(rq); |
1220 | } | |
1221 | ||
43010659 | 1222 | static unsigned long |
bb44e5d1 | 1223 | load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f PW |
1224 | unsigned long max_load_move, |
1225 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1226 | int *all_pinned, int *this_best_prio) | |
bb44e5d1 | 1227 | { |
c7a1e46a SR |
1228 | /* don't touch RT tasks */ |
1229 | return 0; | |
e1d1484f PW |
1230 | } |
1231 | ||
1232 | static int | |
1233 | move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1234 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1235 | { | |
c7a1e46a SR |
1236 | /* don't touch RT tasks */ |
1237 | return 0; | |
bb44e5d1 | 1238 | } |
deeeccd4 | 1239 | |
cd8ba7cd MT |
1240 | static void set_cpus_allowed_rt(struct task_struct *p, |
1241 | const cpumask_t *new_mask) | |
73fe6aae GH |
1242 | { |
1243 | int weight = cpus_weight(*new_mask); | |
1244 | ||
1245 | BUG_ON(!rt_task(p)); | |
1246 | ||
1247 | /* | |
1248 | * Update the migration status of the RQ if we have an RT task | |
1249 | * which is running AND changing its weight value. | |
1250 | */ | |
6f505b16 | 1251 | if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) { |
73fe6aae GH |
1252 | struct rq *rq = task_rq(p); |
1253 | ||
6f505b16 | 1254 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
73fe6aae | 1255 | rq->rt.rt_nr_migratory++; |
6f505b16 | 1256 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
73fe6aae GH |
1257 | BUG_ON(!rq->rt.rt_nr_migratory); |
1258 | rq->rt.rt_nr_migratory--; | |
1259 | } | |
1260 | ||
1261 | update_rt_migration(rq); | |
1262 | } | |
1263 | ||
1264 | p->cpus_allowed = *new_mask; | |
6f505b16 | 1265 | p->rt.nr_cpus_allowed = weight; |
73fe6aae | 1266 | } |
deeeccd4 | 1267 | |
bdd7c81b | 1268 | /* Assumes rq->lock is held */ |
1f11eb6a | 1269 | static void rq_online_rt(struct rq *rq) |
bdd7c81b IM |
1270 | { |
1271 | if (rq->rt.overloaded) | |
1272 | rt_set_overload(rq); | |
6e0534f2 | 1273 | |
7def2be1 PZ |
1274 | __enable_runtime(rq); |
1275 | ||
6e0534f2 | 1276 | cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio); |
bdd7c81b IM |
1277 | } |
1278 | ||
1279 | /* Assumes rq->lock is held */ | |
1f11eb6a | 1280 | static void rq_offline_rt(struct rq *rq) |
bdd7c81b IM |
1281 | { |
1282 | if (rq->rt.overloaded) | |
1283 | rt_clear_overload(rq); | |
6e0534f2 | 1284 | |
7def2be1 PZ |
1285 | __disable_runtime(rq); |
1286 | ||
6e0534f2 | 1287 | cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_INVALID); |
bdd7c81b | 1288 | } |
cb469845 SR |
1289 | |
1290 | /* | |
1291 | * When switch from the rt queue, we bring ourselves to a position | |
1292 | * that we might want to pull RT tasks from other runqueues. | |
1293 | */ | |
1294 | static void switched_from_rt(struct rq *rq, struct task_struct *p, | |
1295 | int running) | |
1296 | { | |
1297 | /* | |
1298 | * If there are other RT tasks then we will reschedule | |
1299 | * and the scheduling of the other RT tasks will handle | |
1300 | * the balancing. But if we are the last RT task | |
1301 | * we may need to handle the pulling of RT tasks | |
1302 | * now. | |
1303 | */ | |
1304 | if (!rq->rt.rt_nr_running) | |
1305 | pull_rt_task(rq); | |
1306 | } | |
1307 | #endif /* CONFIG_SMP */ | |
1308 | ||
1309 | /* | |
1310 | * When switching a task to RT, we may overload the runqueue | |
1311 | * with RT tasks. In this case we try to push them off to | |
1312 | * other runqueues. | |
1313 | */ | |
1314 | static void switched_to_rt(struct rq *rq, struct task_struct *p, | |
1315 | int running) | |
1316 | { | |
1317 | int check_resched = 1; | |
1318 | ||
1319 | /* | |
1320 | * If we are already running, then there's nothing | |
1321 | * that needs to be done. But if we are not running | |
1322 | * we may need to preempt the current running task. | |
1323 | * If that current running task is also an RT task | |
1324 | * then see if we can move to another run queue. | |
1325 | */ | |
1326 | if (!running) { | |
1327 | #ifdef CONFIG_SMP | |
1328 | if (rq->rt.overloaded && push_rt_task(rq) && | |
1329 | /* Don't resched if we changed runqueues */ | |
1330 | rq != task_rq(p)) | |
1331 | check_resched = 0; | |
1332 | #endif /* CONFIG_SMP */ | |
1333 | if (check_resched && p->prio < rq->curr->prio) | |
1334 | resched_task(rq->curr); | |
1335 | } | |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * Priority of the task has changed. This may cause | |
1340 | * us to initiate a push or pull. | |
1341 | */ | |
1342 | static void prio_changed_rt(struct rq *rq, struct task_struct *p, | |
1343 | int oldprio, int running) | |
1344 | { | |
1345 | if (running) { | |
1346 | #ifdef CONFIG_SMP | |
1347 | /* | |
1348 | * If our priority decreases while running, we | |
1349 | * may need to pull tasks to this runqueue. | |
1350 | */ | |
1351 | if (oldprio < p->prio) | |
1352 | pull_rt_task(rq); | |
1353 | /* | |
1354 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
1355 | * then reschedule. Note, the above pull_rt_task |
1356 | * can release the rq lock and p could migrate. | |
1357 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 1358 | */ |
6fa46fa5 | 1359 | if (p->prio > rq->rt.highest_prio && rq->curr == p) |
cb469845 SR |
1360 | resched_task(p); |
1361 | #else | |
1362 | /* For UP simply resched on drop of prio */ | |
1363 | if (oldprio < p->prio) | |
1364 | resched_task(p); | |
e8fa1362 | 1365 | #endif /* CONFIG_SMP */ |
cb469845 SR |
1366 | } else { |
1367 | /* | |
1368 | * This task is not running, but if it is | |
1369 | * greater than the current running task | |
1370 | * then reschedule. | |
1371 | */ | |
1372 | if (p->prio < rq->curr->prio) | |
1373 | resched_task(rq->curr); | |
1374 | } | |
1375 | } | |
1376 | ||
78f2c7db PZ |
1377 | static void watchdog(struct rq *rq, struct task_struct *p) |
1378 | { | |
1379 | unsigned long soft, hard; | |
1380 | ||
1381 | if (!p->signal) | |
1382 | return; | |
1383 | ||
1384 | soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur; | |
1385 | hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max; | |
1386 | ||
1387 | if (soft != RLIM_INFINITY) { | |
1388 | unsigned long next; | |
1389 | ||
1390 | p->rt.timeout++; | |
1391 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); | |
5a52dd50 | 1392 | if (p->rt.timeout > next) |
78f2c7db PZ |
1393 | p->it_sched_expires = p->se.sum_exec_runtime; |
1394 | } | |
1395 | } | |
bb44e5d1 | 1396 | |
8f4d37ec | 1397 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 1398 | { |
67e2be02 PZ |
1399 | update_curr_rt(rq); |
1400 | ||
78f2c7db PZ |
1401 | watchdog(rq, p); |
1402 | ||
bb44e5d1 IM |
1403 | /* |
1404 | * RR tasks need a special form of timeslice management. | |
1405 | * FIFO tasks have no timeslices. | |
1406 | */ | |
1407 | if (p->policy != SCHED_RR) | |
1408 | return; | |
1409 | ||
fa717060 | 1410 | if (--p->rt.time_slice) |
bb44e5d1 IM |
1411 | return; |
1412 | ||
fa717060 | 1413 | p->rt.time_slice = DEF_TIMESLICE; |
bb44e5d1 | 1414 | |
98fbc798 DA |
1415 | /* |
1416 | * Requeue to the end of queue if we are not the only element | |
1417 | * on the queue: | |
1418 | */ | |
fa717060 | 1419 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
98fbc798 DA |
1420 | requeue_task_rt(rq, p); |
1421 | set_tsk_need_resched(p); | |
1422 | } | |
bb44e5d1 IM |
1423 | } |
1424 | ||
83b699ed SV |
1425 | static void set_curr_task_rt(struct rq *rq) |
1426 | { | |
1427 | struct task_struct *p = rq->curr; | |
1428 | ||
1429 | p->se.exec_start = rq->clock; | |
1430 | } | |
1431 | ||
2abdad0a | 1432 | static const struct sched_class rt_sched_class = { |
5522d5d5 | 1433 | .next = &fair_sched_class, |
bb44e5d1 IM |
1434 | .enqueue_task = enqueue_task_rt, |
1435 | .dequeue_task = dequeue_task_rt, | |
1436 | .yield_task = yield_task_rt, | |
e7693a36 GH |
1437 | #ifdef CONFIG_SMP |
1438 | .select_task_rq = select_task_rq_rt, | |
1439 | #endif /* CONFIG_SMP */ | |
bb44e5d1 IM |
1440 | |
1441 | .check_preempt_curr = check_preempt_curr_rt, | |
1442 | ||
1443 | .pick_next_task = pick_next_task_rt, | |
1444 | .put_prev_task = put_prev_task_rt, | |
1445 | ||
681f3e68 | 1446 | #ifdef CONFIG_SMP |
bb44e5d1 | 1447 | .load_balance = load_balance_rt, |
e1d1484f | 1448 | .move_one_task = move_one_task_rt, |
73fe6aae | 1449 | .set_cpus_allowed = set_cpus_allowed_rt, |
1f11eb6a GH |
1450 | .rq_online = rq_online_rt, |
1451 | .rq_offline = rq_offline_rt, | |
9a897c5a SR |
1452 | .pre_schedule = pre_schedule_rt, |
1453 | .post_schedule = post_schedule_rt, | |
1454 | .task_wake_up = task_wake_up_rt, | |
cb469845 | 1455 | .switched_from = switched_from_rt, |
681f3e68 | 1456 | #endif |
bb44e5d1 | 1457 | |
83b699ed | 1458 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 1459 | .task_tick = task_tick_rt, |
cb469845 SR |
1460 | |
1461 | .prio_changed = prio_changed_rt, | |
1462 | .switched_to = switched_to_rt, | |
bb44e5d1 | 1463 | }; |
ada18de2 PZ |
1464 | |
1465 | #ifdef CONFIG_SCHED_DEBUG | |
1466 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
1467 | ||
1468 | static void print_rt_stats(struct seq_file *m, int cpu) | |
1469 | { | |
1470 | struct rt_rq *rt_rq; | |
1471 | ||
1472 | rcu_read_lock(); | |
1473 | for_each_leaf_rt_rq(rt_rq, cpu_rq(cpu)) | |
1474 | print_rt_rq(m, cpu, rt_rq); | |
1475 | rcu_read_unlock(); | |
1476 | } | |
1477 | #endif |