]>
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 | { | |
637f5085 | 15 | cpu_set(rq->cpu, rq->rd->rto_mask); |
4fd29176 SR |
16 | /* |
17 | * Make sure the mask is visible before we set | |
18 | * the overload count. That is checked to determine | |
19 | * if we should look at the mask. It would be a shame | |
20 | * if we looked at the mask, but the mask was not | |
21 | * updated yet. | |
22 | */ | |
23 | wmb(); | |
637f5085 | 24 | atomic_inc(&rq->rd->rto_count); |
4fd29176 | 25 | } |
84de4274 | 26 | |
4fd29176 SR |
27 | static inline void rt_clear_overload(struct rq *rq) |
28 | { | |
29 | /* the order here really doesn't matter */ | |
637f5085 GH |
30 | atomic_dec(&rq->rd->rto_count); |
31 | cpu_clear(rq->cpu, rq->rd->rto_mask); | |
4fd29176 | 32 | } |
73fe6aae GH |
33 | |
34 | static void update_rt_migration(struct rq *rq) | |
35 | { | |
637f5085 | 36 | if (rq->rt.rt_nr_migratory && (rq->rt.rt_nr_running > 1)) { |
cdc8eb98 GH |
37 | if (!rq->rt.overloaded) { |
38 | rt_set_overload(rq); | |
39 | rq->rt.overloaded = 1; | |
40 | } | |
41 | } else if (rq->rt.overloaded) { | |
73fe6aae | 42 | rt_clear_overload(rq); |
637f5085 GH |
43 | rq->rt.overloaded = 0; |
44 | } | |
73fe6aae | 45 | } |
4fd29176 SR |
46 | #endif /* CONFIG_SMP */ |
47 | ||
6f505b16 | 48 | static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) |
fa85ae24 | 49 | { |
6f505b16 PZ |
50 | return container_of(rt_se, struct task_struct, rt); |
51 | } | |
52 | ||
53 | static inline int on_rt_rq(struct sched_rt_entity *rt_se) | |
54 | { | |
55 | return !list_empty(&rt_se->run_list); | |
56 | } | |
57 | ||
052f1dc7 | 58 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 59 | |
9f0c1e56 | 60 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) |
6f505b16 PZ |
61 | { |
62 | if (!rt_rq->tg) | |
9f0c1e56 | 63 | return RUNTIME_INF; |
6f505b16 | 64 | |
d0b27fa7 | 65 | return rt_rq->tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
66 | } |
67 | ||
68 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
69 | list_for_each_entry(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) | |
70 | ||
71 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) | |
72 | { | |
73 | return rt_rq->rq; | |
74 | } | |
75 | ||
76 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
77 | { | |
78 | return rt_se->rt_rq; | |
79 | } | |
80 | ||
81 | #define for_each_sched_rt_entity(rt_se) \ | |
82 | for (; rt_se; rt_se = rt_se->parent) | |
83 | ||
84 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
85 | { | |
86 | return rt_se->my_q; | |
87 | } | |
88 | ||
89 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se); | |
90 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se); | |
91 | ||
9f0c1e56 | 92 | static void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 PZ |
93 | { |
94 | struct sched_rt_entity *rt_se = rt_rq->rt_se; | |
95 | ||
96 | if (rt_se && !on_rt_rq(rt_se) && rt_rq->rt_nr_running) { | |
1020387f PZ |
97 | struct task_struct *curr = rq_of_rt_rq(rt_rq)->curr; |
98 | ||
6f505b16 | 99 | enqueue_rt_entity(rt_se); |
1020387f PZ |
100 | if (rt_rq->highest_prio < curr->prio) |
101 | resched_task(curr); | |
6f505b16 PZ |
102 | } |
103 | } | |
104 | ||
9f0c1e56 | 105 | static void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
106 | { |
107 | struct sched_rt_entity *rt_se = rt_rq->rt_se; | |
108 | ||
109 | if (rt_se && on_rt_rq(rt_se)) | |
110 | dequeue_rt_entity(rt_se); | |
111 | } | |
112 | ||
23b0fdfc PZ |
113 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
114 | { | |
115 | return rt_rq->rt_throttled && !rt_rq->rt_nr_boosted; | |
116 | } | |
117 | ||
118 | static int rt_se_boosted(struct sched_rt_entity *rt_se) | |
119 | { | |
120 | struct rt_rq *rt_rq = group_rt_rq(rt_se); | |
121 | struct task_struct *p; | |
122 | ||
123 | if (rt_rq) | |
124 | return !!rt_rq->rt_nr_boosted; | |
125 | ||
126 | p = rt_task_of(rt_se); | |
127 | return p->prio != p->normal_prio; | |
128 | } | |
129 | ||
d0b27fa7 PZ |
130 | #ifdef CONFIG_SMP |
131 | static inline cpumask_t sched_rt_period_mask(void) | |
132 | { | |
133 | return cpu_rq(smp_processor_id())->rd->span; | |
134 | } | |
6f505b16 | 135 | #else |
d0b27fa7 PZ |
136 | static inline cpumask_t sched_rt_period_mask(void) |
137 | { | |
138 | return cpu_online_map; | |
139 | } | |
140 | #endif | |
6f505b16 | 141 | |
d0b27fa7 PZ |
142 | static inline |
143 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
6f505b16 | 144 | { |
d0b27fa7 PZ |
145 | return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu]; |
146 | } | |
9f0c1e56 | 147 | |
d0b27fa7 PZ |
148 | #else |
149 | ||
150 | static inline u64 sched_rt_runtime(struct rt_rq *rt_rq) | |
151 | { | |
152 | return def_rt_bandwidth.rt_runtime; | |
6f505b16 PZ |
153 | } |
154 | ||
155 | #define for_each_leaf_rt_rq(rt_rq, rq) \ | |
156 | for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) | |
157 | ||
158 | static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq) | |
159 | { | |
160 | return container_of(rt_rq, struct rq, rt); | |
161 | } | |
162 | ||
163 | static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se) | |
164 | { | |
165 | struct task_struct *p = rt_task_of(rt_se); | |
166 | struct rq *rq = task_rq(p); | |
167 | ||
168 | return &rq->rt; | |
169 | } | |
170 | ||
171 | #define for_each_sched_rt_entity(rt_se) \ | |
172 | for (; rt_se; rt_se = NULL) | |
173 | ||
174 | static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se) | |
175 | { | |
176 | return NULL; | |
177 | } | |
178 | ||
9f0c1e56 | 179 | static inline void sched_rt_rq_enqueue(struct rt_rq *rt_rq) |
6f505b16 PZ |
180 | { |
181 | } | |
182 | ||
9f0c1e56 | 183 | static inline void sched_rt_rq_dequeue(struct rt_rq *rt_rq) |
6f505b16 PZ |
184 | { |
185 | } | |
186 | ||
23b0fdfc PZ |
187 | static inline int rt_rq_throttled(struct rt_rq *rt_rq) |
188 | { | |
189 | return rt_rq->rt_throttled; | |
190 | } | |
d0b27fa7 PZ |
191 | |
192 | static inline cpumask_t sched_rt_period_mask(void) | |
193 | { | |
194 | return cpu_online_map; | |
195 | } | |
196 | ||
197 | static inline | |
198 | struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu) | |
199 | { | |
200 | return &cpu_rq(cpu)->rt; | |
201 | } | |
202 | ||
6f505b16 PZ |
203 | #endif |
204 | ||
d0b27fa7 PZ |
205 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) |
206 | { | |
207 | int i, idle = 1; | |
208 | cpumask_t span; | |
209 | ||
210 | if (rt_b->rt_runtime == RUNTIME_INF) | |
211 | return 1; | |
212 | ||
213 | span = sched_rt_period_mask(); | |
214 | for_each_cpu_mask(i, span) { | |
215 | int enqueue = 0; | |
216 | struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i); | |
217 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
218 | ||
219 | spin_lock(&rq->lock); | |
220 | if (rt_rq->rt_time) { | |
221 | u64 runtime = rt_b->rt_runtime; | |
222 | ||
223 | rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime); | |
224 | if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) { | |
225 | rt_rq->rt_throttled = 0; | |
226 | enqueue = 1; | |
227 | } | |
228 | if (rt_rq->rt_time || rt_rq->rt_nr_running) | |
229 | idle = 0; | |
230 | } | |
231 | ||
232 | if (enqueue) | |
233 | sched_rt_rq_enqueue(rt_rq); | |
234 | spin_unlock(&rq->lock); | |
235 | } | |
236 | ||
237 | return idle; | |
238 | } | |
239 | ||
6f505b16 PZ |
240 | static inline int rt_se_prio(struct sched_rt_entity *rt_se) |
241 | { | |
052f1dc7 | 242 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
243 | struct rt_rq *rt_rq = group_rt_rq(rt_se); |
244 | ||
245 | if (rt_rq) | |
246 | return rt_rq->highest_prio; | |
247 | #endif | |
248 | ||
249 | return rt_task_of(rt_se)->prio; | |
250 | } | |
251 | ||
9f0c1e56 | 252 | static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq) |
6f505b16 | 253 | { |
9f0c1e56 | 254 | u64 runtime = sched_rt_runtime(rt_rq); |
fa85ae24 | 255 | |
9f0c1e56 | 256 | if (runtime == RUNTIME_INF) |
fa85ae24 PZ |
257 | return 0; |
258 | ||
259 | if (rt_rq->rt_throttled) | |
23b0fdfc | 260 | return rt_rq_throttled(rt_rq); |
fa85ae24 | 261 | |
9f0c1e56 | 262 | if (rt_rq->rt_time > runtime) { |
6f505b16 | 263 | rt_rq->rt_throttled = 1; |
23b0fdfc | 264 | if (rt_rq_throttled(rt_rq)) { |
9f0c1e56 | 265 | sched_rt_rq_dequeue(rt_rq); |
23b0fdfc PZ |
266 | return 1; |
267 | } | |
fa85ae24 PZ |
268 | } |
269 | ||
270 | return 0; | |
271 | } | |
272 | ||
bb44e5d1 IM |
273 | /* |
274 | * Update the current task's runtime statistics. Skip current tasks that | |
275 | * are not in our scheduling class. | |
276 | */ | |
a9957449 | 277 | static void update_curr_rt(struct rq *rq) |
bb44e5d1 IM |
278 | { |
279 | struct task_struct *curr = rq->curr; | |
6f505b16 PZ |
280 | struct sched_rt_entity *rt_se = &curr->rt; |
281 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
bb44e5d1 IM |
282 | u64 delta_exec; |
283 | ||
284 | if (!task_has_rt_policy(curr)) | |
285 | return; | |
286 | ||
d281918d | 287 | delta_exec = rq->clock - curr->se.exec_start; |
bb44e5d1 IM |
288 | if (unlikely((s64)delta_exec < 0)) |
289 | delta_exec = 0; | |
6cfb0d5d IM |
290 | |
291 | schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec)); | |
bb44e5d1 IM |
292 | |
293 | curr->se.sum_exec_runtime += delta_exec; | |
d281918d | 294 | curr->se.exec_start = rq->clock; |
d842de87 | 295 | cpuacct_charge(curr, delta_exec); |
fa85ae24 | 296 | |
6f505b16 | 297 | rt_rq->rt_time += delta_exec; |
9f0c1e56 | 298 | if (sched_rt_runtime_exceeded(rt_rq)) |
fa85ae24 | 299 | resched_task(curr); |
bb44e5d1 IM |
300 | } |
301 | ||
6f505b16 PZ |
302 | static inline |
303 | void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
63489e45 | 304 | { |
6f505b16 PZ |
305 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
306 | rt_rq->rt_nr_running++; | |
052f1dc7 | 307 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
308 | if (rt_se_prio(rt_se) < rt_rq->highest_prio) |
309 | rt_rq->highest_prio = rt_se_prio(rt_se); | |
310 | #endif | |
764a9d6f | 311 | #ifdef CONFIG_SMP |
6f505b16 PZ |
312 | if (rt_se->nr_cpus_allowed > 1) { |
313 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
73fe6aae | 314 | rq->rt.rt_nr_migratory++; |
6f505b16 | 315 | } |
73fe6aae | 316 | |
6f505b16 PZ |
317 | update_rt_migration(rq_of_rt_rq(rt_rq)); |
318 | #endif | |
052f1dc7 | 319 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
320 | if (rt_se_boosted(rt_se)) |
321 | rt_rq->rt_nr_boosted++; | |
d0b27fa7 PZ |
322 | |
323 | if (rt_rq->tg) | |
324 | start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); | |
325 | #else | |
326 | start_rt_bandwidth(&def_rt_bandwidth); | |
23b0fdfc | 327 | #endif |
63489e45 SR |
328 | } |
329 | ||
6f505b16 PZ |
330 | static inline |
331 | void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) | |
63489e45 | 332 | { |
6f505b16 PZ |
333 | WARN_ON(!rt_prio(rt_se_prio(rt_se))); |
334 | WARN_ON(!rt_rq->rt_nr_running); | |
335 | rt_rq->rt_nr_running--; | |
052f1dc7 | 336 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 | 337 | if (rt_rq->rt_nr_running) { |
764a9d6f SR |
338 | struct rt_prio_array *array; |
339 | ||
6f505b16 PZ |
340 | WARN_ON(rt_se_prio(rt_se) < rt_rq->highest_prio); |
341 | if (rt_se_prio(rt_se) == rt_rq->highest_prio) { | |
764a9d6f | 342 | /* recalculate */ |
6f505b16 PZ |
343 | array = &rt_rq->active; |
344 | rt_rq->highest_prio = | |
764a9d6f SR |
345 | sched_find_first_bit(array->bitmap); |
346 | } /* otherwise leave rq->highest prio alone */ | |
347 | } else | |
6f505b16 PZ |
348 | rt_rq->highest_prio = MAX_RT_PRIO; |
349 | #endif | |
350 | #ifdef CONFIG_SMP | |
351 | if (rt_se->nr_cpus_allowed > 1) { | |
352 | struct rq *rq = rq_of_rt_rq(rt_rq); | |
73fe6aae | 353 | rq->rt.rt_nr_migratory--; |
6f505b16 | 354 | } |
73fe6aae | 355 | |
6f505b16 | 356 | update_rt_migration(rq_of_rt_rq(rt_rq)); |
764a9d6f | 357 | #endif /* CONFIG_SMP */ |
052f1dc7 | 358 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
359 | if (rt_se_boosted(rt_se)) |
360 | rt_rq->rt_nr_boosted--; | |
361 | ||
362 | WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); | |
363 | #endif | |
63489e45 SR |
364 | } |
365 | ||
6f505b16 | 366 | static void enqueue_rt_entity(struct sched_rt_entity *rt_se) |
bb44e5d1 | 367 | { |
6f505b16 PZ |
368 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); |
369 | struct rt_prio_array *array = &rt_rq->active; | |
370 | struct rt_rq *group_rq = group_rt_rq(rt_se); | |
bb44e5d1 | 371 | |
23b0fdfc | 372 | if (group_rq && rt_rq_throttled(group_rq)) |
6f505b16 | 373 | return; |
63489e45 | 374 | |
6f505b16 PZ |
375 | list_add_tail(&rt_se->run_list, array->queue + rt_se_prio(rt_se)); |
376 | __set_bit(rt_se_prio(rt_se), array->bitmap); | |
78f2c7db | 377 | |
6f505b16 PZ |
378 | inc_rt_tasks(rt_se, rt_rq); |
379 | } | |
380 | ||
381 | static void dequeue_rt_entity(struct sched_rt_entity *rt_se) | |
382 | { | |
383 | struct rt_rq *rt_rq = rt_rq_of_se(rt_se); | |
384 | struct rt_prio_array *array = &rt_rq->active; | |
385 | ||
386 | list_del_init(&rt_se->run_list); | |
387 | if (list_empty(array->queue + rt_se_prio(rt_se))) | |
388 | __clear_bit(rt_se_prio(rt_se), array->bitmap); | |
389 | ||
390 | dec_rt_tasks(rt_se, rt_rq); | |
391 | } | |
392 | ||
393 | /* | |
394 | * Because the prio of an upper entry depends on the lower | |
395 | * entries, we must remove entries top - down. | |
396 | * | |
397 | * XXX: O(1/2 h^2) because we can only walk up, not down the chain. | |
398 | * doesn't matter much for now, as h=2 for GROUP_SCHED. | |
399 | */ | |
400 | static void dequeue_rt_stack(struct task_struct *p) | |
401 | { | |
402 | struct sched_rt_entity *rt_se, *top_se; | |
403 | ||
404 | /* | |
405 | * dequeue all, top - down. | |
406 | */ | |
407 | do { | |
408 | rt_se = &p->rt; | |
409 | top_se = NULL; | |
410 | for_each_sched_rt_entity(rt_se) { | |
411 | if (on_rt_rq(rt_se)) | |
412 | top_se = rt_se; | |
413 | } | |
414 | if (top_se) | |
415 | dequeue_rt_entity(top_se); | |
416 | } while (top_se); | |
bb44e5d1 IM |
417 | } |
418 | ||
419 | /* | |
420 | * Adding/removing a task to/from a priority array: | |
421 | */ | |
6f505b16 PZ |
422 | static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup) |
423 | { | |
424 | struct sched_rt_entity *rt_se = &p->rt; | |
425 | ||
426 | if (wakeup) | |
427 | rt_se->timeout = 0; | |
428 | ||
429 | dequeue_rt_stack(p); | |
430 | ||
431 | /* | |
432 | * enqueue everybody, bottom - up. | |
433 | */ | |
434 | for_each_sched_rt_entity(rt_se) | |
435 | enqueue_rt_entity(rt_se); | |
6f505b16 PZ |
436 | } |
437 | ||
f02231e5 | 438 | static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep) |
bb44e5d1 | 439 | { |
6f505b16 PZ |
440 | struct sched_rt_entity *rt_se = &p->rt; |
441 | struct rt_rq *rt_rq; | |
bb44e5d1 | 442 | |
f1e14ef6 | 443 | update_curr_rt(rq); |
bb44e5d1 | 444 | |
6f505b16 PZ |
445 | dequeue_rt_stack(p); |
446 | ||
447 | /* | |
448 | * re-enqueue all non-empty rt_rq entities. | |
449 | */ | |
450 | for_each_sched_rt_entity(rt_se) { | |
451 | rt_rq = group_rt_rq(rt_se); | |
452 | if (rt_rq && rt_rq->rt_nr_running) | |
453 | enqueue_rt_entity(rt_se); | |
454 | } | |
bb44e5d1 IM |
455 | } |
456 | ||
457 | /* | |
458 | * Put task to the end of the run list without the overhead of dequeue | |
459 | * followed by enqueue. | |
460 | */ | |
6f505b16 PZ |
461 | static |
462 | void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se) | |
463 | { | |
464 | struct rt_prio_array *array = &rt_rq->active; | |
465 | ||
466 | list_move_tail(&rt_se->run_list, array->queue + rt_se_prio(rt_se)); | |
467 | } | |
468 | ||
bb44e5d1 IM |
469 | static void requeue_task_rt(struct rq *rq, struct task_struct *p) |
470 | { | |
6f505b16 PZ |
471 | struct sched_rt_entity *rt_se = &p->rt; |
472 | struct rt_rq *rt_rq; | |
bb44e5d1 | 473 | |
6f505b16 PZ |
474 | for_each_sched_rt_entity(rt_se) { |
475 | rt_rq = rt_rq_of_se(rt_se); | |
476 | requeue_rt_entity(rt_rq, rt_se); | |
477 | } | |
bb44e5d1 IM |
478 | } |
479 | ||
6f505b16 | 480 | static void yield_task_rt(struct rq *rq) |
bb44e5d1 | 481 | { |
4530d7ab | 482 | requeue_task_rt(rq, rq->curr); |
bb44e5d1 IM |
483 | } |
484 | ||
e7693a36 | 485 | #ifdef CONFIG_SMP |
318e0893 GH |
486 | static int find_lowest_rq(struct task_struct *task); |
487 | ||
e7693a36 GH |
488 | static int select_task_rq_rt(struct task_struct *p, int sync) |
489 | { | |
318e0893 GH |
490 | struct rq *rq = task_rq(p); |
491 | ||
492 | /* | |
e1f47d89 SR |
493 | * If the current task is an RT task, then |
494 | * try to see if we can wake this RT task up on another | |
495 | * runqueue. Otherwise simply start this RT task | |
496 | * on its current runqueue. | |
497 | * | |
498 | * We want to avoid overloading runqueues. Even if | |
499 | * the RT task is of higher priority than the current RT task. | |
500 | * RT tasks behave differently than other tasks. If | |
501 | * one gets preempted, we try to push it off to another queue. | |
502 | * So trying to keep a preempting RT task on the same | |
503 | * cache hot CPU will force the running RT task to | |
504 | * a cold CPU. So we waste all the cache for the lower | |
505 | * RT task in hopes of saving some of a RT task | |
506 | * that is just being woken and probably will have | |
507 | * cold cache anyway. | |
318e0893 | 508 | */ |
17b3279b | 509 | if (unlikely(rt_task(rq->curr)) && |
6f505b16 | 510 | (p->rt.nr_cpus_allowed > 1)) { |
318e0893 GH |
511 | int cpu = find_lowest_rq(p); |
512 | ||
513 | return (cpu == -1) ? task_cpu(p) : cpu; | |
514 | } | |
515 | ||
516 | /* | |
517 | * Otherwise, just let it ride on the affined RQ and the | |
518 | * post-schedule router will push the preempted task away | |
519 | */ | |
e7693a36 GH |
520 | return task_cpu(p); |
521 | } | |
522 | #endif /* CONFIG_SMP */ | |
523 | ||
bb44e5d1 IM |
524 | /* |
525 | * Preempt the current task with a newly woken task if needed: | |
526 | */ | |
527 | static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p) | |
528 | { | |
529 | if (p->prio < rq->curr->prio) | |
530 | resched_task(rq->curr); | |
531 | } | |
532 | ||
6f505b16 PZ |
533 | static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq, |
534 | struct rt_rq *rt_rq) | |
bb44e5d1 | 535 | { |
6f505b16 PZ |
536 | struct rt_prio_array *array = &rt_rq->active; |
537 | struct sched_rt_entity *next = NULL; | |
bb44e5d1 IM |
538 | struct list_head *queue; |
539 | int idx; | |
540 | ||
541 | idx = sched_find_first_bit(array->bitmap); | |
6f505b16 | 542 | BUG_ON(idx >= MAX_RT_PRIO); |
bb44e5d1 IM |
543 | |
544 | queue = array->queue + idx; | |
6f505b16 | 545 | next = list_entry(queue->next, struct sched_rt_entity, run_list); |
326587b8 | 546 | |
6f505b16 PZ |
547 | return next; |
548 | } | |
bb44e5d1 | 549 | |
6f505b16 PZ |
550 | static struct task_struct *pick_next_task_rt(struct rq *rq) |
551 | { | |
552 | struct sched_rt_entity *rt_se; | |
553 | struct task_struct *p; | |
554 | struct rt_rq *rt_rq; | |
bb44e5d1 | 555 | |
6f505b16 PZ |
556 | rt_rq = &rq->rt; |
557 | ||
558 | if (unlikely(!rt_rq->rt_nr_running)) | |
559 | return NULL; | |
560 | ||
23b0fdfc | 561 | if (rt_rq_throttled(rt_rq)) |
6f505b16 PZ |
562 | return NULL; |
563 | ||
564 | do { | |
565 | rt_se = pick_next_rt_entity(rq, rt_rq); | |
326587b8 | 566 | BUG_ON(!rt_se); |
6f505b16 PZ |
567 | rt_rq = group_rt_rq(rt_se); |
568 | } while (rt_rq); | |
569 | ||
570 | p = rt_task_of(rt_se); | |
571 | p->se.exec_start = rq->clock; | |
572 | return p; | |
bb44e5d1 IM |
573 | } |
574 | ||
31ee529c | 575 | static void put_prev_task_rt(struct rq *rq, struct task_struct *p) |
bb44e5d1 | 576 | { |
f1e14ef6 | 577 | update_curr_rt(rq); |
bb44e5d1 IM |
578 | p->se.exec_start = 0; |
579 | } | |
580 | ||
681f3e68 | 581 | #ifdef CONFIG_SMP |
6f505b16 | 582 | |
e8fa1362 SR |
583 | /* Only try algorithms three times */ |
584 | #define RT_MAX_TRIES 3 | |
585 | ||
586 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest); | |
587 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep); | |
588 | ||
f65eda4f SR |
589 | static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu) |
590 | { | |
591 | if (!task_running(rq, p) && | |
73fe6aae | 592 | (cpu < 0 || cpu_isset(cpu, p->cpus_allowed)) && |
6f505b16 | 593 | (p->rt.nr_cpus_allowed > 1)) |
f65eda4f SR |
594 | return 1; |
595 | return 0; | |
596 | } | |
597 | ||
e8fa1362 | 598 | /* Return the second highest RT task, NULL otherwise */ |
79064fbf | 599 | static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu) |
e8fa1362 | 600 | { |
6f505b16 PZ |
601 | struct task_struct *next = NULL; |
602 | struct sched_rt_entity *rt_se; | |
603 | struct rt_prio_array *array; | |
604 | struct rt_rq *rt_rq; | |
e8fa1362 SR |
605 | int idx; |
606 | ||
6f505b16 PZ |
607 | for_each_leaf_rt_rq(rt_rq, rq) { |
608 | array = &rt_rq->active; | |
609 | idx = sched_find_first_bit(array->bitmap); | |
610 | next_idx: | |
611 | if (idx >= MAX_RT_PRIO) | |
612 | continue; | |
613 | if (next && next->prio < idx) | |
614 | continue; | |
615 | list_for_each_entry(rt_se, array->queue + idx, run_list) { | |
616 | struct task_struct *p = rt_task_of(rt_se); | |
617 | if (pick_rt_task(rq, p, cpu)) { | |
618 | next = p; | |
619 | break; | |
620 | } | |
621 | } | |
622 | if (!next) { | |
623 | idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1); | |
624 | goto next_idx; | |
625 | } | |
f65eda4f SR |
626 | } |
627 | ||
e8fa1362 SR |
628 | return next; |
629 | } | |
630 | ||
631 | static DEFINE_PER_CPU(cpumask_t, local_cpu_mask); | |
632 | ||
6e1254d2 | 633 | static int find_lowest_cpus(struct task_struct *task, cpumask_t *lowest_mask) |
e8fa1362 | 634 | { |
6e1254d2 | 635 | int lowest_prio = -1; |
610bf056 | 636 | int lowest_cpu = -1; |
06f90dbd | 637 | int count = 0; |
610bf056 | 638 | int cpu; |
e8fa1362 | 639 | |
637f5085 | 640 | cpus_and(*lowest_mask, task_rq(task)->rd->online, task->cpus_allowed); |
e8fa1362 | 641 | |
07b4032c GH |
642 | /* |
643 | * Scan each rq for the lowest prio. | |
644 | */ | |
610bf056 | 645 | for_each_cpu_mask(cpu, *lowest_mask) { |
07b4032c | 646 | struct rq *rq = cpu_rq(cpu); |
e8fa1362 | 647 | |
07b4032c GH |
648 | /* We look for lowest RT prio or non-rt CPU */ |
649 | if (rq->rt.highest_prio >= MAX_RT_PRIO) { | |
610bf056 SR |
650 | /* |
651 | * if we already found a low RT queue | |
652 | * and now we found this non-rt queue | |
653 | * clear the mask and set our bit. | |
654 | * Otherwise just return the queue as is | |
655 | * and the count==1 will cause the algorithm | |
656 | * to use the first bit found. | |
657 | */ | |
658 | if (lowest_cpu != -1) { | |
6e1254d2 | 659 | cpus_clear(*lowest_mask); |
610bf056 SR |
660 | cpu_set(rq->cpu, *lowest_mask); |
661 | } | |
6e1254d2 | 662 | return 1; |
07b4032c GH |
663 | } |
664 | ||
665 | /* no locking for now */ | |
6e1254d2 GH |
666 | if ((rq->rt.highest_prio > task->prio) |
667 | && (rq->rt.highest_prio >= lowest_prio)) { | |
668 | if (rq->rt.highest_prio > lowest_prio) { | |
669 | /* new low - clear old data */ | |
670 | lowest_prio = rq->rt.highest_prio; | |
610bf056 SR |
671 | lowest_cpu = cpu; |
672 | count = 0; | |
6e1254d2 | 673 | } |
06f90dbd | 674 | count++; |
610bf056 SR |
675 | } else |
676 | cpu_clear(cpu, *lowest_mask); | |
677 | } | |
678 | ||
679 | /* | |
680 | * Clear out all the set bits that represent | |
681 | * runqueues that were of higher prio than | |
682 | * the lowest_prio. | |
683 | */ | |
684 | if (lowest_cpu > 0) { | |
685 | /* | |
686 | * Perhaps we could add another cpumask op to | |
687 | * zero out bits. Like cpu_zero_bits(cpumask, nrbits); | |
688 | * Then that could be optimized to use memset and such. | |
689 | */ | |
690 | for_each_cpu_mask(cpu, *lowest_mask) { | |
691 | if (cpu >= lowest_cpu) | |
692 | break; | |
693 | cpu_clear(cpu, *lowest_mask); | |
e8fa1362 | 694 | } |
07b4032c GH |
695 | } |
696 | ||
06f90dbd | 697 | return count; |
6e1254d2 GH |
698 | } |
699 | ||
700 | static inline int pick_optimal_cpu(int this_cpu, cpumask_t *mask) | |
701 | { | |
702 | int first; | |
703 | ||
704 | /* "this_cpu" is cheaper to preempt than a remote processor */ | |
705 | if ((this_cpu != -1) && cpu_isset(this_cpu, *mask)) | |
706 | return this_cpu; | |
707 | ||
708 | first = first_cpu(*mask); | |
709 | if (first != NR_CPUS) | |
710 | return first; | |
711 | ||
712 | return -1; | |
713 | } | |
714 | ||
715 | static int find_lowest_rq(struct task_struct *task) | |
716 | { | |
717 | struct sched_domain *sd; | |
718 | cpumask_t *lowest_mask = &__get_cpu_var(local_cpu_mask); | |
719 | int this_cpu = smp_processor_id(); | |
720 | int cpu = task_cpu(task); | |
06f90dbd GH |
721 | int count = find_lowest_cpus(task, lowest_mask); |
722 | ||
723 | if (!count) | |
724 | return -1; /* No targets found */ | |
6e1254d2 | 725 | |
06f90dbd GH |
726 | /* |
727 | * There is no sense in performing an optimal search if only one | |
728 | * target is found. | |
729 | */ | |
730 | if (count == 1) | |
731 | return first_cpu(*lowest_mask); | |
6e1254d2 GH |
732 | |
733 | /* | |
734 | * At this point we have built a mask of cpus representing the | |
735 | * lowest priority tasks in the system. Now we want to elect | |
736 | * the best one based on our affinity and topology. | |
737 | * | |
738 | * We prioritize the last cpu that the task executed on since | |
739 | * it is most likely cache-hot in that location. | |
740 | */ | |
741 | if (cpu_isset(cpu, *lowest_mask)) | |
742 | return cpu; | |
743 | ||
744 | /* | |
745 | * Otherwise, we consult the sched_domains span maps to figure | |
746 | * out which cpu is logically closest to our hot cache data. | |
747 | */ | |
748 | if (this_cpu == cpu) | |
749 | this_cpu = -1; /* Skip this_cpu opt if the same */ | |
750 | ||
751 | for_each_domain(cpu, sd) { | |
752 | if (sd->flags & SD_WAKE_AFFINE) { | |
753 | cpumask_t domain_mask; | |
754 | int best_cpu; | |
755 | ||
756 | cpus_and(domain_mask, sd->span, *lowest_mask); | |
757 | ||
758 | best_cpu = pick_optimal_cpu(this_cpu, | |
759 | &domain_mask); | |
760 | if (best_cpu != -1) | |
761 | return best_cpu; | |
762 | } | |
763 | } | |
764 | ||
765 | /* | |
766 | * And finally, if there were no matches within the domains | |
767 | * just give the caller *something* to work with from the compatible | |
768 | * locations. | |
769 | */ | |
770 | return pick_optimal_cpu(this_cpu, lowest_mask); | |
07b4032c GH |
771 | } |
772 | ||
773 | /* Will lock the rq it finds */ | |
4df64c0b | 774 | static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq) |
07b4032c GH |
775 | { |
776 | struct rq *lowest_rq = NULL; | |
07b4032c | 777 | int tries; |
4df64c0b | 778 | int cpu; |
e8fa1362 | 779 | |
07b4032c GH |
780 | for (tries = 0; tries < RT_MAX_TRIES; tries++) { |
781 | cpu = find_lowest_rq(task); | |
782 | ||
2de0b463 | 783 | if ((cpu == -1) || (cpu == rq->cpu)) |
e8fa1362 SR |
784 | break; |
785 | ||
07b4032c GH |
786 | lowest_rq = cpu_rq(cpu); |
787 | ||
e8fa1362 | 788 | /* if the prio of this runqueue changed, try again */ |
07b4032c | 789 | if (double_lock_balance(rq, lowest_rq)) { |
e8fa1362 SR |
790 | /* |
791 | * We had to unlock the run queue. In | |
792 | * the mean time, task could have | |
793 | * migrated already or had its affinity changed. | |
794 | * Also make sure that it wasn't scheduled on its rq. | |
795 | */ | |
07b4032c | 796 | if (unlikely(task_rq(task) != rq || |
4df64c0b IM |
797 | !cpu_isset(lowest_rq->cpu, |
798 | task->cpus_allowed) || | |
07b4032c | 799 | task_running(rq, task) || |
e8fa1362 | 800 | !task->se.on_rq)) { |
4df64c0b | 801 | |
e8fa1362 SR |
802 | spin_unlock(&lowest_rq->lock); |
803 | lowest_rq = NULL; | |
804 | break; | |
805 | } | |
806 | } | |
807 | ||
808 | /* If this rq is still suitable use it. */ | |
809 | if (lowest_rq->rt.highest_prio > task->prio) | |
810 | break; | |
811 | ||
812 | /* try again */ | |
813 | spin_unlock(&lowest_rq->lock); | |
814 | lowest_rq = NULL; | |
815 | } | |
816 | ||
817 | return lowest_rq; | |
818 | } | |
819 | ||
820 | /* | |
821 | * If the current CPU has more than one RT task, see if the non | |
822 | * running task can migrate over to a CPU that is running a task | |
823 | * of lesser priority. | |
824 | */ | |
697f0a48 | 825 | static int push_rt_task(struct rq *rq) |
e8fa1362 SR |
826 | { |
827 | struct task_struct *next_task; | |
828 | struct rq *lowest_rq; | |
829 | int ret = 0; | |
830 | int paranoid = RT_MAX_TRIES; | |
831 | ||
a22d7fc1 GH |
832 | if (!rq->rt.overloaded) |
833 | return 0; | |
834 | ||
697f0a48 | 835 | next_task = pick_next_highest_task_rt(rq, -1); |
e8fa1362 SR |
836 | if (!next_task) |
837 | return 0; | |
838 | ||
839 | retry: | |
697f0a48 | 840 | if (unlikely(next_task == rq->curr)) { |
f65eda4f | 841 | WARN_ON(1); |
e8fa1362 | 842 | return 0; |
f65eda4f | 843 | } |
e8fa1362 SR |
844 | |
845 | /* | |
846 | * It's possible that the next_task slipped in of | |
847 | * higher priority than current. If that's the case | |
848 | * just reschedule current. | |
849 | */ | |
697f0a48 GH |
850 | if (unlikely(next_task->prio < rq->curr->prio)) { |
851 | resched_task(rq->curr); | |
e8fa1362 SR |
852 | return 0; |
853 | } | |
854 | ||
697f0a48 | 855 | /* We might release rq lock */ |
e8fa1362 SR |
856 | get_task_struct(next_task); |
857 | ||
858 | /* find_lock_lowest_rq locks the rq if found */ | |
697f0a48 | 859 | lowest_rq = find_lock_lowest_rq(next_task, rq); |
e8fa1362 SR |
860 | if (!lowest_rq) { |
861 | struct task_struct *task; | |
862 | /* | |
697f0a48 | 863 | * find lock_lowest_rq releases rq->lock |
e8fa1362 SR |
864 | * so it is possible that next_task has changed. |
865 | * If it has, then try again. | |
866 | */ | |
697f0a48 | 867 | task = pick_next_highest_task_rt(rq, -1); |
e8fa1362 SR |
868 | if (unlikely(task != next_task) && task && paranoid--) { |
869 | put_task_struct(next_task); | |
870 | next_task = task; | |
871 | goto retry; | |
872 | } | |
873 | goto out; | |
874 | } | |
875 | ||
697f0a48 | 876 | deactivate_task(rq, next_task, 0); |
e8fa1362 SR |
877 | set_task_cpu(next_task, lowest_rq->cpu); |
878 | activate_task(lowest_rq, next_task, 0); | |
879 | ||
880 | resched_task(lowest_rq->curr); | |
881 | ||
882 | spin_unlock(&lowest_rq->lock); | |
883 | ||
884 | ret = 1; | |
885 | out: | |
886 | put_task_struct(next_task); | |
887 | ||
888 | return ret; | |
889 | } | |
890 | ||
891 | /* | |
892 | * TODO: Currently we just use the second highest prio task on | |
893 | * the queue, and stop when it can't migrate (or there's | |
894 | * no more RT tasks). There may be a case where a lower | |
895 | * priority RT task has a different affinity than the | |
896 | * higher RT task. In this case the lower RT task could | |
897 | * possibly be able to migrate where as the higher priority | |
898 | * RT task could not. We currently ignore this issue. | |
899 | * Enhancements are welcome! | |
900 | */ | |
901 | static void push_rt_tasks(struct rq *rq) | |
902 | { | |
903 | /* push_rt_task will return true if it moved an RT */ | |
904 | while (push_rt_task(rq)) | |
905 | ; | |
906 | } | |
907 | ||
f65eda4f SR |
908 | static int pull_rt_task(struct rq *this_rq) |
909 | { | |
80bf3171 IM |
910 | int this_cpu = this_rq->cpu, ret = 0, cpu; |
911 | struct task_struct *p, *next; | |
f65eda4f | 912 | struct rq *src_rq; |
f65eda4f | 913 | |
637f5085 | 914 | if (likely(!rt_overloaded(this_rq))) |
f65eda4f SR |
915 | return 0; |
916 | ||
917 | next = pick_next_task_rt(this_rq); | |
918 | ||
637f5085 | 919 | for_each_cpu_mask(cpu, this_rq->rd->rto_mask) { |
f65eda4f SR |
920 | if (this_cpu == cpu) |
921 | continue; | |
922 | ||
923 | src_rq = cpu_rq(cpu); | |
f65eda4f SR |
924 | /* |
925 | * We can potentially drop this_rq's lock in | |
926 | * double_lock_balance, and another CPU could | |
927 | * steal our next task - hence we must cause | |
928 | * the caller to recalculate the next task | |
929 | * in that case: | |
930 | */ | |
931 | if (double_lock_balance(this_rq, src_rq)) { | |
932 | struct task_struct *old_next = next; | |
80bf3171 | 933 | |
f65eda4f SR |
934 | next = pick_next_task_rt(this_rq); |
935 | if (next != old_next) | |
936 | ret = 1; | |
937 | } | |
938 | ||
939 | /* | |
940 | * Are there still pullable RT tasks? | |
941 | */ | |
614ee1f6 MG |
942 | if (src_rq->rt.rt_nr_running <= 1) |
943 | goto skip; | |
f65eda4f | 944 | |
f65eda4f SR |
945 | p = pick_next_highest_task_rt(src_rq, this_cpu); |
946 | ||
947 | /* | |
948 | * Do we have an RT task that preempts | |
949 | * the to-be-scheduled task? | |
950 | */ | |
951 | if (p && (!next || (p->prio < next->prio))) { | |
952 | WARN_ON(p == src_rq->curr); | |
953 | WARN_ON(!p->se.on_rq); | |
954 | ||
955 | /* | |
956 | * There's a chance that p is higher in priority | |
957 | * than what's currently running on its cpu. | |
958 | * This is just that p is wakeing up and hasn't | |
959 | * had a chance to schedule. We only pull | |
960 | * p if it is lower in priority than the | |
961 | * current task on the run queue or | |
962 | * this_rq next task is lower in prio than | |
963 | * the current task on that rq. | |
964 | */ | |
965 | if (p->prio < src_rq->curr->prio || | |
966 | (next && next->prio < src_rq->curr->prio)) | |
614ee1f6 | 967 | goto skip; |
f65eda4f SR |
968 | |
969 | ret = 1; | |
970 | ||
971 | deactivate_task(src_rq, p, 0); | |
972 | set_task_cpu(p, this_cpu); | |
973 | activate_task(this_rq, p, 0); | |
974 | /* | |
975 | * We continue with the search, just in | |
976 | * case there's an even higher prio task | |
977 | * in another runqueue. (low likelyhood | |
978 | * but possible) | |
80bf3171 | 979 | * |
f65eda4f SR |
980 | * Update next so that we won't pick a task |
981 | * on another cpu with a priority lower (or equal) | |
982 | * than the one we just picked. | |
983 | */ | |
984 | next = p; | |
985 | ||
986 | } | |
614ee1f6 | 987 | skip: |
f65eda4f SR |
988 | spin_unlock(&src_rq->lock); |
989 | } | |
990 | ||
991 | return ret; | |
992 | } | |
993 | ||
9a897c5a | 994 | static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) |
f65eda4f SR |
995 | { |
996 | /* Try to pull RT tasks here if we lower this rq's prio */ | |
7f51f298 | 997 | if (unlikely(rt_task(prev)) && rq->rt.highest_prio > prev->prio) |
f65eda4f SR |
998 | pull_rt_task(rq); |
999 | } | |
1000 | ||
9a897c5a | 1001 | static void post_schedule_rt(struct rq *rq) |
e8fa1362 SR |
1002 | { |
1003 | /* | |
1004 | * If we have more than one rt_task queued, then | |
1005 | * see if we can push the other rt_tasks off to other CPUS. | |
1006 | * Note we may release the rq lock, and since | |
1007 | * the lock was owned by prev, we need to release it | |
1008 | * first via finish_lock_switch and then reaquire it here. | |
1009 | */ | |
a22d7fc1 | 1010 | if (unlikely(rq->rt.overloaded)) { |
e8fa1362 SR |
1011 | spin_lock_irq(&rq->lock); |
1012 | push_rt_tasks(rq); | |
1013 | spin_unlock_irq(&rq->lock); | |
1014 | } | |
1015 | } | |
1016 | ||
4642dafd | 1017 | |
9a897c5a | 1018 | static void task_wake_up_rt(struct rq *rq, struct task_struct *p) |
4642dafd | 1019 | { |
9a897c5a | 1020 | if (!task_running(rq, p) && |
a22d7fc1 GH |
1021 | (p->prio >= rq->rt.highest_prio) && |
1022 | rq->rt.overloaded) | |
4642dafd SR |
1023 | push_rt_tasks(rq); |
1024 | } | |
1025 | ||
43010659 | 1026 | static unsigned long |
bb44e5d1 | 1027 | load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f PW |
1028 | unsigned long max_load_move, |
1029 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1030 | int *all_pinned, int *this_best_prio) | |
bb44e5d1 | 1031 | { |
c7a1e46a SR |
1032 | /* don't touch RT tasks */ |
1033 | return 0; | |
e1d1484f PW |
1034 | } |
1035 | ||
1036 | static int | |
1037 | move_one_task_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1038 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1039 | { | |
c7a1e46a SR |
1040 | /* don't touch RT tasks */ |
1041 | return 0; | |
bb44e5d1 | 1042 | } |
deeeccd4 | 1043 | |
73fe6aae GH |
1044 | static void set_cpus_allowed_rt(struct task_struct *p, cpumask_t *new_mask) |
1045 | { | |
1046 | int weight = cpus_weight(*new_mask); | |
1047 | ||
1048 | BUG_ON(!rt_task(p)); | |
1049 | ||
1050 | /* | |
1051 | * Update the migration status of the RQ if we have an RT task | |
1052 | * which is running AND changing its weight value. | |
1053 | */ | |
6f505b16 | 1054 | if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) { |
73fe6aae GH |
1055 | struct rq *rq = task_rq(p); |
1056 | ||
6f505b16 | 1057 | if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { |
73fe6aae | 1058 | rq->rt.rt_nr_migratory++; |
6f505b16 | 1059 | } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { |
73fe6aae GH |
1060 | BUG_ON(!rq->rt.rt_nr_migratory); |
1061 | rq->rt.rt_nr_migratory--; | |
1062 | } | |
1063 | ||
1064 | update_rt_migration(rq); | |
1065 | } | |
1066 | ||
1067 | p->cpus_allowed = *new_mask; | |
6f505b16 | 1068 | p->rt.nr_cpus_allowed = weight; |
73fe6aae | 1069 | } |
deeeccd4 | 1070 | |
bdd7c81b IM |
1071 | /* Assumes rq->lock is held */ |
1072 | static void join_domain_rt(struct rq *rq) | |
1073 | { | |
1074 | if (rq->rt.overloaded) | |
1075 | rt_set_overload(rq); | |
1076 | } | |
1077 | ||
1078 | /* Assumes rq->lock is held */ | |
1079 | static void leave_domain_rt(struct rq *rq) | |
1080 | { | |
1081 | if (rq->rt.overloaded) | |
1082 | rt_clear_overload(rq); | |
1083 | } | |
cb469845 SR |
1084 | |
1085 | /* | |
1086 | * When switch from the rt queue, we bring ourselves to a position | |
1087 | * that we might want to pull RT tasks from other runqueues. | |
1088 | */ | |
1089 | static void switched_from_rt(struct rq *rq, struct task_struct *p, | |
1090 | int running) | |
1091 | { | |
1092 | /* | |
1093 | * If there are other RT tasks then we will reschedule | |
1094 | * and the scheduling of the other RT tasks will handle | |
1095 | * the balancing. But if we are the last RT task | |
1096 | * we may need to handle the pulling of RT tasks | |
1097 | * now. | |
1098 | */ | |
1099 | if (!rq->rt.rt_nr_running) | |
1100 | pull_rt_task(rq); | |
1101 | } | |
1102 | #endif /* CONFIG_SMP */ | |
1103 | ||
1104 | /* | |
1105 | * When switching a task to RT, we may overload the runqueue | |
1106 | * with RT tasks. In this case we try to push them off to | |
1107 | * other runqueues. | |
1108 | */ | |
1109 | static void switched_to_rt(struct rq *rq, struct task_struct *p, | |
1110 | int running) | |
1111 | { | |
1112 | int check_resched = 1; | |
1113 | ||
1114 | /* | |
1115 | * If we are already running, then there's nothing | |
1116 | * that needs to be done. But if we are not running | |
1117 | * we may need to preempt the current running task. | |
1118 | * If that current running task is also an RT task | |
1119 | * then see if we can move to another run queue. | |
1120 | */ | |
1121 | if (!running) { | |
1122 | #ifdef CONFIG_SMP | |
1123 | if (rq->rt.overloaded && push_rt_task(rq) && | |
1124 | /* Don't resched if we changed runqueues */ | |
1125 | rq != task_rq(p)) | |
1126 | check_resched = 0; | |
1127 | #endif /* CONFIG_SMP */ | |
1128 | if (check_resched && p->prio < rq->curr->prio) | |
1129 | resched_task(rq->curr); | |
1130 | } | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * Priority of the task has changed. This may cause | |
1135 | * us to initiate a push or pull. | |
1136 | */ | |
1137 | static void prio_changed_rt(struct rq *rq, struct task_struct *p, | |
1138 | int oldprio, int running) | |
1139 | { | |
1140 | if (running) { | |
1141 | #ifdef CONFIG_SMP | |
1142 | /* | |
1143 | * If our priority decreases while running, we | |
1144 | * may need to pull tasks to this runqueue. | |
1145 | */ | |
1146 | if (oldprio < p->prio) | |
1147 | pull_rt_task(rq); | |
1148 | /* | |
1149 | * If there's a higher priority task waiting to run | |
6fa46fa5 SR |
1150 | * then reschedule. Note, the above pull_rt_task |
1151 | * can release the rq lock and p could migrate. | |
1152 | * Only reschedule if p is still on the same runqueue. | |
cb469845 | 1153 | */ |
6fa46fa5 | 1154 | if (p->prio > rq->rt.highest_prio && rq->curr == p) |
cb469845 SR |
1155 | resched_task(p); |
1156 | #else | |
1157 | /* For UP simply resched on drop of prio */ | |
1158 | if (oldprio < p->prio) | |
1159 | resched_task(p); | |
e8fa1362 | 1160 | #endif /* CONFIG_SMP */ |
cb469845 SR |
1161 | } else { |
1162 | /* | |
1163 | * This task is not running, but if it is | |
1164 | * greater than the current running task | |
1165 | * then reschedule. | |
1166 | */ | |
1167 | if (p->prio < rq->curr->prio) | |
1168 | resched_task(rq->curr); | |
1169 | } | |
1170 | } | |
1171 | ||
78f2c7db PZ |
1172 | static void watchdog(struct rq *rq, struct task_struct *p) |
1173 | { | |
1174 | unsigned long soft, hard; | |
1175 | ||
1176 | if (!p->signal) | |
1177 | return; | |
1178 | ||
1179 | soft = p->signal->rlim[RLIMIT_RTTIME].rlim_cur; | |
1180 | hard = p->signal->rlim[RLIMIT_RTTIME].rlim_max; | |
1181 | ||
1182 | if (soft != RLIM_INFINITY) { | |
1183 | unsigned long next; | |
1184 | ||
1185 | p->rt.timeout++; | |
1186 | next = DIV_ROUND_UP(min(soft, hard), USEC_PER_SEC/HZ); | |
5a52dd50 | 1187 | if (p->rt.timeout > next) |
78f2c7db PZ |
1188 | p->it_sched_expires = p->se.sum_exec_runtime; |
1189 | } | |
1190 | } | |
bb44e5d1 | 1191 | |
8f4d37ec | 1192 | static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued) |
bb44e5d1 | 1193 | { |
67e2be02 PZ |
1194 | update_curr_rt(rq); |
1195 | ||
78f2c7db PZ |
1196 | watchdog(rq, p); |
1197 | ||
bb44e5d1 IM |
1198 | /* |
1199 | * RR tasks need a special form of timeslice management. | |
1200 | * FIFO tasks have no timeslices. | |
1201 | */ | |
1202 | if (p->policy != SCHED_RR) | |
1203 | return; | |
1204 | ||
fa717060 | 1205 | if (--p->rt.time_slice) |
bb44e5d1 IM |
1206 | return; |
1207 | ||
fa717060 | 1208 | p->rt.time_slice = DEF_TIMESLICE; |
bb44e5d1 | 1209 | |
98fbc798 DA |
1210 | /* |
1211 | * Requeue to the end of queue if we are not the only element | |
1212 | * on the queue: | |
1213 | */ | |
fa717060 | 1214 | if (p->rt.run_list.prev != p->rt.run_list.next) { |
98fbc798 DA |
1215 | requeue_task_rt(rq, p); |
1216 | set_tsk_need_resched(p); | |
1217 | } | |
bb44e5d1 IM |
1218 | } |
1219 | ||
83b699ed SV |
1220 | static void set_curr_task_rt(struct rq *rq) |
1221 | { | |
1222 | struct task_struct *p = rq->curr; | |
1223 | ||
1224 | p->se.exec_start = rq->clock; | |
1225 | } | |
1226 | ||
5522d5d5 IM |
1227 | const struct sched_class rt_sched_class = { |
1228 | .next = &fair_sched_class, | |
bb44e5d1 IM |
1229 | .enqueue_task = enqueue_task_rt, |
1230 | .dequeue_task = dequeue_task_rt, | |
1231 | .yield_task = yield_task_rt, | |
e7693a36 GH |
1232 | #ifdef CONFIG_SMP |
1233 | .select_task_rq = select_task_rq_rt, | |
1234 | #endif /* CONFIG_SMP */ | |
bb44e5d1 IM |
1235 | |
1236 | .check_preempt_curr = check_preempt_curr_rt, | |
1237 | ||
1238 | .pick_next_task = pick_next_task_rt, | |
1239 | .put_prev_task = put_prev_task_rt, | |
1240 | ||
681f3e68 | 1241 | #ifdef CONFIG_SMP |
bb44e5d1 | 1242 | .load_balance = load_balance_rt, |
e1d1484f | 1243 | .move_one_task = move_one_task_rt, |
73fe6aae | 1244 | .set_cpus_allowed = set_cpus_allowed_rt, |
bdd7c81b IM |
1245 | .join_domain = join_domain_rt, |
1246 | .leave_domain = leave_domain_rt, | |
9a897c5a SR |
1247 | .pre_schedule = pre_schedule_rt, |
1248 | .post_schedule = post_schedule_rt, | |
1249 | .task_wake_up = task_wake_up_rt, | |
cb469845 | 1250 | .switched_from = switched_from_rt, |
681f3e68 | 1251 | #endif |
bb44e5d1 | 1252 | |
83b699ed | 1253 | .set_curr_task = set_curr_task_rt, |
bb44e5d1 | 1254 | .task_tick = task_tick_rt, |
cb469845 SR |
1255 | |
1256 | .prio_changed = prio_changed_rt, | |
1257 | .switched_to = switched_to_rt, | |
bb44e5d1 | 1258 | }; |