Commit | Line | Data |
---|---|---|
bf0f6f24 IM |
1 | /* |
2 | * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH) | |
3 | * | |
4 | * Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
5 | * | |
6 | * Interactivity improvements by Mike Galbraith | |
7 | * (C) 2007 Mike Galbraith <efault@gmx.de> | |
8 | * | |
9 | * Various enhancements by Dmitry Adamushko. | |
10 | * (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com> | |
11 | * | |
12 | * Group scheduling enhancements by Srivatsa Vaddagiri | |
13 | * Copyright IBM Corporation, 2007 | |
14 | * Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> | |
15 | * | |
16 | * Scaled math optimizations by Thomas Gleixner | |
17 | * Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de> | |
21805085 PZ |
18 | * |
19 | * Adaptive scheduling granularity, math enhancements by Peter Zijlstra | |
20 | * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> | |
bf0f6f24 IM |
21 | */ |
22 | ||
23 | /* | |
21805085 PZ |
24 | * Targeted preemption latency for CPU-bound tasks: |
25 | * (default: 20ms, units: nanoseconds) | |
bf0f6f24 | 26 | * |
21805085 | 27 | * NOTE: this latency value is not the same as the concept of |
d274a4ce IM |
28 | * 'timeslice length' - timeslices in CFS are of variable length |
29 | * and have no persistent notion like in traditional, time-slice | |
30 | * based scheduling concepts. | |
bf0f6f24 | 31 | * |
d274a4ce IM |
32 | * (to see the precise effective timeslice length of your workload, |
33 | * run vmstat and monitor the context-switches (cs) field) | |
bf0f6f24 | 34 | */ |
2bd8e6d4 IM |
35 | const_debug unsigned int sysctl_sched_latency = 20000000ULL; |
36 | ||
37 | /* | |
38 | * After fork, child runs first. (default) If set to 0 then | |
39 | * parent will (try to) run first. | |
40 | */ | |
41 | const_debug unsigned int sysctl_sched_child_runs_first = 1; | |
21805085 PZ |
42 | |
43 | /* | |
44 | * Minimal preemption granularity for CPU-bound tasks: | |
45 | * (default: 2 msec, units: nanoseconds) | |
46 | */ | |
5f6d858e | 47 | const_debug unsigned int sysctl_sched_nr_latency = 20; |
bf0f6f24 | 48 | |
1799e35d IM |
49 | /* |
50 | * sys_sched_yield() compat mode | |
51 | * | |
52 | * This option switches the agressive yield implementation of the | |
53 | * old scheduler back on. | |
54 | */ | |
55 | unsigned int __read_mostly sysctl_sched_compat_yield; | |
56 | ||
bf0f6f24 IM |
57 | /* |
58 | * SCHED_BATCH wake-up granularity. | |
155bb293 | 59 | * (default: 10 msec, units: nanoseconds) |
bf0f6f24 IM |
60 | * |
61 | * This option delays the preemption effects of decoupled workloads | |
62 | * and reduces their over-scheduling. Synchronous workloads will still | |
63 | * have immediate wakeup/sleep latencies. | |
64 | */ | |
155bb293 | 65 | const_debug unsigned int sysctl_sched_batch_wakeup_granularity = 10000000UL; |
bf0f6f24 IM |
66 | |
67 | /* | |
68 | * SCHED_OTHER wake-up granularity. | |
155bb293 | 69 | * (default: 10 msec, units: nanoseconds) |
bf0f6f24 IM |
70 | * |
71 | * This option delays the preemption effects of decoupled workloads | |
72 | * and reduces their over-scheduling. Synchronous workloads will still | |
73 | * have immediate wakeup/sleep latencies. | |
74 | */ | |
155bb293 | 75 | const_debug unsigned int sysctl_sched_wakeup_granularity = 10000000UL; |
bf0f6f24 | 76 | |
da84d961 IM |
77 | const_debug unsigned int sysctl_sched_migration_cost = 500000UL; |
78 | ||
bf0f6f24 IM |
79 | /************************************************************** |
80 | * CFS operations on generic schedulable entities: | |
81 | */ | |
82 | ||
62160e3f | 83 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 84 | |
62160e3f | 85 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
86 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
87 | { | |
62160e3f | 88 | return cfs_rq->rq; |
bf0f6f24 IM |
89 | } |
90 | ||
62160e3f IM |
91 | /* An entity is a task if it doesn't "own" a runqueue */ |
92 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 93 | |
62160e3f | 94 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
bf0f6f24 | 95 | |
62160e3f IM |
96 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
97 | { | |
98 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
99 | } |
100 | ||
101 | #define entity_is_task(se) 1 | |
102 | ||
bf0f6f24 IM |
103 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
104 | ||
105 | static inline struct task_struct *task_of(struct sched_entity *se) | |
106 | { | |
107 | return container_of(se, struct task_struct, se); | |
108 | } | |
109 | ||
110 | ||
111 | /************************************************************** | |
112 | * Scheduling class tree data structure manipulation methods: | |
113 | */ | |
114 | ||
0702e3eb | 115 | static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime) |
02e0431a | 116 | { |
368059a9 PZ |
117 | s64 delta = (s64)(vruntime - min_vruntime); |
118 | if (delta > 0) | |
02e0431a PZ |
119 | min_vruntime = vruntime; |
120 | ||
121 | return min_vruntime; | |
122 | } | |
123 | ||
0702e3eb | 124 | static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime) |
b0ffd246 PZ |
125 | { |
126 | s64 delta = (s64)(vruntime - min_vruntime); | |
127 | if (delta < 0) | |
128 | min_vruntime = vruntime; | |
129 | ||
130 | return min_vruntime; | |
131 | } | |
132 | ||
0702e3eb | 133 | static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
9014623c | 134 | { |
30cfdcfc | 135 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
136 | } |
137 | ||
bf0f6f24 IM |
138 | /* |
139 | * Enqueue an entity into the rb-tree: | |
140 | */ | |
0702e3eb | 141 | static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
142 | { |
143 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
144 | struct rb_node *parent = NULL; | |
145 | struct sched_entity *entry; | |
9014623c | 146 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
147 | int leftmost = 1; |
148 | ||
149 | /* | |
150 | * Find the right place in the rbtree: | |
151 | */ | |
152 | while (*link) { | |
153 | parent = *link; | |
154 | entry = rb_entry(parent, struct sched_entity, run_node); | |
155 | /* | |
156 | * We dont care about collisions. Nodes with | |
157 | * the same key stay together. | |
158 | */ | |
9014623c | 159 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
160 | link = &parent->rb_left; |
161 | } else { | |
162 | link = &parent->rb_right; | |
163 | leftmost = 0; | |
164 | } | |
165 | } | |
166 | ||
167 | /* | |
168 | * Maintain a cache of leftmost tree entries (it is frequently | |
169 | * used): | |
170 | */ | |
171 | if (leftmost) | |
57cb499d | 172 | cfs_rq->rb_leftmost = &se->run_node; |
bf0f6f24 IM |
173 | |
174 | rb_link_node(&se->run_node, parent, link); | |
175 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
176 | } |
177 | ||
0702e3eb | 178 | static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
179 | { |
180 | if (cfs_rq->rb_leftmost == &se->run_node) | |
57cb499d | 181 | cfs_rq->rb_leftmost = rb_next(&se->run_node); |
e9acbff6 | 182 | |
bf0f6f24 | 183 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
184 | } |
185 | ||
186 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
187 | { | |
188 | return cfs_rq->rb_leftmost; | |
189 | } | |
190 | ||
191 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
192 | { | |
193 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
194 | } | |
195 | ||
aeb73b04 PZ |
196 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
197 | { | |
198 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
199 | struct sched_entity *se = NULL; | |
200 | struct rb_node *parent; | |
201 | ||
202 | while (*link) { | |
203 | parent = *link; | |
204 | se = rb_entry(parent, struct sched_entity, run_node); | |
205 | link = &parent->rb_right; | |
206 | } | |
207 | ||
208 | return se; | |
209 | } | |
210 | ||
bf0f6f24 IM |
211 | /************************************************************** |
212 | * Scheduling class statistics methods: | |
213 | */ | |
214 | ||
647e7cac IM |
215 | |
216 | /* | |
217 | * The idea is to set a period in which each task runs once. | |
218 | * | |
219 | * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch | |
220 | * this period because otherwise the slices get too small. | |
221 | * | |
222 | * p = (nr <= nl) ? l : l*nr/nl | |
223 | */ | |
4d78e7b6 PZ |
224 | static u64 __sched_period(unsigned long nr_running) |
225 | { | |
226 | u64 period = sysctl_sched_latency; | |
5f6d858e | 227 | unsigned long nr_latency = sysctl_sched_nr_latency; |
4d78e7b6 PZ |
228 | |
229 | if (unlikely(nr_running > nr_latency)) { | |
230 | period *= nr_running; | |
231 | do_div(period, nr_latency); | |
232 | } | |
233 | ||
234 | return period; | |
235 | } | |
236 | ||
647e7cac IM |
237 | /* |
238 | * We calculate the wall-time slice from the period by taking a part | |
239 | * proportional to the weight. | |
240 | * | |
241 | * s = p*w/rw | |
242 | */ | |
6d0f0ebd | 243 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 244 | { |
647e7cac | 245 | u64 slice = __sched_period(cfs_rq->nr_running); |
21805085 | 246 | |
647e7cac IM |
247 | slice *= se->load.weight; |
248 | do_div(slice, cfs_rq->load.weight); | |
21805085 | 249 | |
647e7cac | 250 | return slice; |
bf0f6f24 IM |
251 | } |
252 | ||
647e7cac IM |
253 | /* |
254 | * We calculate the vruntime slice. | |
255 | * | |
256 | * vs = s/w = p/rw | |
257 | */ | |
258 | static u64 __sched_vslice(unsigned long rq_weight, unsigned long nr_running) | |
67e9fb2a | 259 | { |
647e7cac | 260 | u64 vslice = __sched_period(nr_running); |
67e9fb2a | 261 | |
10b77724 | 262 | vslice *= NICE_0_LOAD; |
647e7cac | 263 | do_div(vslice, rq_weight); |
67e9fb2a | 264 | |
647e7cac IM |
265 | return vslice; |
266 | } | |
5f6d858e | 267 | |
647e7cac IM |
268 | static u64 sched_vslice(struct cfs_rq *cfs_rq) |
269 | { | |
270 | return __sched_vslice(cfs_rq->load.weight, cfs_rq->nr_running); | |
271 | } | |
272 | ||
273 | static u64 sched_vslice_add(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
274 | { | |
275 | return __sched_vslice(cfs_rq->load.weight + se->load.weight, | |
276 | cfs_rq->nr_running + 1); | |
67e9fb2a PZ |
277 | } |
278 | ||
bf0f6f24 IM |
279 | /* |
280 | * Update the current task's runtime statistics. Skip current tasks that | |
281 | * are not in our scheduling class. | |
282 | */ | |
283 | static inline void | |
8ebc91d9 IM |
284 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
285 | unsigned long delta_exec) | |
bf0f6f24 | 286 | { |
bbdba7c0 | 287 | unsigned long delta_exec_weighted; |
b0ffd246 | 288 | u64 vruntime; |
bf0f6f24 | 289 | |
8179ca23 | 290 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
291 | |
292 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 293 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
e9acbff6 IM |
294 | delta_exec_weighted = delta_exec; |
295 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { | |
296 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, | |
297 | &curr->load); | |
298 | } | |
299 | curr->vruntime += delta_exec_weighted; | |
02e0431a PZ |
300 | |
301 | /* | |
302 | * maintain cfs_rq->min_vruntime to be a monotonic increasing | |
303 | * value tracking the leftmost vruntime in the tree. | |
304 | */ | |
305 | if (first_fair(cfs_rq)) { | |
b0ffd246 PZ |
306 | vruntime = min_vruntime(curr->vruntime, |
307 | __pick_next_entity(cfs_rq)->vruntime); | |
02e0431a | 308 | } else |
b0ffd246 | 309 | vruntime = curr->vruntime; |
02e0431a PZ |
310 | |
311 | cfs_rq->min_vruntime = | |
b0ffd246 | 312 | max_vruntime(cfs_rq->min_vruntime, vruntime); |
bf0f6f24 IM |
313 | } |
314 | ||
b7cc0896 | 315 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 316 | { |
429d43bc | 317 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 318 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
319 | unsigned long delta_exec; |
320 | ||
321 | if (unlikely(!curr)) | |
322 | return; | |
323 | ||
324 | /* | |
325 | * Get the amount of time the current task was running | |
326 | * since the last time we changed load (this cannot | |
327 | * overflow on 32 bits): | |
328 | */ | |
8ebc91d9 | 329 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 330 | |
8ebc91d9 IM |
331 | __update_curr(cfs_rq, curr, delta_exec); |
332 | curr->exec_start = now; | |
bf0f6f24 IM |
333 | } |
334 | ||
335 | static inline void | |
5870db5b | 336 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 337 | { |
d281918d | 338 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
339 | } |
340 | ||
bf0f6f24 IM |
341 | /* |
342 | * Task is being enqueued - update stats: | |
343 | */ | |
d2417e5a | 344 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 345 | { |
bf0f6f24 IM |
346 | /* |
347 | * Are we enqueueing a waiting task? (for current tasks | |
348 | * a dequeue/enqueue event is a NOP) | |
349 | */ | |
429d43bc | 350 | if (se != cfs_rq->curr) |
5870db5b | 351 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
352 | } |
353 | ||
bf0f6f24 | 354 | static void |
9ef0a961 | 355 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 356 | { |
bbdba7c0 IM |
357 | schedstat_set(se->wait_max, max(se->wait_max, |
358 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6cfb0d5d | 359 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
360 | } |
361 | ||
362 | static inline void | |
19b6a2e3 | 363 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 364 | { |
bf0f6f24 IM |
365 | /* |
366 | * Mark the end of the wait period if dequeueing a | |
367 | * waiting task: | |
368 | */ | |
429d43bc | 369 | if (se != cfs_rq->curr) |
9ef0a961 | 370 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
371 | } |
372 | ||
373 | /* | |
374 | * We are picking a new current task - update its stats: | |
375 | */ | |
376 | static inline void | |
79303e9e | 377 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
378 | { |
379 | /* | |
380 | * We are starting a new run period: | |
381 | */ | |
d281918d | 382 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
383 | } |
384 | ||
bf0f6f24 IM |
385 | /************************************************** |
386 | * Scheduling class queueing methods: | |
387 | */ | |
388 | ||
30cfdcfc DA |
389 | static void |
390 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
391 | { | |
392 | update_load_add(&cfs_rq->load, se->load.weight); | |
393 | cfs_rq->nr_running++; | |
394 | se->on_rq = 1; | |
395 | } | |
396 | ||
397 | static void | |
398 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
399 | { | |
400 | update_load_sub(&cfs_rq->load, se->load.weight); | |
401 | cfs_rq->nr_running--; | |
402 | se->on_rq = 0; | |
403 | } | |
404 | ||
2396af69 | 405 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 406 | { |
bf0f6f24 IM |
407 | #ifdef CONFIG_SCHEDSTATS |
408 | if (se->sleep_start) { | |
d281918d | 409 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
410 | |
411 | if ((s64)delta < 0) | |
412 | delta = 0; | |
413 | ||
414 | if (unlikely(delta > se->sleep_max)) | |
415 | se->sleep_max = delta; | |
416 | ||
417 | se->sleep_start = 0; | |
418 | se->sum_sleep_runtime += delta; | |
419 | } | |
420 | if (se->block_start) { | |
d281918d | 421 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
422 | |
423 | if ((s64)delta < 0) | |
424 | delta = 0; | |
425 | ||
426 | if (unlikely(delta > se->block_max)) | |
427 | se->block_max = delta; | |
428 | ||
429 | se->block_start = 0; | |
430 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
431 | |
432 | /* | |
433 | * Blocking time is in units of nanosecs, so shift by 20 to | |
434 | * get a milliseconds-range estimation of the amount of | |
435 | * time that the task spent sleeping: | |
436 | */ | |
437 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf IM |
438 | struct task_struct *tsk = task_of(se); |
439 | ||
30084fbd IM |
440 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
441 | delta >> 20); | |
442 | } | |
bf0f6f24 IM |
443 | } |
444 | #endif | |
445 | } | |
446 | ||
ddc97297 PZ |
447 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
448 | { | |
449 | #ifdef CONFIG_SCHED_DEBUG | |
450 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
451 | ||
452 | if (d < 0) | |
453 | d = -d; | |
454 | ||
455 | if (d > 3*sysctl_sched_latency) | |
456 | schedstat_inc(cfs_rq, nr_spread_over); | |
457 | #endif | |
458 | } | |
459 | ||
aeb73b04 PZ |
460 | static void |
461 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
462 | { | |
67e9fb2a | 463 | u64 vruntime; |
aeb73b04 | 464 | |
67e9fb2a | 465 | vruntime = cfs_rq->min_vruntime; |
94dfb5e7 | 466 | |
06877c33 | 467 | if (sched_feat(TREE_AVG)) { |
94dfb5e7 PZ |
468 | struct sched_entity *last = __pick_last_entity(cfs_rq); |
469 | if (last) { | |
67e9fb2a PZ |
470 | vruntime += last->vruntime; |
471 | vruntime >>= 1; | |
94dfb5e7 | 472 | } |
67e9fb2a | 473 | } else if (sched_feat(APPROX_AVG) && cfs_rq->nr_running) |
647e7cac | 474 | vruntime += sched_vslice(cfs_rq)/2; |
94dfb5e7 PZ |
475 | |
476 | if (initial && sched_feat(START_DEBIT)) | |
647e7cac | 477 | vruntime += sched_vslice_add(cfs_rq, se); |
aeb73b04 | 478 | |
8465e792 | 479 | if (!initial) { |
e62dd02e DA |
480 | if (sched_feat(NEW_FAIR_SLEEPERS) && entity_is_task(se) && |
481 | task_of(se)->policy != SCHED_BATCH) | |
94359f05 IM |
482 | vruntime -= sysctl_sched_latency; |
483 | ||
b8487b92 | 484 | vruntime = max_t(s64, vruntime, se->vruntime); |
aeb73b04 PZ |
485 | } |
486 | ||
67e9fb2a PZ |
487 | se->vruntime = vruntime; |
488 | ||
aeb73b04 PZ |
489 | } |
490 | ||
bf0f6f24 | 491 | static void |
83b699ed | 492 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
493 | { |
494 | /* | |
a2a2d680 | 495 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 496 | */ |
b7cc0896 | 497 | update_curr(cfs_rq); |
bf0f6f24 | 498 | |
e9acbff6 | 499 | if (wakeup) { |
aeb73b04 | 500 | place_entity(cfs_rq, se, 0); |
2396af69 | 501 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 502 | } |
bf0f6f24 | 503 | |
d2417e5a | 504 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 505 | check_spread(cfs_rq, se); |
83b699ed SV |
506 | if (se != cfs_rq->curr) |
507 | __enqueue_entity(cfs_rq, se); | |
30cfdcfc | 508 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 IM |
509 | } |
510 | ||
511 | static void | |
525c2716 | 512 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 513 | { |
a2a2d680 DA |
514 | /* |
515 | * Update run-time statistics of the 'current'. | |
516 | */ | |
517 | update_curr(cfs_rq); | |
518 | ||
19b6a2e3 | 519 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 520 | if (sleep) { |
95938a35 | 521 | se->peer_preempt = 0; |
67e9fb2a | 522 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
523 | if (entity_is_task(se)) { |
524 | struct task_struct *tsk = task_of(se); | |
525 | ||
526 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 527 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 528 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 529 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 530 | } |
db36cc7d | 531 | #endif |
67e9fb2a PZ |
532 | } |
533 | ||
83b699ed | 534 | if (se != cfs_rq->curr) |
30cfdcfc DA |
535 | __dequeue_entity(cfs_rq, se); |
536 | account_entity_dequeue(cfs_rq, se); | |
bf0f6f24 IM |
537 | } |
538 | ||
539 | /* | |
540 | * Preempt the current task with a newly woken task if needed: | |
541 | */ | |
7c92e54f | 542 | static void |
2e09bf55 | 543 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 544 | { |
11697830 PZ |
545 | unsigned long ideal_runtime, delta_exec; |
546 | ||
6d0f0ebd | 547 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 | 548 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
95938a35 MG |
549 | if (delta_exec > ideal_runtime || |
550 | (sched_feat(PREEMPT_RESTRICT) && curr->peer_preempt)) | |
bf0f6f24 | 551 | resched_task(rq_of(cfs_rq)->curr); |
95938a35 | 552 | curr->peer_preempt = 0; |
bf0f6f24 IM |
553 | } |
554 | ||
83b699ed | 555 | static void |
8494f412 | 556 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 557 | { |
83b699ed SV |
558 | /* 'current' is not kept within the tree. */ |
559 | if (se->on_rq) { | |
560 | /* | |
561 | * Any task has to be enqueued before it get to execute on | |
562 | * a CPU. So account for the time it spent waiting on the | |
563 | * runqueue. | |
564 | */ | |
565 | update_stats_wait_end(cfs_rq, se); | |
566 | __dequeue_entity(cfs_rq, se); | |
567 | } | |
568 | ||
79303e9e | 569 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 570 | cfs_rq->curr = se; |
eba1ed4b IM |
571 | #ifdef CONFIG_SCHEDSTATS |
572 | /* | |
573 | * Track our maximum slice length, if the CPU's load is at | |
574 | * least twice that of our own weight (i.e. dont track it | |
575 | * when there are only lesser-weight tasks around): | |
576 | */ | |
495eca49 | 577 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
578 | se->slice_max = max(se->slice_max, |
579 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
580 | } | |
581 | #endif | |
4a55b450 | 582 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
583 | } |
584 | ||
9948f4b2 | 585 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 | 586 | { |
08ec3df5 | 587 | struct sched_entity *se = NULL; |
bf0f6f24 | 588 | |
08ec3df5 DA |
589 | if (first_fair(cfs_rq)) { |
590 | se = __pick_next_entity(cfs_rq); | |
591 | set_next_entity(cfs_rq, se); | |
592 | } | |
bf0f6f24 IM |
593 | |
594 | return se; | |
595 | } | |
596 | ||
ab6cde26 | 597 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
598 | { |
599 | /* | |
600 | * If still on the runqueue then deactivate_task() | |
601 | * was not called and update_curr() has to be done: | |
602 | */ | |
603 | if (prev->on_rq) | |
b7cc0896 | 604 | update_curr(cfs_rq); |
bf0f6f24 | 605 | |
ddc97297 | 606 | check_spread(cfs_rq, prev); |
30cfdcfc | 607 | if (prev->on_rq) { |
5870db5b | 608 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
609 | /* Put 'current' back into the tree. */ |
610 | __enqueue_entity(cfs_rq, prev); | |
611 | } | |
429d43bc | 612 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
613 | } |
614 | ||
615 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
616 | { | |
bf0f6f24 | 617 | /* |
30cfdcfc | 618 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 619 | */ |
30cfdcfc | 620 | update_curr(cfs_rq); |
bf0f6f24 | 621 | |
ce6c1311 | 622 | if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT)) |
2e09bf55 | 623 | check_preempt_tick(cfs_rq, curr); |
bf0f6f24 IM |
624 | } |
625 | ||
626 | /************************************************** | |
627 | * CFS operations on tasks: | |
628 | */ | |
629 | ||
630 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
631 | ||
632 | /* Walk up scheduling entities hierarchy */ | |
633 | #define for_each_sched_entity(se) \ | |
634 | for (; se; se = se->parent) | |
635 | ||
636 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
637 | { | |
638 | return p->se.cfs_rq; | |
639 | } | |
640 | ||
641 | /* runqueue on which this entity is (to be) queued */ | |
642 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
643 | { | |
644 | return se->cfs_rq; | |
645 | } | |
646 | ||
647 | /* runqueue "owned" by this group */ | |
648 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
649 | { | |
650 | return grp->my_q; | |
651 | } | |
652 | ||
653 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
654 | * another cpu ('this_cpu') | |
655 | */ | |
656 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
657 | { | |
29f59db3 | 658 | return cfs_rq->tg->cfs_rq[this_cpu]; |
bf0f6f24 IM |
659 | } |
660 | ||
661 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
662 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
663 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
664 | ||
fad095a7 SV |
665 | /* Do the two (enqueued) entities belong to the same group ? */ |
666 | static inline int | |
667 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
bf0f6f24 | 668 | { |
fad095a7 | 669 | if (se->cfs_rq == pse->cfs_rq) |
bf0f6f24 IM |
670 | return 1; |
671 | ||
672 | return 0; | |
673 | } | |
674 | ||
fad095a7 SV |
675 | static inline struct sched_entity *parent_entity(struct sched_entity *se) |
676 | { | |
677 | return se->parent; | |
678 | } | |
679 | ||
bf0f6f24 IM |
680 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
681 | ||
682 | #define for_each_sched_entity(se) \ | |
683 | for (; se; se = NULL) | |
684 | ||
685 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
686 | { | |
687 | return &task_rq(p)->cfs; | |
688 | } | |
689 | ||
690 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
691 | { | |
692 | struct task_struct *p = task_of(se); | |
693 | struct rq *rq = task_rq(p); | |
694 | ||
695 | return &rq->cfs; | |
696 | } | |
697 | ||
698 | /* runqueue "owned" by this group */ | |
699 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
700 | { | |
701 | return NULL; | |
702 | } | |
703 | ||
704 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
705 | { | |
706 | return &cpu_rq(this_cpu)->cfs; | |
707 | } | |
708 | ||
709 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
710 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
711 | ||
fad095a7 SV |
712 | static inline int |
713 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
bf0f6f24 IM |
714 | { |
715 | return 1; | |
716 | } | |
717 | ||
fad095a7 SV |
718 | static inline struct sched_entity *parent_entity(struct sched_entity *se) |
719 | { | |
720 | return NULL; | |
721 | } | |
722 | ||
bf0f6f24 IM |
723 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
724 | ||
725 | /* | |
726 | * The enqueue_task method is called before nr_running is | |
727 | * increased. Here we update the fair scheduling stats and | |
728 | * then put the task into the rbtree: | |
729 | */ | |
fd390f6a | 730 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
731 | { |
732 | struct cfs_rq *cfs_rq; | |
733 | struct sched_entity *se = &p->se; | |
734 | ||
735 | for_each_sched_entity(se) { | |
736 | if (se->on_rq) | |
737 | break; | |
738 | cfs_rq = cfs_rq_of(se); | |
83b699ed | 739 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 740 | wakeup = 1; |
bf0f6f24 IM |
741 | } |
742 | } | |
743 | ||
744 | /* | |
745 | * The dequeue_task method is called before nr_running is | |
746 | * decreased. We remove the task from the rbtree and | |
747 | * update the fair scheduling stats: | |
748 | */ | |
f02231e5 | 749 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
750 | { |
751 | struct cfs_rq *cfs_rq; | |
752 | struct sched_entity *se = &p->se; | |
753 | ||
754 | for_each_sched_entity(se) { | |
755 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 756 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
757 | /* Don't dequeue parent if it has other entities besides us */ |
758 | if (cfs_rq->load.weight) | |
759 | break; | |
b9fa3df3 | 760 | sleep = 1; |
bf0f6f24 IM |
761 | } |
762 | } | |
763 | ||
764 | /* | |
1799e35d IM |
765 | * sched_yield() support is very simple - we dequeue and enqueue. |
766 | * | |
767 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 768 | */ |
4530d7ab | 769 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 770 | { |
72ea22f8 | 771 | struct cfs_rq *cfs_rq = task_cfs_rq(rq->curr); |
4530d7ab | 772 | struct sched_entity *rightmost, *se = &rq->curr->se; |
bf0f6f24 IM |
773 | |
774 | /* | |
1799e35d IM |
775 | * Are we the only task in the tree? |
776 | */ | |
777 | if (unlikely(cfs_rq->nr_running == 1)) | |
778 | return; | |
779 | ||
780 | if (likely(!sysctl_sched_compat_yield)) { | |
781 | __update_rq_clock(rq); | |
782 | /* | |
a2a2d680 | 783 | * Update run-time statistics of the 'current'. |
1799e35d | 784 | */ |
2b1e315d | 785 | update_curr(cfs_rq); |
1799e35d IM |
786 | |
787 | return; | |
788 | } | |
789 | /* | |
790 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 791 | */ |
2b1e315d | 792 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
793 | /* |
794 | * Already in the rightmost position? | |
795 | */ | |
2b1e315d | 796 | if (unlikely(rightmost->vruntime < se->vruntime)) |
1799e35d IM |
797 | return; |
798 | ||
799 | /* | |
800 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
801 | * Upon rescheduling, sched_class::put_prev_task() will place |
802 | * 'current' within the tree based on its new key value. | |
1799e35d | 803 | */ |
30cfdcfc | 804 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
805 | } |
806 | ||
807 | /* | |
808 | * Preempt the current task with a newly woken task if needed: | |
809 | */ | |
2e09bf55 | 810 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
811 | { |
812 | struct task_struct *curr = rq->curr; | |
fad095a7 | 813 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
8651a86c | 814 | struct sched_entity *se = &curr->se, *pse = &p->se; |
810e95cc | 815 | s64 delta, gran; |
bf0f6f24 IM |
816 | |
817 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 818 | update_rq_clock(rq); |
b7cc0896 | 819 | update_curr(cfs_rq); |
bf0f6f24 IM |
820 | resched_task(curr); |
821 | return; | |
822 | } | |
91c234b4 IM |
823 | /* |
824 | * Batch tasks do not preempt (their preemption is driven by | |
825 | * the tick): | |
826 | */ | |
827 | if (unlikely(p->policy == SCHED_BATCH)) | |
828 | return; | |
bf0f6f24 | 829 | |
ce6c1311 PZ |
830 | if (sched_feat(WAKEUP_PREEMPT)) { |
831 | while (!is_same_group(se, pse)) { | |
832 | se = parent_entity(se); | |
833 | pse = parent_entity(pse); | |
834 | } | |
8651a86c | 835 | |
ce6c1311 PZ |
836 | delta = se->vruntime - pse->vruntime; |
837 | gran = sysctl_sched_wakeup_granularity; | |
838 | if (unlikely(se->load.weight != NICE_0_LOAD)) | |
839 | gran = calc_delta_fair(gran, &se->load); | |
8651a86c | 840 | |
95938a35 MG |
841 | if (delta > gran) { |
842 | int now = !sched_feat(PREEMPT_RESTRICT); | |
843 | ||
844 | if (now || p->prio < curr->prio || !se->peer_preempt++) | |
845 | resched_task(curr); | |
846 | } | |
ce6c1311 | 847 | } |
bf0f6f24 IM |
848 | } |
849 | ||
fb8d4724 | 850 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
851 | { |
852 | struct cfs_rq *cfs_rq = &rq->cfs; | |
853 | struct sched_entity *se; | |
854 | ||
855 | if (unlikely(!cfs_rq->nr_running)) | |
856 | return NULL; | |
857 | ||
858 | do { | |
9948f4b2 | 859 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
860 | cfs_rq = group_cfs_rq(se); |
861 | } while (cfs_rq); | |
862 | ||
863 | return task_of(se); | |
864 | } | |
865 | ||
866 | /* | |
867 | * Account for a descheduled task: | |
868 | */ | |
31ee529c | 869 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
870 | { |
871 | struct sched_entity *se = &prev->se; | |
872 | struct cfs_rq *cfs_rq; | |
873 | ||
874 | for_each_sched_entity(se) { | |
875 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 876 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
877 | } |
878 | } | |
879 | ||
681f3e68 | 880 | #ifdef CONFIG_SMP |
bf0f6f24 IM |
881 | /************************************************** |
882 | * Fair scheduling class load-balancing methods: | |
883 | */ | |
884 | ||
885 | /* | |
886 | * Load-balancing iterator. Note: while the runqueue stays locked | |
887 | * during the whole iteration, the current task might be | |
888 | * dequeued so the iterator has to be dequeue-safe. Here we | |
889 | * achieve that by always pre-iterating before returning | |
890 | * the current task: | |
891 | */ | |
a9957449 | 892 | static struct task_struct * |
bf0f6f24 IM |
893 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) |
894 | { | |
895 | struct task_struct *p; | |
896 | ||
897 | if (!curr) | |
898 | return NULL; | |
899 | ||
900 | p = rb_entry(curr, struct task_struct, se.run_node); | |
901 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
902 | ||
903 | return p; | |
904 | } | |
905 | ||
906 | static struct task_struct *load_balance_start_fair(void *arg) | |
907 | { | |
908 | struct cfs_rq *cfs_rq = arg; | |
909 | ||
910 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
911 | } | |
912 | ||
913 | static struct task_struct *load_balance_next_fair(void *arg) | |
914 | { | |
915 | struct cfs_rq *cfs_rq = arg; | |
916 | ||
917 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
918 | } | |
919 | ||
a4ac01c3 | 920 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
921 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
922 | { | |
923 | struct sched_entity *curr; | |
924 | struct task_struct *p; | |
925 | ||
926 | if (!cfs_rq->nr_running) | |
927 | return MAX_PRIO; | |
928 | ||
9b5b7751 SV |
929 | curr = cfs_rq->curr; |
930 | if (!curr) | |
931 | curr = __pick_next_entity(cfs_rq); | |
932 | ||
bf0f6f24 IM |
933 | p = task_of(curr); |
934 | ||
935 | return p->prio; | |
936 | } | |
a4ac01c3 | 937 | #endif |
bf0f6f24 | 938 | |
43010659 | 939 | static unsigned long |
bf0f6f24 | 940 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f | 941 | unsigned long max_load_move, |
a4ac01c3 PW |
942 | struct sched_domain *sd, enum cpu_idle_type idle, |
943 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
944 | { |
945 | struct cfs_rq *busy_cfs_rq; | |
bf0f6f24 IM |
946 | long rem_load_move = max_load_move; |
947 | struct rq_iterator cfs_rq_iterator; | |
948 | ||
949 | cfs_rq_iterator.start = load_balance_start_fair; | |
950 | cfs_rq_iterator.next = load_balance_next_fair; | |
951 | ||
952 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 953 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 954 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 955 | long imbalance; |
bf0f6f24 | 956 | unsigned long maxload; |
bf0f6f24 IM |
957 | |
958 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
959 | ||
e56f31aa | 960 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
961 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
962 | if (imbalance <= 0) | |
963 | continue; | |
964 | ||
965 | /* Don't pull more than imbalance/2 */ | |
966 | imbalance /= 2; | |
967 | maxload = min(rem_load_move, imbalance); | |
968 | ||
a4ac01c3 PW |
969 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
970 | #else | |
e56f31aa | 971 | # define maxload rem_load_move |
a4ac01c3 | 972 | #endif |
e1d1484f PW |
973 | /* |
974 | * pass busy_cfs_rq argument into | |
bf0f6f24 IM |
975 | * load_balance_[start|next]_fair iterators |
976 | */ | |
977 | cfs_rq_iterator.arg = busy_cfs_rq; | |
e1d1484f PW |
978 | rem_load_move -= balance_tasks(this_rq, this_cpu, busiest, |
979 | maxload, sd, idle, all_pinned, | |
980 | this_best_prio, | |
981 | &cfs_rq_iterator); | |
bf0f6f24 | 982 | |
e1d1484f | 983 | if (rem_load_move <= 0) |
bf0f6f24 IM |
984 | break; |
985 | } | |
986 | ||
43010659 | 987 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
988 | } |
989 | ||
e1d1484f PW |
990 | static int |
991 | move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
992 | struct sched_domain *sd, enum cpu_idle_type idle) | |
993 | { | |
994 | struct cfs_rq *busy_cfs_rq; | |
995 | struct rq_iterator cfs_rq_iterator; | |
996 | ||
997 | cfs_rq_iterator.start = load_balance_start_fair; | |
998 | cfs_rq_iterator.next = load_balance_next_fair; | |
999 | ||
1000 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
1001 | /* | |
1002 | * pass busy_cfs_rq argument into | |
1003 | * load_balance_[start|next]_fair iterators | |
1004 | */ | |
1005 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1006 | if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle, | |
1007 | &cfs_rq_iterator)) | |
1008 | return 1; | |
1009 | } | |
1010 | ||
1011 | return 0; | |
1012 | } | |
681f3e68 | 1013 | #endif |
e1d1484f | 1014 | |
bf0f6f24 IM |
1015 | /* |
1016 | * scheduler tick hitting a task of our scheduling class: | |
1017 | */ | |
1018 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
1019 | { | |
1020 | struct cfs_rq *cfs_rq; | |
1021 | struct sched_entity *se = &curr->se; | |
1022 | ||
1023 | for_each_sched_entity(se) { | |
1024 | cfs_rq = cfs_rq_of(se); | |
1025 | entity_tick(cfs_rq, se); | |
1026 | } | |
1027 | } | |
1028 | ||
8eb172d9 | 1029 | #define swap(a, b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
4d78e7b6 | 1030 | |
bf0f6f24 IM |
1031 | /* |
1032 | * Share the fairness runtime between parent and child, thus the | |
1033 | * total amount of pressure for CPU stays equal - new tasks | |
1034 | * get a chance to run but frequent forkers are not allowed to | |
1035 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1036 | * the child is not running yet. | |
1037 | */ | |
ee0827d8 | 1038 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1039 | { |
1040 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1041 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
00bf7bfc | 1042 | int this_cpu = smp_processor_id(); |
bf0f6f24 IM |
1043 | |
1044 | sched_info_queued(p); | |
1045 | ||
7109c442 | 1046 | update_curr(cfs_rq); |
aeb73b04 | 1047 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1048 | |
00bf7bfc | 1049 | if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) && |
4d78e7b6 | 1050 | curr->vruntime < se->vruntime) { |
87fefa38 | 1051 | /* |
edcb60a3 IM |
1052 | * Upon rescheduling, sched_class::put_prev_task() will place |
1053 | * 'current' within the tree based on its new key value. | |
1054 | */ | |
4d78e7b6 | 1055 | swap(curr->vruntime, se->vruntime); |
4d78e7b6 | 1056 | } |
bf0f6f24 | 1057 | |
95938a35 | 1058 | se->peer_preempt = 0; |
b9dca1e0 | 1059 | enqueue_task_fair(rq, p, 0); |
bb61c210 | 1060 | resched_task(rq->curr); |
bf0f6f24 IM |
1061 | } |
1062 | ||
83b699ed SV |
1063 | /* Account for a task changing its policy or group. |
1064 | * | |
1065 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1066 | * migrates between groups/classes. | |
1067 | */ | |
1068 | static void set_curr_task_fair(struct rq *rq) | |
1069 | { | |
1070 | struct sched_entity *se = &rq->curr->se; | |
1071 | ||
1072 | for_each_sched_entity(se) | |
1073 | set_next_entity(cfs_rq_of(se), se); | |
1074 | } | |
1075 | ||
bf0f6f24 IM |
1076 | /* |
1077 | * All the scheduling class methods: | |
1078 | */ | |
5522d5d5 IM |
1079 | static const struct sched_class fair_sched_class = { |
1080 | .next = &idle_sched_class, | |
bf0f6f24 IM |
1081 | .enqueue_task = enqueue_task_fair, |
1082 | .dequeue_task = dequeue_task_fair, | |
1083 | .yield_task = yield_task_fair, | |
1084 | ||
2e09bf55 | 1085 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1086 | |
1087 | .pick_next_task = pick_next_task_fair, | |
1088 | .put_prev_task = put_prev_task_fair, | |
1089 | ||
681f3e68 | 1090 | #ifdef CONFIG_SMP |
bf0f6f24 | 1091 | .load_balance = load_balance_fair, |
e1d1484f | 1092 | .move_one_task = move_one_task_fair, |
681f3e68 | 1093 | #endif |
bf0f6f24 | 1094 | |
83b699ed | 1095 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 IM |
1096 | .task_tick = task_tick_fair, |
1097 | .task_new = task_new_fair, | |
1098 | }; | |
1099 | ||
1100 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1101 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1102 | { |
bf0f6f24 IM |
1103 | struct cfs_rq *cfs_rq; |
1104 | ||
75c28ace SV |
1105 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1106 | print_cfs_rq(m, cpu, &cpu_rq(cpu)->cfs); | |
1107 | #endif | |
c3b64f1e | 1108 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1109 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1110 | } |
1111 | #endif |