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 PZ |
27 | * NOTE: this latency value is not the same as the concept of |
28 | * 'timeslice length' - timeslices in CFS are of variable length. | |
29 | * (to see the precise effective timeslice length of your workload, | |
30 | * run vmstat and monitor the context-switches field) | |
bf0f6f24 IM |
31 | * |
32 | * On SMP systems the value of this is multiplied by the log2 of the | |
33 | * number of CPUs. (i.e. factor 2x on 2-way systems, 3x on 4-way | |
34 | * systems, 4x on 8-way systems, 5x on 16-way systems, etc.) | |
21805085 | 35 | * Targeted preemption latency for CPU-bound tasks: |
bf0f6f24 | 36 | */ |
2bd8e6d4 IM |
37 | const_debug unsigned int sysctl_sched_latency = 20000000ULL; |
38 | ||
39 | /* | |
40 | * After fork, child runs first. (default) If set to 0 then | |
41 | * parent will (try to) run first. | |
42 | */ | |
43 | const_debug unsigned int sysctl_sched_child_runs_first = 1; | |
21805085 PZ |
44 | |
45 | /* | |
46 | * Minimal preemption granularity for CPU-bound tasks: | |
47 | * (default: 2 msec, units: nanoseconds) | |
48 | */ | |
5f6d858e | 49 | const_debug unsigned int sysctl_sched_nr_latency = 20; |
bf0f6f24 | 50 | |
1799e35d IM |
51 | /* |
52 | * sys_sched_yield() compat mode | |
53 | * | |
54 | * This option switches the agressive yield implementation of the | |
55 | * old scheduler back on. | |
56 | */ | |
57 | unsigned int __read_mostly sysctl_sched_compat_yield; | |
58 | ||
bf0f6f24 IM |
59 | /* |
60 | * SCHED_BATCH wake-up granularity. | |
155bb293 | 61 | * (default: 10 msec, units: nanoseconds) |
bf0f6f24 IM |
62 | * |
63 | * This option delays the preemption effects of decoupled workloads | |
64 | * and reduces their over-scheduling. Synchronous workloads will still | |
65 | * have immediate wakeup/sleep latencies. | |
66 | */ | |
155bb293 | 67 | const_debug unsigned int sysctl_sched_batch_wakeup_granularity = 10000000UL; |
bf0f6f24 IM |
68 | |
69 | /* | |
70 | * SCHED_OTHER wake-up granularity. | |
155bb293 | 71 | * (default: 10 msec, units: nanoseconds) |
bf0f6f24 IM |
72 | * |
73 | * This option delays the preemption effects of decoupled workloads | |
74 | * and reduces their over-scheduling. Synchronous workloads will still | |
75 | * have immediate wakeup/sleep latencies. | |
76 | */ | |
155bb293 | 77 | const_debug unsigned int sysctl_sched_wakeup_granularity = 10000000UL; |
bf0f6f24 | 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 | ||
02e0431a PZ |
115 | static inline u64 |
116 | max_vruntime(u64 min_vruntime, u64 vruntime) | |
117 | { | |
368059a9 PZ |
118 | s64 delta = (s64)(vruntime - min_vruntime); |
119 | if (delta > 0) | |
02e0431a PZ |
120 | min_vruntime = vruntime; |
121 | ||
122 | return min_vruntime; | |
123 | } | |
124 | ||
b0ffd246 PZ |
125 | static inline u64 |
126 | min_vruntime(u64 min_vruntime, u64 vruntime) | |
127 | { | |
128 | s64 delta = (s64)(vruntime - min_vruntime); | |
129 | if (delta < 0) | |
130 | min_vruntime = vruntime; | |
131 | ||
132 | return min_vruntime; | |
133 | } | |
134 | ||
02e0431a PZ |
135 | static inline s64 |
136 | entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
9014623c | 137 | { |
30cfdcfc | 138 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
139 | } |
140 | ||
bf0f6f24 IM |
141 | /* |
142 | * Enqueue an entity into the rb-tree: | |
143 | */ | |
19ccd97a | 144 | static void |
bf0f6f24 IM |
145 | __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
146 | { | |
147 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
148 | struct rb_node *parent = NULL; | |
149 | struct sched_entity *entry; | |
9014623c | 150 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
151 | int leftmost = 1; |
152 | ||
153 | /* | |
154 | * Find the right place in the rbtree: | |
155 | */ | |
156 | while (*link) { | |
157 | parent = *link; | |
158 | entry = rb_entry(parent, struct sched_entity, run_node); | |
159 | /* | |
160 | * We dont care about collisions. Nodes with | |
161 | * the same key stay together. | |
162 | */ | |
9014623c | 163 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
164 | link = &parent->rb_left; |
165 | } else { | |
166 | link = &parent->rb_right; | |
167 | leftmost = 0; | |
168 | } | |
169 | } | |
170 | ||
171 | /* | |
172 | * Maintain a cache of leftmost tree entries (it is frequently | |
173 | * used): | |
174 | */ | |
175 | if (leftmost) | |
57cb499d | 176 | cfs_rq->rb_leftmost = &se->run_node; |
bf0f6f24 IM |
177 | |
178 | rb_link_node(&se->run_node, parent, link); | |
179 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
180 | } |
181 | ||
19ccd97a | 182 | static void |
bf0f6f24 IM |
183 | __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
184 | { | |
185 | if (cfs_rq->rb_leftmost == &se->run_node) | |
57cb499d | 186 | cfs_rq->rb_leftmost = rb_next(&se->run_node); |
e9acbff6 | 187 | |
bf0f6f24 | 188 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
189 | } |
190 | ||
191 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
192 | { | |
193 | return cfs_rq->rb_leftmost; | |
194 | } | |
195 | ||
196 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
197 | { | |
198 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
199 | } | |
200 | ||
aeb73b04 PZ |
201 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
202 | { | |
203 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
204 | struct sched_entity *se = NULL; | |
205 | struct rb_node *parent; | |
206 | ||
207 | while (*link) { | |
208 | parent = *link; | |
209 | se = rb_entry(parent, struct sched_entity, run_node); | |
210 | link = &parent->rb_right; | |
211 | } | |
212 | ||
213 | return se; | |
214 | } | |
215 | ||
bf0f6f24 IM |
216 | /************************************************************** |
217 | * Scheduling class statistics methods: | |
218 | */ | |
219 | ||
647e7cac IM |
220 | |
221 | /* | |
222 | * The idea is to set a period in which each task runs once. | |
223 | * | |
224 | * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch | |
225 | * this period because otherwise the slices get too small. | |
226 | * | |
227 | * p = (nr <= nl) ? l : l*nr/nl | |
228 | */ | |
4d78e7b6 PZ |
229 | static u64 __sched_period(unsigned long nr_running) |
230 | { | |
231 | u64 period = sysctl_sched_latency; | |
5f6d858e | 232 | unsigned long nr_latency = sysctl_sched_nr_latency; |
4d78e7b6 PZ |
233 | |
234 | if (unlikely(nr_running > nr_latency)) { | |
235 | period *= nr_running; | |
236 | do_div(period, nr_latency); | |
237 | } | |
238 | ||
239 | return period; | |
240 | } | |
241 | ||
647e7cac IM |
242 | /* |
243 | * We calculate the wall-time slice from the period by taking a part | |
244 | * proportional to the weight. | |
245 | * | |
246 | * s = p*w/rw | |
247 | */ | |
6d0f0ebd | 248 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 249 | { |
647e7cac | 250 | u64 slice = __sched_period(cfs_rq->nr_running); |
21805085 | 251 | |
647e7cac IM |
252 | slice *= se->load.weight; |
253 | do_div(slice, cfs_rq->load.weight); | |
21805085 | 254 | |
647e7cac | 255 | return slice; |
bf0f6f24 IM |
256 | } |
257 | ||
647e7cac IM |
258 | /* |
259 | * We calculate the vruntime slice. | |
260 | * | |
261 | * vs = s/w = p/rw | |
262 | */ | |
263 | static u64 __sched_vslice(unsigned long rq_weight, unsigned long nr_running) | |
67e9fb2a | 264 | { |
647e7cac | 265 | u64 vslice = __sched_period(nr_running); |
67e9fb2a | 266 | |
647e7cac | 267 | do_div(vslice, rq_weight); |
67e9fb2a | 268 | |
647e7cac IM |
269 | return vslice; |
270 | } | |
5f6d858e | 271 | |
647e7cac IM |
272 | static u64 sched_vslice(struct cfs_rq *cfs_rq) |
273 | { | |
274 | return __sched_vslice(cfs_rq->load.weight, cfs_rq->nr_running); | |
275 | } | |
276 | ||
277 | static u64 sched_vslice_add(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
278 | { | |
279 | return __sched_vslice(cfs_rq->load.weight + se->load.weight, | |
280 | cfs_rq->nr_running + 1); | |
67e9fb2a PZ |
281 | } |
282 | ||
bf0f6f24 IM |
283 | /* |
284 | * Update the current task's runtime statistics. Skip current tasks that | |
285 | * are not in our scheduling class. | |
286 | */ | |
287 | static inline void | |
8ebc91d9 IM |
288 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
289 | unsigned long delta_exec) | |
bf0f6f24 | 290 | { |
bbdba7c0 | 291 | unsigned long delta_exec_weighted; |
b0ffd246 | 292 | u64 vruntime; |
bf0f6f24 | 293 | |
8179ca23 | 294 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
295 | |
296 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 297 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
e9acbff6 IM |
298 | delta_exec_weighted = delta_exec; |
299 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { | |
300 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, | |
301 | &curr->load); | |
302 | } | |
303 | curr->vruntime += delta_exec_weighted; | |
02e0431a PZ |
304 | |
305 | /* | |
306 | * maintain cfs_rq->min_vruntime to be a monotonic increasing | |
307 | * value tracking the leftmost vruntime in the tree. | |
308 | */ | |
309 | if (first_fair(cfs_rq)) { | |
b0ffd246 PZ |
310 | vruntime = min_vruntime(curr->vruntime, |
311 | __pick_next_entity(cfs_rq)->vruntime); | |
02e0431a | 312 | } else |
b0ffd246 | 313 | vruntime = curr->vruntime; |
02e0431a PZ |
314 | |
315 | cfs_rq->min_vruntime = | |
b0ffd246 | 316 | max_vruntime(cfs_rq->min_vruntime, vruntime); |
bf0f6f24 IM |
317 | } |
318 | ||
b7cc0896 | 319 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 320 | { |
429d43bc | 321 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 322 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
323 | unsigned long delta_exec; |
324 | ||
325 | if (unlikely(!curr)) | |
326 | return; | |
327 | ||
328 | /* | |
329 | * Get the amount of time the current task was running | |
330 | * since the last time we changed load (this cannot | |
331 | * overflow on 32 bits): | |
332 | */ | |
8ebc91d9 | 333 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 334 | |
8ebc91d9 IM |
335 | __update_curr(cfs_rq, curr, delta_exec); |
336 | curr->exec_start = now; | |
bf0f6f24 IM |
337 | } |
338 | ||
339 | static inline void | |
5870db5b | 340 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 341 | { |
d281918d | 342 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
343 | } |
344 | ||
bf0f6f24 | 345 | static inline unsigned long |
08e2388a | 346 | calc_weighted(unsigned long delta, struct sched_entity *se) |
bf0f6f24 | 347 | { |
08e2388a | 348 | unsigned long weight = se->load.weight; |
bf0f6f24 | 349 | |
08e2388a IM |
350 | if (unlikely(weight != NICE_0_LOAD)) |
351 | return (u64)delta * se->load.weight >> NICE_0_SHIFT; | |
352 | else | |
353 | return delta; | |
bf0f6f24 | 354 | } |
bf0f6f24 IM |
355 | |
356 | /* | |
357 | * Task is being enqueued - update stats: | |
358 | */ | |
d2417e5a | 359 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 360 | { |
bf0f6f24 IM |
361 | /* |
362 | * Are we enqueueing a waiting task? (for current tasks | |
363 | * a dequeue/enqueue event is a NOP) | |
364 | */ | |
429d43bc | 365 | if (se != cfs_rq->curr) |
5870db5b | 366 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
367 | } |
368 | ||
bf0f6f24 | 369 | static void |
9ef0a961 | 370 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 371 | { |
bbdba7c0 IM |
372 | schedstat_set(se->wait_max, max(se->wait_max, |
373 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6cfb0d5d | 374 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
375 | } |
376 | ||
377 | static inline void | |
19b6a2e3 | 378 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 379 | { |
b7cc0896 | 380 | update_curr(cfs_rq); |
bf0f6f24 IM |
381 | /* |
382 | * Mark the end of the wait period if dequeueing a | |
383 | * waiting task: | |
384 | */ | |
429d43bc | 385 | if (se != cfs_rq->curr) |
9ef0a961 | 386 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
387 | } |
388 | ||
389 | /* | |
390 | * We are picking a new current task - update its stats: | |
391 | */ | |
392 | static inline void | |
79303e9e | 393 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
394 | { |
395 | /* | |
396 | * We are starting a new run period: | |
397 | */ | |
d281918d | 398 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
399 | } |
400 | ||
401 | /* | |
402 | * We are descheduling a task - update its stats: | |
403 | */ | |
404 | static inline void | |
c7e9b5b2 | 405 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
406 | { |
407 | se->exec_start = 0; | |
408 | } | |
409 | ||
410 | /************************************************** | |
411 | * Scheduling class queueing methods: | |
412 | */ | |
413 | ||
30cfdcfc DA |
414 | static void |
415 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
416 | { | |
417 | update_load_add(&cfs_rq->load, se->load.weight); | |
418 | cfs_rq->nr_running++; | |
419 | se->on_rq = 1; | |
420 | } | |
421 | ||
422 | static void | |
423 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
424 | { | |
425 | update_load_sub(&cfs_rq->load, se->load.weight); | |
426 | cfs_rq->nr_running--; | |
427 | se->on_rq = 0; | |
428 | } | |
429 | ||
2396af69 | 430 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 431 | { |
bf0f6f24 IM |
432 | #ifdef CONFIG_SCHEDSTATS |
433 | if (se->sleep_start) { | |
d281918d | 434 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
435 | |
436 | if ((s64)delta < 0) | |
437 | delta = 0; | |
438 | ||
439 | if (unlikely(delta > se->sleep_max)) | |
440 | se->sleep_max = delta; | |
441 | ||
442 | se->sleep_start = 0; | |
443 | se->sum_sleep_runtime += delta; | |
444 | } | |
445 | if (se->block_start) { | |
d281918d | 446 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
447 | |
448 | if ((s64)delta < 0) | |
449 | delta = 0; | |
450 | ||
451 | if (unlikely(delta > se->block_max)) | |
452 | se->block_max = delta; | |
453 | ||
454 | se->block_start = 0; | |
455 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
456 | |
457 | /* | |
458 | * Blocking time is in units of nanosecs, so shift by 20 to | |
459 | * get a milliseconds-range estimation of the amount of | |
460 | * time that the task spent sleeping: | |
461 | */ | |
462 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf IM |
463 | struct task_struct *tsk = task_of(se); |
464 | ||
30084fbd IM |
465 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
466 | delta >> 20); | |
467 | } | |
bf0f6f24 IM |
468 | } |
469 | #endif | |
470 | } | |
471 | ||
ddc97297 PZ |
472 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
473 | { | |
474 | #ifdef CONFIG_SCHED_DEBUG | |
475 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
476 | ||
477 | if (d < 0) | |
478 | d = -d; | |
479 | ||
480 | if (d > 3*sysctl_sched_latency) | |
481 | schedstat_inc(cfs_rq, nr_spread_over); | |
482 | #endif | |
483 | } | |
484 | ||
aeb73b04 PZ |
485 | static void |
486 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
487 | { | |
67e9fb2a | 488 | u64 vruntime; |
aeb73b04 | 489 | |
67e9fb2a | 490 | vruntime = cfs_rq->min_vruntime; |
94dfb5e7 PZ |
491 | |
492 | if (sched_feat(USE_TREE_AVG)) { | |
493 | struct sched_entity *last = __pick_last_entity(cfs_rq); | |
494 | if (last) { | |
67e9fb2a PZ |
495 | vruntime += last->vruntime; |
496 | vruntime >>= 1; | |
94dfb5e7 | 497 | } |
67e9fb2a | 498 | } else if (sched_feat(APPROX_AVG) && cfs_rq->nr_running) |
647e7cac | 499 | vruntime += sched_vslice(cfs_rq)/2; |
94dfb5e7 PZ |
500 | |
501 | if (initial && sched_feat(START_DEBIT)) | |
647e7cac | 502 | vruntime += sched_vslice_add(cfs_rq, se); |
aeb73b04 | 503 | |
8465e792 | 504 | if (!initial) { |
94359f05 IM |
505 | if (sched_feat(NEW_FAIR_SLEEPERS)) |
506 | vruntime -= sysctl_sched_latency; | |
507 | ||
b8487b92 | 508 | vruntime = max_t(s64, vruntime, se->vruntime); |
aeb73b04 PZ |
509 | } |
510 | ||
67e9fb2a PZ |
511 | se->vruntime = vruntime; |
512 | ||
aeb73b04 PZ |
513 | } |
514 | ||
bf0f6f24 | 515 | static void |
83b699ed | 516 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
517 | { |
518 | /* | |
519 | * Update the fair clock. | |
520 | */ | |
b7cc0896 | 521 | update_curr(cfs_rq); |
bf0f6f24 | 522 | |
e9acbff6 | 523 | if (wakeup) { |
aeb73b04 | 524 | place_entity(cfs_rq, se, 0); |
2396af69 | 525 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 526 | } |
bf0f6f24 | 527 | |
d2417e5a | 528 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 529 | check_spread(cfs_rq, se); |
83b699ed SV |
530 | if (se != cfs_rq->curr) |
531 | __enqueue_entity(cfs_rq, se); | |
30cfdcfc | 532 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 IM |
533 | } |
534 | ||
535 | static void | |
525c2716 | 536 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 537 | { |
19b6a2e3 | 538 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 539 | if (sleep) { |
67e9fb2a | 540 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
541 | if (entity_is_task(se)) { |
542 | struct task_struct *tsk = task_of(se); | |
543 | ||
544 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 545 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 546 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 547 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 548 | } |
db36cc7d | 549 | #endif |
67e9fb2a PZ |
550 | } |
551 | ||
83b699ed | 552 | if (se != cfs_rq->curr) |
30cfdcfc DA |
553 | __dequeue_entity(cfs_rq, se); |
554 | account_entity_dequeue(cfs_rq, se); | |
bf0f6f24 IM |
555 | } |
556 | ||
557 | /* | |
558 | * Preempt the current task with a newly woken task if needed: | |
559 | */ | |
7c92e54f | 560 | static void |
2e09bf55 | 561 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 562 | { |
11697830 PZ |
563 | unsigned long ideal_runtime, delta_exec; |
564 | ||
6d0f0ebd | 565 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 PZ |
566 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
567 | if (delta_exec > ideal_runtime) | |
bf0f6f24 IM |
568 | resched_task(rq_of(cfs_rq)->curr); |
569 | } | |
570 | ||
83b699ed | 571 | static void |
8494f412 | 572 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 573 | { |
83b699ed SV |
574 | /* 'current' is not kept within the tree. */ |
575 | if (se->on_rq) { | |
576 | /* | |
577 | * Any task has to be enqueued before it get to execute on | |
578 | * a CPU. So account for the time it spent waiting on the | |
579 | * runqueue. | |
580 | */ | |
581 | update_stats_wait_end(cfs_rq, se); | |
582 | __dequeue_entity(cfs_rq, se); | |
583 | } | |
584 | ||
79303e9e | 585 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 586 | cfs_rq->curr = se; |
eba1ed4b IM |
587 | #ifdef CONFIG_SCHEDSTATS |
588 | /* | |
589 | * Track our maximum slice length, if the CPU's load is at | |
590 | * least twice that of our own weight (i.e. dont track it | |
591 | * when there are only lesser-weight tasks around): | |
592 | */ | |
495eca49 | 593 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
594 | se->slice_max = max(se->slice_max, |
595 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
596 | } | |
597 | #endif | |
4a55b450 | 598 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
599 | } |
600 | ||
9948f4b2 | 601 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 | 602 | { |
08ec3df5 | 603 | struct sched_entity *se = NULL; |
bf0f6f24 | 604 | |
08ec3df5 DA |
605 | if (first_fair(cfs_rq)) { |
606 | se = __pick_next_entity(cfs_rq); | |
607 | set_next_entity(cfs_rq, se); | |
608 | } | |
bf0f6f24 IM |
609 | |
610 | return se; | |
611 | } | |
612 | ||
ab6cde26 | 613 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
614 | { |
615 | /* | |
616 | * If still on the runqueue then deactivate_task() | |
617 | * was not called and update_curr() has to be done: | |
618 | */ | |
619 | if (prev->on_rq) | |
b7cc0896 | 620 | update_curr(cfs_rq); |
bf0f6f24 | 621 | |
c7e9b5b2 | 622 | update_stats_curr_end(cfs_rq, prev); |
bf0f6f24 | 623 | |
ddc97297 | 624 | check_spread(cfs_rq, prev); |
30cfdcfc | 625 | if (prev->on_rq) { |
5870db5b | 626 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
627 | /* Put 'current' back into the tree. */ |
628 | __enqueue_entity(cfs_rq, prev); | |
629 | } | |
429d43bc | 630 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
631 | } |
632 | ||
633 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
634 | { | |
bf0f6f24 | 635 | /* |
30cfdcfc | 636 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 637 | */ |
30cfdcfc | 638 | update_curr(cfs_rq); |
bf0f6f24 | 639 | |
2e09bf55 IM |
640 | if (cfs_rq->nr_running > 1) |
641 | check_preempt_tick(cfs_rq, curr); | |
bf0f6f24 IM |
642 | } |
643 | ||
644 | /************************************************** | |
645 | * CFS operations on tasks: | |
646 | */ | |
647 | ||
648 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
649 | ||
650 | /* Walk up scheduling entities hierarchy */ | |
651 | #define for_each_sched_entity(se) \ | |
652 | for (; se; se = se->parent) | |
653 | ||
654 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
655 | { | |
656 | return p->se.cfs_rq; | |
657 | } | |
658 | ||
659 | /* runqueue on which this entity is (to be) queued */ | |
660 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
661 | { | |
662 | return se->cfs_rq; | |
663 | } | |
664 | ||
665 | /* runqueue "owned" by this group */ | |
666 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
667 | { | |
668 | return grp->my_q; | |
669 | } | |
670 | ||
671 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
672 | * another cpu ('this_cpu') | |
673 | */ | |
674 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
675 | { | |
29f59db3 | 676 | return cfs_rq->tg->cfs_rq[this_cpu]; |
bf0f6f24 IM |
677 | } |
678 | ||
679 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
680 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
681 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
682 | ||
fad095a7 SV |
683 | /* Do the two (enqueued) entities belong to the same group ? */ |
684 | static inline int | |
685 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
bf0f6f24 | 686 | { |
fad095a7 | 687 | if (se->cfs_rq == pse->cfs_rq) |
bf0f6f24 IM |
688 | return 1; |
689 | ||
690 | return 0; | |
691 | } | |
692 | ||
fad095a7 SV |
693 | static inline struct sched_entity *parent_entity(struct sched_entity *se) |
694 | { | |
695 | return se->parent; | |
696 | } | |
697 | ||
bf0f6f24 IM |
698 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
699 | ||
700 | #define for_each_sched_entity(se) \ | |
701 | for (; se; se = NULL) | |
702 | ||
703 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
704 | { | |
705 | return &task_rq(p)->cfs; | |
706 | } | |
707 | ||
708 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
709 | { | |
710 | struct task_struct *p = task_of(se); | |
711 | struct rq *rq = task_rq(p); | |
712 | ||
713 | return &rq->cfs; | |
714 | } | |
715 | ||
716 | /* runqueue "owned" by this group */ | |
717 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
718 | { | |
719 | return NULL; | |
720 | } | |
721 | ||
722 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
723 | { | |
724 | return &cpu_rq(this_cpu)->cfs; | |
725 | } | |
726 | ||
727 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
728 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
729 | ||
fad095a7 SV |
730 | static inline int |
731 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
bf0f6f24 IM |
732 | { |
733 | return 1; | |
734 | } | |
735 | ||
fad095a7 SV |
736 | static inline struct sched_entity *parent_entity(struct sched_entity *se) |
737 | { | |
738 | return NULL; | |
739 | } | |
740 | ||
bf0f6f24 IM |
741 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
742 | ||
743 | /* | |
744 | * The enqueue_task method is called before nr_running is | |
745 | * increased. Here we update the fair scheduling stats and | |
746 | * then put the task into the rbtree: | |
747 | */ | |
fd390f6a | 748 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
749 | { |
750 | struct cfs_rq *cfs_rq; | |
751 | struct sched_entity *se = &p->se; | |
752 | ||
753 | for_each_sched_entity(se) { | |
754 | if (se->on_rq) | |
755 | break; | |
756 | cfs_rq = cfs_rq_of(se); | |
83b699ed | 757 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 758 | wakeup = 1; |
bf0f6f24 IM |
759 | } |
760 | } | |
761 | ||
762 | /* | |
763 | * The dequeue_task method is called before nr_running is | |
764 | * decreased. We remove the task from the rbtree and | |
765 | * update the fair scheduling stats: | |
766 | */ | |
f02231e5 | 767 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
768 | { |
769 | struct cfs_rq *cfs_rq; | |
770 | struct sched_entity *se = &p->se; | |
771 | ||
772 | for_each_sched_entity(se) { | |
773 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 774 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
775 | /* Don't dequeue parent if it has other entities besides us */ |
776 | if (cfs_rq->load.weight) | |
777 | break; | |
b9fa3df3 | 778 | sleep = 1; |
bf0f6f24 IM |
779 | } |
780 | } | |
781 | ||
782 | /* | |
1799e35d IM |
783 | * sched_yield() support is very simple - we dequeue and enqueue. |
784 | * | |
785 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 786 | */ |
4530d7ab | 787 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 788 | { |
72ea22f8 | 789 | struct cfs_rq *cfs_rq = task_cfs_rq(rq->curr); |
4530d7ab | 790 | struct sched_entity *rightmost, *se = &rq->curr->se; |
bf0f6f24 IM |
791 | |
792 | /* | |
1799e35d IM |
793 | * Are we the only task in the tree? |
794 | */ | |
795 | if (unlikely(cfs_rq->nr_running == 1)) | |
796 | return; | |
797 | ||
798 | if (likely(!sysctl_sched_compat_yield)) { | |
799 | __update_rq_clock(rq); | |
800 | /* | |
801 | * Dequeue and enqueue the task to update its | |
802 | * position within the tree: | |
803 | */ | |
2b1e315d | 804 | update_curr(cfs_rq); |
1799e35d IM |
805 | |
806 | return; | |
807 | } | |
808 | /* | |
809 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 810 | */ |
2b1e315d | 811 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
812 | /* |
813 | * Already in the rightmost position? | |
814 | */ | |
2b1e315d | 815 | if (unlikely(rightmost->vruntime < se->vruntime)) |
1799e35d IM |
816 | return; |
817 | ||
818 | /* | |
819 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
820 | * Upon rescheduling, sched_class::put_prev_task() will place |
821 | * 'current' within the tree based on its new key value. | |
1799e35d | 822 | */ |
30cfdcfc | 823 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
824 | } |
825 | ||
826 | /* | |
827 | * Preempt the current task with a newly woken task if needed: | |
828 | */ | |
2e09bf55 | 829 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
830 | { |
831 | struct task_struct *curr = rq->curr; | |
fad095a7 | 832 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
8651a86c | 833 | struct sched_entity *se = &curr->se, *pse = &p->se; |
fad095a7 | 834 | s64 delta; |
bf0f6f24 IM |
835 | |
836 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 837 | update_rq_clock(rq); |
b7cc0896 | 838 | update_curr(cfs_rq); |
bf0f6f24 IM |
839 | resched_task(curr); |
840 | return; | |
841 | } | |
842 | ||
fad095a7 SV |
843 | while (!is_same_group(se, pse)) { |
844 | se = parent_entity(se); | |
845 | pse = parent_entity(pse); | |
846 | } | |
8651a86c | 847 | |
fad095a7 | 848 | delta = se->vruntime - pse->vruntime; |
8651a86c | 849 | |
fad095a7 SV |
850 | if (delta > (s64)sysctl_sched_wakeup_granularity) |
851 | resched_task(curr); | |
bf0f6f24 IM |
852 | } |
853 | ||
fb8d4724 | 854 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
855 | { |
856 | struct cfs_rq *cfs_rq = &rq->cfs; | |
857 | struct sched_entity *se; | |
858 | ||
859 | if (unlikely(!cfs_rq->nr_running)) | |
860 | return NULL; | |
861 | ||
862 | do { | |
9948f4b2 | 863 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
864 | cfs_rq = group_cfs_rq(se); |
865 | } while (cfs_rq); | |
866 | ||
867 | return task_of(se); | |
868 | } | |
869 | ||
870 | /* | |
871 | * Account for a descheduled task: | |
872 | */ | |
31ee529c | 873 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
874 | { |
875 | struct sched_entity *se = &prev->se; | |
876 | struct cfs_rq *cfs_rq; | |
877 | ||
878 | for_each_sched_entity(se) { | |
879 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 880 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
881 | } |
882 | } | |
883 | ||
884 | /************************************************** | |
885 | * Fair scheduling class load-balancing methods: | |
886 | */ | |
887 | ||
888 | /* | |
889 | * Load-balancing iterator. Note: while the runqueue stays locked | |
890 | * during the whole iteration, the current task might be | |
891 | * dequeued so the iterator has to be dequeue-safe. Here we | |
892 | * achieve that by always pre-iterating before returning | |
893 | * the current task: | |
894 | */ | |
a9957449 | 895 | static struct task_struct * |
bf0f6f24 IM |
896 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) |
897 | { | |
898 | struct task_struct *p; | |
899 | ||
900 | if (!curr) | |
901 | return NULL; | |
902 | ||
903 | p = rb_entry(curr, struct task_struct, se.run_node); | |
904 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
905 | ||
906 | return p; | |
907 | } | |
908 | ||
909 | static struct task_struct *load_balance_start_fair(void *arg) | |
910 | { | |
911 | struct cfs_rq *cfs_rq = arg; | |
912 | ||
913 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
914 | } | |
915 | ||
916 | static struct task_struct *load_balance_next_fair(void *arg) | |
917 | { | |
918 | struct cfs_rq *cfs_rq = arg; | |
919 | ||
920 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
921 | } | |
922 | ||
a4ac01c3 | 923 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
924 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
925 | { | |
926 | struct sched_entity *curr; | |
927 | struct task_struct *p; | |
928 | ||
929 | if (!cfs_rq->nr_running) | |
930 | return MAX_PRIO; | |
931 | ||
9b5b7751 SV |
932 | curr = cfs_rq->curr; |
933 | if (!curr) | |
934 | curr = __pick_next_entity(cfs_rq); | |
935 | ||
bf0f6f24 IM |
936 | p = task_of(curr); |
937 | ||
938 | return p->prio; | |
939 | } | |
a4ac01c3 | 940 | #endif |
bf0f6f24 | 941 | |
43010659 | 942 | static unsigned long |
bf0f6f24 | 943 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
a4ac01c3 PW |
944 | unsigned long max_nr_move, unsigned long max_load_move, |
945 | struct sched_domain *sd, enum cpu_idle_type idle, | |
946 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
947 | { |
948 | struct cfs_rq *busy_cfs_rq; | |
949 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
950 | long rem_load_move = max_load_move; | |
951 | struct rq_iterator cfs_rq_iterator; | |
952 | ||
953 | cfs_rq_iterator.start = load_balance_start_fair; | |
954 | cfs_rq_iterator.next = load_balance_next_fair; | |
955 | ||
956 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 957 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 958 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 959 | long imbalance; |
bf0f6f24 | 960 | unsigned long maxload; |
bf0f6f24 IM |
961 | |
962 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
963 | ||
e56f31aa | 964 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
965 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
966 | if (imbalance <= 0) | |
967 | continue; | |
968 | ||
969 | /* Don't pull more than imbalance/2 */ | |
970 | imbalance /= 2; | |
971 | maxload = min(rem_load_move, imbalance); | |
972 | ||
a4ac01c3 PW |
973 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
974 | #else | |
e56f31aa | 975 | # define maxload rem_load_move |
a4ac01c3 | 976 | #endif |
bf0f6f24 IM |
977 | /* pass busy_cfs_rq argument into |
978 | * load_balance_[start|next]_fair iterators | |
979 | */ | |
980 | cfs_rq_iterator.arg = busy_cfs_rq; | |
981 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, | |
982 | max_nr_move, maxload, sd, idle, all_pinned, | |
a4ac01c3 | 983 | &load_moved, this_best_prio, &cfs_rq_iterator); |
bf0f6f24 IM |
984 | |
985 | total_nr_moved += nr_moved; | |
986 | max_nr_move -= nr_moved; | |
987 | rem_load_move -= load_moved; | |
988 | ||
989 | if (max_nr_move <= 0 || rem_load_move <= 0) | |
990 | break; | |
991 | } | |
992 | ||
43010659 | 993 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
994 | } |
995 | ||
996 | /* | |
997 | * scheduler tick hitting a task of our scheduling class: | |
998 | */ | |
999 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
1000 | { | |
1001 | struct cfs_rq *cfs_rq; | |
1002 | struct sched_entity *se = &curr->se; | |
1003 | ||
1004 | for_each_sched_entity(se) { | |
1005 | cfs_rq = cfs_rq_of(se); | |
1006 | entity_tick(cfs_rq, se); | |
1007 | } | |
1008 | } | |
1009 | ||
4d78e7b6 PZ |
1010 | #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
1011 | ||
bf0f6f24 IM |
1012 | /* |
1013 | * Share the fairness runtime between parent and child, thus the | |
1014 | * total amount of pressure for CPU stays equal - new tasks | |
1015 | * get a chance to run but frequent forkers are not allowed to | |
1016 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1017 | * the child is not running yet. | |
1018 | */ | |
ee0827d8 | 1019 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1020 | { |
1021 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1022 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
bf0f6f24 IM |
1023 | |
1024 | sched_info_queued(p); | |
1025 | ||
7109c442 | 1026 | update_curr(cfs_rq); |
aeb73b04 | 1027 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1028 | |
4d78e7b6 PZ |
1029 | if (sysctl_sched_child_runs_first && |
1030 | curr->vruntime < se->vruntime) { | |
87fefa38 | 1031 | /* |
edcb60a3 IM |
1032 | * Upon rescheduling, sched_class::put_prev_task() will place |
1033 | * 'current' within the tree based on its new key value. | |
1034 | */ | |
4d78e7b6 | 1035 | swap(curr->vruntime, se->vruntime); |
4d78e7b6 | 1036 | } |
bf0f6f24 | 1037 | |
e9acbff6 | 1038 | update_stats_enqueue(cfs_rq, se); |
ddc97297 PZ |
1039 | check_spread(cfs_rq, se); |
1040 | check_spread(cfs_rq, curr); | |
bf0f6f24 | 1041 | __enqueue_entity(cfs_rq, se); |
30cfdcfc | 1042 | account_entity_enqueue(cfs_rq, se); |
bb61c210 | 1043 | resched_task(rq->curr); |
bf0f6f24 IM |
1044 | } |
1045 | ||
83b699ed SV |
1046 | /* Account for a task changing its policy or group. |
1047 | * | |
1048 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1049 | * migrates between groups/classes. | |
1050 | */ | |
1051 | static void set_curr_task_fair(struct rq *rq) | |
1052 | { | |
1053 | struct sched_entity *se = &rq->curr->se; | |
1054 | ||
1055 | for_each_sched_entity(se) | |
1056 | set_next_entity(cfs_rq_of(se), se); | |
1057 | } | |
1058 | ||
bf0f6f24 IM |
1059 | /* |
1060 | * All the scheduling class methods: | |
1061 | */ | |
5522d5d5 IM |
1062 | static const struct sched_class fair_sched_class = { |
1063 | .next = &idle_sched_class, | |
bf0f6f24 IM |
1064 | .enqueue_task = enqueue_task_fair, |
1065 | .dequeue_task = dequeue_task_fair, | |
1066 | .yield_task = yield_task_fair, | |
1067 | ||
2e09bf55 | 1068 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1069 | |
1070 | .pick_next_task = pick_next_task_fair, | |
1071 | .put_prev_task = put_prev_task_fair, | |
1072 | ||
1073 | .load_balance = load_balance_fair, | |
1074 | ||
83b699ed | 1075 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 IM |
1076 | .task_tick = task_tick_fair, |
1077 | .task_new = task_new_fair, | |
1078 | }; | |
1079 | ||
1080 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1081 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1082 | { |
bf0f6f24 IM |
1083 | struct cfs_rq *cfs_rq; |
1084 | ||
75c28ace SV |
1085 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1086 | print_cfs_rq(m, cpu, &cpu_rq(cpu)->cfs); | |
1087 | #endif | |
c3b64f1e | 1088 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1089 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1090 | } |
1091 | #endif |