]>
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 | */ | |
172ac3db | 49 | unsigned int sysctl_sched_min_granularity __read_mostly = 2000000ULL; |
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. | |
71fd3714 | 61 | * (default: 25 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 | */ | |
2bd8e6d4 | 67 | const_debug unsigned int sysctl_sched_batch_wakeup_granularity = 25000000UL; |
bf0f6f24 IM |
68 | |
69 | /* | |
70 | * SCHED_OTHER wake-up granularity. | |
71 | * (default: 1 msec, units: nanoseconds) | |
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 | */ | |
2e09bf55 | 77 | const_debug unsigned int sysctl_sched_wakeup_granularity = 2000000UL; |
bf0f6f24 | 78 | |
bf0f6f24 IM |
79 | unsigned int sysctl_sched_runtime_limit __read_mostly; |
80 | ||
bf0f6f24 IM |
81 | extern struct sched_class fair_sched_class; |
82 | ||
83 | /************************************************************** | |
84 | * CFS operations on generic schedulable entities: | |
85 | */ | |
86 | ||
62160e3f | 87 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 88 | |
62160e3f | 89 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
90 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
91 | { | |
62160e3f | 92 | return cfs_rq->rq; |
bf0f6f24 IM |
93 | } |
94 | ||
62160e3f IM |
95 | /* An entity is a task if it doesn't "own" a runqueue */ |
96 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 97 | |
62160e3f | 98 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
bf0f6f24 | 99 | |
62160e3f IM |
100 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
101 | { | |
102 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
103 | } |
104 | ||
105 | #define entity_is_task(se) 1 | |
106 | ||
bf0f6f24 IM |
107 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
108 | ||
109 | static inline struct task_struct *task_of(struct sched_entity *se) | |
110 | { | |
111 | return container_of(se, struct task_struct, se); | |
112 | } | |
113 | ||
114 | ||
115 | /************************************************************** | |
116 | * Scheduling class tree data structure manipulation methods: | |
117 | */ | |
118 | ||
e9acbff6 IM |
119 | static inline void |
120 | set_leftmost(struct cfs_rq *cfs_rq, struct rb_node *leftmost) | |
121 | { | |
122 | struct sched_entity *se; | |
123 | ||
124 | cfs_rq->rb_leftmost = leftmost; | |
125 | if (leftmost) { | |
126 | se = rb_entry(leftmost, struct sched_entity, run_node); | |
127 | cfs_rq->min_vruntime = max(se->vruntime, | |
128 | cfs_rq->min_vruntime); | |
129 | } | |
130 | } | |
131 | ||
bf0f6f24 IM |
132 | /* |
133 | * Enqueue an entity into the rb-tree: | |
134 | */ | |
19ccd97a | 135 | static void |
bf0f6f24 IM |
136 | __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
137 | { | |
138 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
139 | struct rb_node *parent = NULL; | |
140 | struct sched_entity *entry; | |
141 | s64 key = se->fair_key; | |
142 | int leftmost = 1; | |
143 | ||
144 | /* | |
145 | * Find the right place in the rbtree: | |
146 | */ | |
147 | while (*link) { | |
148 | parent = *link; | |
149 | entry = rb_entry(parent, struct sched_entity, run_node); | |
150 | /* | |
151 | * We dont care about collisions. Nodes with | |
152 | * the same key stay together. | |
153 | */ | |
154 | if (key - entry->fair_key < 0) { | |
155 | link = &parent->rb_left; | |
156 | } else { | |
157 | link = &parent->rb_right; | |
158 | leftmost = 0; | |
159 | } | |
160 | } | |
161 | ||
162 | /* | |
163 | * Maintain a cache of leftmost tree entries (it is frequently | |
164 | * used): | |
165 | */ | |
166 | if (leftmost) | |
e9acbff6 | 167 | set_leftmost(cfs_rq, &se->run_node); |
bf0f6f24 IM |
168 | |
169 | rb_link_node(&se->run_node, parent, link); | |
170 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
171 | update_load_add(&cfs_rq->load, se->load.weight); | |
172 | cfs_rq->nr_running++; | |
173 | se->on_rq = 1; | |
a206c072 IM |
174 | |
175 | schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); | |
bf0f6f24 IM |
176 | } |
177 | ||
19ccd97a | 178 | static void |
bf0f6f24 IM |
179 | __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
180 | { | |
181 | if (cfs_rq->rb_leftmost == &se->run_node) | |
e9acbff6 IM |
182 | set_leftmost(cfs_rq, rb_next(&se->run_node)); |
183 | ||
bf0f6f24 IM |
184 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
185 | update_load_sub(&cfs_rq->load, se->load.weight); | |
186 | cfs_rq->nr_running--; | |
187 | se->on_rq = 0; | |
a206c072 IM |
188 | |
189 | schedstat_add(cfs_rq, wait_runtime, -se->wait_runtime); | |
bf0f6f24 IM |
190 | } |
191 | ||
192 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
193 | { | |
194 | return cfs_rq->rb_leftmost; | |
195 | } | |
196 | ||
197 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
198 | { | |
199 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
200 | } | |
201 | ||
202 | /************************************************************** | |
203 | * Scheduling class statistics methods: | |
204 | */ | |
205 | ||
4d78e7b6 PZ |
206 | static u64 __sched_period(unsigned long nr_running) |
207 | { | |
208 | u64 period = sysctl_sched_latency; | |
209 | unsigned long nr_latency = | |
210 | sysctl_sched_latency / sysctl_sched_min_granularity; | |
211 | ||
212 | if (unlikely(nr_running > nr_latency)) { | |
213 | period *= nr_running; | |
214 | do_div(period, nr_latency); | |
215 | } | |
216 | ||
217 | return period; | |
218 | } | |
219 | ||
6d0f0ebd | 220 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 221 | { |
6d0f0ebd | 222 | u64 period = __sched_period(cfs_rq->nr_running); |
21805085 | 223 | |
6d0f0ebd PZ |
224 | period *= se->load.weight; |
225 | do_div(period, cfs_rq->load.weight); | |
21805085 | 226 | |
6d0f0ebd | 227 | return period; |
bf0f6f24 IM |
228 | } |
229 | ||
230 | static inline void | |
231 | limit_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
232 | { | |
233 | long limit = sysctl_sched_runtime_limit; | |
234 | ||
235 | /* | |
236 | * Niced tasks have the same history dynamic range as | |
237 | * non-niced tasks: | |
238 | */ | |
239 | if (unlikely(se->wait_runtime > limit)) { | |
240 | se->wait_runtime = limit; | |
241 | schedstat_inc(se, wait_runtime_overruns); | |
242 | schedstat_inc(cfs_rq, wait_runtime_overruns); | |
243 | } | |
244 | if (unlikely(se->wait_runtime < -limit)) { | |
245 | se->wait_runtime = -limit; | |
246 | schedstat_inc(se, wait_runtime_underruns); | |
247 | schedstat_inc(cfs_rq, wait_runtime_underruns); | |
248 | } | |
249 | } | |
250 | ||
251 | static inline void | |
252 | __add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
253 | { | |
254 | se->wait_runtime += delta; | |
255 | schedstat_add(se, sum_wait_runtime, delta); | |
256 | limit_wait_runtime(cfs_rq, se); | |
257 | } | |
258 | ||
259 | static void | |
260 | add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
261 | { | |
262 | schedstat_add(cfs_rq, wait_runtime, -se->wait_runtime); | |
263 | __add_wait_runtime(cfs_rq, se, delta); | |
264 | schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); | |
265 | } | |
266 | ||
267 | /* | |
268 | * Update the current task's runtime statistics. Skip current tasks that | |
269 | * are not in our scheduling class. | |
270 | */ | |
271 | static inline void | |
8ebc91d9 IM |
272 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
273 | unsigned long delta_exec) | |
bf0f6f24 | 274 | { |
e9acbff6 | 275 | unsigned long delta, delta_fair, delta_mine, delta_exec_weighted; |
bf0f6f24 IM |
276 | struct load_weight *lw = &cfs_rq->load; |
277 | unsigned long load = lw->weight; | |
278 | ||
8179ca23 | 279 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
280 | |
281 | curr->sum_exec_runtime += delta_exec; | |
282 | cfs_rq->exec_clock += delta_exec; | |
e9acbff6 IM |
283 | delta_exec_weighted = delta_exec; |
284 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { | |
285 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, | |
286 | &curr->load); | |
287 | } | |
288 | curr->vruntime += delta_exec_weighted; | |
bf0f6f24 | 289 | |
6cb58195 IM |
290 | if (!sched_feat(FAIR_SLEEPERS)) |
291 | return; | |
292 | ||
fd8bb43e IM |
293 | if (unlikely(!load)) |
294 | return; | |
295 | ||
bf0f6f24 IM |
296 | delta_fair = calc_delta_fair(delta_exec, lw); |
297 | delta_mine = calc_delta_mine(delta_exec, curr->load.weight, lw); | |
298 | ||
5f01d519 | 299 | if (cfs_rq->sleeper_bonus > sysctl_sched_min_granularity) { |
ea0aa3b2 | 300 | delta = min((u64)delta_mine, cfs_rq->sleeper_bonus); |
b2133c8b IM |
301 | delta = min(delta, (unsigned long)( |
302 | (long)sysctl_sched_runtime_limit - curr->wait_runtime)); | |
bf0f6f24 IM |
303 | cfs_rq->sleeper_bonus -= delta; |
304 | delta_mine -= delta; | |
305 | } | |
306 | ||
307 | cfs_rq->fair_clock += delta_fair; | |
308 | /* | |
309 | * We executed delta_exec amount of time on the CPU, | |
310 | * but we were only entitled to delta_mine amount of | |
311 | * time during that period (if nr_running == 1 then | |
312 | * the two values are equal) | |
313 | * [Note: delta_mine - delta_exec is negative]: | |
314 | */ | |
315 | add_wait_runtime(cfs_rq, curr, delta_mine - delta_exec); | |
316 | } | |
317 | ||
b7cc0896 | 318 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 319 | { |
429d43bc | 320 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 321 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
322 | unsigned long delta_exec; |
323 | ||
324 | if (unlikely(!curr)) | |
325 | return; | |
326 | ||
327 | /* | |
328 | * Get the amount of time the current task was running | |
329 | * since the last time we changed load (this cannot | |
330 | * overflow on 32 bits): | |
331 | */ | |
8ebc91d9 | 332 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 333 | |
8ebc91d9 IM |
334 | __update_curr(cfs_rq, curr, delta_exec); |
335 | curr->exec_start = now; | |
bf0f6f24 IM |
336 | } |
337 | ||
338 | static inline void | |
5870db5b | 339 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
340 | { |
341 | se->wait_start_fair = cfs_rq->fair_clock; | |
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 | * Update the key: | |
369 | */ | |
e9acbff6 | 370 | se->fair_key = se->vruntime; |
bf0f6f24 IM |
371 | } |
372 | ||
373 | /* | |
374 | * Note: must be called with a freshly updated rq->fair_clock. | |
375 | */ | |
376 | static inline void | |
8ebc91d9 IM |
377 | __update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, |
378 | unsigned long delta_fair) | |
bf0f6f24 | 379 | { |
d281918d IM |
380 | schedstat_set(se->wait_max, max(se->wait_max, |
381 | rq_of(cfs_rq)->clock - se->wait_start)); | |
bf0f6f24 | 382 | |
08e2388a | 383 | delta_fair = calc_weighted(delta_fair, se); |
bf0f6f24 IM |
384 | |
385 | add_wait_runtime(cfs_rq, se, delta_fair); | |
386 | } | |
387 | ||
388 | static void | |
9ef0a961 | 389 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
390 | { |
391 | unsigned long delta_fair; | |
392 | ||
b77d69db IM |
393 | if (unlikely(!se->wait_start_fair)) |
394 | return; | |
395 | ||
bf0f6f24 IM |
396 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), |
397 | (u64)(cfs_rq->fair_clock - se->wait_start_fair)); | |
398 | ||
8ebc91d9 | 399 | __update_stats_wait_end(cfs_rq, se, delta_fair); |
bf0f6f24 IM |
400 | |
401 | se->wait_start_fair = 0; | |
6cfb0d5d | 402 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
403 | } |
404 | ||
405 | static inline void | |
19b6a2e3 | 406 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 407 | { |
b7cc0896 | 408 | update_curr(cfs_rq); |
bf0f6f24 IM |
409 | /* |
410 | * Mark the end of the wait period if dequeueing a | |
411 | * waiting task: | |
412 | */ | |
429d43bc | 413 | if (se != cfs_rq->curr) |
9ef0a961 | 414 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
415 | } |
416 | ||
417 | /* | |
418 | * We are picking a new current task - update its stats: | |
419 | */ | |
420 | static inline void | |
79303e9e | 421 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
422 | { |
423 | /* | |
424 | * We are starting a new run period: | |
425 | */ | |
d281918d | 426 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
427 | } |
428 | ||
429 | /* | |
430 | * We are descheduling a task - update its stats: | |
431 | */ | |
432 | static inline void | |
c7e9b5b2 | 433 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
434 | { |
435 | se->exec_start = 0; | |
436 | } | |
437 | ||
438 | /************************************************** | |
439 | * Scheduling class queueing methods: | |
440 | */ | |
441 | ||
8ebc91d9 IM |
442 | static void __enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, |
443 | unsigned long delta_fair) | |
bf0f6f24 | 444 | { |
8ebc91d9 | 445 | unsigned long load = cfs_rq->load.weight; |
bf0f6f24 IM |
446 | long prev_runtime; |
447 | ||
b2133c8b IM |
448 | /* |
449 | * Do not boost sleepers if there's too much bonus 'in flight' | |
450 | * already: | |
451 | */ | |
452 | if (unlikely(cfs_rq->sleeper_bonus > sysctl_sched_runtime_limit)) | |
453 | return; | |
454 | ||
e59c80c5 | 455 | if (sched_feat(SLEEPER_LOAD_AVG)) |
bf0f6f24 IM |
456 | load = rq_of(cfs_rq)->cpu_load[2]; |
457 | ||
bf0f6f24 IM |
458 | /* |
459 | * Fix up delta_fair with the effect of us running | |
460 | * during the whole sleep period: | |
461 | */ | |
e59c80c5 | 462 | if (sched_feat(SLEEPER_AVG)) |
bf0f6f24 IM |
463 | delta_fair = div64_likely32((u64)delta_fair * load, |
464 | load + se->load.weight); | |
465 | ||
08e2388a | 466 | delta_fair = calc_weighted(delta_fair, se); |
bf0f6f24 IM |
467 | |
468 | prev_runtime = se->wait_runtime; | |
469 | __add_wait_runtime(cfs_rq, se, delta_fair); | |
470 | delta_fair = se->wait_runtime - prev_runtime; | |
471 | ||
472 | /* | |
473 | * Track the amount of bonus we've given to sleepers: | |
474 | */ | |
475 | cfs_rq->sleeper_bonus += delta_fair; | |
bf0f6f24 IM |
476 | } |
477 | ||
2396af69 | 478 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
479 | { |
480 | struct task_struct *tsk = task_of(se); | |
481 | unsigned long delta_fair; | |
482 | ||
483 | if ((entity_is_task(se) && tsk->policy == SCHED_BATCH) || | |
e59c80c5 | 484 | !sched_feat(FAIR_SLEEPERS)) |
bf0f6f24 IM |
485 | return; |
486 | ||
487 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), | |
488 | (u64)(cfs_rq->fair_clock - se->sleep_start_fair)); | |
489 | ||
8ebc91d9 | 490 | __enqueue_sleeper(cfs_rq, se, delta_fair); |
bf0f6f24 IM |
491 | |
492 | se->sleep_start_fair = 0; | |
493 | ||
494 | #ifdef CONFIG_SCHEDSTATS | |
495 | if (se->sleep_start) { | |
d281918d | 496 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
497 | |
498 | if ((s64)delta < 0) | |
499 | delta = 0; | |
500 | ||
501 | if (unlikely(delta > se->sleep_max)) | |
502 | se->sleep_max = delta; | |
503 | ||
504 | se->sleep_start = 0; | |
505 | se->sum_sleep_runtime += delta; | |
506 | } | |
507 | if (se->block_start) { | |
d281918d | 508 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
509 | |
510 | if ((s64)delta < 0) | |
511 | delta = 0; | |
512 | ||
513 | if (unlikely(delta > se->block_max)) | |
514 | se->block_max = delta; | |
515 | ||
516 | se->block_start = 0; | |
517 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
518 | |
519 | /* | |
520 | * Blocking time is in units of nanosecs, so shift by 20 to | |
521 | * get a milliseconds-range estimation of the amount of | |
522 | * time that the task spent sleeping: | |
523 | */ | |
524 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
525 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), | |
526 | delta >> 20); | |
527 | } | |
bf0f6f24 IM |
528 | } |
529 | #endif | |
530 | } | |
531 | ||
532 | static void | |
668031ca | 533 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
534 | { |
535 | /* | |
536 | * Update the fair clock. | |
537 | */ | |
b7cc0896 | 538 | update_curr(cfs_rq); |
bf0f6f24 | 539 | |
e9acbff6 IM |
540 | if (wakeup) { |
541 | u64 min_runtime, latency; | |
542 | ||
543 | min_runtime = cfs_rq->min_vruntime; | |
544 | min_runtime += sysctl_sched_latency/2; | |
545 | ||
546 | if (sched_feat(NEW_FAIR_SLEEPERS)) { | |
547 | latency = calc_weighted(sysctl_sched_latency, se); | |
548 | if (min_runtime > latency) | |
549 | min_runtime -= latency; | |
550 | } | |
551 | ||
552 | se->vruntime = max(se->vruntime, min_runtime); | |
553 | ||
2396af69 | 554 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 555 | } |
bf0f6f24 | 556 | |
d2417e5a | 557 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 IM |
558 | __enqueue_entity(cfs_rq, se); |
559 | } | |
560 | ||
561 | static void | |
525c2716 | 562 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 563 | { |
19b6a2e3 | 564 | update_stats_dequeue(cfs_rq, se); |
bf0f6f24 IM |
565 | if (sleep) { |
566 | se->sleep_start_fair = cfs_rq->fair_clock; | |
567 | #ifdef CONFIG_SCHEDSTATS | |
568 | if (entity_is_task(se)) { | |
569 | struct task_struct *tsk = task_of(se); | |
570 | ||
571 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 572 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 573 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 574 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 575 | } |
bf0f6f24 IM |
576 | #endif |
577 | } | |
578 | __dequeue_entity(cfs_rq, se); | |
579 | } | |
580 | ||
581 | /* | |
582 | * Preempt the current task with a newly woken task if needed: | |
583 | */ | |
7c92e54f | 584 | static void |
2e09bf55 | 585 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 586 | { |
11697830 PZ |
587 | unsigned long ideal_runtime, delta_exec; |
588 | ||
6d0f0ebd | 589 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 PZ |
590 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
591 | if (delta_exec > ideal_runtime) | |
bf0f6f24 IM |
592 | resched_task(rq_of(cfs_rq)->curr); |
593 | } | |
594 | ||
595 | static inline void | |
8494f412 | 596 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
597 | { |
598 | /* | |
599 | * Any task has to be enqueued before it get to execute on | |
600 | * a CPU. So account for the time it spent waiting on the | |
601 | * runqueue. (note, here we rely on pick_next_task() having | |
602 | * done a put_prev_task_fair() shortly before this, which | |
603 | * updated rq->fair_clock - used by update_stats_wait_end()) | |
604 | */ | |
9ef0a961 | 605 | update_stats_wait_end(cfs_rq, se); |
79303e9e | 606 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 607 | cfs_rq->curr = se; |
eba1ed4b IM |
608 | #ifdef CONFIG_SCHEDSTATS |
609 | /* | |
610 | * Track our maximum slice length, if the CPU's load is at | |
611 | * least twice that of our own weight (i.e. dont track it | |
612 | * when there are only lesser-weight tasks around): | |
613 | */ | |
614 | if (rq_of(cfs_rq)->ls.load.weight >= 2*se->load.weight) { | |
615 | se->slice_max = max(se->slice_max, | |
616 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
617 | } | |
618 | #endif | |
4a55b450 | 619 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
620 | } |
621 | ||
9948f4b2 | 622 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 IM |
623 | { |
624 | struct sched_entity *se = __pick_next_entity(cfs_rq); | |
625 | ||
8494f412 | 626 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
627 | |
628 | return se; | |
629 | } | |
630 | ||
ab6cde26 | 631 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
632 | { |
633 | /* | |
634 | * If still on the runqueue then deactivate_task() | |
635 | * was not called and update_curr() has to be done: | |
636 | */ | |
637 | if (prev->on_rq) | |
b7cc0896 | 638 | update_curr(cfs_rq); |
bf0f6f24 | 639 | |
c7e9b5b2 | 640 | update_stats_curr_end(cfs_rq, prev); |
bf0f6f24 IM |
641 | |
642 | if (prev->on_rq) | |
5870db5b | 643 | update_stats_wait_start(cfs_rq, prev); |
429d43bc | 644 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
645 | } |
646 | ||
647 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
648 | { | |
bf0f6f24 IM |
649 | /* |
650 | * Dequeue and enqueue the task to update its | |
651 | * position within the tree: | |
652 | */ | |
525c2716 | 653 | dequeue_entity(cfs_rq, curr, 0); |
668031ca | 654 | enqueue_entity(cfs_rq, curr, 0); |
bf0f6f24 | 655 | |
2e09bf55 IM |
656 | if (cfs_rq->nr_running > 1) |
657 | check_preempt_tick(cfs_rq, curr); | |
bf0f6f24 IM |
658 | } |
659 | ||
660 | /************************************************** | |
661 | * CFS operations on tasks: | |
662 | */ | |
663 | ||
664 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
665 | ||
666 | /* Walk up scheduling entities hierarchy */ | |
667 | #define for_each_sched_entity(se) \ | |
668 | for (; se; se = se->parent) | |
669 | ||
670 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
671 | { | |
672 | return p->se.cfs_rq; | |
673 | } | |
674 | ||
675 | /* runqueue on which this entity is (to be) queued */ | |
676 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
677 | { | |
678 | return se->cfs_rq; | |
679 | } | |
680 | ||
681 | /* runqueue "owned" by this group */ | |
682 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
683 | { | |
684 | return grp->my_q; | |
685 | } | |
686 | ||
687 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
688 | * another cpu ('this_cpu') | |
689 | */ | |
690 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
691 | { | |
692 | /* A later patch will take group into account */ | |
693 | return &cpu_rq(this_cpu)->cfs; | |
694 | } | |
695 | ||
696 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
697 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
698 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
699 | ||
700 | /* Do the two (enqueued) tasks belong to the same group ? */ | |
701 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
702 | { | |
703 | if (curr->se.cfs_rq == p->se.cfs_rq) | |
704 | return 1; | |
705 | ||
706 | return 0; | |
707 | } | |
708 | ||
709 | #else /* CONFIG_FAIR_GROUP_SCHED */ | |
710 | ||
711 | #define for_each_sched_entity(se) \ | |
712 | for (; se; se = NULL) | |
713 | ||
714 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
715 | { | |
716 | return &task_rq(p)->cfs; | |
717 | } | |
718 | ||
719 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
720 | { | |
721 | struct task_struct *p = task_of(se); | |
722 | struct rq *rq = task_rq(p); | |
723 | ||
724 | return &rq->cfs; | |
725 | } | |
726 | ||
727 | /* runqueue "owned" by this group */ | |
728 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
729 | { | |
730 | return NULL; | |
731 | } | |
732 | ||
733 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
734 | { | |
735 | return &cpu_rq(this_cpu)->cfs; | |
736 | } | |
737 | ||
738 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
739 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
740 | ||
741 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
742 | { | |
743 | return 1; | |
744 | } | |
745 | ||
746 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
747 | ||
748 | /* | |
749 | * The enqueue_task method is called before nr_running is | |
750 | * increased. Here we update the fair scheduling stats and | |
751 | * then put the task into the rbtree: | |
752 | */ | |
fd390f6a | 753 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
754 | { |
755 | struct cfs_rq *cfs_rq; | |
756 | struct sched_entity *se = &p->se; | |
757 | ||
758 | for_each_sched_entity(se) { | |
759 | if (se->on_rq) | |
760 | break; | |
761 | cfs_rq = cfs_rq_of(se); | |
668031ca | 762 | enqueue_entity(cfs_rq, se, wakeup); |
bf0f6f24 IM |
763 | } |
764 | } | |
765 | ||
766 | /* | |
767 | * The dequeue_task method is called before nr_running is | |
768 | * decreased. We remove the task from the rbtree and | |
769 | * update the fair scheduling stats: | |
770 | */ | |
f02231e5 | 771 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
772 | { |
773 | struct cfs_rq *cfs_rq; | |
774 | struct sched_entity *se = &p->se; | |
775 | ||
776 | for_each_sched_entity(se) { | |
777 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 778 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
779 | /* Don't dequeue parent if it has other entities besides us */ |
780 | if (cfs_rq->load.weight) | |
781 | break; | |
782 | } | |
783 | } | |
784 | ||
785 | /* | |
1799e35d IM |
786 | * sched_yield() support is very simple - we dequeue and enqueue. |
787 | * | |
788 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 IM |
789 | */ |
790 | static void yield_task_fair(struct rq *rq, struct task_struct *p) | |
791 | { | |
792 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1799e35d IM |
793 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
794 | struct sched_entity *rightmost, *se = &p->se; | |
795 | struct rb_node *parent; | |
bf0f6f24 IM |
796 | |
797 | /* | |
1799e35d IM |
798 | * Are we the only task in the tree? |
799 | */ | |
800 | if (unlikely(cfs_rq->nr_running == 1)) | |
801 | return; | |
802 | ||
803 | if (likely(!sysctl_sched_compat_yield)) { | |
804 | __update_rq_clock(rq); | |
805 | /* | |
806 | * Dequeue and enqueue the task to update its | |
807 | * position within the tree: | |
808 | */ | |
809 | dequeue_entity(cfs_rq, &p->se, 0); | |
810 | enqueue_entity(cfs_rq, &p->se, 0); | |
811 | ||
812 | return; | |
813 | } | |
814 | /* | |
815 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 816 | */ |
1799e35d IM |
817 | do { |
818 | parent = *link; | |
819 | link = &parent->rb_right; | |
820 | } while (*link); | |
821 | ||
822 | rightmost = rb_entry(parent, struct sched_entity, run_node); | |
823 | /* | |
824 | * Already in the rightmost position? | |
825 | */ | |
826 | if (unlikely(rightmost == se)) | |
827 | return; | |
828 | ||
829 | /* | |
830 | * Minimally necessary key value to be last in the tree: | |
831 | */ | |
832 | se->fair_key = rightmost->fair_key + 1; | |
833 | ||
834 | if (cfs_rq->rb_leftmost == &se->run_node) | |
835 | cfs_rq->rb_leftmost = rb_next(&se->run_node); | |
836 | /* | |
837 | * Relink the task to the rightmost position: | |
838 | */ | |
839 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); | |
840 | rb_link_node(&se->run_node, parent, link); | |
841 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
842 | } |
843 | ||
844 | /* | |
845 | * Preempt the current task with a newly woken task if needed: | |
846 | */ | |
2e09bf55 | 847 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
848 | { |
849 | struct task_struct *curr = rq->curr; | |
850 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
bf0f6f24 IM |
851 | |
852 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 853 | update_rq_clock(rq); |
b7cc0896 | 854 | update_curr(cfs_rq); |
bf0f6f24 IM |
855 | resched_task(curr); |
856 | return; | |
857 | } | |
2e09bf55 IM |
858 | if (is_same_group(curr, p)) { |
859 | s64 delta = curr->se.vruntime - p->se.vruntime; | |
bf0f6f24 | 860 | |
2e09bf55 IM |
861 | if (delta > (s64)sysctl_sched_wakeup_granularity) |
862 | resched_task(curr); | |
863 | } | |
bf0f6f24 IM |
864 | } |
865 | ||
fb8d4724 | 866 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
867 | { |
868 | struct cfs_rq *cfs_rq = &rq->cfs; | |
869 | struct sched_entity *se; | |
870 | ||
871 | if (unlikely(!cfs_rq->nr_running)) | |
872 | return NULL; | |
873 | ||
874 | do { | |
9948f4b2 | 875 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
876 | cfs_rq = group_cfs_rq(se); |
877 | } while (cfs_rq); | |
878 | ||
879 | return task_of(se); | |
880 | } | |
881 | ||
882 | /* | |
883 | * Account for a descheduled task: | |
884 | */ | |
31ee529c | 885 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
886 | { |
887 | struct sched_entity *se = &prev->se; | |
888 | struct cfs_rq *cfs_rq; | |
889 | ||
890 | for_each_sched_entity(se) { | |
891 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 892 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
893 | } |
894 | } | |
895 | ||
896 | /************************************************** | |
897 | * Fair scheduling class load-balancing methods: | |
898 | */ | |
899 | ||
900 | /* | |
901 | * Load-balancing iterator. Note: while the runqueue stays locked | |
902 | * during the whole iteration, the current task might be | |
903 | * dequeued so the iterator has to be dequeue-safe. Here we | |
904 | * achieve that by always pre-iterating before returning | |
905 | * the current task: | |
906 | */ | |
907 | static inline struct task_struct * | |
908 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) | |
909 | { | |
910 | struct task_struct *p; | |
911 | ||
912 | if (!curr) | |
913 | return NULL; | |
914 | ||
915 | p = rb_entry(curr, struct task_struct, se.run_node); | |
916 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
917 | ||
918 | return p; | |
919 | } | |
920 | ||
921 | static struct task_struct *load_balance_start_fair(void *arg) | |
922 | { | |
923 | struct cfs_rq *cfs_rq = arg; | |
924 | ||
925 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
926 | } | |
927 | ||
928 | static struct task_struct *load_balance_next_fair(void *arg) | |
929 | { | |
930 | struct cfs_rq *cfs_rq = arg; | |
931 | ||
932 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
933 | } | |
934 | ||
a4ac01c3 | 935 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
936 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
937 | { | |
938 | struct sched_entity *curr; | |
939 | struct task_struct *p; | |
940 | ||
941 | if (!cfs_rq->nr_running) | |
942 | return MAX_PRIO; | |
943 | ||
944 | curr = __pick_next_entity(cfs_rq); | |
945 | p = task_of(curr); | |
946 | ||
947 | return p->prio; | |
948 | } | |
a4ac01c3 | 949 | #endif |
bf0f6f24 | 950 | |
43010659 | 951 | static unsigned long |
bf0f6f24 | 952 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
a4ac01c3 PW |
953 | unsigned long max_nr_move, unsigned long max_load_move, |
954 | struct sched_domain *sd, enum cpu_idle_type idle, | |
955 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
956 | { |
957 | struct cfs_rq *busy_cfs_rq; | |
958 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
959 | long rem_load_move = max_load_move; | |
960 | struct rq_iterator cfs_rq_iterator; | |
961 | ||
962 | cfs_rq_iterator.start = load_balance_start_fair; | |
963 | cfs_rq_iterator.next = load_balance_next_fair; | |
964 | ||
965 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 966 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 967 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 968 | long imbalance; |
bf0f6f24 | 969 | unsigned long maxload; |
bf0f6f24 IM |
970 | |
971 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
972 | ||
e56f31aa | 973 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
974 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
975 | if (imbalance <= 0) | |
976 | continue; | |
977 | ||
978 | /* Don't pull more than imbalance/2 */ | |
979 | imbalance /= 2; | |
980 | maxload = min(rem_load_move, imbalance); | |
981 | ||
a4ac01c3 PW |
982 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
983 | #else | |
e56f31aa | 984 | # define maxload rem_load_move |
a4ac01c3 | 985 | #endif |
bf0f6f24 IM |
986 | /* pass busy_cfs_rq argument into |
987 | * load_balance_[start|next]_fair iterators | |
988 | */ | |
989 | cfs_rq_iterator.arg = busy_cfs_rq; | |
990 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, | |
991 | max_nr_move, maxload, sd, idle, all_pinned, | |
a4ac01c3 | 992 | &load_moved, this_best_prio, &cfs_rq_iterator); |
bf0f6f24 IM |
993 | |
994 | total_nr_moved += nr_moved; | |
995 | max_nr_move -= nr_moved; | |
996 | rem_load_move -= load_moved; | |
997 | ||
998 | if (max_nr_move <= 0 || rem_load_move <= 0) | |
999 | break; | |
1000 | } | |
1001 | ||
43010659 | 1002 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
1003 | } |
1004 | ||
1005 | /* | |
1006 | * scheduler tick hitting a task of our scheduling class: | |
1007 | */ | |
1008 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
1009 | { | |
1010 | struct cfs_rq *cfs_rq; | |
1011 | struct sched_entity *se = &curr->se; | |
1012 | ||
1013 | for_each_sched_entity(se) { | |
1014 | cfs_rq = cfs_rq_of(se); | |
1015 | entity_tick(cfs_rq, se); | |
1016 | } | |
1017 | } | |
1018 | ||
4d78e7b6 PZ |
1019 | #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
1020 | ||
bf0f6f24 IM |
1021 | /* |
1022 | * Share the fairness runtime between parent and child, thus the | |
1023 | * total amount of pressure for CPU stays equal - new tasks | |
1024 | * get a chance to run but frequent forkers are not allowed to | |
1025 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1026 | * the child is not running yet. | |
1027 | */ | |
ee0827d8 | 1028 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1029 | { |
1030 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1031 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
bf0f6f24 IM |
1032 | |
1033 | sched_info_queued(p); | |
1034 | ||
7109c442 | 1035 | update_curr(cfs_rq); |
4d78e7b6 | 1036 | se->vruntime = cfs_rq->min_vruntime; |
d2417e5a | 1037 | update_stats_enqueue(cfs_rq, se); |
4d78e7b6 | 1038 | |
bf0f6f24 IM |
1039 | /* |
1040 | * The first wait is dominated by the child-runs-first logic, | |
1041 | * so do not credit it with that waiting time yet: | |
1042 | */ | |
e59c80c5 | 1043 | if (sched_feat(SKIP_INITIAL)) |
9f508f82 | 1044 | se->wait_start_fair = 0; |
bf0f6f24 IM |
1045 | |
1046 | /* | |
1047 | * The statistical average of wait_runtime is about | |
1048 | * -granularity/2, so initialize the task with that: | |
1049 | */ | |
e59c80c5 | 1050 | if (sched_feat(START_DEBIT)) |
4d78e7b6 PZ |
1051 | se->wait_runtime = -(__sched_period(cfs_rq->nr_running+1) / 2); |
1052 | ||
1053 | if (sysctl_sched_child_runs_first && | |
1054 | curr->vruntime < se->vruntime) { | |
1055 | ||
1056 | dequeue_entity(cfs_rq, curr, 0); | |
1057 | swap(curr->vruntime, se->vruntime); | |
1058 | enqueue_entity(cfs_rq, curr, 0); | |
1059 | } | |
bf0f6f24 | 1060 | |
e9acbff6 | 1061 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 | 1062 | __enqueue_entity(cfs_rq, se); |
bb61c210 | 1063 | resched_task(rq->curr); |
bf0f6f24 IM |
1064 | } |
1065 | ||
1066 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1067 | /* Account for a task changing its policy or group. | |
1068 | * | |
1069 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1070 | * migrates between groups/classes. | |
1071 | */ | |
1072 | static void set_curr_task_fair(struct rq *rq) | |
1073 | { | |
7c6c16f3 | 1074 | struct sched_entity *se = &rq->curr->se; |
a8e504d2 | 1075 | |
c3b64f1e IM |
1076 | for_each_sched_entity(se) |
1077 | set_next_entity(cfs_rq_of(se), se); | |
bf0f6f24 IM |
1078 | } |
1079 | #else | |
1080 | static void set_curr_task_fair(struct rq *rq) | |
1081 | { | |
1082 | } | |
1083 | #endif | |
1084 | ||
1085 | /* | |
1086 | * All the scheduling class methods: | |
1087 | */ | |
1088 | struct sched_class fair_sched_class __read_mostly = { | |
1089 | .enqueue_task = enqueue_task_fair, | |
1090 | .dequeue_task = dequeue_task_fair, | |
1091 | .yield_task = yield_task_fair, | |
1092 | ||
2e09bf55 | 1093 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1094 | |
1095 | .pick_next_task = pick_next_task_fair, | |
1096 | .put_prev_task = put_prev_task_fair, | |
1097 | ||
1098 | .load_balance = load_balance_fair, | |
1099 | ||
1100 | .set_curr_task = set_curr_task_fair, | |
1101 | .task_tick = task_tick_fair, | |
1102 | .task_new = task_new_fair, | |
1103 | }; | |
1104 | ||
1105 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1106 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1107 | { |
bf0f6f24 IM |
1108 | struct cfs_rq *cfs_rq; |
1109 | ||
c3b64f1e | 1110 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1111 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1112 | } |
1113 | #endif |