]>
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 | ||
aeb73b04 PZ |
202 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
203 | { | |
204 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
205 | struct sched_entity *se = NULL; | |
206 | struct rb_node *parent; | |
207 | ||
208 | while (*link) { | |
209 | parent = *link; | |
210 | se = rb_entry(parent, struct sched_entity, run_node); | |
211 | link = &parent->rb_right; | |
212 | } | |
213 | ||
214 | return se; | |
215 | } | |
216 | ||
bf0f6f24 IM |
217 | /************************************************************** |
218 | * Scheduling class statistics methods: | |
219 | */ | |
220 | ||
4d78e7b6 PZ |
221 | static u64 __sched_period(unsigned long nr_running) |
222 | { | |
223 | u64 period = sysctl_sched_latency; | |
224 | unsigned long nr_latency = | |
225 | sysctl_sched_latency / sysctl_sched_min_granularity; | |
226 | ||
227 | if (unlikely(nr_running > nr_latency)) { | |
228 | period *= nr_running; | |
229 | do_div(period, nr_latency); | |
230 | } | |
231 | ||
232 | return period; | |
233 | } | |
234 | ||
6d0f0ebd | 235 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 236 | { |
6d0f0ebd | 237 | u64 period = __sched_period(cfs_rq->nr_running); |
21805085 | 238 | |
6d0f0ebd PZ |
239 | period *= se->load.weight; |
240 | do_div(period, cfs_rq->load.weight); | |
21805085 | 241 | |
6d0f0ebd | 242 | return period; |
bf0f6f24 IM |
243 | } |
244 | ||
245 | static inline void | |
246 | limit_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
247 | { | |
248 | long limit = sysctl_sched_runtime_limit; | |
249 | ||
250 | /* | |
251 | * Niced tasks have the same history dynamic range as | |
252 | * non-niced tasks: | |
253 | */ | |
254 | if (unlikely(se->wait_runtime > limit)) { | |
255 | se->wait_runtime = limit; | |
256 | schedstat_inc(se, wait_runtime_overruns); | |
257 | schedstat_inc(cfs_rq, wait_runtime_overruns); | |
258 | } | |
259 | if (unlikely(se->wait_runtime < -limit)) { | |
260 | se->wait_runtime = -limit; | |
261 | schedstat_inc(se, wait_runtime_underruns); | |
262 | schedstat_inc(cfs_rq, wait_runtime_underruns); | |
263 | } | |
264 | } | |
265 | ||
266 | static inline void | |
267 | __add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
268 | { | |
269 | se->wait_runtime += delta; | |
270 | schedstat_add(se, sum_wait_runtime, delta); | |
271 | limit_wait_runtime(cfs_rq, se); | |
272 | } | |
273 | ||
274 | static void | |
275 | add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
276 | { | |
277 | schedstat_add(cfs_rq, wait_runtime, -se->wait_runtime); | |
278 | __add_wait_runtime(cfs_rq, se, delta); | |
279 | schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); | |
280 | } | |
281 | ||
282 | /* | |
283 | * Update the current task's runtime statistics. Skip current tasks that | |
284 | * are not in our scheduling class. | |
285 | */ | |
286 | static inline void | |
8ebc91d9 IM |
287 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
288 | unsigned long delta_exec) | |
bf0f6f24 | 289 | { |
e9acbff6 | 290 | unsigned long delta, delta_fair, delta_mine, delta_exec_weighted; |
bf0f6f24 IM |
291 | struct load_weight *lw = &cfs_rq->load; |
292 | unsigned long load = lw->weight; | |
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; | |
297 | 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; | |
bf0f6f24 | 304 | |
6cb58195 IM |
305 | if (!sched_feat(FAIR_SLEEPERS)) |
306 | return; | |
307 | ||
fd8bb43e IM |
308 | if (unlikely(!load)) |
309 | return; | |
310 | ||
bf0f6f24 IM |
311 | delta_fair = calc_delta_fair(delta_exec, lw); |
312 | delta_mine = calc_delta_mine(delta_exec, curr->load.weight, lw); | |
313 | ||
5f01d519 | 314 | if (cfs_rq->sleeper_bonus > sysctl_sched_min_granularity) { |
ea0aa3b2 | 315 | delta = min((u64)delta_mine, cfs_rq->sleeper_bonus); |
b2133c8b IM |
316 | delta = min(delta, (unsigned long)( |
317 | (long)sysctl_sched_runtime_limit - curr->wait_runtime)); | |
bf0f6f24 IM |
318 | cfs_rq->sleeper_bonus -= delta; |
319 | delta_mine -= delta; | |
320 | } | |
321 | ||
322 | cfs_rq->fair_clock += delta_fair; | |
323 | /* | |
324 | * We executed delta_exec amount of time on the CPU, | |
325 | * but we were only entitled to delta_mine amount of | |
326 | * time during that period (if nr_running == 1 then | |
327 | * the two values are equal) | |
328 | * [Note: delta_mine - delta_exec is negative]: | |
329 | */ | |
330 | add_wait_runtime(cfs_rq, curr, delta_mine - delta_exec); | |
331 | } | |
332 | ||
b7cc0896 | 333 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 334 | { |
429d43bc | 335 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 336 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
337 | unsigned long delta_exec; |
338 | ||
339 | if (unlikely(!curr)) | |
340 | return; | |
341 | ||
342 | /* | |
343 | * Get the amount of time the current task was running | |
344 | * since the last time we changed load (this cannot | |
345 | * overflow on 32 bits): | |
346 | */ | |
8ebc91d9 | 347 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 348 | |
8ebc91d9 IM |
349 | __update_curr(cfs_rq, curr, delta_exec); |
350 | curr->exec_start = now; | |
bf0f6f24 IM |
351 | } |
352 | ||
353 | static inline void | |
5870db5b | 354 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
355 | { |
356 | se->wait_start_fair = cfs_rq->fair_clock; | |
d281918d | 357 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
358 | } |
359 | ||
bf0f6f24 | 360 | static inline unsigned long |
08e2388a | 361 | calc_weighted(unsigned long delta, struct sched_entity *se) |
bf0f6f24 | 362 | { |
08e2388a | 363 | unsigned long weight = se->load.weight; |
bf0f6f24 | 364 | |
08e2388a IM |
365 | if (unlikely(weight != NICE_0_LOAD)) |
366 | return (u64)delta * se->load.weight >> NICE_0_SHIFT; | |
367 | else | |
368 | return delta; | |
bf0f6f24 | 369 | } |
bf0f6f24 IM |
370 | |
371 | /* | |
372 | * Task is being enqueued - update stats: | |
373 | */ | |
d2417e5a | 374 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 375 | { |
bf0f6f24 IM |
376 | /* |
377 | * Are we enqueueing a waiting task? (for current tasks | |
378 | * a dequeue/enqueue event is a NOP) | |
379 | */ | |
429d43bc | 380 | if (se != cfs_rq->curr) |
5870db5b | 381 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
382 | /* |
383 | * Update the key: | |
384 | */ | |
e9acbff6 | 385 | se->fair_key = se->vruntime; |
bf0f6f24 IM |
386 | } |
387 | ||
388 | /* | |
389 | * Note: must be called with a freshly updated rq->fair_clock. | |
390 | */ | |
391 | static inline void | |
8ebc91d9 IM |
392 | __update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, |
393 | unsigned long delta_fair) | |
bf0f6f24 | 394 | { |
d281918d IM |
395 | schedstat_set(se->wait_max, max(se->wait_max, |
396 | rq_of(cfs_rq)->clock - se->wait_start)); | |
bf0f6f24 | 397 | |
08e2388a | 398 | delta_fair = calc_weighted(delta_fair, se); |
bf0f6f24 IM |
399 | |
400 | add_wait_runtime(cfs_rq, se, delta_fair); | |
401 | } | |
402 | ||
403 | static void | |
9ef0a961 | 404 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
405 | { |
406 | unsigned long delta_fair; | |
407 | ||
b77d69db IM |
408 | if (unlikely(!se->wait_start_fair)) |
409 | return; | |
410 | ||
bf0f6f24 IM |
411 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), |
412 | (u64)(cfs_rq->fair_clock - se->wait_start_fair)); | |
413 | ||
8ebc91d9 | 414 | __update_stats_wait_end(cfs_rq, se, delta_fair); |
bf0f6f24 IM |
415 | |
416 | se->wait_start_fair = 0; | |
6cfb0d5d | 417 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
418 | } |
419 | ||
420 | static inline void | |
19b6a2e3 | 421 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 422 | { |
b7cc0896 | 423 | update_curr(cfs_rq); |
bf0f6f24 IM |
424 | /* |
425 | * Mark the end of the wait period if dequeueing a | |
426 | * waiting task: | |
427 | */ | |
429d43bc | 428 | if (se != cfs_rq->curr) |
9ef0a961 | 429 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
430 | } |
431 | ||
432 | /* | |
433 | * We are picking a new current task - update its stats: | |
434 | */ | |
435 | static inline void | |
79303e9e | 436 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
437 | { |
438 | /* | |
439 | * We are starting a new run period: | |
440 | */ | |
d281918d | 441 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
442 | } |
443 | ||
444 | /* | |
445 | * We are descheduling a task - update its stats: | |
446 | */ | |
447 | static inline void | |
c7e9b5b2 | 448 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
449 | { |
450 | se->exec_start = 0; | |
451 | } | |
452 | ||
453 | /************************************************** | |
454 | * Scheduling class queueing methods: | |
455 | */ | |
456 | ||
8ebc91d9 IM |
457 | static void __enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, |
458 | unsigned long delta_fair) | |
bf0f6f24 | 459 | { |
8ebc91d9 | 460 | unsigned long load = cfs_rq->load.weight; |
bf0f6f24 IM |
461 | long prev_runtime; |
462 | ||
b2133c8b IM |
463 | /* |
464 | * Do not boost sleepers if there's too much bonus 'in flight' | |
465 | * already: | |
466 | */ | |
467 | if (unlikely(cfs_rq->sleeper_bonus > sysctl_sched_runtime_limit)) | |
468 | return; | |
469 | ||
e59c80c5 | 470 | if (sched_feat(SLEEPER_LOAD_AVG)) |
bf0f6f24 IM |
471 | load = rq_of(cfs_rq)->cpu_load[2]; |
472 | ||
bf0f6f24 IM |
473 | /* |
474 | * Fix up delta_fair with the effect of us running | |
475 | * during the whole sleep period: | |
476 | */ | |
e59c80c5 | 477 | if (sched_feat(SLEEPER_AVG)) |
bf0f6f24 IM |
478 | delta_fair = div64_likely32((u64)delta_fair * load, |
479 | load + se->load.weight); | |
480 | ||
08e2388a | 481 | delta_fair = calc_weighted(delta_fair, se); |
bf0f6f24 IM |
482 | |
483 | prev_runtime = se->wait_runtime; | |
484 | __add_wait_runtime(cfs_rq, se, delta_fair); | |
485 | delta_fair = se->wait_runtime - prev_runtime; | |
486 | ||
487 | /* | |
488 | * Track the amount of bonus we've given to sleepers: | |
489 | */ | |
490 | cfs_rq->sleeper_bonus += delta_fair; | |
bf0f6f24 IM |
491 | } |
492 | ||
2396af69 | 493 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
494 | { |
495 | struct task_struct *tsk = task_of(se); | |
496 | unsigned long delta_fair; | |
497 | ||
498 | if ((entity_is_task(se) && tsk->policy == SCHED_BATCH) || | |
e59c80c5 | 499 | !sched_feat(FAIR_SLEEPERS)) |
bf0f6f24 IM |
500 | return; |
501 | ||
502 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), | |
503 | (u64)(cfs_rq->fair_clock - se->sleep_start_fair)); | |
504 | ||
8ebc91d9 | 505 | __enqueue_sleeper(cfs_rq, se, delta_fair); |
bf0f6f24 IM |
506 | |
507 | se->sleep_start_fair = 0; | |
508 | ||
509 | #ifdef CONFIG_SCHEDSTATS | |
510 | if (se->sleep_start) { | |
d281918d | 511 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
512 | |
513 | if ((s64)delta < 0) | |
514 | delta = 0; | |
515 | ||
516 | if (unlikely(delta > se->sleep_max)) | |
517 | se->sleep_max = delta; | |
518 | ||
519 | se->sleep_start = 0; | |
520 | se->sum_sleep_runtime += delta; | |
521 | } | |
522 | if (se->block_start) { | |
d281918d | 523 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
524 | |
525 | if ((s64)delta < 0) | |
526 | delta = 0; | |
527 | ||
528 | if (unlikely(delta > se->block_max)) | |
529 | se->block_max = delta; | |
530 | ||
531 | se->block_start = 0; | |
532 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
533 | |
534 | /* | |
535 | * Blocking time is in units of nanosecs, so shift by 20 to | |
536 | * get a milliseconds-range estimation of the amount of | |
537 | * time that the task spent sleeping: | |
538 | */ | |
539 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
540 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), | |
541 | delta >> 20); | |
542 | } | |
bf0f6f24 IM |
543 | } |
544 | #endif | |
545 | } | |
546 | ||
aeb73b04 PZ |
547 | static void |
548 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
549 | { | |
aeb73b04 PZ |
550 | u64 min_runtime, latency; |
551 | ||
552 | min_runtime = cfs_rq->min_vruntime; | |
94dfb5e7 PZ |
553 | |
554 | if (sched_feat(USE_TREE_AVG)) { | |
555 | struct sched_entity *last = __pick_last_entity(cfs_rq); | |
556 | if (last) { | |
557 | min_runtime = __pick_next_entity(cfs_rq)->vruntime; | |
558 | min_runtime += last->vruntime; | |
559 | min_runtime >>= 1; | |
560 | } | |
561 | } else if (sched_feat(APPROX_AVG)) | |
562 | min_runtime += sysctl_sched_latency/2; | |
563 | ||
564 | if (initial && sched_feat(START_DEBIT)) | |
565 | min_runtime += sched_slice(cfs_rq, se); | |
aeb73b04 PZ |
566 | |
567 | if (!initial && sched_feat(NEW_FAIR_SLEEPERS)) { | |
568 | latency = sysctl_sched_latency; | |
569 | if (min_runtime > latency) | |
570 | min_runtime -= latency; | |
571 | else | |
572 | min_runtime = 0; | |
573 | } | |
574 | ||
575 | se->vruntime = max(se->vruntime, min_runtime); | |
576 | } | |
577 | ||
bf0f6f24 | 578 | static void |
668031ca | 579 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
580 | { |
581 | /* | |
582 | * Update the fair clock. | |
583 | */ | |
b7cc0896 | 584 | update_curr(cfs_rq); |
bf0f6f24 | 585 | |
e9acbff6 | 586 | if (wakeup) { |
aeb73b04 | 587 | place_entity(cfs_rq, se, 0); |
2396af69 | 588 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 589 | } |
bf0f6f24 | 590 | |
d2417e5a | 591 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 IM |
592 | __enqueue_entity(cfs_rq, se); |
593 | } | |
594 | ||
595 | static void | |
525c2716 | 596 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 597 | { |
19b6a2e3 | 598 | update_stats_dequeue(cfs_rq, se); |
bf0f6f24 IM |
599 | if (sleep) { |
600 | se->sleep_start_fair = cfs_rq->fair_clock; | |
601 | #ifdef CONFIG_SCHEDSTATS | |
602 | if (entity_is_task(se)) { | |
603 | struct task_struct *tsk = task_of(se); | |
604 | ||
605 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 606 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 607 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 608 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 609 | } |
bf0f6f24 IM |
610 | #endif |
611 | } | |
612 | __dequeue_entity(cfs_rq, se); | |
613 | } | |
614 | ||
615 | /* | |
616 | * Preempt the current task with a newly woken task if needed: | |
617 | */ | |
7c92e54f | 618 | static void |
2e09bf55 | 619 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 620 | { |
11697830 PZ |
621 | unsigned long ideal_runtime, delta_exec; |
622 | ||
6d0f0ebd | 623 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 PZ |
624 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
625 | if (delta_exec > ideal_runtime) | |
bf0f6f24 IM |
626 | resched_task(rq_of(cfs_rq)->curr); |
627 | } | |
628 | ||
629 | static inline void | |
8494f412 | 630 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
631 | { |
632 | /* | |
633 | * Any task has to be enqueued before it get to execute on | |
634 | * a CPU. So account for the time it spent waiting on the | |
635 | * runqueue. (note, here we rely on pick_next_task() having | |
636 | * done a put_prev_task_fair() shortly before this, which | |
637 | * updated rq->fair_clock - used by update_stats_wait_end()) | |
638 | */ | |
9ef0a961 | 639 | update_stats_wait_end(cfs_rq, se); |
79303e9e | 640 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 641 | cfs_rq->curr = se; |
eba1ed4b IM |
642 | #ifdef CONFIG_SCHEDSTATS |
643 | /* | |
644 | * Track our maximum slice length, if the CPU's load is at | |
645 | * least twice that of our own weight (i.e. dont track it | |
646 | * when there are only lesser-weight tasks around): | |
647 | */ | |
648 | if (rq_of(cfs_rq)->ls.load.weight >= 2*se->load.weight) { | |
649 | se->slice_max = max(se->slice_max, | |
650 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
651 | } | |
652 | #endif | |
4a55b450 | 653 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
654 | } |
655 | ||
9948f4b2 | 656 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 IM |
657 | { |
658 | struct sched_entity *se = __pick_next_entity(cfs_rq); | |
659 | ||
8494f412 | 660 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
661 | |
662 | return se; | |
663 | } | |
664 | ||
ab6cde26 | 665 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
666 | { |
667 | /* | |
668 | * If still on the runqueue then deactivate_task() | |
669 | * was not called and update_curr() has to be done: | |
670 | */ | |
671 | if (prev->on_rq) | |
b7cc0896 | 672 | update_curr(cfs_rq); |
bf0f6f24 | 673 | |
c7e9b5b2 | 674 | update_stats_curr_end(cfs_rq, prev); |
bf0f6f24 IM |
675 | |
676 | if (prev->on_rq) | |
5870db5b | 677 | update_stats_wait_start(cfs_rq, prev); |
429d43bc | 678 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
679 | } |
680 | ||
681 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
682 | { | |
bf0f6f24 IM |
683 | /* |
684 | * Dequeue and enqueue the task to update its | |
685 | * position within the tree: | |
686 | */ | |
525c2716 | 687 | dequeue_entity(cfs_rq, curr, 0); |
668031ca | 688 | enqueue_entity(cfs_rq, curr, 0); |
bf0f6f24 | 689 | |
2e09bf55 IM |
690 | if (cfs_rq->nr_running > 1) |
691 | check_preempt_tick(cfs_rq, curr); | |
bf0f6f24 IM |
692 | } |
693 | ||
694 | /************************************************** | |
695 | * CFS operations on tasks: | |
696 | */ | |
697 | ||
698 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
699 | ||
700 | /* Walk up scheduling entities hierarchy */ | |
701 | #define for_each_sched_entity(se) \ | |
702 | for (; se; se = se->parent) | |
703 | ||
704 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
705 | { | |
706 | return p->se.cfs_rq; | |
707 | } | |
708 | ||
709 | /* runqueue on which this entity is (to be) queued */ | |
710 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
711 | { | |
712 | return se->cfs_rq; | |
713 | } | |
714 | ||
715 | /* runqueue "owned" by this group */ | |
716 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
717 | { | |
718 | return grp->my_q; | |
719 | } | |
720 | ||
721 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
722 | * another cpu ('this_cpu') | |
723 | */ | |
724 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
725 | { | |
726 | /* A later patch will take group into account */ | |
727 | return &cpu_rq(this_cpu)->cfs; | |
728 | } | |
729 | ||
730 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
731 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
732 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
733 | ||
734 | /* Do the two (enqueued) tasks belong to the same group ? */ | |
735 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
736 | { | |
737 | if (curr->se.cfs_rq == p->se.cfs_rq) | |
738 | return 1; | |
739 | ||
740 | return 0; | |
741 | } | |
742 | ||
743 | #else /* CONFIG_FAIR_GROUP_SCHED */ | |
744 | ||
745 | #define for_each_sched_entity(se) \ | |
746 | for (; se; se = NULL) | |
747 | ||
748 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
749 | { | |
750 | return &task_rq(p)->cfs; | |
751 | } | |
752 | ||
753 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
754 | { | |
755 | struct task_struct *p = task_of(se); | |
756 | struct rq *rq = task_rq(p); | |
757 | ||
758 | return &rq->cfs; | |
759 | } | |
760 | ||
761 | /* runqueue "owned" by this group */ | |
762 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
763 | { | |
764 | return NULL; | |
765 | } | |
766 | ||
767 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
768 | { | |
769 | return &cpu_rq(this_cpu)->cfs; | |
770 | } | |
771 | ||
772 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
773 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
774 | ||
775 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
776 | { | |
777 | return 1; | |
778 | } | |
779 | ||
780 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
781 | ||
782 | /* | |
783 | * The enqueue_task method is called before nr_running is | |
784 | * increased. Here we update the fair scheduling stats and | |
785 | * then put the task into the rbtree: | |
786 | */ | |
fd390f6a | 787 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
788 | { |
789 | struct cfs_rq *cfs_rq; | |
790 | struct sched_entity *se = &p->se; | |
791 | ||
792 | for_each_sched_entity(se) { | |
793 | if (se->on_rq) | |
794 | break; | |
795 | cfs_rq = cfs_rq_of(se); | |
668031ca | 796 | enqueue_entity(cfs_rq, se, wakeup); |
bf0f6f24 IM |
797 | } |
798 | } | |
799 | ||
800 | /* | |
801 | * The dequeue_task method is called before nr_running is | |
802 | * decreased. We remove the task from the rbtree and | |
803 | * update the fair scheduling stats: | |
804 | */ | |
f02231e5 | 805 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
806 | { |
807 | struct cfs_rq *cfs_rq; | |
808 | struct sched_entity *se = &p->se; | |
809 | ||
810 | for_each_sched_entity(se) { | |
811 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 812 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
813 | /* Don't dequeue parent if it has other entities besides us */ |
814 | if (cfs_rq->load.weight) | |
815 | break; | |
816 | } | |
817 | } | |
818 | ||
819 | /* | |
1799e35d IM |
820 | * sched_yield() support is very simple - we dequeue and enqueue. |
821 | * | |
822 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 IM |
823 | */ |
824 | static void yield_task_fair(struct rq *rq, struct task_struct *p) | |
825 | { | |
826 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1799e35d IM |
827 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
828 | struct sched_entity *rightmost, *se = &p->se; | |
829 | struct rb_node *parent; | |
bf0f6f24 IM |
830 | |
831 | /* | |
1799e35d IM |
832 | * Are we the only task in the tree? |
833 | */ | |
834 | if (unlikely(cfs_rq->nr_running == 1)) | |
835 | return; | |
836 | ||
837 | if (likely(!sysctl_sched_compat_yield)) { | |
838 | __update_rq_clock(rq); | |
839 | /* | |
840 | * Dequeue and enqueue the task to update its | |
841 | * position within the tree: | |
842 | */ | |
843 | dequeue_entity(cfs_rq, &p->se, 0); | |
844 | enqueue_entity(cfs_rq, &p->se, 0); | |
845 | ||
846 | return; | |
847 | } | |
848 | /* | |
849 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 850 | */ |
1799e35d IM |
851 | do { |
852 | parent = *link; | |
853 | link = &parent->rb_right; | |
854 | } while (*link); | |
855 | ||
856 | rightmost = rb_entry(parent, struct sched_entity, run_node); | |
857 | /* | |
858 | * Already in the rightmost position? | |
859 | */ | |
860 | if (unlikely(rightmost == se)) | |
861 | return; | |
862 | ||
863 | /* | |
864 | * Minimally necessary key value to be last in the tree: | |
865 | */ | |
866 | se->fair_key = rightmost->fair_key + 1; | |
867 | ||
868 | if (cfs_rq->rb_leftmost == &se->run_node) | |
869 | cfs_rq->rb_leftmost = rb_next(&se->run_node); | |
870 | /* | |
871 | * Relink the task to the rightmost position: | |
872 | */ | |
873 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); | |
874 | rb_link_node(&se->run_node, parent, link); | |
875 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
876 | } |
877 | ||
878 | /* | |
879 | * Preempt the current task with a newly woken task if needed: | |
880 | */ | |
2e09bf55 | 881 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
882 | { |
883 | struct task_struct *curr = rq->curr; | |
884 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
bf0f6f24 IM |
885 | |
886 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 887 | update_rq_clock(rq); |
b7cc0896 | 888 | update_curr(cfs_rq); |
bf0f6f24 IM |
889 | resched_task(curr); |
890 | return; | |
891 | } | |
2e09bf55 IM |
892 | if (is_same_group(curr, p)) { |
893 | s64 delta = curr->se.vruntime - p->se.vruntime; | |
bf0f6f24 | 894 | |
2e09bf55 IM |
895 | if (delta > (s64)sysctl_sched_wakeup_granularity) |
896 | resched_task(curr); | |
897 | } | |
bf0f6f24 IM |
898 | } |
899 | ||
fb8d4724 | 900 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
901 | { |
902 | struct cfs_rq *cfs_rq = &rq->cfs; | |
903 | struct sched_entity *se; | |
904 | ||
905 | if (unlikely(!cfs_rq->nr_running)) | |
906 | return NULL; | |
907 | ||
908 | do { | |
9948f4b2 | 909 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
910 | cfs_rq = group_cfs_rq(se); |
911 | } while (cfs_rq); | |
912 | ||
913 | return task_of(se); | |
914 | } | |
915 | ||
916 | /* | |
917 | * Account for a descheduled task: | |
918 | */ | |
31ee529c | 919 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
920 | { |
921 | struct sched_entity *se = &prev->se; | |
922 | struct cfs_rq *cfs_rq; | |
923 | ||
924 | for_each_sched_entity(se) { | |
925 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 926 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
927 | } |
928 | } | |
929 | ||
930 | /************************************************** | |
931 | * Fair scheduling class load-balancing methods: | |
932 | */ | |
933 | ||
934 | /* | |
935 | * Load-balancing iterator. Note: while the runqueue stays locked | |
936 | * during the whole iteration, the current task might be | |
937 | * dequeued so the iterator has to be dequeue-safe. Here we | |
938 | * achieve that by always pre-iterating before returning | |
939 | * the current task: | |
940 | */ | |
941 | static inline struct task_struct * | |
942 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) | |
943 | { | |
944 | struct task_struct *p; | |
945 | ||
946 | if (!curr) | |
947 | return NULL; | |
948 | ||
949 | p = rb_entry(curr, struct task_struct, se.run_node); | |
950 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
951 | ||
952 | return p; | |
953 | } | |
954 | ||
955 | static struct task_struct *load_balance_start_fair(void *arg) | |
956 | { | |
957 | struct cfs_rq *cfs_rq = arg; | |
958 | ||
959 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
960 | } | |
961 | ||
962 | static struct task_struct *load_balance_next_fair(void *arg) | |
963 | { | |
964 | struct cfs_rq *cfs_rq = arg; | |
965 | ||
966 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
967 | } | |
968 | ||
a4ac01c3 | 969 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
970 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
971 | { | |
972 | struct sched_entity *curr; | |
973 | struct task_struct *p; | |
974 | ||
975 | if (!cfs_rq->nr_running) | |
976 | return MAX_PRIO; | |
977 | ||
978 | curr = __pick_next_entity(cfs_rq); | |
979 | p = task_of(curr); | |
980 | ||
981 | return p->prio; | |
982 | } | |
a4ac01c3 | 983 | #endif |
bf0f6f24 | 984 | |
43010659 | 985 | static unsigned long |
bf0f6f24 | 986 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
a4ac01c3 PW |
987 | unsigned long max_nr_move, unsigned long max_load_move, |
988 | struct sched_domain *sd, enum cpu_idle_type idle, | |
989 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
990 | { |
991 | struct cfs_rq *busy_cfs_rq; | |
992 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
993 | long rem_load_move = max_load_move; | |
994 | struct rq_iterator cfs_rq_iterator; | |
995 | ||
996 | cfs_rq_iterator.start = load_balance_start_fair; | |
997 | cfs_rq_iterator.next = load_balance_next_fair; | |
998 | ||
999 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 1000 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 1001 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 1002 | long imbalance; |
bf0f6f24 | 1003 | unsigned long maxload; |
bf0f6f24 IM |
1004 | |
1005 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
1006 | ||
e56f31aa | 1007 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
1008 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
1009 | if (imbalance <= 0) | |
1010 | continue; | |
1011 | ||
1012 | /* Don't pull more than imbalance/2 */ | |
1013 | imbalance /= 2; | |
1014 | maxload = min(rem_load_move, imbalance); | |
1015 | ||
a4ac01c3 PW |
1016 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
1017 | #else | |
e56f31aa | 1018 | # define maxload rem_load_move |
a4ac01c3 | 1019 | #endif |
bf0f6f24 IM |
1020 | /* pass busy_cfs_rq argument into |
1021 | * load_balance_[start|next]_fair iterators | |
1022 | */ | |
1023 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1024 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, | |
1025 | max_nr_move, maxload, sd, idle, all_pinned, | |
a4ac01c3 | 1026 | &load_moved, this_best_prio, &cfs_rq_iterator); |
bf0f6f24 IM |
1027 | |
1028 | total_nr_moved += nr_moved; | |
1029 | max_nr_move -= nr_moved; | |
1030 | rem_load_move -= load_moved; | |
1031 | ||
1032 | if (max_nr_move <= 0 || rem_load_move <= 0) | |
1033 | break; | |
1034 | } | |
1035 | ||
43010659 | 1036 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
1037 | } |
1038 | ||
1039 | /* | |
1040 | * scheduler tick hitting a task of our scheduling class: | |
1041 | */ | |
1042 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
1043 | { | |
1044 | struct cfs_rq *cfs_rq; | |
1045 | struct sched_entity *se = &curr->se; | |
1046 | ||
1047 | for_each_sched_entity(se) { | |
1048 | cfs_rq = cfs_rq_of(se); | |
1049 | entity_tick(cfs_rq, se); | |
1050 | } | |
1051 | } | |
1052 | ||
4d78e7b6 PZ |
1053 | #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
1054 | ||
bf0f6f24 IM |
1055 | /* |
1056 | * Share the fairness runtime between parent and child, thus the | |
1057 | * total amount of pressure for CPU stays equal - new tasks | |
1058 | * get a chance to run but frequent forkers are not allowed to | |
1059 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1060 | * the child is not running yet. | |
1061 | */ | |
ee0827d8 | 1062 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1063 | { |
1064 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1065 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
bf0f6f24 IM |
1066 | |
1067 | sched_info_queued(p); | |
1068 | ||
7109c442 | 1069 | update_curr(cfs_rq); |
aeb73b04 | 1070 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1071 | |
bf0f6f24 IM |
1072 | /* |
1073 | * The statistical average of wait_runtime is about | |
1074 | * -granularity/2, so initialize the task with that: | |
1075 | */ | |
e59c80c5 | 1076 | if (sched_feat(START_DEBIT)) |
4d78e7b6 PZ |
1077 | se->wait_runtime = -(__sched_period(cfs_rq->nr_running+1) / 2); |
1078 | ||
1079 | if (sysctl_sched_child_runs_first && | |
1080 | curr->vruntime < se->vruntime) { | |
1081 | ||
1082 | dequeue_entity(cfs_rq, curr, 0); | |
1083 | swap(curr->vruntime, se->vruntime); | |
1084 | enqueue_entity(cfs_rq, curr, 0); | |
1085 | } | |
bf0f6f24 | 1086 | |
e9acbff6 | 1087 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 | 1088 | __enqueue_entity(cfs_rq, se); |
bb61c210 | 1089 | resched_task(rq->curr); |
bf0f6f24 IM |
1090 | } |
1091 | ||
1092 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1093 | /* Account for a task changing its policy or group. | |
1094 | * | |
1095 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1096 | * migrates between groups/classes. | |
1097 | */ | |
1098 | static void set_curr_task_fair(struct rq *rq) | |
1099 | { | |
7c6c16f3 | 1100 | struct sched_entity *se = &rq->curr->se; |
a8e504d2 | 1101 | |
c3b64f1e IM |
1102 | for_each_sched_entity(se) |
1103 | set_next_entity(cfs_rq_of(se), se); | |
bf0f6f24 IM |
1104 | } |
1105 | #else | |
1106 | static void set_curr_task_fair(struct rq *rq) | |
1107 | { | |
1108 | } | |
1109 | #endif | |
1110 | ||
1111 | /* | |
1112 | * All the scheduling class methods: | |
1113 | */ | |
1114 | struct sched_class fair_sched_class __read_mostly = { | |
1115 | .enqueue_task = enqueue_task_fair, | |
1116 | .dequeue_task = dequeue_task_fair, | |
1117 | .yield_task = yield_task_fair, | |
1118 | ||
2e09bf55 | 1119 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1120 | |
1121 | .pick_next_task = pick_next_task_fair, | |
1122 | .put_prev_task = put_prev_task_fair, | |
1123 | ||
1124 | .load_balance = load_balance_fair, | |
1125 | ||
1126 | .set_curr_task = set_curr_task_fair, | |
1127 | .task_tick = task_tick_fair, | |
1128 | .task_new = task_new_fair, | |
1129 | }; | |
1130 | ||
1131 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1132 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1133 | { |
bf0f6f24 IM |
1134 | struct cfs_rq *cfs_rq; |
1135 | ||
c3b64f1e | 1136 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1137 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1138 | } |
1139 | #endif |