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); | |
9014623c PZ |
127 | if ((se->vruntime > cfs_rq->min_vruntime) || |
128 | (cfs_rq->min_vruntime > (1ULL << 61) && | |
129 | se->vruntime < (1ULL << 50))) | |
130 | cfs_rq->min_vruntime = se->vruntime; | |
e9acbff6 IM |
131 | } |
132 | } | |
133 | ||
9014623c PZ |
134 | s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
135 | { | |
136 | return se->fair_key - cfs_rq->min_vruntime; | |
137 | } | |
138 | ||
bf0f6f24 IM |
139 | /* |
140 | * Enqueue an entity into the rb-tree: | |
141 | */ | |
19ccd97a | 142 | static void |
bf0f6f24 IM |
143 | __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
144 | { | |
145 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
146 | struct rb_node *parent = NULL; | |
147 | struct sched_entity *entry; | |
9014623c | 148 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
149 | int leftmost = 1; |
150 | ||
151 | /* | |
152 | * Find the right place in the rbtree: | |
153 | */ | |
154 | while (*link) { | |
155 | parent = *link; | |
156 | entry = rb_entry(parent, struct sched_entity, run_node); | |
157 | /* | |
158 | * We dont care about collisions. Nodes with | |
159 | * the same key stay together. | |
160 | */ | |
9014623c | 161 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
162 | link = &parent->rb_left; |
163 | } else { | |
164 | link = &parent->rb_right; | |
165 | leftmost = 0; | |
166 | } | |
167 | } | |
168 | ||
169 | /* | |
170 | * Maintain a cache of leftmost tree entries (it is frequently | |
171 | * used): | |
172 | */ | |
173 | if (leftmost) | |
e9acbff6 | 174 | set_leftmost(cfs_rq, &se->run_node); |
bf0f6f24 IM |
175 | |
176 | rb_link_node(&se->run_node, parent, link); | |
177 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
178 | update_load_add(&cfs_rq->load, se->load.weight); | |
179 | cfs_rq->nr_running++; | |
180 | se->on_rq = 1; | |
a206c072 IM |
181 | |
182 | schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); | |
bf0f6f24 IM |
183 | } |
184 | ||
19ccd97a | 185 | static void |
bf0f6f24 IM |
186 | __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
187 | { | |
188 | if (cfs_rq->rb_leftmost == &se->run_node) | |
e9acbff6 IM |
189 | set_leftmost(cfs_rq, rb_next(&se->run_node)); |
190 | ||
bf0f6f24 IM |
191 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
192 | update_load_sub(&cfs_rq->load, se->load.weight); | |
193 | cfs_rq->nr_running--; | |
194 | se->on_rq = 0; | |
a206c072 IM |
195 | |
196 | schedstat_add(cfs_rq, wait_runtime, -se->wait_runtime); | |
bf0f6f24 IM |
197 | } |
198 | ||
199 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
200 | { | |
201 | return cfs_rq->rb_leftmost; | |
202 | } | |
203 | ||
204 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
205 | { | |
206 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
207 | } | |
208 | ||
aeb73b04 PZ |
209 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
210 | { | |
211 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
212 | struct sched_entity *se = NULL; | |
213 | struct rb_node *parent; | |
214 | ||
215 | while (*link) { | |
216 | parent = *link; | |
217 | se = rb_entry(parent, struct sched_entity, run_node); | |
218 | link = &parent->rb_right; | |
219 | } | |
220 | ||
221 | return se; | |
222 | } | |
223 | ||
bf0f6f24 IM |
224 | /************************************************************** |
225 | * Scheduling class statistics methods: | |
226 | */ | |
227 | ||
4d78e7b6 PZ |
228 | static u64 __sched_period(unsigned long nr_running) |
229 | { | |
230 | u64 period = sysctl_sched_latency; | |
231 | unsigned long nr_latency = | |
232 | sysctl_sched_latency / sysctl_sched_min_granularity; | |
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 | ||
6d0f0ebd | 242 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 243 | { |
6d0f0ebd | 244 | u64 period = __sched_period(cfs_rq->nr_running); |
21805085 | 245 | |
6d0f0ebd PZ |
246 | period *= se->load.weight; |
247 | do_div(period, cfs_rq->load.weight); | |
21805085 | 248 | |
6d0f0ebd | 249 | return period; |
bf0f6f24 IM |
250 | } |
251 | ||
252 | static inline void | |
253 | limit_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
254 | { | |
255 | long limit = sysctl_sched_runtime_limit; | |
256 | ||
257 | /* | |
258 | * Niced tasks have the same history dynamic range as | |
259 | * non-niced tasks: | |
260 | */ | |
261 | if (unlikely(se->wait_runtime > limit)) { | |
262 | se->wait_runtime = limit; | |
263 | schedstat_inc(se, wait_runtime_overruns); | |
264 | schedstat_inc(cfs_rq, wait_runtime_overruns); | |
265 | } | |
266 | if (unlikely(se->wait_runtime < -limit)) { | |
267 | se->wait_runtime = -limit; | |
268 | schedstat_inc(se, wait_runtime_underruns); | |
269 | schedstat_inc(cfs_rq, wait_runtime_underruns); | |
270 | } | |
271 | } | |
272 | ||
273 | static inline void | |
274 | __add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
275 | { | |
276 | se->wait_runtime += delta; | |
277 | schedstat_add(se, sum_wait_runtime, delta); | |
278 | limit_wait_runtime(cfs_rq, se); | |
279 | } | |
280 | ||
281 | static void | |
282 | add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
283 | { | |
284 | schedstat_add(cfs_rq, wait_runtime, -se->wait_runtime); | |
285 | __add_wait_runtime(cfs_rq, se, delta); | |
286 | schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); | |
287 | } | |
288 | ||
289 | /* | |
290 | * Update the current task's runtime statistics. Skip current tasks that | |
291 | * are not in our scheduling class. | |
292 | */ | |
293 | static inline void | |
8ebc91d9 IM |
294 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
295 | unsigned long delta_exec) | |
bf0f6f24 | 296 | { |
e9acbff6 | 297 | unsigned long delta, delta_fair, delta_mine, delta_exec_weighted; |
bf0f6f24 IM |
298 | struct load_weight *lw = &cfs_rq->load; |
299 | unsigned long load = lw->weight; | |
300 | ||
8179ca23 | 301 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
302 | |
303 | curr->sum_exec_runtime += delta_exec; | |
304 | cfs_rq->exec_clock += delta_exec; | |
e9acbff6 IM |
305 | delta_exec_weighted = delta_exec; |
306 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { | |
307 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, | |
308 | &curr->load); | |
309 | } | |
310 | curr->vruntime += delta_exec_weighted; | |
bf0f6f24 | 311 | |
6cb58195 IM |
312 | if (!sched_feat(FAIR_SLEEPERS)) |
313 | return; | |
314 | ||
fd8bb43e IM |
315 | if (unlikely(!load)) |
316 | return; | |
317 | ||
bf0f6f24 IM |
318 | delta_fair = calc_delta_fair(delta_exec, lw); |
319 | delta_mine = calc_delta_mine(delta_exec, curr->load.weight, lw); | |
320 | ||
5f01d519 | 321 | if (cfs_rq->sleeper_bonus > sysctl_sched_min_granularity) { |
ea0aa3b2 | 322 | delta = min((u64)delta_mine, cfs_rq->sleeper_bonus); |
b2133c8b IM |
323 | delta = min(delta, (unsigned long)( |
324 | (long)sysctl_sched_runtime_limit - curr->wait_runtime)); | |
bf0f6f24 IM |
325 | cfs_rq->sleeper_bonus -= delta; |
326 | delta_mine -= delta; | |
327 | } | |
328 | ||
329 | cfs_rq->fair_clock += delta_fair; | |
330 | /* | |
331 | * We executed delta_exec amount of time on the CPU, | |
332 | * but we were only entitled to delta_mine amount of | |
333 | * time during that period (if nr_running == 1 then | |
334 | * the two values are equal) | |
335 | * [Note: delta_mine - delta_exec is negative]: | |
336 | */ | |
337 | add_wait_runtime(cfs_rq, curr, delta_mine - delta_exec); | |
338 | } | |
339 | ||
b7cc0896 | 340 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 341 | { |
429d43bc | 342 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 343 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
344 | unsigned long delta_exec; |
345 | ||
346 | if (unlikely(!curr)) | |
347 | return; | |
348 | ||
349 | /* | |
350 | * Get the amount of time the current task was running | |
351 | * since the last time we changed load (this cannot | |
352 | * overflow on 32 bits): | |
353 | */ | |
8ebc91d9 | 354 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 355 | |
8ebc91d9 IM |
356 | __update_curr(cfs_rq, curr, delta_exec); |
357 | curr->exec_start = now; | |
bf0f6f24 IM |
358 | } |
359 | ||
360 | static inline void | |
5870db5b | 361 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
362 | { |
363 | se->wait_start_fair = cfs_rq->fair_clock; | |
d281918d | 364 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
365 | } |
366 | ||
bf0f6f24 | 367 | static inline unsigned long |
08e2388a | 368 | calc_weighted(unsigned long delta, struct sched_entity *se) |
bf0f6f24 | 369 | { |
08e2388a | 370 | unsigned long weight = se->load.weight; |
bf0f6f24 | 371 | |
08e2388a IM |
372 | if (unlikely(weight != NICE_0_LOAD)) |
373 | return (u64)delta * se->load.weight >> NICE_0_SHIFT; | |
374 | else | |
375 | return delta; | |
bf0f6f24 | 376 | } |
bf0f6f24 IM |
377 | |
378 | /* | |
379 | * Task is being enqueued - update stats: | |
380 | */ | |
d2417e5a | 381 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 382 | { |
bf0f6f24 IM |
383 | /* |
384 | * Are we enqueueing a waiting task? (for current tasks | |
385 | * a dequeue/enqueue event is a NOP) | |
386 | */ | |
429d43bc | 387 | if (se != cfs_rq->curr) |
5870db5b | 388 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
389 | /* |
390 | * Update the key: | |
391 | */ | |
e9acbff6 | 392 | se->fair_key = se->vruntime; |
bf0f6f24 IM |
393 | } |
394 | ||
395 | /* | |
396 | * Note: must be called with a freshly updated rq->fair_clock. | |
397 | */ | |
398 | static inline void | |
8ebc91d9 IM |
399 | __update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, |
400 | unsigned long delta_fair) | |
bf0f6f24 | 401 | { |
d281918d IM |
402 | schedstat_set(se->wait_max, max(se->wait_max, |
403 | rq_of(cfs_rq)->clock - se->wait_start)); | |
bf0f6f24 | 404 | |
08e2388a | 405 | delta_fair = calc_weighted(delta_fair, se); |
bf0f6f24 IM |
406 | |
407 | add_wait_runtime(cfs_rq, se, delta_fair); | |
408 | } | |
409 | ||
410 | static void | |
9ef0a961 | 411 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
412 | { |
413 | unsigned long delta_fair; | |
414 | ||
b77d69db IM |
415 | if (unlikely(!se->wait_start_fair)) |
416 | return; | |
417 | ||
bf0f6f24 IM |
418 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), |
419 | (u64)(cfs_rq->fair_clock - se->wait_start_fair)); | |
420 | ||
8ebc91d9 | 421 | __update_stats_wait_end(cfs_rq, se, delta_fair); |
bf0f6f24 IM |
422 | |
423 | se->wait_start_fair = 0; | |
6cfb0d5d | 424 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
425 | } |
426 | ||
427 | static inline void | |
19b6a2e3 | 428 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 429 | { |
b7cc0896 | 430 | update_curr(cfs_rq); |
bf0f6f24 IM |
431 | /* |
432 | * Mark the end of the wait period if dequeueing a | |
433 | * waiting task: | |
434 | */ | |
429d43bc | 435 | if (se != cfs_rq->curr) |
9ef0a961 | 436 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
437 | } |
438 | ||
439 | /* | |
440 | * We are picking a new current task - update its stats: | |
441 | */ | |
442 | static inline void | |
79303e9e | 443 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
444 | { |
445 | /* | |
446 | * We are starting a new run period: | |
447 | */ | |
d281918d | 448 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
449 | } |
450 | ||
451 | /* | |
452 | * We are descheduling a task - update its stats: | |
453 | */ | |
454 | static inline void | |
c7e9b5b2 | 455 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
456 | { |
457 | se->exec_start = 0; | |
458 | } | |
459 | ||
460 | /************************************************** | |
461 | * Scheduling class queueing methods: | |
462 | */ | |
463 | ||
8ebc91d9 IM |
464 | static void __enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, |
465 | unsigned long delta_fair) | |
bf0f6f24 | 466 | { |
8ebc91d9 | 467 | unsigned long load = cfs_rq->load.weight; |
bf0f6f24 IM |
468 | long prev_runtime; |
469 | ||
b2133c8b IM |
470 | /* |
471 | * Do not boost sleepers if there's too much bonus 'in flight' | |
472 | * already: | |
473 | */ | |
474 | if (unlikely(cfs_rq->sleeper_bonus > sysctl_sched_runtime_limit)) | |
475 | return; | |
476 | ||
e59c80c5 | 477 | if (sched_feat(SLEEPER_LOAD_AVG)) |
bf0f6f24 IM |
478 | load = rq_of(cfs_rq)->cpu_load[2]; |
479 | ||
bf0f6f24 IM |
480 | /* |
481 | * Fix up delta_fair with the effect of us running | |
482 | * during the whole sleep period: | |
483 | */ | |
e59c80c5 | 484 | if (sched_feat(SLEEPER_AVG)) |
bf0f6f24 IM |
485 | delta_fair = div64_likely32((u64)delta_fair * load, |
486 | load + se->load.weight); | |
487 | ||
08e2388a | 488 | delta_fair = calc_weighted(delta_fair, se); |
bf0f6f24 IM |
489 | |
490 | prev_runtime = se->wait_runtime; | |
491 | __add_wait_runtime(cfs_rq, se, delta_fair); | |
492 | delta_fair = se->wait_runtime - prev_runtime; | |
493 | ||
494 | /* | |
495 | * Track the amount of bonus we've given to sleepers: | |
496 | */ | |
497 | cfs_rq->sleeper_bonus += delta_fair; | |
bf0f6f24 IM |
498 | } |
499 | ||
2396af69 | 500 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
501 | { |
502 | struct task_struct *tsk = task_of(se); | |
503 | unsigned long delta_fair; | |
504 | ||
505 | if ((entity_is_task(se) && tsk->policy == SCHED_BATCH) || | |
e59c80c5 | 506 | !sched_feat(FAIR_SLEEPERS)) |
bf0f6f24 IM |
507 | return; |
508 | ||
509 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), | |
510 | (u64)(cfs_rq->fair_clock - se->sleep_start_fair)); | |
511 | ||
8ebc91d9 | 512 | __enqueue_sleeper(cfs_rq, se, delta_fair); |
bf0f6f24 IM |
513 | |
514 | se->sleep_start_fair = 0; | |
515 | ||
516 | #ifdef CONFIG_SCHEDSTATS | |
517 | if (se->sleep_start) { | |
d281918d | 518 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
519 | |
520 | if ((s64)delta < 0) | |
521 | delta = 0; | |
522 | ||
523 | if (unlikely(delta > se->sleep_max)) | |
524 | se->sleep_max = delta; | |
525 | ||
526 | se->sleep_start = 0; | |
527 | se->sum_sleep_runtime += delta; | |
528 | } | |
529 | if (se->block_start) { | |
d281918d | 530 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
531 | |
532 | if ((s64)delta < 0) | |
533 | delta = 0; | |
534 | ||
535 | if (unlikely(delta > se->block_max)) | |
536 | se->block_max = delta; | |
537 | ||
538 | se->block_start = 0; | |
539 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
540 | |
541 | /* | |
542 | * Blocking time is in units of nanosecs, so shift by 20 to | |
543 | * get a milliseconds-range estimation of the amount of | |
544 | * time that the task spent sleeping: | |
545 | */ | |
546 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
547 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), | |
548 | delta >> 20); | |
549 | } | |
bf0f6f24 IM |
550 | } |
551 | #endif | |
552 | } | |
553 | ||
aeb73b04 PZ |
554 | static void |
555 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
556 | { | |
aeb73b04 PZ |
557 | u64 min_runtime, latency; |
558 | ||
559 | min_runtime = cfs_rq->min_vruntime; | |
94dfb5e7 PZ |
560 | |
561 | if (sched_feat(USE_TREE_AVG)) { | |
562 | struct sched_entity *last = __pick_last_entity(cfs_rq); | |
563 | if (last) { | |
564 | min_runtime = __pick_next_entity(cfs_rq)->vruntime; | |
565 | min_runtime += last->vruntime; | |
566 | min_runtime >>= 1; | |
567 | } | |
568 | } else if (sched_feat(APPROX_AVG)) | |
569 | min_runtime += sysctl_sched_latency/2; | |
570 | ||
571 | if (initial && sched_feat(START_DEBIT)) | |
572 | min_runtime += sched_slice(cfs_rq, se); | |
aeb73b04 PZ |
573 | |
574 | if (!initial && sched_feat(NEW_FAIR_SLEEPERS)) { | |
575 | latency = sysctl_sched_latency; | |
576 | if (min_runtime > latency) | |
577 | min_runtime -= latency; | |
578 | else | |
579 | min_runtime = 0; | |
580 | } | |
581 | ||
582 | se->vruntime = max(se->vruntime, min_runtime); | |
583 | } | |
584 | ||
bf0f6f24 | 585 | static void |
668031ca | 586 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
587 | { |
588 | /* | |
589 | * Update the fair clock. | |
590 | */ | |
b7cc0896 | 591 | update_curr(cfs_rq); |
bf0f6f24 | 592 | |
e9acbff6 | 593 | if (wakeup) { |
aeb73b04 | 594 | place_entity(cfs_rq, se, 0); |
2396af69 | 595 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 596 | } |
bf0f6f24 | 597 | |
d2417e5a | 598 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 IM |
599 | __enqueue_entity(cfs_rq, se); |
600 | } | |
601 | ||
602 | static void | |
525c2716 | 603 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 604 | { |
19b6a2e3 | 605 | update_stats_dequeue(cfs_rq, se); |
bf0f6f24 IM |
606 | if (sleep) { |
607 | se->sleep_start_fair = cfs_rq->fair_clock; | |
608 | #ifdef CONFIG_SCHEDSTATS | |
609 | if (entity_is_task(se)) { | |
610 | struct task_struct *tsk = task_of(se); | |
611 | ||
612 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 613 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 614 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 615 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 616 | } |
bf0f6f24 IM |
617 | #endif |
618 | } | |
619 | __dequeue_entity(cfs_rq, se); | |
620 | } | |
621 | ||
622 | /* | |
623 | * Preempt the current task with a newly woken task if needed: | |
624 | */ | |
7c92e54f | 625 | static void |
2e09bf55 | 626 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 627 | { |
11697830 PZ |
628 | unsigned long ideal_runtime, delta_exec; |
629 | ||
6d0f0ebd | 630 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 PZ |
631 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
632 | if (delta_exec > ideal_runtime) | |
bf0f6f24 IM |
633 | resched_task(rq_of(cfs_rq)->curr); |
634 | } | |
635 | ||
636 | static inline void | |
8494f412 | 637 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
638 | { |
639 | /* | |
640 | * Any task has to be enqueued before it get to execute on | |
641 | * a CPU. So account for the time it spent waiting on the | |
642 | * runqueue. (note, here we rely on pick_next_task() having | |
643 | * done a put_prev_task_fair() shortly before this, which | |
644 | * updated rq->fair_clock - used by update_stats_wait_end()) | |
645 | */ | |
9ef0a961 | 646 | update_stats_wait_end(cfs_rq, se); |
79303e9e | 647 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 648 | cfs_rq->curr = se; |
eba1ed4b IM |
649 | #ifdef CONFIG_SCHEDSTATS |
650 | /* | |
651 | * Track our maximum slice length, if the CPU's load is at | |
652 | * least twice that of our own weight (i.e. dont track it | |
653 | * when there are only lesser-weight tasks around): | |
654 | */ | |
655 | if (rq_of(cfs_rq)->ls.load.weight >= 2*se->load.weight) { | |
656 | se->slice_max = max(se->slice_max, | |
657 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
658 | } | |
659 | #endif | |
4a55b450 | 660 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
661 | } |
662 | ||
9948f4b2 | 663 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 IM |
664 | { |
665 | struct sched_entity *se = __pick_next_entity(cfs_rq); | |
666 | ||
8494f412 | 667 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
668 | |
669 | return se; | |
670 | } | |
671 | ||
ab6cde26 | 672 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
673 | { |
674 | /* | |
675 | * If still on the runqueue then deactivate_task() | |
676 | * was not called and update_curr() has to be done: | |
677 | */ | |
678 | if (prev->on_rq) | |
b7cc0896 | 679 | update_curr(cfs_rq); |
bf0f6f24 | 680 | |
c7e9b5b2 | 681 | update_stats_curr_end(cfs_rq, prev); |
bf0f6f24 IM |
682 | |
683 | if (prev->on_rq) | |
5870db5b | 684 | update_stats_wait_start(cfs_rq, prev); |
429d43bc | 685 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
686 | } |
687 | ||
688 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
689 | { | |
bf0f6f24 IM |
690 | /* |
691 | * Dequeue and enqueue the task to update its | |
692 | * position within the tree: | |
693 | */ | |
525c2716 | 694 | dequeue_entity(cfs_rq, curr, 0); |
668031ca | 695 | enqueue_entity(cfs_rq, curr, 0); |
bf0f6f24 | 696 | |
2e09bf55 IM |
697 | if (cfs_rq->nr_running > 1) |
698 | check_preempt_tick(cfs_rq, curr); | |
bf0f6f24 IM |
699 | } |
700 | ||
701 | /************************************************** | |
702 | * CFS operations on tasks: | |
703 | */ | |
704 | ||
705 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
706 | ||
707 | /* Walk up scheduling entities hierarchy */ | |
708 | #define for_each_sched_entity(se) \ | |
709 | for (; se; se = se->parent) | |
710 | ||
711 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
712 | { | |
713 | return p->se.cfs_rq; | |
714 | } | |
715 | ||
716 | /* runqueue on which this entity is (to be) queued */ | |
717 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
718 | { | |
719 | return se->cfs_rq; | |
720 | } | |
721 | ||
722 | /* runqueue "owned" by this group */ | |
723 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
724 | { | |
725 | return grp->my_q; | |
726 | } | |
727 | ||
728 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
729 | * another cpu ('this_cpu') | |
730 | */ | |
731 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
732 | { | |
733 | /* A later patch will take group into account */ | |
734 | return &cpu_rq(this_cpu)->cfs; | |
735 | } | |
736 | ||
737 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
738 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
739 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
740 | ||
741 | /* Do the two (enqueued) tasks belong to the same group ? */ | |
742 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
743 | { | |
744 | if (curr->se.cfs_rq == p->se.cfs_rq) | |
745 | return 1; | |
746 | ||
747 | return 0; | |
748 | } | |
749 | ||
750 | #else /* CONFIG_FAIR_GROUP_SCHED */ | |
751 | ||
752 | #define for_each_sched_entity(se) \ | |
753 | for (; se; se = NULL) | |
754 | ||
755 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
756 | { | |
757 | return &task_rq(p)->cfs; | |
758 | } | |
759 | ||
760 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
761 | { | |
762 | struct task_struct *p = task_of(se); | |
763 | struct rq *rq = task_rq(p); | |
764 | ||
765 | return &rq->cfs; | |
766 | } | |
767 | ||
768 | /* runqueue "owned" by this group */ | |
769 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
770 | { | |
771 | return NULL; | |
772 | } | |
773 | ||
774 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
775 | { | |
776 | return &cpu_rq(this_cpu)->cfs; | |
777 | } | |
778 | ||
779 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
780 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
781 | ||
782 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
783 | { | |
784 | return 1; | |
785 | } | |
786 | ||
787 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
788 | ||
789 | /* | |
790 | * The enqueue_task method is called before nr_running is | |
791 | * increased. Here we update the fair scheduling stats and | |
792 | * then put the task into the rbtree: | |
793 | */ | |
fd390f6a | 794 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
795 | { |
796 | struct cfs_rq *cfs_rq; | |
797 | struct sched_entity *se = &p->se; | |
798 | ||
799 | for_each_sched_entity(se) { | |
800 | if (se->on_rq) | |
801 | break; | |
802 | cfs_rq = cfs_rq_of(se); | |
668031ca | 803 | enqueue_entity(cfs_rq, se, wakeup); |
bf0f6f24 IM |
804 | } |
805 | } | |
806 | ||
807 | /* | |
808 | * The dequeue_task method is called before nr_running is | |
809 | * decreased. We remove the task from the rbtree and | |
810 | * update the fair scheduling stats: | |
811 | */ | |
f02231e5 | 812 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
813 | { |
814 | struct cfs_rq *cfs_rq; | |
815 | struct sched_entity *se = &p->se; | |
816 | ||
817 | for_each_sched_entity(se) { | |
818 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 819 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
820 | /* Don't dequeue parent if it has other entities besides us */ |
821 | if (cfs_rq->load.weight) | |
822 | break; | |
823 | } | |
824 | } | |
825 | ||
826 | /* | |
1799e35d IM |
827 | * sched_yield() support is very simple - we dequeue and enqueue. |
828 | * | |
829 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 IM |
830 | */ |
831 | static void yield_task_fair(struct rq *rq, struct task_struct *p) | |
832 | { | |
833 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1799e35d IM |
834 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
835 | struct sched_entity *rightmost, *se = &p->se; | |
836 | struct rb_node *parent; | |
bf0f6f24 IM |
837 | |
838 | /* | |
1799e35d IM |
839 | * Are we the only task in the tree? |
840 | */ | |
841 | if (unlikely(cfs_rq->nr_running == 1)) | |
842 | return; | |
843 | ||
844 | if (likely(!sysctl_sched_compat_yield)) { | |
845 | __update_rq_clock(rq); | |
846 | /* | |
847 | * Dequeue and enqueue the task to update its | |
848 | * position within the tree: | |
849 | */ | |
850 | dequeue_entity(cfs_rq, &p->se, 0); | |
851 | enqueue_entity(cfs_rq, &p->se, 0); | |
852 | ||
853 | return; | |
854 | } | |
855 | /* | |
856 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 857 | */ |
1799e35d IM |
858 | do { |
859 | parent = *link; | |
860 | link = &parent->rb_right; | |
861 | } while (*link); | |
862 | ||
863 | rightmost = rb_entry(parent, struct sched_entity, run_node); | |
864 | /* | |
865 | * Already in the rightmost position? | |
866 | */ | |
867 | if (unlikely(rightmost == se)) | |
868 | return; | |
869 | ||
870 | /* | |
871 | * Minimally necessary key value to be last in the tree: | |
872 | */ | |
873 | se->fair_key = rightmost->fair_key + 1; | |
874 | ||
875 | if (cfs_rq->rb_leftmost == &se->run_node) | |
876 | cfs_rq->rb_leftmost = rb_next(&se->run_node); | |
877 | /* | |
878 | * Relink the task to the rightmost position: | |
879 | */ | |
880 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); | |
881 | rb_link_node(&se->run_node, parent, link); | |
882 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
883 | } |
884 | ||
885 | /* | |
886 | * Preempt the current task with a newly woken task if needed: | |
887 | */ | |
2e09bf55 | 888 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
889 | { |
890 | struct task_struct *curr = rq->curr; | |
891 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
bf0f6f24 IM |
892 | |
893 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 894 | update_rq_clock(rq); |
b7cc0896 | 895 | update_curr(cfs_rq); |
bf0f6f24 IM |
896 | resched_task(curr); |
897 | return; | |
898 | } | |
2e09bf55 IM |
899 | if (is_same_group(curr, p)) { |
900 | s64 delta = curr->se.vruntime - p->se.vruntime; | |
bf0f6f24 | 901 | |
2e09bf55 IM |
902 | if (delta > (s64)sysctl_sched_wakeup_granularity) |
903 | resched_task(curr); | |
904 | } | |
bf0f6f24 IM |
905 | } |
906 | ||
fb8d4724 | 907 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
908 | { |
909 | struct cfs_rq *cfs_rq = &rq->cfs; | |
910 | struct sched_entity *se; | |
911 | ||
912 | if (unlikely(!cfs_rq->nr_running)) | |
913 | return NULL; | |
914 | ||
915 | do { | |
9948f4b2 | 916 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
917 | cfs_rq = group_cfs_rq(se); |
918 | } while (cfs_rq); | |
919 | ||
920 | return task_of(se); | |
921 | } | |
922 | ||
923 | /* | |
924 | * Account for a descheduled task: | |
925 | */ | |
31ee529c | 926 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
927 | { |
928 | struct sched_entity *se = &prev->se; | |
929 | struct cfs_rq *cfs_rq; | |
930 | ||
931 | for_each_sched_entity(se) { | |
932 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 933 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
934 | } |
935 | } | |
936 | ||
937 | /************************************************** | |
938 | * Fair scheduling class load-balancing methods: | |
939 | */ | |
940 | ||
941 | /* | |
942 | * Load-balancing iterator. Note: while the runqueue stays locked | |
943 | * during the whole iteration, the current task might be | |
944 | * dequeued so the iterator has to be dequeue-safe. Here we | |
945 | * achieve that by always pre-iterating before returning | |
946 | * the current task: | |
947 | */ | |
948 | static inline struct task_struct * | |
949 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) | |
950 | { | |
951 | struct task_struct *p; | |
952 | ||
953 | if (!curr) | |
954 | return NULL; | |
955 | ||
956 | p = rb_entry(curr, struct task_struct, se.run_node); | |
957 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
958 | ||
959 | return p; | |
960 | } | |
961 | ||
962 | static struct task_struct *load_balance_start_fair(void *arg) | |
963 | { | |
964 | struct cfs_rq *cfs_rq = arg; | |
965 | ||
966 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
967 | } | |
968 | ||
969 | static struct task_struct *load_balance_next_fair(void *arg) | |
970 | { | |
971 | struct cfs_rq *cfs_rq = arg; | |
972 | ||
973 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
974 | } | |
975 | ||
a4ac01c3 | 976 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
977 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
978 | { | |
979 | struct sched_entity *curr; | |
980 | struct task_struct *p; | |
981 | ||
982 | if (!cfs_rq->nr_running) | |
983 | return MAX_PRIO; | |
984 | ||
985 | curr = __pick_next_entity(cfs_rq); | |
986 | p = task_of(curr); | |
987 | ||
988 | return p->prio; | |
989 | } | |
a4ac01c3 | 990 | #endif |
bf0f6f24 | 991 | |
43010659 | 992 | static unsigned long |
bf0f6f24 | 993 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
a4ac01c3 PW |
994 | unsigned long max_nr_move, unsigned long max_load_move, |
995 | struct sched_domain *sd, enum cpu_idle_type idle, | |
996 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
997 | { |
998 | struct cfs_rq *busy_cfs_rq; | |
999 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
1000 | long rem_load_move = max_load_move; | |
1001 | struct rq_iterator cfs_rq_iterator; | |
1002 | ||
1003 | cfs_rq_iterator.start = load_balance_start_fair; | |
1004 | cfs_rq_iterator.next = load_balance_next_fair; | |
1005 | ||
1006 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 1007 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 1008 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 1009 | long imbalance; |
bf0f6f24 | 1010 | unsigned long maxload; |
bf0f6f24 IM |
1011 | |
1012 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
1013 | ||
e56f31aa | 1014 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
1015 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
1016 | if (imbalance <= 0) | |
1017 | continue; | |
1018 | ||
1019 | /* Don't pull more than imbalance/2 */ | |
1020 | imbalance /= 2; | |
1021 | maxload = min(rem_load_move, imbalance); | |
1022 | ||
a4ac01c3 PW |
1023 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
1024 | #else | |
e56f31aa | 1025 | # define maxload rem_load_move |
a4ac01c3 | 1026 | #endif |
bf0f6f24 IM |
1027 | /* pass busy_cfs_rq argument into |
1028 | * load_balance_[start|next]_fair iterators | |
1029 | */ | |
1030 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1031 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, | |
1032 | max_nr_move, maxload, sd, idle, all_pinned, | |
a4ac01c3 | 1033 | &load_moved, this_best_prio, &cfs_rq_iterator); |
bf0f6f24 IM |
1034 | |
1035 | total_nr_moved += nr_moved; | |
1036 | max_nr_move -= nr_moved; | |
1037 | rem_load_move -= load_moved; | |
1038 | ||
1039 | if (max_nr_move <= 0 || rem_load_move <= 0) | |
1040 | break; | |
1041 | } | |
1042 | ||
43010659 | 1043 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
1044 | } |
1045 | ||
1046 | /* | |
1047 | * scheduler tick hitting a task of our scheduling class: | |
1048 | */ | |
1049 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
1050 | { | |
1051 | struct cfs_rq *cfs_rq; | |
1052 | struct sched_entity *se = &curr->se; | |
1053 | ||
1054 | for_each_sched_entity(se) { | |
1055 | cfs_rq = cfs_rq_of(se); | |
1056 | entity_tick(cfs_rq, se); | |
1057 | } | |
1058 | } | |
1059 | ||
4d78e7b6 PZ |
1060 | #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
1061 | ||
bf0f6f24 IM |
1062 | /* |
1063 | * Share the fairness runtime between parent and child, thus the | |
1064 | * total amount of pressure for CPU stays equal - new tasks | |
1065 | * get a chance to run but frequent forkers are not allowed to | |
1066 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1067 | * the child is not running yet. | |
1068 | */ | |
ee0827d8 | 1069 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1070 | { |
1071 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1072 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
bf0f6f24 IM |
1073 | |
1074 | sched_info_queued(p); | |
1075 | ||
7109c442 | 1076 | update_curr(cfs_rq); |
aeb73b04 | 1077 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1078 | |
bf0f6f24 IM |
1079 | /* |
1080 | * The statistical average of wait_runtime is about | |
1081 | * -granularity/2, so initialize the task with that: | |
1082 | */ | |
e59c80c5 | 1083 | if (sched_feat(START_DEBIT)) |
4d78e7b6 PZ |
1084 | se->wait_runtime = -(__sched_period(cfs_rq->nr_running+1) / 2); |
1085 | ||
1086 | if (sysctl_sched_child_runs_first && | |
1087 | curr->vruntime < se->vruntime) { | |
1088 | ||
1089 | dequeue_entity(cfs_rq, curr, 0); | |
1090 | swap(curr->vruntime, se->vruntime); | |
1091 | enqueue_entity(cfs_rq, curr, 0); | |
1092 | } | |
bf0f6f24 | 1093 | |
e9acbff6 | 1094 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 | 1095 | __enqueue_entity(cfs_rq, se); |
bb61c210 | 1096 | resched_task(rq->curr); |
bf0f6f24 IM |
1097 | } |
1098 | ||
1099 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1100 | /* Account for a task changing its policy or group. | |
1101 | * | |
1102 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1103 | * migrates between groups/classes. | |
1104 | */ | |
1105 | static void set_curr_task_fair(struct rq *rq) | |
1106 | { | |
7c6c16f3 | 1107 | struct sched_entity *se = &rq->curr->se; |
a8e504d2 | 1108 | |
c3b64f1e IM |
1109 | for_each_sched_entity(se) |
1110 | set_next_entity(cfs_rq_of(se), se); | |
bf0f6f24 IM |
1111 | } |
1112 | #else | |
1113 | static void set_curr_task_fair(struct rq *rq) | |
1114 | { | |
1115 | } | |
1116 | #endif | |
1117 | ||
1118 | /* | |
1119 | * All the scheduling class methods: | |
1120 | */ | |
1121 | struct sched_class fair_sched_class __read_mostly = { | |
1122 | .enqueue_task = enqueue_task_fair, | |
1123 | .dequeue_task = dequeue_task_fair, | |
1124 | .yield_task = yield_task_fair, | |
1125 | ||
2e09bf55 | 1126 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1127 | |
1128 | .pick_next_task = pick_next_task_fair, | |
1129 | .put_prev_task = put_prev_task_fair, | |
1130 | ||
1131 | .load_balance = load_balance_fair, | |
1132 | ||
1133 | .set_curr_task = set_curr_task_fair, | |
1134 | .task_tick = task_tick_fair, | |
1135 | .task_new = task_new_fair, | |
1136 | }; | |
1137 | ||
1138 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1139 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1140 | { |
bf0f6f24 IM |
1141 | struct cfs_rq *cfs_rq; |
1142 | ||
c3b64f1e | 1143 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1144 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1145 | } |
1146 | #endif |