]>
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 | */ |
21805085 PZ |
37 | unsigned int sysctl_sched_latency __read_mostly = 20000000ULL; |
38 | ||
39 | /* | |
40 | * Minimal preemption granularity for CPU-bound tasks: | |
41 | * (default: 2 msec, units: nanoseconds) | |
42 | */ | |
172ac3db | 43 | unsigned int sysctl_sched_min_granularity __read_mostly = 2000000ULL; |
bf0f6f24 IM |
44 | |
45 | /* | |
46 | * SCHED_BATCH wake-up granularity. | |
71fd3714 | 47 | * (default: 25 msec, units: nanoseconds) |
bf0f6f24 IM |
48 | * |
49 | * This option delays the preemption effects of decoupled workloads | |
50 | * and reduces their over-scheduling. Synchronous workloads will still | |
51 | * have immediate wakeup/sleep latencies. | |
52 | */ | |
71fd3714 | 53 | unsigned int sysctl_sched_batch_wakeup_granularity __read_mostly = 25000000UL; |
bf0f6f24 IM |
54 | |
55 | /* | |
56 | * SCHED_OTHER wake-up granularity. | |
57 | * (default: 1 msec, units: nanoseconds) | |
58 | * | |
59 | * This option delays the preemption effects of decoupled workloads | |
60 | * and reduces their over-scheduling. Synchronous workloads will still | |
61 | * have immediate wakeup/sleep latencies. | |
62 | */ | |
71fd3714 | 63 | unsigned int sysctl_sched_wakeup_granularity __read_mostly = 1000000UL; |
bf0f6f24 IM |
64 | |
65 | unsigned int sysctl_sched_stat_granularity __read_mostly; | |
66 | ||
67 | /* | |
71fd3714 | 68 | * Initialized in sched_init_granularity() [to 5 times the base granularity]: |
bf0f6f24 IM |
69 | */ |
70 | unsigned int sysctl_sched_runtime_limit __read_mostly; | |
71 | ||
72 | /* | |
73 | * Debugging: various feature bits | |
74 | */ | |
75 | enum { | |
76 | SCHED_FEAT_FAIR_SLEEPERS = 1, | |
77 | SCHED_FEAT_SLEEPER_AVG = 2, | |
78 | SCHED_FEAT_SLEEPER_LOAD_AVG = 4, | |
79 | SCHED_FEAT_PRECISE_CPU_LOAD = 8, | |
80 | SCHED_FEAT_START_DEBIT = 16, | |
81 | SCHED_FEAT_SKIP_INITIAL = 32, | |
82 | }; | |
83 | ||
84 | unsigned int sysctl_sched_features __read_mostly = | |
85 | SCHED_FEAT_FAIR_SLEEPERS *1 | | |
5d2b3d36 | 86 | SCHED_FEAT_SLEEPER_AVG *0 | |
bf0f6f24 IM |
87 | SCHED_FEAT_SLEEPER_LOAD_AVG *1 | |
88 | SCHED_FEAT_PRECISE_CPU_LOAD *1 | | |
89 | SCHED_FEAT_START_DEBIT *1 | | |
90 | SCHED_FEAT_SKIP_INITIAL *0; | |
91 | ||
92 | extern struct sched_class fair_sched_class; | |
93 | ||
94 | /************************************************************** | |
95 | * CFS operations on generic schedulable entities: | |
96 | */ | |
97 | ||
98 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
99 | ||
100 | /* cpu runqueue to which this cfs_rq is attached */ | |
101 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
102 | { | |
103 | return cfs_rq->rq; | |
104 | } | |
105 | ||
106 | /* currently running entity (if any) on this cfs_rq */ | |
107 | static inline struct sched_entity *cfs_rq_curr(struct cfs_rq *cfs_rq) | |
108 | { | |
109 | return cfs_rq->curr; | |
110 | } | |
111 | ||
112 | /* An entity is a task if it doesn't "own" a runqueue */ | |
113 | #define entity_is_task(se) (!se->my_q) | |
114 | ||
115 | static inline void | |
116 | set_cfs_rq_curr(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
117 | { | |
118 | cfs_rq->curr = se; | |
119 | } | |
120 | ||
121 | #else /* CONFIG_FAIR_GROUP_SCHED */ | |
122 | ||
123 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
124 | { | |
125 | return container_of(cfs_rq, struct rq, cfs); | |
126 | } | |
127 | ||
128 | static inline struct sched_entity *cfs_rq_curr(struct cfs_rq *cfs_rq) | |
129 | { | |
130 | struct rq *rq = rq_of(cfs_rq); | |
131 | ||
132 | if (unlikely(rq->curr->sched_class != &fair_sched_class)) | |
133 | return NULL; | |
134 | ||
135 | return &rq->curr->se; | |
136 | } | |
137 | ||
138 | #define entity_is_task(se) 1 | |
139 | ||
140 | static inline void | |
141 | set_cfs_rq_curr(struct cfs_rq *cfs_rq, struct sched_entity *se) { } | |
142 | ||
143 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
144 | ||
145 | static inline struct task_struct *task_of(struct sched_entity *se) | |
146 | { | |
147 | return container_of(se, struct task_struct, se); | |
148 | } | |
149 | ||
150 | ||
151 | /************************************************************** | |
152 | * Scheduling class tree data structure manipulation methods: | |
153 | */ | |
154 | ||
155 | /* | |
156 | * Enqueue an entity into the rb-tree: | |
157 | */ | |
158 | static inline void | |
159 | __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
160 | { | |
161 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
162 | struct rb_node *parent = NULL; | |
163 | struct sched_entity *entry; | |
164 | s64 key = se->fair_key; | |
165 | int leftmost = 1; | |
166 | ||
167 | /* | |
168 | * Find the right place in the rbtree: | |
169 | */ | |
170 | while (*link) { | |
171 | parent = *link; | |
172 | entry = rb_entry(parent, struct sched_entity, run_node); | |
173 | /* | |
174 | * We dont care about collisions. Nodes with | |
175 | * the same key stay together. | |
176 | */ | |
177 | if (key - entry->fair_key < 0) { | |
178 | link = &parent->rb_left; | |
179 | } else { | |
180 | link = &parent->rb_right; | |
181 | leftmost = 0; | |
182 | } | |
183 | } | |
184 | ||
185 | /* | |
186 | * Maintain a cache of leftmost tree entries (it is frequently | |
187 | * used): | |
188 | */ | |
189 | if (leftmost) | |
190 | cfs_rq->rb_leftmost = &se->run_node; | |
191 | ||
192 | rb_link_node(&se->run_node, parent, link); | |
193 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
194 | update_load_add(&cfs_rq->load, se->load.weight); | |
195 | cfs_rq->nr_running++; | |
196 | se->on_rq = 1; | |
197 | } | |
198 | ||
199 | static inline void | |
200 | __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
201 | { | |
202 | if (cfs_rq->rb_leftmost == &se->run_node) | |
203 | cfs_rq->rb_leftmost = rb_next(&se->run_node); | |
204 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); | |
205 | update_load_sub(&cfs_rq->load, se->load.weight); | |
206 | cfs_rq->nr_running--; | |
207 | se->on_rq = 0; | |
208 | } | |
209 | ||
210 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
211 | { | |
212 | return cfs_rq->rb_leftmost; | |
213 | } | |
214 | ||
215 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
216 | { | |
217 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
218 | } | |
219 | ||
220 | /************************************************************** | |
221 | * Scheduling class statistics methods: | |
222 | */ | |
223 | ||
21805085 PZ |
224 | /* |
225 | * Calculate the preemption granularity needed to schedule every | |
226 | * runnable task once per sysctl_sched_latency amount of time. | |
227 | * (down to a sensible low limit on granularity) | |
228 | * | |
229 | * For example, if there are 2 tasks running and latency is 10 msecs, | |
230 | * we switch tasks every 5 msecs. If we have 3 tasks running, we have | |
231 | * to switch tasks every 3.33 msecs to get a 10 msecs observed latency | |
232 | * for each task. We do finer and finer scheduling up to until we | |
233 | * reach the minimum granularity value. | |
234 | * | |
235 | * To achieve this we use the following dynamic-granularity rule: | |
236 | * | |
237 | * gran = lat/nr - lat/nr/nr | |
238 | * | |
239 | * This comes out of the following equations: | |
240 | * | |
241 | * kA1 + gran = kB1 | |
242 | * kB2 + gran = kA2 | |
243 | * kA2 = kA1 | |
244 | * kB2 = kB1 - d + d/nr | |
245 | * lat = d * nr | |
246 | * | |
247 | * Where 'k' is key, 'A' is task A (waiting), 'B' is task B (running), | |
248 | * '1' is start of time, '2' is end of time, 'd' is delay between | |
249 | * 1 and 2 (during which task B was running), 'nr' is number of tasks | |
250 | * running, 'lat' is the the period of each task. ('lat' is the | |
251 | * sched_latency that we aim for.) | |
252 | */ | |
253 | static long | |
254 | sched_granularity(struct cfs_rq *cfs_rq) | |
255 | { | |
256 | unsigned int gran = sysctl_sched_latency; | |
257 | unsigned int nr = cfs_rq->nr_running; | |
258 | ||
259 | if (nr > 1) { | |
260 | gran = gran/nr - gran/nr/nr; | |
172ac3db | 261 | gran = max(gran, sysctl_sched_min_granularity); |
21805085 PZ |
262 | } |
263 | ||
264 | return gran; | |
265 | } | |
266 | ||
bf0f6f24 IM |
267 | /* |
268 | * We rescale the rescheduling granularity of tasks according to their | |
269 | * nice level, but only linearly, not exponentially: | |
270 | */ | |
271 | static long | |
272 | niced_granularity(struct sched_entity *curr, unsigned long granularity) | |
273 | { | |
274 | u64 tmp; | |
275 | ||
7cff8cf6 IM |
276 | if (likely(curr->load.weight == NICE_0_LOAD)) |
277 | return granularity; | |
bf0f6f24 | 278 | /* |
7cff8cf6 | 279 | * Positive nice levels get the same granularity as nice-0: |
bf0f6f24 | 280 | */ |
7cff8cf6 IM |
281 | if (likely(curr->load.weight < NICE_0_LOAD)) { |
282 | tmp = curr->load.weight * (u64)granularity; | |
283 | return (long) (tmp >> NICE_0_SHIFT); | |
284 | } | |
bf0f6f24 | 285 | /* |
7cff8cf6 | 286 | * Negative nice level tasks get linearly finer |
bf0f6f24 IM |
287 | * granularity: |
288 | */ | |
7cff8cf6 | 289 | tmp = curr->load.inv_weight * (u64)granularity; |
bf0f6f24 IM |
290 | |
291 | /* | |
292 | * It will always fit into 'long': | |
293 | */ | |
7cff8cf6 | 294 | return (long) (tmp >> WMULT_SHIFT); |
bf0f6f24 IM |
295 | } |
296 | ||
297 | static inline void | |
298 | limit_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
299 | { | |
300 | long limit = sysctl_sched_runtime_limit; | |
301 | ||
302 | /* | |
303 | * Niced tasks have the same history dynamic range as | |
304 | * non-niced tasks: | |
305 | */ | |
306 | if (unlikely(se->wait_runtime > limit)) { | |
307 | se->wait_runtime = limit; | |
308 | schedstat_inc(se, wait_runtime_overruns); | |
309 | schedstat_inc(cfs_rq, wait_runtime_overruns); | |
310 | } | |
311 | if (unlikely(se->wait_runtime < -limit)) { | |
312 | se->wait_runtime = -limit; | |
313 | schedstat_inc(se, wait_runtime_underruns); | |
314 | schedstat_inc(cfs_rq, wait_runtime_underruns); | |
315 | } | |
316 | } | |
317 | ||
318 | static inline void | |
319 | __add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
320 | { | |
321 | se->wait_runtime += delta; | |
322 | schedstat_add(se, sum_wait_runtime, delta); | |
323 | limit_wait_runtime(cfs_rq, se); | |
324 | } | |
325 | ||
326 | static void | |
327 | add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta) | |
328 | { | |
329 | schedstat_add(cfs_rq, wait_runtime, -se->wait_runtime); | |
330 | __add_wait_runtime(cfs_rq, se, delta); | |
331 | schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); | |
332 | } | |
333 | ||
334 | /* | |
335 | * Update the current task's runtime statistics. Skip current tasks that | |
336 | * are not in our scheduling class. | |
337 | */ | |
338 | static inline void | |
b7cc0896 | 339 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 340 | { |
c5dcfe72 | 341 | unsigned long delta, delta_exec, delta_fair, delta_mine; |
bf0f6f24 IM |
342 | struct load_weight *lw = &cfs_rq->load; |
343 | unsigned long load = lw->weight; | |
344 | ||
bf0f6f24 | 345 | delta_exec = curr->delta_exec; |
8179ca23 | 346 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
347 | |
348 | curr->sum_exec_runtime += delta_exec; | |
349 | cfs_rq->exec_clock += delta_exec; | |
350 | ||
fd8bb43e IM |
351 | if (unlikely(!load)) |
352 | return; | |
353 | ||
bf0f6f24 IM |
354 | delta_fair = calc_delta_fair(delta_exec, lw); |
355 | delta_mine = calc_delta_mine(delta_exec, curr->load.weight, lw); | |
356 | ||
5f01d519 | 357 | if (cfs_rq->sleeper_bonus > sysctl_sched_min_granularity) { |
ea0aa3b2 | 358 | delta = min((u64)delta_mine, cfs_rq->sleeper_bonus); |
b2133c8b IM |
359 | delta = min(delta, (unsigned long)( |
360 | (long)sysctl_sched_runtime_limit - curr->wait_runtime)); | |
bf0f6f24 IM |
361 | cfs_rq->sleeper_bonus -= delta; |
362 | delta_mine -= delta; | |
363 | } | |
364 | ||
365 | cfs_rq->fair_clock += delta_fair; | |
366 | /* | |
367 | * We executed delta_exec amount of time on the CPU, | |
368 | * but we were only entitled to delta_mine amount of | |
369 | * time during that period (if nr_running == 1 then | |
370 | * the two values are equal) | |
371 | * [Note: delta_mine - delta_exec is negative]: | |
372 | */ | |
373 | add_wait_runtime(cfs_rq, curr, delta_mine - delta_exec); | |
374 | } | |
375 | ||
b7cc0896 | 376 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 IM |
377 | { |
378 | struct sched_entity *curr = cfs_rq_curr(cfs_rq); | |
379 | unsigned long delta_exec; | |
380 | ||
381 | if (unlikely(!curr)) | |
382 | return; | |
383 | ||
384 | /* | |
385 | * Get the amount of time the current task was running | |
386 | * since the last time we changed load (this cannot | |
387 | * overflow on 32 bits): | |
388 | */ | |
d281918d | 389 | delta_exec = (unsigned long)(rq_of(cfs_rq)->clock - curr->exec_start); |
bf0f6f24 IM |
390 | |
391 | curr->delta_exec += delta_exec; | |
392 | ||
393 | if (unlikely(curr->delta_exec > sysctl_sched_stat_granularity)) { | |
b7cc0896 | 394 | __update_curr(cfs_rq, curr); |
bf0f6f24 IM |
395 | curr->delta_exec = 0; |
396 | } | |
d281918d | 397 | curr->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
398 | } |
399 | ||
400 | static inline void | |
5870db5b | 401 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
402 | { |
403 | se->wait_start_fair = cfs_rq->fair_clock; | |
d281918d | 404 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
405 | } |
406 | ||
407 | /* | |
408 | * We calculate fair deltas here, so protect against the random effects | |
409 | * of a multiplication overflow by capping it to the runtime limit: | |
410 | */ | |
411 | #if BITS_PER_LONG == 32 | |
412 | static inline unsigned long | |
413 | calc_weighted(unsigned long delta, unsigned long weight, int shift) | |
414 | { | |
415 | u64 tmp = (u64)delta * weight >> shift; | |
416 | ||
417 | if (unlikely(tmp > sysctl_sched_runtime_limit*2)) | |
418 | return sysctl_sched_runtime_limit*2; | |
419 | return tmp; | |
420 | } | |
421 | #else | |
422 | static inline unsigned long | |
423 | calc_weighted(unsigned long delta, unsigned long weight, int shift) | |
424 | { | |
425 | return delta * weight >> shift; | |
426 | } | |
427 | #endif | |
428 | ||
429 | /* | |
430 | * Task is being enqueued - update stats: | |
431 | */ | |
d2417e5a | 432 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
433 | { |
434 | s64 key; | |
435 | ||
436 | /* | |
437 | * Are we enqueueing a waiting task? (for current tasks | |
438 | * a dequeue/enqueue event is a NOP) | |
439 | */ | |
440 | if (se != cfs_rq_curr(cfs_rq)) | |
5870db5b | 441 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
442 | /* |
443 | * Update the key: | |
444 | */ | |
445 | key = cfs_rq->fair_clock; | |
446 | ||
447 | /* | |
448 | * Optimize the common nice 0 case: | |
449 | */ | |
450 | if (likely(se->load.weight == NICE_0_LOAD)) { | |
451 | key -= se->wait_runtime; | |
452 | } else { | |
453 | u64 tmp; | |
454 | ||
455 | if (se->wait_runtime < 0) { | |
456 | tmp = -se->wait_runtime; | |
457 | key += (tmp * se->load.inv_weight) >> | |
458 | (WMULT_SHIFT - NICE_0_SHIFT); | |
459 | } else { | |
460 | tmp = se->wait_runtime; | |
a69edb55 IM |
461 | key -= (tmp * se->load.inv_weight) >> |
462 | (WMULT_SHIFT - NICE_0_SHIFT); | |
bf0f6f24 IM |
463 | } |
464 | } | |
465 | ||
466 | se->fair_key = key; | |
467 | } | |
468 | ||
469 | /* | |
470 | * Note: must be called with a freshly updated rq->fair_clock. | |
471 | */ | |
472 | static inline void | |
eac55ea3 | 473 | __update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
474 | { |
475 | unsigned long delta_fair = se->delta_fair_run; | |
476 | ||
d281918d IM |
477 | schedstat_set(se->wait_max, max(se->wait_max, |
478 | rq_of(cfs_rq)->clock - se->wait_start)); | |
bf0f6f24 IM |
479 | |
480 | if (unlikely(se->load.weight != NICE_0_LOAD)) | |
481 | delta_fair = calc_weighted(delta_fair, se->load.weight, | |
482 | NICE_0_SHIFT); | |
483 | ||
484 | add_wait_runtime(cfs_rq, se, delta_fair); | |
485 | } | |
486 | ||
487 | static void | |
9ef0a961 | 488 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
489 | { |
490 | unsigned long delta_fair; | |
491 | ||
492 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), | |
493 | (u64)(cfs_rq->fair_clock - se->wait_start_fair)); | |
494 | ||
495 | se->delta_fair_run += delta_fair; | |
496 | if (unlikely(abs(se->delta_fair_run) >= | |
497 | sysctl_sched_stat_granularity)) { | |
eac55ea3 | 498 | __update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
499 | se->delta_fair_run = 0; |
500 | } | |
501 | ||
502 | se->wait_start_fair = 0; | |
6cfb0d5d | 503 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
504 | } |
505 | ||
506 | static inline void | |
19b6a2e3 | 507 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 508 | { |
b7cc0896 | 509 | update_curr(cfs_rq); |
bf0f6f24 IM |
510 | /* |
511 | * Mark the end of the wait period if dequeueing a | |
512 | * waiting task: | |
513 | */ | |
514 | if (se != cfs_rq_curr(cfs_rq)) | |
9ef0a961 | 515 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
516 | } |
517 | ||
518 | /* | |
519 | * We are picking a new current task - update its stats: | |
520 | */ | |
521 | static inline void | |
79303e9e | 522 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
523 | { |
524 | /* | |
525 | * We are starting a new run period: | |
526 | */ | |
d281918d | 527 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
528 | } |
529 | ||
530 | /* | |
531 | * We are descheduling a task - update its stats: | |
532 | */ | |
533 | static inline void | |
c7e9b5b2 | 534 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
535 | { |
536 | se->exec_start = 0; | |
537 | } | |
538 | ||
539 | /************************************************** | |
540 | * Scheduling class queueing methods: | |
541 | */ | |
542 | ||
dfdc119e | 543 | static void __enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
544 | { |
545 | unsigned long load = cfs_rq->load.weight, delta_fair; | |
546 | long prev_runtime; | |
547 | ||
b2133c8b IM |
548 | /* |
549 | * Do not boost sleepers if there's too much bonus 'in flight' | |
550 | * already: | |
551 | */ | |
552 | if (unlikely(cfs_rq->sleeper_bonus > sysctl_sched_runtime_limit)) | |
553 | return; | |
554 | ||
bf0f6f24 IM |
555 | if (sysctl_sched_features & SCHED_FEAT_SLEEPER_LOAD_AVG) |
556 | load = rq_of(cfs_rq)->cpu_load[2]; | |
557 | ||
558 | delta_fair = se->delta_fair_sleep; | |
559 | ||
560 | /* | |
561 | * Fix up delta_fair with the effect of us running | |
562 | * during the whole sleep period: | |
563 | */ | |
564 | if (sysctl_sched_features & SCHED_FEAT_SLEEPER_AVG) | |
565 | delta_fair = div64_likely32((u64)delta_fair * load, | |
566 | load + se->load.weight); | |
567 | ||
568 | if (unlikely(se->load.weight != NICE_0_LOAD)) | |
569 | delta_fair = calc_weighted(delta_fair, se->load.weight, | |
570 | NICE_0_SHIFT); | |
571 | ||
572 | prev_runtime = se->wait_runtime; | |
573 | __add_wait_runtime(cfs_rq, se, delta_fair); | |
b2133c8b | 574 | schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); |
bf0f6f24 IM |
575 | delta_fair = se->wait_runtime - prev_runtime; |
576 | ||
577 | /* | |
578 | * Track the amount of bonus we've given to sleepers: | |
579 | */ | |
580 | cfs_rq->sleeper_bonus += delta_fair; | |
bf0f6f24 IM |
581 | } |
582 | ||
2396af69 | 583 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
584 | { |
585 | struct task_struct *tsk = task_of(se); | |
586 | unsigned long delta_fair; | |
587 | ||
588 | if ((entity_is_task(se) && tsk->policy == SCHED_BATCH) || | |
589 | !(sysctl_sched_features & SCHED_FEAT_FAIR_SLEEPERS)) | |
590 | return; | |
591 | ||
592 | delta_fair = (unsigned long)min((u64)(2*sysctl_sched_runtime_limit), | |
593 | (u64)(cfs_rq->fair_clock - se->sleep_start_fair)); | |
594 | ||
595 | se->delta_fair_sleep += delta_fair; | |
596 | if (unlikely(abs(se->delta_fair_sleep) >= | |
597 | sysctl_sched_stat_granularity)) { | |
dfdc119e | 598 | __enqueue_sleeper(cfs_rq, se); |
bf0f6f24 IM |
599 | se->delta_fair_sleep = 0; |
600 | } | |
601 | ||
602 | se->sleep_start_fair = 0; | |
603 | ||
604 | #ifdef CONFIG_SCHEDSTATS | |
605 | if (se->sleep_start) { | |
d281918d | 606 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
607 | |
608 | if ((s64)delta < 0) | |
609 | delta = 0; | |
610 | ||
611 | if (unlikely(delta > se->sleep_max)) | |
612 | se->sleep_max = delta; | |
613 | ||
614 | se->sleep_start = 0; | |
615 | se->sum_sleep_runtime += delta; | |
616 | } | |
617 | if (se->block_start) { | |
d281918d | 618 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
619 | |
620 | if ((s64)delta < 0) | |
621 | delta = 0; | |
622 | ||
623 | if (unlikely(delta > se->block_max)) | |
624 | se->block_max = delta; | |
625 | ||
626 | se->block_start = 0; | |
627 | se->sum_sleep_runtime += delta; | |
628 | } | |
629 | #endif | |
630 | } | |
631 | ||
632 | static void | |
668031ca | 633 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
634 | { |
635 | /* | |
636 | * Update the fair clock. | |
637 | */ | |
b7cc0896 | 638 | update_curr(cfs_rq); |
bf0f6f24 IM |
639 | |
640 | if (wakeup) | |
2396af69 | 641 | enqueue_sleeper(cfs_rq, se); |
bf0f6f24 | 642 | |
d2417e5a | 643 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 IM |
644 | __enqueue_entity(cfs_rq, se); |
645 | } | |
646 | ||
647 | static void | |
525c2716 | 648 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 649 | { |
19b6a2e3 | 650 | update_stats_dequeue(cfs_rq, se); |
bf0f6f24 IM |
651 | if (sleep) { |
652 | se->sleep_start_fair = cfs_rq->fair_clock; | |
653 | #ifdef CONFIG_SCHEDSTATS | |
654 | if (entity_is_task(se)) { | |
655 | struct task_struct *tsk = task_of(se); | |
656 | ||
657 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 658 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 659 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 660 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
661 | } |
662 | cfs_rq->wait_runtime -= se->wait_runtime; | |
663 | #endif | |
664 | } | |
665 | __dequeue_entity(cfs_rq, se); | |
666 | } | |
667 | ||
668 | /* | |
669 | * Preempt the current task with a newly woken task if needed: | |
670 | */ | |
f6cf891c | 671 | static int |
bf0f6f24 IM |
672 | __check_preempt_curr_fair(struct cfs_rq *cfs_rq, struct sched_entity *se, |
673 | struct sched_entity *curr, unsigned long granularity) | |
674 | { | |
675 | s64 __delta = curr->fair_key - se->fair_key; | |
676 | ||
677 | /* | |
678 | * Take scheduling granularity into account - do not | |
679 | * preempt the current task unless the best task has | |
680 | * a larger than sched_granularity fairness advantage: | |
681 | */ | |
f6cf891c | 682 | if (__delta > niced_granularity(curr, granularity)) { |
bf0f6f24 | 683 | resched_task(rq_of(cfs_rq)->curr); |
f6cf891c IM |
684 | return 1; |
685 | } | |
686 | return 0; | |
bf0f6f24 IM |
687 | } |
688 | ||
689 | static inline void | |
8494f412 | 690 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
691 | { |
692 | /* | |
693 | * Any task has to be enqueued before it get to execute on | |
694 | * a CPU. So account for the time it spent waiting on the | |
695 | * runqueue. (note, here we rely on pick_next_task() having | |
696 | * done a put_prev_task_fair() shortly before this, which | |
697 | * updated rq->fair_clock - used by update_stats_wait_end()) | |
698 | */ | |
9ef0a961 | 699 | update_stats_wait_end(cfs_rq, se); |
79303e9e | 700 | update_stats_curr_start(cfs_rq, se); |
bf0f6f24 IM |
701 | set_cfs_rq_curr(cfs_rq, se); |
702 | } | |
703 | ||
9948f4b2 | 704 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 IM |
705 | { |
706 | struct sched_entity *se = __pick_next_entity(cfs_rq); | |
707 | ||
8494f412 | 708 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
709 | |
710 | return se; | |
711 | } | |
712 | ||
ab6cde26 | 713 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
714 | { |
715 | /* | |
716 | * If still on the runqueue then deactivate_task() | |
717 | * was not called and update_curr() has to be done: | |
718 | */ | |
719 | if (prev->on_rq) | |
b7cc0896 | 720 | update_curr(cfs_rq); |
bf0f6f24 | 721 | |
c7e9b5b2 | 722 | update_stats_curr_end(cfs_rq, prev); |
bf0f6f24 IM |
723 | |
724 | if (prev->on_rq) | |
5870db5b | 725 | update_stats_wait_start(cfs_rq, prev); |
bf0f6f24 IM |
726 | set_cfs_rq_curr(cfs_rq, NULL); |
727 | } | |
728 | ||
729 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
730 | { | |
f6cf891c | 731 | unsigned long gran, ideal_runtime, delta_exec; |
bf0f6f24 | 732 | struct sched_entity *next; |
c1b3da3e | 733 | |
bf0f6f24 IM |
734 | /* |
735 | * Dequeue and enqueue the task to update its | |
736 | * position within the tree: | |
737 | */ | |
525c2716 | 738 | dequeue_entity(cfs_rq, curr, 0); |
668031ca | 739 | enqueue_entity(cfs_rq, curr, 0); |
bf0f6f24 IM |
740 | |
741 | /* | |
742 | * Reschedule if another task tops the current one. | |
743 | */ | |
744 | next = __pick_next_entity(cfs_rq); | |
745 | if (next == curr) | |
746 | return; | |
747 | ||
f6cf891c IM |
748 | gran = sched_granularity(cfs_rq); |
749 | ideal_runtime = niced_granularity(curr, | |
750 | max(sysctl_sched_latency / cfs_rq->nr_running, | |
751 | (unsigned long)sysctl_sched_min_granularity)); | |
752 | /* | |
753 | * If we executed more than what the latency constraint suggests, | |
754 | * reduce the rescheduling granularity. This way the total latency | |
755 | * of how much a task is not scheduled converges to | |
756 | * sysctl_sched_latency: | |
757 | */ | |
758 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; | |
759 | if (delta_exec > ideal_runtime) | |
760 | gran = 0; | |
761 | ||
762 | if (__check_preempt_curr_fair(cfs_rq, next, curr, gran)) | |
763 | curr->prev_sum_exec_runtime = curr->sum_exec_runtime; | |
bf0f6f24 IM |
764 | } |
765 | ||
766 | /************************************************** | |
767 | * CFS operations on tasks: | |
768 | */ | |
769 | ||
770 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
771 | ||
772 | /* Walk up scheduling entities hierarchy */ | |
773 | #define for_each_sched_entity(se) \ | |
774 | for (; se; se = se->parent) | |
775 | ||
776 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
777 | { | |
778 | return p->se.cfs_rq; | |
779 | } | |
780 | ||
781 | /* runqueue on which this entity is (to be) queued */ | |
782 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
783 | { | |
784 | return se->cfs_rq; | |
785 | } | |
786 | ||
787 | /* runqueue "owned" by this group */ | |
788 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
789 | { | |
790 | return grp->my_q; | |
791 | } | |
792 | ||
793 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
794 | * another cpu ('this_cpu') | |
795 | */ | |
796 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
797 | { | |
798 | /* A later patch will take group into account */ | |
799 | return &cpu_rq(this_cpu)->cfs; | |
800 | } | |
801 | ||
802 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
803 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
804 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
805 | ||
806 | /* Do the two (enqueued) tasks belong to the same group ? */ | |
807 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
808 | { | |
809 | if (curr->se.cfs_rq == p->se.cfs_rq) | |
810 | return 1; | |
811 | ||
812 | return 0; | |
813 | } | |
814 | ||
815 | #else /* CONFIG_FAIR_GROUP_SCHED */ | |
816 | ||
817 | #define for_each_sched_entity(se) \ | |
818 | for (; se; se = NULL) | |
819 | ||
820 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
821 | { | |
822 | return &task_rq(p)->cfs; | |
823 | } | |
824 | ||
825 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
826 | { | |
827 | struct task_struct *p = task_of(se); | |
828 | struct rq *rq = task_rq(p); | |
829 | ||
830 | return &rq->cfs; | |
831 | } | |
832 | ||
833 | /* runqueue "owned" by this group */ | |
834 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
835 | { | |
836 | return NULL; | |
837 | } | |
838 | ||
839 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
840 | { | |
841 | return &cpu_rq(this_cpu)->cfs; | |
842 | } | |
843 | ||
844 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
845 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
846 | ||
847 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
848 | { | |
849 | return 1; | |
850 | } | |
851 | ||
852 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
853 | ||
854 | /* | |
855 | * The enqueue_task method is called before nr_running is | |
856 | * increased. Here we update the fair scheduling stats and | |
857 | * then put the task into the rbtree: | |
858 | */ | |
fd390f6a | 859 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
860 | { |
861 | struct cfs_rq *cfs_rq; | |
862 | struct sched_entity *se = &p->se; | |
863 | ||
864 | for_each_sched_entity(se) { | |
865 | if (se->on_rq) | |
866 | break; | |
867 | cfs_rq = cfs_rq_of(se); | |
668031ca | 868 | enqueue_entity(cfs_rq, se, wakeup); |
bf0f6f24 IM |
869 | } |
870 | } | |
871 | ||
872 | /* | |
873 | * The dequeue_task method is called before nr_running is | |
874 | * decreased. We remove the task from the rbtree and | |
875 | * update the fair scheduling stats: | |
876 | */ | |
f02231e5 | 877 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
878 | { |
879 | struct cfs_rq *cfs_rq; | |
880 | struct sched_entity *se = &p->se; | |
881 | ||
882 | for_each_sched_entity(se) { | |
883 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 884 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
885 | /* Don't dequeue parent if it has other entities besides us */ |
886 | if (cfs_rq->load.weight) | |
887 | break; | |
888 | } | |
889 | } | |
890 | ||
891 | /* | |
892 | * sched_yield() support is very simple - we dequeue and enqueue | |
893 | */ | |
894 | static void yield_task_fair(struct rq *rq, struct task_struct *p) | |
895 | { | |
896 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
bf0f6f24 | 897 | |
c1b3da3e | 898 | __update_rq_clock(rq); |
bf0f6f24 IM |
899 | /* |
900 | * Dequeue and enqueue the task to update its | |
901 | * position within the tree: | |
902 | */ | |
525c2716 | 903 | dequeue_entity(cfs_rq, &p->se, 0); |
668031ca | 904 | enqueue_entity(cfs_rq, &p->se, 0); |
bf0f6f24 IM |
905 | } |
906 | ||
907 | /* | |
908 | * Preempt the current task with a newly woken task if needed: | |
909 | */ | |
910 | static void check_preempt_curr_fair(struct rq *rq, struct task_struct *p) | |
911 | { | |
912 | struct task_struct *curr = rq->curr; | |
913 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
914 | unsigned long gran; | |
915 | ||
916 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 917 | update_rq_clock(rq); |
b7cc0896 | 918 | update_curr(cfs_rq); |
bf0f6f24 IM |
919 | resched_task(curr); |
920 | return; | |
921 | } | |
922 | ||
923 | gran = sysctl_sched_wakeup_granularity; | |
924 | /* | |
925 | * Batch tasks prefer throughput over latency: | |
926 | */ | |
927 | if (unlikely(p->policy == SCHED_BATCH)) | |
928 | gran = sysctl_sched_batch_wakeup_granularity; | |
929 | ||
930 | if (is_same_group(curr, p)) | |
931 | __check_preempt_curr_fair(cfs_rq, &p->se, &curr->se, gran); | |
932 | } | |
933 | ||
fb8d4724 | 934 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
935 | { |
936 | struct cfs_rq *cfs_rq = &rq->cfs; | |
937 | struct sched_entity *se; | |
938 | ||
939 | if (unlikely(!cfs_rq->nr_running)) | |
940 | return NULL; | |
941 | ||
942 | do { | |
9948f4b2 | 943 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
944 | cfs_rq = group_cfs_rq(se); |
945 | } while (cfs_rq); | |
946 | ||
947 | return task_of(se); | |
948 | } | |
949 | ||
950 | /* | |
951 | * Account for a descheduled task: | |
952 | */ | |
31ee529c | 953 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
954 | { |
955 | struct sched_entity *se = &prev->se; | |
956 | struct cfs_rq *cfs_rq; | |
957 | ||
958 | for_each_sched_entity(se) { | |
959 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 960 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
961 | } |
962 | } | |
963 | ||
964 | /************************************************** | |
965 | * Fair scheduling class load-balancing methods: | |
966 | */ | |
967 | ||
968 | /* | |
969 | * Load-balancing iterator. Note: while the runqueue stays locked | |
970 | * during the whole iteration, the current task might be | |
971 | * dequeued so the iterator has to be dequeue-safe. Here we | |
972 | * achieve that by always pre-iterating before returning | |
973 | * the current task: | |
974 | */ | |
975 | static inline struct task_struct * | |
976 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) | |
977 | { | |
978 | struct task_struct *p; | |
979 | ||
980 | if (!curr) | |
981 | return NULL; | |
982 | ||
983 | p = rb_entry(curr, struct task_struct, se.run_node); | |
984 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
985 | ||
986 | return p; | |
987 | } | |
988 | ||
989 | static struct task_struct *load_balance_start_fair(void *arg) | |
990 | { | |
991 | struct cfs_rq *cfs_rq = arg; | |
992 | ||
993 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
994 | } | |
995 | ||
996 | static struct task_struct *load_balance_next_fair(void *arg) | |
997 | { | |
998 | struct cfs_rq *cfs_rq = arg; | |
999 | ||
1000 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
1001 | } | |
1002 | ||
a4ac01c3 | 1003 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
1004 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
1005 | { | |
1006 | struct sched_entity *curr; | |
1007 | struct task_struct *p; | |
1008 | ||
1009 | if (!cfs_rq->nr_running) | |
1010 | return MAX_PRIO; | |
1011 | ||
1012 | curr = __pick_next_entity(cfs_rq); | |
1013 | p = task_of(curr); | |
1014 | ||
1015 | return p->prio; | |
1016 | } | |
a4ac01c3 | 1017 | #endif |
bf0f6f24 | 1018 | |
43010659 | 1019 | static unsigned long |
bf0f6f24 | 1020 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
a4ac01c3 PW |
1021 | unsigned long max_nr_move, unsigned long max_load_move, |
1022 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1023 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
1024 | { |
1025 | struct cfs_rq *busy_cfs_rq; | |
1026 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
1027 | long rem_load_move = max_load_move; | |
1028 | struct rq_iterator cfs_rq_iterator; | |
1029 | ||
1030 | cfs_rq_iterator.start = load_balance_start_fair; | |
1031 | cfs_rq_iterator.next = load_balance_next_fair; | |
1032 | ||
1033 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 1034 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 1035 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 1036 | long imbalance; |
bf0f6f24 | 1037 | unsigned long maxload; |
bf0f6f24 IM |
1038 | |
1039 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
1040 | ||
e56f31aa | 1041 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
1042 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
1043 | if (imbalance <= 0) | |
1044 | continue; | |
1045 | ||
1046 | /* Don't pull more than imbalance/2 */ | |
1047 | imbalance /= 2; | |
1048 | maxload = min(rem_load_move, imbalance); | |
1049 | ||
a4ac01c3 PW |
1050 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
1051 | #else | |
e56f31aa | 1052 | # define maxload rem_load_move |
a4ac01c3 | 1053 | #endif |
bf0f6f24 IM |
1054 | /* pass busy_cfs_rq argument into |
1055 | * load_balance_[start|next]_fair iterators | |
1056 | */ | |
1057 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1058 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, | |
1059 | max_nr_move, maxload, sd, idle, all_pinned, | |
a4ac01c3 | 1060 | &load_moved, this_best_prio, &cfs_rq_iterator); |
bf0f6f24 IM |
1061 | |
1062 | total_nr_moved += nr_moved; | |
1063 | max_nr_move -= nr_moved; | |
1064 | rem_load_move -= load_moved; | |
1065 | ||
1066 | if (max_nr_move <= 0 || rem_load_move <= 0) | |
1067 | break; | |
1068 | } | |
1069 | ||
43010659 | 1070 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
1071 | } |
1072 | ||
1073 | /* | |
1074 | * scheduler tick hitting a task of our scheduling class: | |
1075 | */ | |
1076 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
1077 | { | |
1078 | struct cfs_rq *cfs_rq; | |
1079 | struct sched_entity *se = &curr->se; | |
1080 | ||
1081 | for_each_sched_entity(se) { | |
1082 | cfs_rq = cfs_rq_of(se); | |
1083 | entity_tick(cfs_rq, se); | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * Share the fairness runtime between parent and child, thus the | |
1089 | * total amount of pressure for CPU stays equal - new tasks | |
1090 | * get a chance to run but frequent forkers are not allowed to | |
1091 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1092 | * the child is not running yet. | |
1093 | */ | |
ee0827d8 | 1094 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1095 | { |
1096 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1097 | struct sched_entity *se = &p->se; | |
bf0f6f24 IM |
1098 | |
1099 | sched_info_queued(p); | |
1100 | ||
d2417e5a | 1101 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 IM |
1102 | /* |
1103 | * Child runs first: we let it run before the parent | |
1104 | * until it reschedules once. We set up the key so that | |
1105 | * it will preempt the parent: | |
1106 | */ | |
1107 | p->se.fair_key = current->se.fair_key - | |
21805085 | 1108 | niced_granularity(&rq->curr->se, sched_granularity(cfs_rq)) - 1; |
bf0f6f24 IM |
1109 | /* |
1110 | * The first wait is dominated by the child-runs-first logic, | |
1111 | * so do not credit it with that waiting time yet: | |
1112 | */ | |
1113 | if (sysctl_sched_features & SCHED_FEAT_SKIP_INITIAL) | |
1114 | p->se.wait_start_fair = 0; | |
1115 | ||
1116 | /* | |
1117 | * The statistical average of wait_runtime is about | |
1118 | * -granularity/2, so initialize the task with that: | |
1119 | */ | |
1120 | if (sysctl_sched_features & SCHED_FEAT_START_DEBIT) | |
21805085 | 1121 | p->se.wait_runtime = -(sched_granularity(cfs_rq) / 2); |
bf0f6f24 IM |
1122 | |
1123 | __enqueue_entity(cfs_rq, se); | |
bf0f6f24 IM |
1124 | } |
1125 | ||
1126 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1127 | /* Account for a task changing its policy or group. | |
1128 | * | |
1129 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1130 | * migrates between groups/classes. | |
1131 | */ | |
1132 | static void set_curr_task_fair(struct rq *rq) | |
1133 | { | |
7c6c16f3 | 1134 | struct sched_entity *se = &rq->curr->se; |
a8e504d2 | 1135 | |
c3b64f1e IM |
1136 | for_each_sched_entity(se) |
1137 | set_next_entity(cfs_rq_of(se), se); | |
bf0f6f24 IM |
1138 | } |
1139 | #else | |
1140 | static void set_curr_task_fair(struct rq *rq) | |
1141 | { | |
1142 | } | |
1143 | #endif | |
1144 | ||
1145 | /* | |
1146 | * All the scheduling class methods: | |
1147 | */ | |
1148 | struct sched_class fair_sched_class __read_mostly = { | |
1149 | .enqueue_task = enqueue_task_fair, | |
1150 | .dequeue_task = dequeue_task_fair, | |
1151 | .yield_task = yield_task_fair, | |
1152 | ||
1153 | .check_preempt_curr = check_preempt_curr_fair, | |
1154 | ||
1155 | .pick_next_task = pick_next_task_fair, | |
1156 | .put_prev_task = put_prev_task_fair, | |
1157 | ||
1158 | .load_balance = load_balance_fair, | |
1159 | ||
1160 | .set_curr_task = set_curr_task_fair, | |
1161 | .task_tick = task_tick_fair, | |
1162 | .task_new = task_new_fair, | |
1163 | }; | |
1164 | ||
1165 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1166 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1167 | { |
bf0f6f24 IM |
1168 | struct cfs_rq *cfs_rq; |
1169 | ||
c3b64f1e | 1170 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1171 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1172 | } |
1173 | #endif |