]>
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 | 24 | * Targeted preemption latency for CPU-bound tasks: |
722aab0c | 25 | * (default: 20ms * (1 + ilog(ncpus)), units: nanoseconds) |
bf0f6f24 | 26 | * |
21805085 | 27 | * NOTE: this latency value is not the same as the concept of |
d274a4ce IM |
28 | * 'timeslice length' - timeslices in CFS are of variable length |
29 | * and have no persistent notion like in traditional, time-slice | |
30 | * based scheduling concepts. | |
bf0f6f24 | 31 | * |
d274a4ce IM |
32 | * (to see the precise effective timeslice length of your workload, |
33 | * run vmstat and monitor the context-switches (cs) field) | |
bf0f6f24 | 34 | */ |
19978ca6 | 35 | unsigned int sysctl_sched_latency = 20000000ULL; |
2bd8e6d4 IM |
36 | |
37 | /* | |
b2be5e96 | 38 | * Minimal preemption granularity for CPU-bound tasks: |
722aab0c | 39 | * (default: 4 msec * (1 + ilog(ncpus)), units: nanoseconds) |
2bd8e6d4 | 40 | */ |
722aab0c | 41 | unsigned int sysctl_sched_min_granularity = 4000000ULL; |
21805085 PZ |
42 | |
43 | /* | |
b2be5e96 PZ |
44 | * is kept at sysctl_sched_latency / sysctl_sched_min_granularity |
45 | */ | |
722aab0c | 46 | static unsigned int sched_nr_latency = 5; |
b2be5e96 PZ |
47 | |
48 | /* | |
49 | * After fork, child runs first. (default) If set to 0 then | |
50 | * parent will (try to) run first. | |
21805085 | 51 | */ |
b2be5e96 | 52 | const_debug unsigned int sysctl_sched_child_runs_first = 1; |
bf0f6f24 | 53 | |
1799e35d IM |
54 | /* |
55 | * sys_sched_yield() compat mode | |
56 | * | |
57 | * This option switches the agressive yield implementation of the | |
58 | * old scheduler back on. | |
59 | */ | |
60 | unsigned int __read_mostly sysctl_sched_compat_yield; | |
61 | ||
bf0f6f24 IM |
62 | /* |
63 | * SCHED_BATCH wake-up granularity. | |
722aab0c | 64 | * (default: 10 msec * (1 + ilog(ncpus)), units: nanoseconds) |
bf0f6f24 IM |
65 | * |
66 | * This option delays the preemption effects of decoupled workloads | |
67 | * and reduces their over-scheduling. Synchronous workloads will still | |
68 | * have immediate wakeup/sleep latencies. | |
69 | */ | |
19978ca6 | 70 | unsigned int sysctl_sched_batch_wakeup_granularity = 10000000UL; |
bf0f6f24 IM |
71 | |
72 | /* | |
73 | * SCHED_OTHER wake-up granularity. | |
722aab0c | 74 | * (default: 10 msec * (1 + ilog(ncpus)), units: nanoseconds) |
bf0f6f24 IM |
75 | * |
76 | * This option delays the preemption effects of decoupled workloads | |
77 | * and reduces their over-scheduling. Synchronous workloads will still | |
78 | * have immediate wakeup/sleep latencies. | |
79 | */ | |
19978ca6 | 80 | unsigned int sysctl_sched_wakeup_granularity = 10000000UL; |
bf0f6f24 | 81 | |
da84d961 IM |
82 | const_debug unsigned int sysctl_sched_migration_cost = 500000UL; |
83 | ||
bf0f6f24 IM |
84 | /************************************************************** |
85 | * CFS operations on generic schedulable entities: | |
86 | */ | |
87 | ||
62160e3f | 88 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 89 | |
62160e3f | 90 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
91 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
92 | { | |
62160e3f | 93 | return cfs_rq->rq; |
bf0f6f24 IM |
94 | } |
95 | ||
62160e3f IM |
96 | /* An entity is a task if it doesn't "own" a runqueue */ |
97 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 98 | |
62160e3f | 99 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
bf0f6f24 | 100 | |
62160e3f IM |
101 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
102 | { | |
103 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
104 | } |
105 | ||
106 | #define entity_is_task(se) 1 | |
107 | ||
bf0f6f24 IM |
108 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
109 | ||
110 | static inline struct task_struct *task_of(struct sched_entity *se) | |
111 | { | |
112 | return container_of(se, struct task_struct, se); | |
113 | } | |
114 | ||
115 | ||
116 | /************************************************************** | |
117 | * Scheduling class tree data structure manipulation methods: | |
118 | */ | |
119 | ||
0702e3eb | 120 | static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime) |
02e0431a | 121 | { |
368059a9 PZ |
122 | s64 delta = (s64)(vruntime - min_vruntime); |
123 | if (delta > 0) | |
02e0431a PZ |
124 | min_vruntime = vruntime; |
125 | ||
126 | return min_vruntime; | |
127 | } | |
128 | ||
0702e3eb | 129 | static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime) |
b0ffd246 PZ |
130 | { |
131 | s64 delta = (s64)(vruntime - min_vruntime); | |
132 | if (delta < 0) | |
133 | min_vruntime = vruntime; | |
134 | ||
135 | return min_vruntime; | |
136 | } | |
137 | ||
0702e3eb | 138 | static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
9014623c | 139 | { |
30cfdcfc | 140 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
141 | } |
142 | ||
bf0f6f24 IM |
143 | /* |
144 | * Enqueue an entity into the rb-tree: | |
145 | */ | |
0702e3eb | 146 | static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
147 | { |
148 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
149 | struct rb_node *parent = NULL; | |
150 | struct sched_entity *entry; | |
9014623c | 151 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
152 | int leftmost = 1; |
153 | ||
154 | /* | |
155 | * Find the right place in the rbtree: | |
156 | */ | |
157 | while (*link) { | |
158 | parent = *link; | |
159 | entry = rb_entry(parent, struct sched_entity, run_node); | |
160 | /* | |
161 | * We dont care about collisions. Nodes with | |
162 | * the same key stay together. | |
163 | */ | |
9014623c | 164 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
165 | link = &parent->rb_left; |
166 | } else { | |
167 | link = &parent->rb_right; | |
168 | leftmost = 0; | |
169 | } | |
170 | } | |
171 | ||
172 | /* | |
173 | * Maintain a cache of leftmost tree entries (it is frequently | |
174 | * used): | |
175 | */ | |
176 | if (leftmost) | |
57cb499d | 177 | cfs_rq->rb_leftmost = &se->run_node; |
bf0f6f24 IM |
178 | |
179 | rb_link_node(&se->run_node, parent, link); | |
180 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
181 | } |
182 | ||
0702e3eb | 183 | static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
184 | { |
185 | if (cfs_rq->rb_leftmost == &se->run_node) | |
57cb499d | 186 | cfs_rq->rb_leftmost = rb_next(&se->run_node); |
e9acbff6 | 187 | |
bf0f6f24 | 188 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
189 | } |
190 | ||
191 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
192 | { | |
193 | return cfs_rq->rb_leftmost; | |
194 | } | |
195 | ||
196 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
197 | { | |
198 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
199 | } | |
200 | ||
aeb73b04 PZ |
201 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
202 | { | |
203 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
204 | struct sched_entity *se = NULL; | |
205 | struct rb_node *parent; | |
206 | ||
207 | while (*link) { | |
208 | parent = *link; | |
209 | se = rb_entry(parent, struct sched_entity, run_node); | |
210 | link = &parent->rb_right; | |
211 | } | |
212 | ||
213 | return se; | |
214 | } | |
215 | ||
bf0f6f24 IM |
216 | /************************************************************** |
217 | * Scheduling class statistics methods: | |
218 | */ | |
219 | ||
b2be5e96 PZ |
220 | #ifdef CONFIG_SCHED_DEBUG |
221 | int sched_nr_latency_handler(struct ctl_table *table, int write, | |
222 | struct file *filp, void __user *buffer, size_t *lenp, | |
223 | loff_t *ppos) | |
224 | { | |
225 | int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); | |
226 | ||
227 | if (ret || !write) | |
228 | return ret; | |
229 | ||
230 | sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency, | |
231 | sysctl_sched_min_granularity); | |
232 | ||
233 | return 0; | |
234 | } | |
235 | #endif | |
647e7cac IM |
236 | |
237 | /* | |
238 | * The idea is to set a period in which each task runs once. | |
239 | * | |
240 | * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch | |
241 | * this period because otherwise the slices get too small. | |
242 | * | |
243 | * p = (nr <= nl) ? l : l*nr/nl | |
244 | */ | |
4d78e7b6 PZ |
245 | static u64 __sched_period(unsigned long nr_running) |
246 | { | |
247 | u64 period = sysctl_sched_latency; | |
b2be5e96 | 248 | unsigned long nr_latency = sched_nr_latency; |
4d78e7b6 PZ |
249 | |
250 | if (unlikely(nr_running > nr_latency)) { | |
4bf0b771 | 251 | period = sysctl_sched_min_granularity; |
4d78e7b6 | 252 | period *= nr_running; |
4d78e7b6 PZ |
253 | } |
254 | ||
255 | return period; | |
256 | } | |
257 | ||
647e7cac IM |
258 | /* |
259 | * We calculate the wall-time slice from the period by taking a part | |
260 | * proportional to the weight. | |
261 | * | |
262 | * s = p*w/rw | |
263 | */ | |
6d0f0ebd | 264 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 265 | { |
647e7cac | 266 | u64 slice = __sched_period(cfs_rq->nr_running); |
21805085 | 267 | |
647e7cac IM |
268 | slice *= se->load.weight; |
269 | do_div(slice, cfs_rq->load.weight); | |
21805085 | 270 | |
647e7cac | 271 | return slice; |
bf0f6f24 IM |
272 | } |
273 | ||
647e7cac IM |
274 | /* |
275 | * We calculate the vruntime slice. | |
276 | * | |
277 | * vs = s/w = p/rw | |
278 | */ | |
279 | static u64 __sched_vslice(unsigned long rq_weight, unsigned long nr_running) | |
67e9fb2a | 280 | { |
647e7cac | 281 | u64 vslice = __sched_period(nr_running); |
67e9fb2a | 282 | |
10b77724 | 283 | vslice *= NICE_0_LOAD; |
647e7cac | 284 | do_div(vslice, rq_weight); |
67e9fb2a | 285 | |
647e7cac IM |
286 | return vslice; |
287 | } | |
5f6d858e | 288 | |
647e7cac IM |
289 | static u64 sched_vslice(struct cfs_rq *cfs_rq) |
290 | { | |
291 | return __sched_vslice(cfs_rq->load.weight, cfs_rq->nr_running); | |
292 | } | |
293 | ||
294 | static u64 sched_vslice_add(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
295 | { | |
296 | return __sched_vslice(cfs_rq->load.weight + se->load.weight, | |
297 | cfs_rq->nr_running + 1); | |
67e9fb2a PZ |
298 | } |
299 | ||
bf0f6f24 IM |
300 | /* |
301 | * Update the current task's runtime statistics. Skip current tasks that | |
302 | * are not in our scheduling class. | |
303 | */ | |
304 | static inline void | |
8ebc91d9 IM |
305 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
306 | unsigned long delta_exec) | |
bf0f6f24 | 307 | { |
bbdba7c0 | 308 | unsigned long delta_exec_weighted; |
b0ffd246 | 309 | u64 vruntime; |
bf0f6f24 | 310 | |
8179ca23 | 311 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
312 | |
313 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 314 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
e9acbff6 IM |
315 | delta_exec_weighted = delta_exec; |
316 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { | |
317 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, | |
318 | &curr->load); | |
319 | } | |
320 | curr->vruntime += delta_exec_weighted; | |
02e0431a PZ |
321 | |
322 | /* | |
323 | * maintain cfs_rq->min_vruntime to be a monotonic increasing | |
324 | * value tracking the leftmost vruntime in the tree. | |
325 | */ | |
326 | if (first_fair(cfs_rq)) { | |
b0ffd246 PZ |
327 | vruntime = min_vruntime(curr->vruntime, |
328 | __pick_next_entity(cfs_rq)->vruntime); | |
02e0431a | 329 | } else |
b0ffd246 | 330 | vruntime = curr->vruntime; |
02e0431a PZ |
331 | |
332 | cfs_rq->min_vruntime = | |
b0ffd246 | 333 | max_vruntime(cfs_rq->min_vruntime, vruntime); |
bf0f6f24 IM |
334 | } |
335 | ||
b7cc0896 | 336 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 337 | { |
429d43bc | 338 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 339 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
340 | unsigned long delta_exec; |
341 | ||
342 | if (unlikely(!curr)) | |
343 | return; | |
344 | ||
345 | /* | |
346 | * Get the amount of time the current task was running | |
347 | * since the last time we changed load (this cannot | |
348 | * overflow on 32 bits): | |
349 | */ | |
8ebc91d9 | 350 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 351 | |
8ebc91d9 IM |
352 | __update_curr(cfs_rq, curr, delta_exec); |
353 | curr->exec_start = now; | |
d842de87 SV |
354 | |
355 | if (entity_is_task(curr)) { | |
356 | struct task_struct *curtask = task_of(curr); | |
357 | ||
358 | cpuacct_charge(curtask, delta_exec); | |
359 | } | |
bf0f6f24 IM |
360 | } |
361 | ||
362 | static inline void | |
5870db5b | 363 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 364 | { |
d281918d | 365 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
366 | } |
367 | ||
bf0f6f24 IM |
368 | /* |
369 | * Task is being enqueued - update stats: | |
370 | */ | |
d2417e5a | 371 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 372 | { |
bf0f6f24 IM |
373 | /* |
374 | * Are we enqueueing a waiting task? (for current tasks | |
375 | * a dequeue/enqueue event is a NOP) | |
376 | */ | |
429d43bc | 377 | if (se != cfs_rq->curr) |
5870db5b | 378 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
379 | } |
380 | ||
bf0f6f24 | 381 | static void |
9ef0a961 | 382 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 383 | { |
bbdba7c0 IM |
384 | schedstat_set(se->wait_max, max(se->wait_max, |
385 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6cfb0d5d | 386 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
387 | } |
388 | ||
389 | static inline void | |
19b6a2e3 | 390 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 391 | { |
bf0f6f24 IM |
392 | /* |
393 | * Mark the end of the wait period if dequeueing a | |
394 | * waiting task: | |
395 | */ | |
429d43bc | 396 | if (se != cfs_rq->curr) |
9ef0a961 | 397 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
398 | } |
399 | ||
400 | /* | |
401 | * We are picking a new current task - update its stats: | |
402 | */ | |
403 | static inline void | |
79303e9e | 404 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
405 | { |
406 | /* | |
407 | * We are starting a new run period: | |
408 | */ | |
d281918d | 409 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
410 | } |
411 | ||
bf0f6f24 IM |
412 | /************************************************** |
413 | * Scheduling class queueing methods: | |
414 | */ | |
415 | ||
30cfdcfc DA |
416 | static void |
417 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
418 | { | |
419 | update_load_add(&cfs_rq->load, se->load.weight); | |
420 | cfs_rq->nr_running++; | |
421 | se->on_rq = 1; | |
422 | } | |
423 | ||
424 | static void | |
425 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
426 | { | |
427 | update_load_sub(&cfs_rq->load, se->load.weight); | |
428 | cfs_rq->nr_running--; | |
429 | se->on_rq = 0; | |
430 | } | |
431 | ||
2396af69 | 432 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 433 | { |
bf0f6f24 IM |
434 | #ifdef CONFIG_SCHEDSTATS |
435 | if (se->sleep_start) { | |
d281918d | 436 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
437 | |
438 | if ((s64)delta < 0) | |
439 | delta = 0; | |
440 | ||
441 | if (unlikely(delta > se->sleep_max)) | |
442 | se->sleep_max = delta; | |
443 | ||
444 | se->sleep_start = 0; | |
445 | se->sum_sleep_runtime += delta; | |
446 | } | |
447 | if (se->block_start) { | |
d281918d | 448 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
449 | |
450 | if ((s64)delta < 0) | |
451 | delta = 0; | |
452 | ||
453 | if (unlikely(delta > se->block_max)) | |
454 | se->block_max = delta; | |
455 | ||
456 | se->block_start = 0; | |
457 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
458 | |
459 | /* | |
460 | * Blocking time is in units of nanosecs, so shift by 20 to | |
461 | * get a milliseconds-range estimation of the amount of | |
462 | * time that the task spent sleeping: | |
463 | */ | |
464 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf IM |
465 | struct task_struct *tsk = task_of(se); |
466 | ||
30084fbd IM |
467 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
468 | delta >> 20); | |
469 | } | |
bf0f6f24 IM |
470 | } |
471 | #endif | |
472 | } | |
473 | ||
ddc97297 PZ |
474 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
475 | { | |
476 | #ifdef CONFIG_SCHED_DEBUG | |
477 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
478 | ||
479 | if (d < 0) | |
480 | d = -d; | |
481 | ||
482 | if (d > 3*sysctl_sched_latency) | |
483 | schedstat_inc(cfs_rq, nr_spread_over); | |
484 | #endif | |
485 | } | |
486 | ||
aeb73b04 PZ |
487 | static void |
488 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
489 | { | |
67e9fb2a | 490 | u64 vruntime; |
aeb73b04 | 491 | |
67e9fb2a | 492 | vruntime = cfs_rq->min_vruntime; |
94dfb5e7 | 493 | |
06877c33 | 494 | if (sched_feat(TREE_AVG)) { |
94dfb5e7 PZ |
495 | struct sched_entity *last = __pick_last_entity(cfs_rq); |
496 | if (last) { | |
67e9fb2a PZ |
497 | vruntime += last->vruntime; |
498 | vruntime >>= 1; | |
94dfb5e7 | 499 | } |
67e9fb2a | 500 | } else if (sched_feat(APPROX_AVG) && cfs_rq->nr_running) |
647e7cac | 501 | vruntime += sched_vslice(cfs_rq)/2; |
94dfb5e7 | 502 | |
2cb8600e PZ |
503 | /* |
504 | * The 'current' period is already promised to the current tasks, | |
505 | * however the extra weight of the new task will slow them down a | |
506 | * little, place the new task so that it fits in the slot that | |
507 | * stays open at the end. | |
508 | */ | |
94dfb5e7 | 509 | if (initial && sched_feat(START_DEBIT)) |
647e7cac | 510 | vruntime += sched_vslice_add(cfs_rq, se); |
aeb73b04 | 511 | |
8465e792 | 512 | if (!initial) { |
2cb8600e | 513 | /* sleeps upto a single latency don't count. */ |
6cbf1c12 | 514 | if (sched_feat(NEW_FAIR_SLEEPERS) && entity_is_task(se)) |
94359f05 IM |
515 | vruntime -= sysctl_sched_latency; |
516 | ||
2cb8600e PZ |
517 | /* ensure we never gain time by being placed backwards. */ |
518 | vruntime = max_vruntime(se->vruntime, vruntime); | |
aeb73b04 PZ |
519 | } |
520 | ||
67e9fb2a | 521 | se->vruntime = vruntime; |
aeb73b04 PZ |
522 | } |
523 | ||
bf0f6f24 | 524 | static void |
83b699ed | 525 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
526 | { |
527 | /* | |
a2a2d680 | 528 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 529 | */ |
b7cc0896 | 530 | update_curr(cfs_rq); |
bf0f6f24 | 531 | |
e9acbff6 | 532 | if (wakeup) { |
aeb73b04 | 533 | place_entity(cfs_rq, se, 0); |
2396af69 | 534 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 535 | } |
bf0f6f24 | 536 | |
d2417e5a | 537 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 538 | check_spread(cfs_rq, se); |
83b699ed SV |
539 | if (se != cfs_rq->curr) |
540 | __enqueue_entity(cfs_rq, se); | |
30cfdcfc | 541 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 IM |
542 | } |
543 | ||
544 | static void | |
525c2716 | 545 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 546 | { |
a2a2d680 DA |
547 | /* |
548 | * Update run-time statistics of the 'current'. | |
549 | */ | |
550 | update_curr(cfs_rq); | |
551 | ||
19b6a2e3 | 552 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 553 | if (sleep) { |
67e9fb2a | 554 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
555 | if (entity_is_task(se)) { |
556 | struct task_struct *tsk = task_of(se); | |
557 | ||
558 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 559 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 560 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 561 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 562 | } |
db36cc7d | 563 | #endif |
67e9fb2a PZ |
564 | } |
565 | ||
83b699ed | 566 | if (se != cfs_rq->curr) |
30cfdcfc DA |
567 | __dequeue_entity(cfs_rq, se); |
568 | account_entity_dequeue(cfs_rq, se); | |
bf0f6f24 IM |
569 | } |
570 | ||
571 | /* | |
572 | * Preempt the current task with a newly woken task if needed: | |
573 | */ | |
7c92e54f | 574 | static void |
2e09bf55 | 575 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 576 | { |
11697830 PZ |
577 | unsigned long ideal_runtime, delta_exec; |
578 | ||
6d0f0ebd | 579 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 | 580 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
3e3e13f3 | 581 | if (delta_exec > ideal_runtime) |
bf0f6f24 IM |
582 | resched_task(rq_of(cfs_rq)->curr); |
583 | } | |
584 | ||
83b699ed | 585 | static void |
8494f412 | 586 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 587 | { |
83b699ed SV |
588 | /* 'current' is not kept within the tree. */ |
589 | if (se->on_rq) { | |
590 | /* | |
591 | * Any task has to be enqueued before it get to execute on | |
592 | * a CPU. So account for the time it spent waiting on the | |
593 | * runqueue. | |
594 | */ | |
595 | update_stats_wait_end(cfs_rq, se); | |
596 | __dequeue_entity(cfs_rq, se); | |
597 | } | |
598 | ||
79303e9e | 599 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 600 | cfs_rq->curr = se; |
eba1ed4b IM |
601 | #ifdef CONFIG_SCHEDSTATS |
602 | /* | |
603 | * Track our maximum slice length, if the CPU's load is at | |
604 | * least twice that of our own weight (i.e. dont track it | |
605 | * when there are only lesser-weight tasks around): | |
606 | */ | |
495eca49 | 607 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
608 | se->slice_max = max(se->slice_max, |
609 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
610 | } | |
611 | #endif | |
4a55b450 | 612 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
613 | } |
614 | ||
9948f4b2 | 615 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 | 616 | { |
08ec3df5 | 617 | struct sched_entity *se = NULL; |
bf0f6f24 | 618 | |
08ec3df5 DA |
619 | if (first_fair(cfs_rq)) { |
620 | se = __pick_next_entity(cfs_rq); | |
621 | set_next_entity(cfs_rq, se); | |
622 | } | |
bf0f6f24 IM |
623 | |
624 | return se; | |
625 | } | |
626 | ||
ab6cde26 | 627 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
628 | { |
629 | /* | |
630 | * If still on the runqueue then deactivate_task() | |
631 | * was not called and update_curr() has to be done: | |
632 | */ | |
633 | if (prev->on_rq) | |
b7cc0896 | 634 | update_curr(cfs_rq); |
bf0f6f24 | 635 | |
ddc97297 | 636 | check_spread(cfs_rq, prev); |
30cfdcfc | 637 | if (prev->on_rq) { |
5870db5b | 638 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
639 | /* Put 'current' back into the tree. */ |
640 | __enqueue_entity(cfs_rq, prev); | |
641 | } | |
429d43bc | 642 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
643 | } |
644 | ||
8f4d37ec PZ |
645 | static void |
646 | entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) | |
bf0f6f24 | 647 | { |
bf0f6f24 | 648 | /* |
30cfdcfc | 649 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 650 | */ |
30cfdcfc | 651 | update_curr(cfs_rq); |
bf0f6f24 | 652 | |
8f4d37ec PZ |
653 | #ifdef CONFIG_SCHED_HRTICK |
654 | /* | |
655 | * queued ticks are scheduled to match the slice, so don't bother | |
656 | * validating it and just reschedule. | |
657 | */ | |
658 | if (queued) | |
659 | return resched_task(rq_of(cfs_rq)->curr); | |
660 | /* | |
661 | * don't let the period tick interfere with the hrtick preemption | |
662 | */ | |
663 | if (!sched_feat(DOUBLE_TICK) && | |
664 | hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) | |
665 | return; | |
666 | #endif | |
667 | ||
ce6c1311 | 668 | if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT)) |
2e09bf55 | 669 | check_preempt_tick(cfs_rq, curr); |
bf0f6f24 IM |
670 | } |
671 | ||
672 | /************************************************** | |
673 | * CFS operations on tasks: | |
674 | */ | |
675 | ||
676 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
677 | ||
678 | /* Walk up scheduling entities hierarchy */ | |
679 | #define for_each_sched_entity(se) \ | |
680 | for (; se; se = se->parent) | |
681 | ||
682 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
683 | { | |
684 | return p->se.cfs_rq; | |
685 | } | |
686 | ||
687 | /* runqueue on which this entity is (to be) queued */ | |
688 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
689 | { | |
690 | return se->cfs_rq; | |
691 | } | |
692 | ||
693 | /* runqueue "owned" by this group */ | |
694 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
695 | { | |
696 | return grp->my_q; | |
697 | } | |
698 | ||
699 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
700 | * another cpu ('this_cpu') | |
701 | */ | |
702 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
703 | { | |
29f59db3 | 704 | return cfs_rq->tg->cfs_rq[this_cpu]; |
bf0f6f24 IM |
705 | } |
706 | ||
707 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
708 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
ec2c507f | 709 | list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) |
bf0f6f24 | 710 | |
fad095a7 SV |
711 | /* Do the two (enqueued) entities belong to the same group ? */ |
712 | static inline int | |
713 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
bf0f6f24 | 714 | { |
fad095a7 | 715 | if (se->cfs_rq == pse->cfs_rq) |
bf0f6f24 IM |
716 | return 1; |
717 | ||
718 | return 0; | |
719 | } | |
720 | ||
fad095a7 SV |
721 | static inline struct sched_entity *parent_entity(struct sched_entity *se) |
722 | { | |
723 | return se->parent; | |
724 | } | |
725 | ||
6b2d7700 SV |
726 | #define GROUP_IMBALANCE_PCT 20 |
727 | ||
bf0f6f24 IM |
728 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
729 | ||
730 | #define for_each_sched_entity(se) \ | |
731 | for (; se; se = NULL) | |
732 | ||
733 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
734 | { | |
735 | return &task_rq(p)->cfs; | |
736 | } | |
737 | ||
738 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
739 | { | |
740 | struct task_struct *p = task_of(se); | |
741 | struct rq *rq = task_rq(p); | |
742 | ||
743 | return &rq->cfs; | |
744 | } | |
745 | ||
746 | /* runqueue "owned" by this group */ | |
747 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
748 | { | |
749 | return NULL; | |
750 | } | |
751 | ||
752 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
753 | { | |
754 | return &cpu_rq(this_cpu)->cfs; | |
755 | } | |
756 | ||
757 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
758 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
759 | ||
fad095a7 SV |
760 | static inline int |
761 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
bf0f6f24 IM |
762 | { |
763 | return 1; | |
764 | } | |
765 | ||
fad095a7 SV |
766 | static inline struct sched_entity *parent_entity(struct sched_entity *se) |
767 | { | |
768 | return NULL; | |
769 | } | |
770 | ||
bf0f6f24 IM |
771 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
772 | ||
8f4d37ec PZ |
773 | #ifdef CONFIG_SCHED_HRTICK |
774 | static void hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
775 | { | |
776 | int requeue = rq->curr == p; | |
777 | struct sched_entity *se = &p->se; | |
778 | struct cfs_rq *cfs_rq = cfs_rq_of(se); | |
779 | ||
780 | WARN_ON(task_rq(p) != rq); | |
781 | ||
782 | if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) { | |
783 | u64 slice = sched_slice(cfs_rq, se); | |
784 | u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime; | |
785 | s64 delta = slice - ran; | |
786 | ||
787 | if (delta < 0) { | |
788 | if (rq->curr == p) | |
789 | resched_task(p); | |
790 | return; | |
791 | } | |
792 | ||
793 | /* | |
794 | * Don't schedule slices shorter than 10000ns, that just | |
795 | * doesn't make sense. Rely on vruntime for fairness. | |
796 | */ | |
797 | if (!requeue) | |
798 | delta = max(10000LL, delta); | |
799 | ||
800 | hrtick_start(rq, delta, requeue); | |
801 | } | |
802 | } | |
803 | #else | |
804 | static inline void | |
805 | hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
806 | { | |
807 | } | |
808 | #endif | |
809 | ||
bf0f6f24 IM |
810 | /* |
811 | * The enqueue_task method is called before nr_running is | |
812 | * increased. Here we update the fair scheduling stats and | |
813 | * then put the task into the rbtree: | |
814 | */ | |
fd390f6a | 815 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
816 | { |
817 | struct cfs_rq *cfs_rq; | |
58e2d4ca SV |
818 | struct sched_entity *se = &p->se, |
819 | *topse = NULL; /* Highest schedulable entity */ | |
820 | int incload = 1; | |
bf0f6f24 IM |
821 | |
822 | for_each_sched_entity(se) { | |
58e2d4ca SV |
823 | topse = se; |
824 | if (se->on_rq) { | |
825 | incload = 0; | |
bf0f6f24 | 826 | break; |
58e2d4ca | 827 | } |
bf0f6f24 | 828 | cfs_rq = cfs_rq_of(se); |
83b699ed | 829 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 830 | wakeup = 1; |
bf0f6f24 | 831 | } |
58e2d4ca SV |
832 | /* Increment cpu load if we just enqueued the first task of a group on |
833 | * 'rq->cpu'. 'topse' represents the group to which task 'p' belongs | |
834 | * at the highest grouping level. | |
835 | */ | |
836 | if (incload) | |
837 | inc_cpu_load(rq, topse->load.weight); | |
8f4d37ec PZ |
838 | |
839 | hrtick_start_fair(rq, rq->curr); | |
bf0f6f24 IM |
840 | } |
841 | ||
842 | /* | |
843 | * The dequeue_task method is called before nr_running is | |
844 | * decreased. We remove the task from the rbtree and | |
845 | * update the fair scheduling stats: | |
846 | */ | |
f02231e5 | 847 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
848 | { |
849 | struct cfs_rq *cfs_rq; | |
58e2d4ca SV |
850 | struct sched_entity *se = &p->se, |
851 | *topse = NULL; /* Highest schedulable entity */ | |
852 | int decload = 1; | |
bf0f6f24 IM |
853 | |
854 | for_each_sched_entity(se) { | |
58e2d4ca | 855 | topse = se; |
bf0f6f24 | 856 | cfs_rq = cfs_rq_of(se); |
525c2716 | 857 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 | 858 | /* Don't dequeue parent if it has other entities besides us */ |
58e2d4ca SV |
859 | if (cfs_rq->load.weight) { |
860 | if (parent_entity(se)) | |
861 | decload = 0; | |
bf0f6f24 | 862 | break; |
58e2d4ca | 863 | } |
b9fa3df3 | 864 | sleep = 1; |
bf0f6f24 | 865 | } |
58e2d4ca SV |
866 | /* Decrement cpu load if we just dequeued the last task of a group on |
867 | * 'rq->cpu'. 'topse' represents the group to which task 'p' belongs | |
868 | * at the highest grouping level. | |
869 | */ | |
870 | if (decload) | |
871 | dec_cpu_load(rq, topse->load.weight); | |
8f4d37ec PZ |
872 | |
873 | hrtick_start_fair(rq, rq->curr); | |
bf0f6f24 IM |
874 | } |
875 | ||
876 | /* | |
1799e35d IM |
877 | * sched_yield() support is very simple - we dequeue and enqueue. |
878 | * | |
879 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 880 | */ |
4530d7ab | 881 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 882 | { |
db292ca3 IM |
883 | struct task_struct *curr = rq->curr; |
884 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
885 | struct sched_entity *rightmost, *se = &curr->se; | |
bf0f6f24 IM |
886 | |
887 | /* | |
1799e35d IM |
888 | * Are we the only task in the tree? |
889 | */ | |
890 | if (unlikely(cfs_rq->nr_running == 1)) | |
891 | return; | |
892 | ||
db292ca3 | 893 | if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) { |
1799e35d IM |
894 | __update_rq_clock(rq); |
895 | /* | |
a2a2d680 | 896 | * Update run-time statistics of the 'current'. |
1799e35d | 897 | */ |
2b1e315d | 898 | update_curr(cfs_rq); |
1799e35d IM |
899 | |
900 | return; | |
901 | } | |
902 | /* | |
903 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 904 | */ |
2b1e315d | 905 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
906 | /* |
907 | * Already in the rightmost position? | |
908 | */ | |
2b1e315d | 909 | if (unlikely(rightmost->vruntime < se->vruntime)) |
1799e35d IM |
910 | return; |
911 | ||
912 | /* | |
913 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
914 | * Upon rescheduling, sched_class::put_prev_task() will place |
915 | * 'current' within the tree based on its new key value. | |
1799e35d | 916 | */ |
30cfdcfc | 917 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
918 | } |
919 | ||
e7693a36 GH |
920 | /* |
921 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
922 | * not idle and an idle cpu is available. The span of cpus to | |
923 | * search starts with cpus closest then further out as needed, | |
924 | * so we always favor a closer, idle cpu. | |
925 | * | |
926 | * Returns the CPU we should wake onto. | |
927 | */ | |
928 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
929 | static int wake_idle(int cpu, struct task_struct *p) | |
930 | { | |
931 | cpumask_t tmp; | |
932 | struct sched_domain *sd; | |
933 | int i; | |
934 | ||
935 | /* | |
936 | * If it is idle, then it is the best cpu to run this task. | |
937 | * | |
938 | * This cpu is also the best, if it has more than one task already. | |
939 | * Siblings must be also busy(in most cases) as they didn't already | |
940 | * pickup the extra load from this cpu and hence we need not check | |
941 | * sibling runqueue info. This will avoid the checks and cache miss | |
942 | * penalities associated with that. | |
943 | */ | |
944 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
945 | return cpu; | |
946 | ||
947 | for_each_domain(cpu, sd) { | |
948 | if (sd->flags & SD_WAKE_IDLE) { | |
949 | cpus_and(tmp, sd->span, p->cpus_allowed); | |
950 | for_each_cpu_mask(i, tmp) { | |
951 | if (idle_cpu(i)) { | |
952 | if (i != task_cpu(p)) { | |
953 | schedstat_inc(p, | |
954 | se.nr_wakeups_idle); | |
955 | } | |
956 | return i; | |
957 | } | |
958 | } | |
959 | } else { | |
960 | break; | |
961 | } | |
962 | } | |
963 | return cpu; | |
964 | } | |
965 | #else | |
966 | static inline int wake_idle(int cpu, struct task_struct *p) | |
967 | { | |
968 | return cpu; | |
969 | } | |
970 | #endif | |
971 | ||
972 | #ifdef CONFIG_SMP | |
973 | static int select_task_rq_fair(struct task_struct *p, int sync) | |
974 | { | |
975 | int cpu, this_cpu; | |
976 | struct rq *rq; | |
977 | struct sched_domain *sd, *this_sd = NULL; | |
978 | int new_cpu; | |
979 | ||
980 | cpu = task_cpu(p); | |
981 | rq = task_rq(p); | |
982 | this_cpu = smp_processor_id(); | |
983 | new_cpu = cpu; | |
984 | ||
9ec3b77e DA |
985 | if (cpu == this_cpu) |
986 | goto out_set_cpu; | |
987 | ||
e7693a36 GH |
988 | for_each_domain(this_cpu, sd) { |
989 | if (cpu_isset(cpu, sd->span)) { | |
990 | this_sd = sd; | |
991 | break; | |
992 | } | |
993 | } | |
994 | ||
995 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) | |
996 | goto out_set_cpu; | |
997 | ||
998 | /* | |
999 | * Check for affine wakeup and passive balancing possibilities. | |
1000 | */ | |
1001 | if (this_sd) { | |
1002 | int idx = this_sd->wake_idx; | |
1003 | unsigned int imbalance; | |
1004 | unsigned long load, this_load; | |
1005 | ||
1006 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; | |
1007 | ||
1008 | load = source_load(cpu, idx); | |
1009 | this_load = target_load(this_cpu, idx); | |
1010 | ||
1011 | new_cpu = this_cpu; /* Wake to this CPU if we can */ | |
1012 | ||
1013 | if (this_sd->flags & SD_WAKE_AFFINE) { | |
1014 | unsigned long tl = this_load; | |
1015 | unsigned long tl_per_task; | |
1016 | ||
1017 | /* | |
1018 | * Attract cache-cold tasks on sync wakeups: | |
1019 | */ | |
1020 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1021 | goto out_set_cpu; | |
1022 | ||
1023 | schedstat_inc(p, se.nr_wakeups_affine_attempts); | |
1024 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
1025 | ||
1026 | /* | |
1027 | * If sync wakeup then subtract the (maximum possible) | |
1028 | * effect of the currently running task from the load | |
1029 | * of the current CPU: | |
1030 | */ | |
1031 | if (sync) | |
1032 | tl -= current->se.load.weight; | |
1033 | ||
1034 | if ((tl <= load && | |
1035 | tl + target_load(cpu, idx) <= tl_per_task) || | |
1036 | 100*(tl + p->se.load.weight) <= imbalance*load) { | |
1037 | /* | |
1038 | * This domain has SD_WAKE_AFFINE and | |
1039 | * p is cache cold in this domain, and | |
1040 | * there is no bad imbalance. | |
1041 | */ | |
1042 | schedstat_inc(this_sd, ttwu_move_affine); | |
1043 | schedstat_inc(p, se.nr_wakeups_affine); | |
1044 | goto out_set_cpu; | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * Start passive balancing when half the imbalance_pct | |
1050 | * limit is reached. | |
1051 | */ | |
1052 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1053 | if (imbalance*this_load <= 100*load) { | |
1054 | schedstat_inc(this_sd, ttwu_move_balance); | |
1055 | schedstat_inc(p, se.nr_wakeups_passive); | |
1056 | goto out_set_cpu; | |
1057 | } | |
1058 | } | |
1059 | } | |
1060 | ||
1061 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1062 | out_set_cpu: | |
1063 | return wake_idle(new_cpu, p); | |
1064 | } | |
1065 | #endif /* CONFIG_SMP */ | |
1066 | ||
1067 | ||
bf0f6f24 IM |
1068 | /* |
1069 | * Preempt the current task with a newly woken task if needed: | |
1070 | */ | |
2e09bf55 | 1071 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1072 | { |
1073 | struct task_struct *curr = rq->curr; | |
fad095a7 | 1074 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
8651a86c | 1075 | struct sched_entity *se = &curr->se, *pse = &p->se; |
502d26b5 | 1076 | unsigned long gran; |
bf0f6f24 IM |
1077 | |
1078 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 1079 | update_rq_clock(rq); |
b7cc0896 | 1080 | update_curr(cfs_rq); |
bf0f6f24 IM |
1081 | resched_task(curr); |
1082 | return; | |
1083 | } | |
91c234b4 IM |
1084 | /* |
1085 | * Batch tasks do not preempt (their preemption is driven by | |
1086 | * the tick): | |
1087 | */ | |
1088 | if (unlikely(p->policy == SCHED_BATCH)) | |
1089 | return; | |
bf0f6f24 | 1090 | |
77d9cc44 IM |
1091 | if (!sched_feat(WAKEUP_PREEMPT)) |
1092 | return; | |
8651a86c | 1093 | |
77d9cc44 IM |
1094 | while (!is_same_group(se, pse)) { |
1095 | se = parent_entity(se); | |
1096 | pse = parent_entity(pse); | |
ce6c1311 | 1097 | } |
77d9cc44 | 1098 | |
77d9cc44 IM |
1099 | gran = sysctl_sched_wakeup_granularity; |
1100 | if (unlikely(se->load.weight != NICE_0_LOAD)) | |
1101 | gran = calc_delta_fair(gran, &se->load); | |
1102 | ||
502d26b5 | 1103 | if (pse->vruntime + gran < se->vruntime) |
77d9cc44 | 1104 | resched_task(curr); |
bf0f6f24 IM |
1105 | } |
1106 | ||
fb8d4724 | 1107 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 | 1108 | { |
8f4d37ec | 1109 | struct task_struct *p; |
bf0f6f24 IM |
1110 | struct cfs_rq *cfs_rq = &rq->cfs; |
1111 | struct sched_entity *se; | |
1112 | ||
1113 | if (unlikely(!cfs_rq->nr_running)) | |
1114 | return NULL; | |
1115 | ||
1116 | do { | |
9948f4b2 | 1117 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
1118 | cfs_rq = group_cfs_rq(se); |
1119 | } while (cfs_rq); | |
1120 | ||
8f4d37ec PZ |
1121 | p = task_of(se); |
1122 | hrtick_start_fair(rq, p); | |
1123 | ||
1124 | return p; | |
bf0f6f24 IM |
1125 | } |
1126 | ||
1127 | /* | |
1128 | * Account for a descheduled task: | |
1129 | */ | |
31ee529c | 1130 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
1131 | { |
1132 | struct sched_entity *se = &prev->se; | |
1133 | struct cfs_rq *cfs_rq; | |
1134 | ||
1135 | for_each_sched_entity(se) { | |
1136 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 1137 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
1138 | } |
1139 | } | |
1140 | ||
681f3e68 | 1141 | #ifdef CONFIG_SMP |
bf0f6f24 IM |
1142 | /************************************************** |
1143 | * Fair scheduling class load-balancing methods: | |
1144 | */ | |
1145 | ||
1146 | /* | |
1147 | * Load-balancing iterator. Note: while the runqueue stays locked | |
1148 | * during the whole iteration, the current task might be | |
1149 | * dequeued so the iterator has to be dequeue-safe. Here we | |
1150 | * achieve that by always pre-iterating before returning | |
1151 | * the current task: | |
1152 | */ | |
a9957449 | 1153 | static struct task_struct * |
bf0f6f24 IM |
1154 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) |
1155 | { | |
1156 | struct task_struct *p; | |
1157 | ||
1158 | if (!curr) | |
1159 | return NULL; | |
1160 | ||
1161 | p = rb_entry(curr, struct task_struct, se.run_node); | |
1162 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
1163 | ||
1164 | return p; | |
1165 | } | |
1166 | ||
1167 | static struct task_struct *load_balance_start_fair(void *arg) | |
1168 | { | |
1169 | struct cfs_rq *cfs_rq = arg; | |
1170 | ||
1171 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
1172 | } | |
1173 | ||
1174 | static struct task_struct *load_balance_next_fair(void *arg) | |
1175 | { | |
1176 | struct cfs_rq *cfs_rq = arg; | |
1177 | ||
1178 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
1179 | } | |
1180 | ||
43010659 | 1181 | static unsigned long |
bf0f6f24 | 1182 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f | 1183 | unsigned long max_load_move, |
a4ac01c3 PW |
1184 | struct sched_domain *sd, enum cpu_idle_type idle, |
1185 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
1186 | { |
1187 | struct cfs_rq *busy_cfs_rq; | |
bf0f6f24 IM |
1188 | long rem_load_move = max_load_move; |
1189 | struct rq_iterator cfs_rq_iterator; | |
6b2d7700 | 1190 | unsigned long load_moved; |
bf0f6f24 IM |
1191 | |
1192 | cfs_rq_iterator.start = load_balance_start_fair; | |
1193 | cfs_rq_iterator.next = load_balance_next_fair; | |
1194 | ||
1195 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 1196 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6b2d7700 SV |
1197 | struct cfs_rq *this_cfs_rq = busy_cfs_rq->tg->cfs_rq[this_cpu]; |
1198 | unsigned long maxload, task_load, group_weight; | |
1199 | unsigned long thisload, per_task_load; | |
1200 | struct sched_entity *se = busy_cfs_rq->tg->se[busiest->cpu]; | |
bf0f6f24 | 1201 | |
6b2d7700 SV |
1202 | task_load = busy_cfs_rq->load.weight; |
1203 | group_weight = se->load.weight; | |
bf0f6f24 | 1204 | |
6b2d7700 SV |
1205 | /* |
1206 | * 'group_weight' is contributed by tasks of total weight | |
1207 | * 'task_load'. To move 'rem_load_move' worth of weight only, | |
1208 | * we need to move a maximum task load of: | |
1209 | * | |
1210 | * maxload = (remload / group_weight) * task_load; | |
1211 | */ | |
1212 | maxload = (rem_load_move * task_load) / group_weight; | |
1213 | ||
1214 | if (!maxload || !task_load) | |
bf0f6f24 IM |
1215 | continue; |
1216 | ||
6b2d7700 SV |
1217 | per_task_load = task_load / busy_cfs_rq->nr_running; |
1218 | /* | |
1219 | * balance_tasks will try to forcibly move atleast one task if | |
1220 | * possible (because of SCHED_LOAD_SCALE_FUZZ). Avoid that if | |
1221 | * maxload is less than GROUP_IMBALANCE_FUZZ% the per_task_load. | |
1222 | */ | |
1223 | if (100 * maxload < GROUP_IMBALANCE_PCT * per_task_load) | |
1224 | continue; | |
bf0f6f24 | 1225 | |
6b2d7700 SV |
1226 | /* Disable priority-based load balance */ |
1227 | *this_best_prio = 0; | |
1228 | thisload = this_cfs_rq->load.weight; | |
a4ac01c3 | 1229 | #else |
e56f31aa | 1230 | # define maxload rem_load_move |
a4ac01c3 | 1231 | #endif |
e1d1484f PW |
1232 | /* |
1233 | * pass busy_cfs_rq argument into | |
bf0f6f24 IM |
1234 | * load_balance_[start|next]_fair iterators |
1235 | */ | |
1236 | cfs_rq_iterator.arg = busy_cfs_rq; | |
6b2d7700 | 1237 | load_moved = balance_tasks(this_rq, this_cpu, busiest, |
e1d1484f PW |
1238 | maxload, sd, idle, all_pinned, |
1239 | this_best_prio, | |
1240 | &cfs_rq_iterator); | |
bf0f6f24 | 1241 | |
6b2d7700 SV |
1242 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1243 | /* | |
1244 | * load_moved holds the task load that was moved. The | |
1245 | * effective (group) weight moved would be: | |
1246 | * load_moved_eff = load_moved/task_load * group_weight; | |
1247 | */ | |
1248 | load_moved = (group_weight * load_moved) / task_load; | |
1249 | ||
1250 | /* Adjust shares on both cpus to reflect load_moved */ | |
1251 | group_weight -= load_moved; | |
1252 | set_se_shares(se, group_weight); | |
1253 | ||
1254 | se = busy_cfs_rq->tg->se[this_cpu]; | |
1255 | if (!thisload) | |
1256 | group_weight = load_moved; | |
1257 | else | |
1258 | group_weight = se->load.weight + load_moved; | |
1259 | set_se_shares(se, group_weight); | |
1260 | #endif | |
1261 | ||
1262 | rem_load_move -= load_moved; | |
1263 | ||
e1d1484f | 1264 | if (rem_load_move <= 0) |
bf0f6f24 IM |
1265 | break; |
1266 | } | |
1267 | ||
43010659 | 1268 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
1269 | } |
1270 | ||
e1d1484f PW |
1271 | static int |
1272 | move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1273 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1274 | { | |
1275 | struct cfs_rq *busy_cfs_rq; | |
1276 | struct rq_iterator cfs_rq_iterator; | |
1277 | ||
1278 | cfs_rq_iterator.start = load_balance_start_fair; | |
1279 | cfs_rq_iterator.next = load_balance_next_fair; | |
1280 | ||
1281 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
1282 | /* | |
1283 | * pass busy_cfs_rq argument into | |
1284 | * load_balance_[start|next]_fair iterators | |
1285 | */ | |
1286 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1287 | if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle, | |
1288 | &cfs_rq_iterator)) | |
1289 | return 1; | |
1290 | } | |
1291 | ||
1292 | return 0; | |
1293 | } | |
681f3e68 | 1294 | #endif |
e1d1484f | 1295 | |
bf0f6f24 IM |
1296 | /* |
1297 | * scheduler tick hitting a task of our scheduling class: | |
1298 | */ | |
8f4d37ec | 1299 | static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) |
bf0f6f24 IM |
1300 | { |
1301 | struct cfs_rq *cfs_rq; | |
1302 | struct sched_entity *se = &curr->se; | |
1303 | ||
1304 | for_each_sched_entity(se) { | |
1305 | cfs_rq = cfs_rq_of(se); | |
8f4d37ec | 1306 | entity_tick(cfs_rq, se, queued); |
bf0f6f24 IM |
1307 | } |
1308 | } | |
1309 | ||
8eb172d9 | 1310 | #define swap(a, b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
4d78e7b6 | 1311 | |
bf0f6f24 IM |
1312 | /* |
1313 | * Share the fairness runtime between parent and child, thus the | |
1314 | * total amount of pressure for CPU stays equal - new tasks | |
1315 | * get a chance to run but frequent forkers are not allowed to | |
1316 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1317 | * the child is not running yet. | |
1318 | */ | |
ee0827d8 | 1319 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1320 | { |
1321 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1322 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
00bf7bfc | 1323 | int this_cpu = smp_processor_id(); |
bf0f6f24 IM |
1324 | |
1325 | sched_info_queued(p); | |
1326 | ||
7109c442 | 1327 | update_curr(cfs_rq); |
aeb73b04 | 1328 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1329 | |
3c90e6e9 | 1330 | /* 'curr' will be NULL if the child belongs to a different group */ |
00bf7bfc | 1331 | if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) && |
3c90e6e9 | 1332 | curr && curr->vruntime < se->vruntime) { |
87fefa38 | 1333 | /* |
edcb60a3 IM |
1334 | * Upon rescheduling, sched_class::put_prev_task() will place |
1335 | * 'current' within the tree based on its new key value. | |
1336 | */ | |
4d78e7b6 | 1337 | swap(curr->vruntime, se->vruntime); |
4d78e7b6 | 1338 | } |
bf0f6f24 | 1339 | |
b9dca1e0 | 1340 | enqueue_task_fair(rq, p, 0); |
bb61c210 | 1341 | resched_task(rq->curr); |
bf0f6f24 IM |
1342 | } |
1343 | ||
cb469845 SR |
1344 | /* |
1345 | * Priority of the task has changed. Check to see if we preempt | |
1346 | * the current task. | |
1347 | */ | |
1348 | static void prio_changed_fair(struct rq *rq, struct task_struct *p, | |
1349 | int oldprio, int running) | |
1350 | { | |
1351 | /* | |
1352 | * Reschedule if we are currently running on this runqueue and | |
1353 | * our priority decreased, or if we are not currently running on | |
1354 | * this runqueue and our priority is higher than the current's | |
1355 | */ | |
1356 | if (running) { | |
1357 | if (p->prio > oldprio) | |
1358 | resched_task(rq->curr); | |
1359 | } else | |
1360 | check_preempt_curr(rq, p); | |
1361 | } | |
1362 | ||
1363 | /* | |
1364 | * We switched to the sched_fair class. | |
1365 | */ | |
1366 | static void switched_to_fair(struct rq *rq, struct task_struct *p, | |
1367 | int running) | |
1368 | { | |
1369 | /* | |
1370 | * We were most likely switched from sched_rt, so | |
1371 | * kick off the schedule if running, otherwise just see | |
1372 | * if we can still preempt the current task. | |
1373 | */ | |
1374 | if (running) | |
1375 | resched_task(rq->curr); | |
1376 | else | |
1377 | check_preempt_curr(rq, p); | |
1378 | } | |
1379 | ||
83b699ed SV |
1380 | /* Account for a task changing its policy or group. |
1381 | * | |
1382 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1383 | * migrates between groups/classes. | |
1384 | */ | |
1385 | static void set_curr_task_fair(struct rq *rq) | |
1386 | { | |
1387 | struct sched_entity *se = &rq->curr->se; | |
1388 | ||
1389 | for_each_sched_entity(se) | |
1390 | set_next_entity(cfs_rq_of(se), se); | |
1391 | } | |
1392 | ||
bf0f6f24 IM |
1393 | /* |
1394 | * All the scheduling class methods: | |
1395 | */ | |
5522d5d5 IM |
1396 | static const struct sched_class fair_sched_class = { |
1397 | .next = &idle_sched_class, | |
bf0f6f24 IM |
1398 | .enqueue_task = enqueue_task_fair, |
1399 | .dequeue_task = dequeue_task_fair, | |
1400 | .yield_task = yield_task_fair, | |
e7693a36 GH |
1401 | #ifdef CONFIG_SMP |
1402 | .select_task_rq = select_task_rq_fair, | |
1403 | #endif /* CONFIG_SMP */ | |
bf0f6f24 | 1404 | |
2e09bf55 | 1405 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1406 | |
1407 | .pick_next_task = pick_next_task_fair, | |
1408 | .put_prev_task = put_prev_task_fair, | |
1409 | ||
681f3e68 | 1410 | #ifdef CONFIG_SMP |
bf0f6f24 | 1411 | .load_balance = load_balance_fair, |
e1d1484f | 1412 | .move_one_task = move_one_task_fair, |
681f3e68 | 1413 | #endif |
bf0f6f24 | 1414 | |
83b699ed | 1415 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 IM |
1416 | .task_tick = task_tick_fair, |
1417 | .task_new = task_new_fair, | |
cb469845 SR |
1418 | |
1419 | .prio_changed = prio_changed_fair, | |
1420 | .switched_to = switched_to_fair, | |
bf0f6f24 IM |
1421 | }; |
1422 | ||
1423 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1424 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1425 | { |
bf0f6f24 IM |
1426 | struct cfs_rq *cfs_rq; |
1427 | ||
75c28ace SV |
1428 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1429 | print_cfs_rq(m, cpu, &cpu_rq(cpu)->cfs); | |
1430 | #endif | |
ec2c507f | 1431 | lock_task_group_list(); |
c3b64f1e | 1432 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1433 | print_cfs_rq(m, cpu, cfs_rq); |
ec2c507f | 1434 | unlock_task_group_list(); |
bf0f6f24 IM |
1435 | } |
1436 | #endif |