]>
Commit | Line | Data |
---|---|---|
bf0f6f24 IM |
1 | /* |
2 | * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH) | |
3 | * | |
4 | * Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
5 | * | |
6 | * Interactivity improvements by Mike Galbraith | |
7 | * (C) 2007 Mike Galbraith <efault@gmx.de> | |
8 | * | |
9 | * Various enhancements by Dmitry Adamushko. | |
10 | * (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com> | |
11 | * | |
12 | * Group scheduling enhancements by Srivatsa Vaddagiri | |
13 | * Copyright IBM Corporation, 2007 | |
14 | * Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> | |
15 | * | |
16 | * Scaled math optimizations by Thomas Gleixner | |
17 | * Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de> | |
21805085 PZ |
18 | * |
19 | * Adaptive scheduling granularity, math enhancements by Peter Zijlstra | |
20 | * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> | |
bf0f6f24 IM |
21 | */ |
22 | ||
23 | /* | |
21805085 PZ |
24 | * Targeted preemption latency for CPU-bound tasks: |
25 | * (default: 20ms, units: nanoseconds) | |
bf0f6f24 | 26 | * |
21805085 PZ |
27 | * NOTE: this latency value is not the same as the concept of |
28 | * 'timeslice length' - timeslices in CFS are of variable length. | |
29 | * (to see the precise effective timeslice length of your workload, | |
30 | * run vmstat and monitor the context-switches field) | |
bf0f6f24 IM |
31 | * |
32 | * On SMP systems the value of this is multiplied by the log2 of the | |
33 | * number of CPUs. (i.e. factor 2x on 2-way systems, 3x on 4-way | |
34 | * systems, 4x on 8-way systems, 5x on 16-way systems, etc.) | |
21805085 | 35 | * Targeted preemption latency for CPU-bound tasks: |
bf0f6f24 | 36 | */ |
2bd8e6d4 IM |
37 | const_debug unsigned int sysctl_sched_latency = 20000000ULL; |
38 | ||
39 | /* | |
40 | * After fork, child runs first. (default) If set to 0 then | |
41 | * parent will (try to) run first. | |
42 | */ | |
43 | const_debug unsigned int sysctl_sched_child_runs_first = 1; | |
21805085 PZ |
44 | |
45 | /* | |
46 | * Minimal preemption granularity for CPU-bound tasks: | |
47 | * (default: 2 msec, units: nanoseconds) | |
48 | */ | |
172ac3db | 49 | unsigned int sysctl_sched_min_granularity __read_mostly = 2000000ULL; |
bf0f6f24 | 50 | |
1799e35d IM |
51 | /* |
52 | * sys_sched_yield() compat mode | |
53 | * | |
54 | * This option switches the agressive yield implementation of the | |
55 | * old scheduler back on. | |
56 | */ | |
57 | unsigned int __read_mostly sysctl_sched_compat_yield; | |
58 | ||
bf0f6f24 IM |
59 | /* |
60 | * SCHED_BATCH wake-up granularity. | |
71fd3714 | 61 | * (default: 25 msec, units: nanoseconds) |
bf0f6f24 IM |
62 | * |
63 | * This option delays the preemption effects of decoupled workloads | |
64 | * and reduces their over-scheduling. Synchronous workloads will still | |
65 | * have immediate wakeup/sleep latencies. | |
66 | */ | |
2bd8e6d4 | 67 | const_debug unsigned int sysctl_sched_batch_wakeup_granularity = 25000000UL; |
bf0f6f24 IM |
68 | |
69 | /* | |
70 | * SCHED_OTHER wake-up granularity. | |
71 | * (default: 1 msec, units: nanoseconds) | |
72 | * | |
73 | * This option delays the preemption effects of decoupled workloads | |
74 | * and reduces their over-scheduling. Synchronous workloads will still | |
75 | * have immediate wakeup/sleep latencies. | |
76 | */ | |
2e09bf55 | 77 | const_debug unsigned int sysctl_sched_wakeup_granularity = 2000000UL; |
bf0f6f24 | 78 | |
bf0f6f24 IM |
79 | unsigned int sysctl_sched_runtime_limit __read_mostly; |
80 | ||
bf0f6f24 IM |
81 | extern struct sched_class fair_sched_class; |
82 | ||
83 | /************************************************************** | |
84 | * CFS operations on generic schedulable entities: | |
85 | */ | |
86 | ||
62160e3f | 87 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 88 | |
62160e3f | 89 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
90 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
91 | { | |
62160e3f | 92 | return cfs_rq->rq; |
bf0f6f24 IM |
93 | } |
94 | ||
62160e3f IM |
95 | /* An entity is a task if it doesn't "own" a runqueue */ |
96 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 97 | |
62160e3f | 98 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
bf0f6f24 | 99 | |
62160e3f IM |
100 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
101 | { | |
102 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
103 | } |
104 | ||
105 | #define entity_is_task(se) 1 | |
106 | ||
bf0f6f24 IM |
107 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
108 | ||
109 | static inline struct task_struct *task_of(struct sched_entity *se) | |
110 | { | |
111 | return container_of(se, struct task_struct, se); | |
112 | } | |
113 | ||
114 | ||
115 | /************************************************************** | |
116 | * Scheduling class tree data structure manipulation methods: | |
117 | */ | |
118 | ||
02e0431a PZ |
119 | static inline u64 |
120 | max_vruntime(u64 min_vruntime, u64 vruntime) | |
121 | { | |
122 | if ((vruntime > min_vruntime) || | |
123 | (min_vruntime > (1ULL << 61) && vruntime < (1ULL << 50))) | |
124 | min_vruntime = vruntime; | |
125 | ||
126 | return min_vruntime; | |
127 | } | |
128 | ||
e9acbff6 IM |
129 | static inline void |
130 | set_leftmost(struct cfs_rq *cfs_rq, struct rb_node *leftmost) | |
131 | { | |
132 | struct sched_entity *se; | |
133 | ||
134 | cfs_rq->rb_leftmost = leftmost; | |
02e0431a | 135 | if (leftmost) |
e9acbff6 | 136 | se = rb_entry(leftmost, struct sched_entity, run_node); |
e9acbff6 IM |
137 | } |
138 | ||
02e0431a PZ |
139 | static inline s64 |
140 | entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
9014623c PZ |
141 | { |
142 | return se->fair_key - cfs_rq->min_vruntime; | |
143 | } | |
144 | ||
bf0f6f24 IM |
145 | /* |
146 | * Enqueue an entity into the rb-tree: | |
147 | */ | |
19ccd97a | 148 | static void |
bf0f6f24 IM |
149 | __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
150 | { | |
151 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
152 | struct rb_node *parent = NULL; | |
153 | struct sched_entity *entry; | |
9014623c | 154 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
155 | int leftmost = 1; |
156 | ||
157 | /* | |
158 | * Find the right place in the rbtree: | |
159 | */ | |
160 | while (*link) { | |
161 | parent = *link; | |
162 | entry = rb_entry(parent, struct sched_entity, run_node); | |
163 | /* | |
164 | * We dont care about collisions. Nodes with | |
165 | * the same key stay together. | |
166 | */ | |
9014623c | 167 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
168 | link = &parent->rb_left; |
169 | } else { | |
170 | link = &parent->rb_right; | |
171 | leftmost = 0; | |
172 | } | |
173 | } | |
174 | ||
175 | /* | |
176 | * Maintain a cache of leftmost tree entries (it is frequently | |
177 | * used): | |
178 | */ | |
179 | if (leftmost) | |
e9acbff6 | 180 | set_leftmost(cfs_rq, &se->run_node); |
bf0f6f24 IM |
181 | |
182 | rb_link_node(&se->run_node, parent, link); | |
183 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
184 | update_load_add(&cfs_rq->load, se->load.weight); | |
185 | cfs_rq->nr_running++; | |
186 | se->on_rq = 1; | |
187 | } | |
188 | ||
19ccd97a | 189 | static void |
bf0f6f24 IM |
190 | __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
191 | { | |
192 | if (cfs_rq->rb_leftmost == &se->run_node) | |
e9acbff6 IM |
193 | set_leftmost(cfs_rq, rb_next(&se->run_node)); |
194 | ||
bf0f6f24 IM |
195 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
196 | update_load_sub(&cfs_rq->load, se->load.weight); | |
197 | cfs_rq->nr_running--; | |
198 | se->on_rq = 0; | |
199 | } | |
200 | ||
201 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
202 | { | |
203 | return cfs_rq->rb_leftmost; | |
204 | } | |
205 | ||
206 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
207 | { | |
208 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
209 | } | |
210 | ||
aeb73b04 PZ |
211 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
212 | { | |
213 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
214 | struct sched_entity *se = NULL; | |
215 | struct rb_node *parent; | |
216 | ||
217 | while (*link) { | |
218 | parent = *link; | |
219 | se = rb_entry(parent, struct sched_entity, run_node); | |
220 | link = &parent->rb_right; | |
221 | } | |
222 | ||
223 | return se; | |
224 | } | |
225 | ||
bf0f6f24 IM |
226 | /************************************************************** |
227 | * Scheduling class statistics methods: | |
228 | */ | |
229 | ||
4d78e7b6 PZ |
230 | static u64 __sched_period(unsigned long nr_running) |
231 | { | |
232 | u64 period = sysctl_sched_latency; | |
233 | unsigned long nr_latency = | |
234 | sysctl_sched_latency / sysctl_sched_min_granularity; | |
235 | ||
236 | if (unlikely(nr_running > nr_latency)) { | |
237 | period *= nr_running; | |
238 | do_div(period, nr_latency); | |
239 | } | |
240 | ||
241 | return period; | |
242 | } | |
243 | ||
6d0f0ebd | 244 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 245 | { |
6d0f0ebd | 246 | u64 period = __sched_period(cfs_rq->nr_running); |
21805085 | 247 | |
6d0f0ebd PZ |
248 | period *= se->load.weight; |
249 | do_div(period, cfs_rq->load.weight); | |
21805085 | 250 | |
6d0f0ebd | 251 | return period; |
bf0f6f24 IM |
252 | } |
253 | ||
bf0f6f24 IM |
254 | /* |
255 | * Update the current task's runtime statistics. Skip current tasks that | |
256 | * are not in our scheduling class. | |
257 | */ | |
258 | static inline void | |
8ebc91d9 IM |
259 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
260 | unsigned long delta_exec) | |
bf0f6f24 | 261 | { |
bbdba7c0 | 262 | unsigned long delta_exec_weighted; |
02e0431a | 263 | u64 next_vruntime, min_vruntime; |
bf0f6f24 | 264 | |
8179ca23 | 265 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
266 | |
267 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 268 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
e9acbff6 IM |
269 | delta_exec_weighted = delta_exec; |
270 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { | |
271 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, | |
272 | &curr->load); | |
273 | } | |
274 | curr->vruntime += delta_exec_weighted; | |
02e0431a PZ |
275 | |
276 | /* | |
277 | * maintain cfs_rq->min_vruntime to be a monotonic increasing | |
278 | * value tracking the leftmost vruntime in the tree. | |
279 | */ | |
280 | if (first_fair(cfs_rq)) { | |
281 | next_vruntime = __pick_next_entity(cfs_rq)->vruntime; | |
282 | ||
283 | /* min_vruntime() := !max_vruntime() */ | |
284 | min_vruntime = max_vruntime(curr->vruntime, next_vruntime); | |
285 | if (min_vruntime == next_vruntime) | |
286 | min_vruntime = curr->vruntime; | |
287 | else | |
288 | min_vruntime = next_vruntime; | |
289 | } else | |
290 | min_vruntime = curr->vruntime; | |
291 | ||
292 | cfs_rq->min_vruntime = | |
293 | max_vruntime(cfs_rq->min_vruntime, min_vruntime); | |
bf0f6f24 IM |
294 | } |
295 | ||
b7cc0896 | 296 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 297 | { |
429d43bc | 298 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 299 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
300 | unsigned long delta_exec; |
301 | ||
302 | if (unlikely(!curr)) | |
303 | return; | |
304 | ||
305 | /* | |
306 | * Get the amount of time the current task was running | |
307 | * since the last time we changed load (this cannot | |
308 | * overflow on 32 bits): | |
309 | */ | |
8ebc91d9 | 310 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 311 | |
8ebc91d9 IM |
312 | __update_curr(cfs_rq, curr, delta_exec); |
313 | curr->exec_start = now; | |
bf0f6f24 IM |
314 | } |
315 | ||
316 | static inline void | |
5870db5b | 317 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 318 | { |
d281918d | 319 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
320 | } |
321 | ||
bf0f6f24 | 322 | static inline unsigned long |
08e2388a | 323 | calc_weighted(unsigned long delta, struct sched_entity *se) |
bf0f6f24 | 324 | { |
08e2388a | 325 | unsigned long weight = se->load.weight; |
bf0f6f24 | 326 | |
08e2388a IM |
327 | if (unlikely(weight != NICE_0_LOAD)) |
328 | return (u64)delta * se->load.weight >> NICE_0_SHIFT; | |
329 | else | |
330 | return delta; | |
bf0f6f24 | 331 | } |
bf0f6f24 IM |
332 | |
333 | /* | |
334 | * Task is being enqueued - update stats: | |
335 | */ | |
d2417e5a | 336 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 337 | { |
bf0f6f24 IM |
338 | /* |
339 | * Are we enqueueing a waiting task? (for current tasks | |
340 | * a dequeue/enqueue event is a NOP) | |
341 | */ | |
429d43bc | 342 | if (se != cfs_rq->curr) |
5870db5b | 343 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
344 | /* |
345 | * Update the key: | |
346 | */ | |
e9acbff6 | 347 | se->fair_key = se->vruntime; |
bf0f6f24 IM |
348 | } |
349 | ||
bf0f6f24 | 350 | static void |
9ef0a961 | 351 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 352 | { |
bbdba7c0 IM |
353 | schedstat_set(se->wait_max, max(se->wait_max, |
354 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6cfb0d5d | 355 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
356 | } |
357 | ||
358 | static inline void | |
19b6a2e3 | 359 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 360 | { |
b7cc0896 | 361 | update_curr(cfs_rq); |
bf0f6f24 IM |
362 | /* |
363 | * Mark the end of the wait period if dequeueing a | |
364 | * waiting task: | |
365 | */ | |
429d43bc | 366 | if (se != cfs_rq->curr) |
9ef0a961 | 367 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
368 | } |
369 | ||
370 | /* | |
371 | * We are picking a new current task - update its stats: | |
372 | */ | |
373 | static inline void | |
79303e9e | 374 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
375 | { |
376 | /* | |
377 | * We are starting a new run period: | |
378 | */ | |
d281918d | 379 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
380 | } |
381 | ||
382 | /* | |
383 | * We are descheduling a task - update its stats: | |
384 | */ | |
385 | static inline void | |
c7e9b5b2 | 386 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
387 | { |
388 | se->exec_start = 0; | |
389 | } | |
390 | ||
391 | /************************************************** | |
392 | * Scheduling class queueing methods: | |
393 | */ | |
394 | ||
2396af69 | 395 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 396 | { |
bf0f6f24 IM |
397 | #ifdef CONFIG_SCHEDSTATS |
398 | if (se->sleep_start) { | |
d281918d | 399 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
400 | |
401 | if ((s64)delta < 0) | |
402 | delta = 0; | |
403 | ||
404 | if (unlikely(delta > se->sleep_max)) | |
405 | se->sleep_max = delta; | |
406 | ||
407 | se->sleep_start = 0; | |
408 | se->sum_sleep_runtime += delta; | |
409 | } | |
410 | if (se->block_start) { | |
d281918d | 411 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
412 | |
413 | if ((s64)delta < 0) | |
414 | delta = 0; | |
415 | ||
416 | if (unlikely(delta > se->block_max)) | |
417 | se->block_max = delta; | |
418 | ||
419 | se->block_start = 0; | |
420 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
421 | |
422 | /* | |
423 | * Blocking time is in units of nanosecs, so shift by 20 to | |
424 | * get a milliseconds-range estimation of the amount of | |
425 | * time that the task spent sleeping: | |
426 | */ | |
427 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf IM |
428 | struct task_struct *tsk = task_of(se); |
429 | ||
30084fbd IM |
430 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
431 | delta >> 20); | |
432 | } | |
bf0f6f24 IM |
433 | } |
434 | #endif | |
435 | } | |
436 | ||
aeb73b04 PZ |
437 | static void |
438 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
439 | { | |
aeb73b04 PZ |
440 | u64 min_runtime, latency; |
441 | ||
442 | min_runtime = cfs_rq->min_vruntime; | |
94dfb5e7 PZ |
443 | |
444 | if (sched_feat(USE_TREE_AVG)) { | |
445 | struct sched_entity *last = __pick_last_entity(cfs_rq); | |
446 | if (last) { | |
447 | min_runtime = __pick_next_entity(cfs_rq)->vruntime; | |
448 | min_runtime += last->vruntime; | |
449 | min_runtime >>= 1; | |
450 | } | |
451 | } else if (sched_feat(APPROX_AVG)) | |
452 | min_runtime += sysctl_sched_latency/2; | |
453 | ||
454 | if (initial && sched_feat(START_DEBIT)) | |
455 | min_runtime += sched_slice(cfs_rq, se); | |
aeb73b04 PZ |
456 | |
457 | if (!initial && sched_feat(NEW_FAIR_SLEEPERS)) { | |
458 | latency = sysctl_sched_latency; | |
459 | if (min_runtime > latency) | |
460 | min_runtime -= latency; | |
461 | else | |
462 | min_runtime = 0; | |
463 | } | |
464 | ||
465 | se->vruntime = max(se->vruntime, min_runtime); | |
466 | } | |
467 | ||
bf0f6f24 | 468 | static void |
668031ca | 469 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
470 | { |
471 | /* | |
472 | * Update the fair clock. | |
473 | */ | |
b7cc0896 | 474 | update_curr(cfs_rq); |
bf0f6f24 | 475 | |
e9acbff6 | 476 | if (wakeup) { |
aeb73b04 | 477 | place_entity(cfs_rq, se, 0); |
2396af69 | 478 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 479 | } |
bf0f6f24 | 480 | |
d2417e5a | 481 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 IM |
482 | __enqueue_entity(cfs_rq, se); |
483 | } | |
484 | ||
485 | static void | |
525c2716 | 486 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 487 | { |
19b6a2e3 | 488 | update_stats_dequeue(cfs_rq, se); |
bf0f6f24 | 489 | #ifdef CONFIG_SCHEDSTATS |
db36cc7d | 490 | if (sleep) { |
bf0f6f24 IM |
491 | if (entity_is_task(se)) { |
492 | struct task_struct *tsk = task_of(se); | |
493 | ||
494 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 495 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 496 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 497 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 498 | } |
bf0f6f24 | 499 | } |
db36cc7d | 500 | #endif |
bf0f6f24 IM |
501 | __dequeue_entity(cfs_rq, se); |
502 | } | |
503 | ||
504 | /* | |
505 | * Preempt the current task with a newly woken task if needed: | |
506 | */ | |
7c92e54f | 507 | static void |
2e09bf55 | 508 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 509 | { |
11697830 PZ |
510 | unsigned long ideal_runtime, delta_exec; |
511 | ||
6d0f0ebd | 512 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 PZ |
513 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
514 | if (delta_exec > ideal_runtime) | |
bf0f6f24 IM |
515 | resched_task(rq_of(cfs_rq)->curr); |
516 | } | |
517 | ||
518 | static inline void | |
8494f412 | 519 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
520 | { |
521 | /* | |
522 | * Any task has to be enqueued before it get to execute on | |
523 | * a CPU. So account for the time it spent waiting on the | |
bbdba7c0 | 524 | * runqueue. |
bf0f6f24 | 525 | */ |
9ef0a961 | 526 | update_stats_wait_end(cfs_rq, se); |
79303e9e | 527 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 528 | cfs_rq->curr = se; |
eba1ed4b IM |
529 | #ifdef CONFIG_SCHEDSTATS |
530 | /* | |
531 | * Track our maximum slice length, if the CPU's load is at | |
532 | * least twice that of our own weight (i.e. dont track it | |
533 | * when there are only lesser-weight tasks around): | |
534 | */ | |
495eca49 | 535 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
536 | se->slice_max = max(se->slice_max, |
537 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
538 | } | |
539 | #endif | |
4a55b450 | 540 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
541 | } |
542 | ||
9948f4b2 | 543 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 IM |
544 | { |
545 | struct sched_entity *se = __pick_next_entity(cfs_rq); | |
546 | ||
8494f412 | 547 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
548 | |
549 | return se; | |
550 | } | |
551 | ||
ab6cde26 | 552 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
553 | { |
554 | /* | |
555 | * If still on the runqueue then deactivate_task() | |
556 | * was not called and update_curr() has to be done: | |
557 | */ | |
558 | if (prev->on_rq) | |
b7cc0896 | 559 | update_curr(cfs_rq); |
bf0f6f24 | 560 | |
c7e9b5b2 | 561 | update_stats_curr_end(cfs_rq, prev); |
bf0f6f24 IM |
562 | |
563 | if (prev->on_rq) | |
5870db5b | 564 | update_stats_wait_start(cfs_rq, prev); |
429d43bc | 565 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
566 | } |
567 | ||
568 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
569 | { | |
bf0f6f24 IM |
570 | /* |
571 | * Dequeue and enqueue the task to update its | |
572 | * position within the tree: | |
573 | */ | |
525c2716 | 574 | dequeue_entity(cfs_rq, curr, 0); |
668031ca | 575 | enqueue_entity(cfs_rq, curr, 0); |
bf0f6f24 | 576 | |
2e09bf55 IM |
577 | if (cfs_rq->nr_running > 1) |
578 | check_preempt_tick(cfs_rq, curr); | |
bf0f6f24 IM |
579 | } |
580 | ||
581 | /************************************************** | |
582 | * CFS operations on tasks: | |
583 | */ | |
584 | ||
585 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
586 | ||
587 | /* Walk up scheduling entities hierarchy */ | |
588 | #define for_each_sched_entity(se) \ | |
589 | for (; se; se = se->parent) | |
590 | ||
591 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
592 | { | |
593 | return p->se.cfs_rq; | |
594 | } | |
595 | ||
596 | /* runqueue on which this entity is (to be) queued */ | |
597 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
598 | { | |
599 | return se->cfs_rq; | |
600 | } | |
601 | ||
602 | /* runqueue "owned" by this group */ | |
603 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
604 | { | |
605 | return grp->my_q; | |
606 | } | |
607 | ||
608 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
609 | * another cpu ('this_cpu') | |
610 | */ | |
611 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
612 | { | |
613 | /* A later patch will take group into account */ | |
614 | return &cpu_rq(this_cpu)->cfs; | |
615 | } | |
616 | ||
617 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
618 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
619 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
620 | ||
621 | /* Do the two (enqueued) tasks belong to the same group ? */ | |
622 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
623 | { | |
624 | if (curr->se.cfs_rq == p->se.cfs_rq) | |
625 | return 1; | |
626 | ||
627 | return 0; | |
628 | } | |
629 | ||
630 | #else /* CONFIG_FAIR_GROUP_SCHED */ | |
631 | ||
632 | #define for_each_sched_entity(se) \ | |
633 | for (; se; se = NULL) | |
634 | ||
635 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
636 | { | |
637 | return &task_rq(p)->cfs; | |
638 | } | |
639 | ||
640 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
641 | { | |
642 | struct task_struct *p = task_of(se); | |
643 | struct rq *rq = task_rq(p); | |
644 | ||
645 | return &rq->cfs; | |
646 | } | |
647 | ||
648 | /* runqueue "owned" by this group */ | |
649 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
650 | { | |
651 | return NULL; | |
652 | } | |
653 | ||
654 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
655 | { | |
656 | return &cpu_rq(this_cpu)->cfs; | |
657 | } | |
658 | ||
659 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
660 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
661 | ||
662 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
663 | { | |
664 | return 1; | |
665 | } | |
666 | ||
667 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
668 | ||
669 | /* | |
670 | * The enqueue_task method is called before nr_running is | |
671 | * increased. Here we update the fair scheduling stats and | |
672 | * then put the task into the rbtree: | |
673 | */ | |
fd390f6a | 674 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
675 | { |
676 | struct cfs_rq *cfs_rq; | |
677 | struct sched_entity *se = &p->se; | |
678 | ||
679 | for_each_sched_entity(se) { | |
680 | if (se->on_rq) | |
681 | break; | |
682 | cfs_rq = cfs_rq_of(se); | |
668031ca | 683 | enqueue_entity(cfs_rq, se, wakeup); |
bf0f6f24 IM |
684 | } |
685 | } | |
686 | ||
687 | /* | |
688 | * The dequeue_task method is called before nr_running is | |
689 | * decreased. We remove the task from the rbtree and | |
690 | * update the fair scheduling stats: | |
691 | */ | |
f02231e5 | 692 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
693 | { |
694 | struct cfs_rq *cfs_rq; | |
695 | struct sched_entity *se = &p->se; | |
696 | ||
697 | for_each_sched_entity(se) { | |
698 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 699 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
700 | /* Don't dequeue parent if it has other entities besides us */ |
701 | if (cfs_rq->load.weight) | |
702 | break; | |
703 | } | |
704 | } | |
705 | ||
706 | /* | |
1799e35d IM |
707 | * sched_yield() support is very simple - we dequeue and enqueue. |
708 | * | |
709 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 IM |
710 | */ |
711 | static void yield_task_fair(struct rq *rq, struct task_struct *p) | |
712 | { | |
713 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1799e35d IM |
714 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
715 | struct sched_entity *rightmost, *se = &p->se; | |
716 | struct rb_node *parent; | |
bf0f6f24 IM |
717 | |
718 | /* | |
1799e35d IM |
719 | * Are we the only task in the tree? |
720 | */ | |
721 | if (unlikely(cfs_rq->nr_running == 1)) | |
722 | return; | |
723 | ||
724 | if (likely(!sysctl_sched_compat_yield)) { | |
725 | __update_rq_clock(rq); | |
726 | /* | |
727 | * Dequeue and enqueue the task to update its | |
728 | * position within the tree: | |
729 | */ | |
730 | dequeue_entity(cfs_rq, &p->se, 0); | |
731 | enqueue_entity(cfs_rq, &p->se, 0); | |
732 | ||
733 | return; | |
734 | } | |
735 | /* | |
736 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 737 | */ |
1799e35d IM |
738 | do { |
739 | parent = *link; | |
740 | link = &parent->rb_right; | |
741 | } while (*link); | |
742 | ||
743 | rightmost = rb_entry(parent, struct sched_entity, run_node); | |
744 | /* | |
745 | * Already in the rightmost position? | |
746 | */ | |
747 | if (unlikely(rightmost == se)) | |
748 | return; | |
749 | ||
750 | /* | |
751 | * Minimally necessary key value to be last in the tree: | |
752 | */ | |
753 | se->fair_key = rightmost->fair_key + 1; | |
754 | ||
755 | if (cfs_rq->rb_leftmost == &se->run_node) | |
756 | cfs_rq->rb_leftmost = rb_next(&se->run_node); | |
757 | /* | |
758 | * Relink the task to the rightmost position: | |
759 | */ | |
760 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); | |
761 | rb_link_node(&se->run_node, parent, link); | |
762 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
763 | } |
764 | ||
765 | /* | |
766 | * Preempt the current task with a newly woken task if needed: | |
767 | */ | |
2e09bf55 | 768 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
769 | { |
770 | struct task_struct *curr = rq->curr; | |
771 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
bf0f6f24 IM |
772 | |
773 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 774 | update_rq_clock(rq); |
b7cc0896 | 775 | update_curr(cfs_rq); |
bf0f6f24 IM |
776 | resched_task(curr); |
777 | return; | |
778 | } | |
2e09bf55 IM |
779 | if (is_same_group(curr, p)) { |
780 | s64 delta = curr->se.vruntime - p->se.vruntime; | |
bf0f6f24 | 781 | |
2e09bf55 IM |
782 | if (delta > (s64)sysctl_sched_wakeup_granularity) |
783 | resched_task(curr); | |
784 | } | |
bf0f6f24 IM |
785 | } |
786 | ||
fb8d4724 | 787 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
788 | { |
789 | struct cfs_rq *cfs_rq = &rq->cfs; | |
790 | struct sched_entity *se; | |
791 | ||
792 | if (unlikely(!cfs_rq->nr_running)) | |
793 | return NULL; | |
794 | ||
795 | do { | |
9948f4b2 | 796 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
797 | cfs_rq = group_cfs_rq(se); |
798 | } while (cfs_rq); | |
799 | ||
800 | return task_of(se); | |
801 | } | |
802 | ||
803 | /* | |
804 | * Account for a descheduled task: | |
805 | */ | |
31ee529c | 806 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
807 | { |
808 | struct sched_entity *se = &prev->se; | |
809 | struct cfs_rq *cfs_rq; | |
810 | ||
811 | for_each_sched_entity(se) { | |
812 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 813 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
814 | } |
815 | } | |
816 | ||
817 | /************************************************** | |
818 | * Fair scheduling class load-balancing methods: | |
819 | */ | |
820 | ||
821 | /* | |
822 | * Load-balancing iterator. Note: while the runqueue stays locked | |
823 | * during the whole iteration, the current task might be | |
824 | * dequeued so the iterator has to be dequeue-safe. Here we | |
825 | * achieve that by always pre-iterating before returning | |
826 | * the current task: | |
827 | */ | |
828 | static inline struct task_struct * | |
829 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) | |
830 | { | |
831 | struct task_struct *p; | |
832 | ||
833 | if (!curr) | |
834 | return NULL; | |
835 | ||
836 | p = rb_entry(curr, struct task_struct, se.run_node); | |
837 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
838 | ||
839 | return p; | |
840 | } | |
841 | ||
842 | static struct task_struct *load_balance_start_fair(void *arg) | |
843 | { | |
844 | struct cfs_rq *cfs_rq = arg; | |
845 | ||
846 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
847 | } | |
848 | ||
849 | static struct task_struct *load_balance_next_fair(void *arg) | |
850 | { | |
851 | struct cfs_rq *cfs_rq = arg; | |
852 | ||
853 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
854 | } | |
855 | ||
a4ac01c3 | 856 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
857 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
858 | { | |
859 | struct sched_entity *curr; | |
860 | struct task_struct *p; | |
861 | ||
862 | if (!cfs_rq->nr_running) | |
863 | return MAX_PRIO; | |
864 | ||
865 | curr = __pick_next_entity(cfs_rq); | |
866 | p = task_of(curr); | |
867 | ||
868 | return p->prio; | |
869 | } | |
a4ac01c3 | 870 | #endif |
bf0f6f24 | 871 | |
43010659 | 872 | static unsigned long |
bf0f6f24 | 873 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
a4ac01c3 PW |
874 | unsigned long max_nr_move, unsigned long max_load_move, |
875 | struct sched_domain *sd, enum cpu_idle_type idle, | |
876 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
877 | { |
878 | struct cfs_rq *busy_cfs_rq; | |
879 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
880 | long rem_load_move = max_load_move; | |
881 | struct rq_iterator cfs_rq_iterator; | |
882 | ||
883 | cfs_rq_iterator.start = load_balance_start_fair; | |
884 | cfs_rq_iterator.next = load_balance_next_fair; | |
885 | ||
886 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 887 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 888 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 889 | long imbalance; |
bf0f6f24 | 890 | unsigned long maxload; |
bf0f6f24 IM |
891 | |
892 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
893 | ||
e56f31aa | 894 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
895 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
896 | if (imbalance <= 0) | |
897 | continue; | |
898 | ||
899 | /* Don't pull more than imbalance/2 */ | |
900 | imbalance /= 2; | |
901 | maxload = min(rem_load_move, imbalance); | |
902 | ||
a4ac01c3 PW |
903 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
904 | #else | |
e56f31aa | 905 | # define maxload rem_load_move |
a4ac01c3 | 906 | #endif |
bf0f6f24 IM |
907 | /* pass busy_cfs_rq argument into |
908 | * load_balance_[start|next]_fair iterators | |
909 | */ | |
910 | cfs_rq_iterator.arg = busy_cfs_rq; | |
911 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, | |
912 | max_nr_move, maxload, sd, idle, all_pinned, | |
a4ac01c3 | 913 | &load_moved, this_best_prio, &cfs_rq_iterator); |
bf0f6f24 IM |
914 | |
915 | total_nr_moved += nr_moved; | |
916 | max_nr_move -= nr_moved; | |
917 | rem_load_move -= load_moved; | |
918 | ||
919 | if (max_nr_move <= 0 || rem_load_move <= 0) | |
920 | break; | |
921 | } | |
922 | ||
43010659 | 923 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
924 | } |
925 | ||
926 | /* | |
927 | * scheduler tick hitting a task of our scheduling class: | |
928 | */ | |
929 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
930 | { | |
931 | struct cfs_rq *cfs_rq; | |
932 | struct sched_entity *se = &curr->se; | |
933 | ||
934 | for_each_sched_entity(se) { | |
935 | cfs_rq = cfs_rq_of(se); | |
936 | entity_tick(cfs_rq, se); | |
937 | } | |
938 | } | |
939 | ||
4d78e7b6 PZ |
940 | #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
941 | ||
bf0f6f24 IM |
942 | /* |
943 | * Share the fairness runtime between parent and child, thus the | |
944 | * total amount of pressure for CPU stays equal - new tasks | |
945 | * get a chance to run but frequent forkers are not allowed to | |
946 | * monopolize the CPU. Note: the parent runqueue is locked, | |
947 | * the child is not running yet. | |
948 | */ | |
ee0827d8 | 949 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
950 | { |
951 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 952 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
bf0f6f24 IM |
953 | |
954 | sched_info_queued(p); | |
955 | ||
7109c442 | 956 | update_curr(cfs_rq); |
aeb73b04 | 957 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 958 | |
4d78e7b6 PZ |
959 | if (sysctl_sched_child_runs_first && |
960 | curr->vruntime < se->vruntime) { | |
961 | ||
962 | dequeue_entity(cfs_rq, curr, 0); | |
963 | swap(curr->vruntime, se->vruntime); | |
964 | enqueue_entity(cfs_rq, curr, 0); | |
965 | } | |
bf0f6f24 | 966 | |
e9acbff6 | 967 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 | 968 | __enqueue_entity(cfs_rq, se); |
bb61c210 | 969 | resched_task(rq->curr); |
bf0f6f24 IM |
970 | } |
971 | ||
972 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
973 | /* Account for a task changing its policy or group. | |
974 | * | |
975 | * This routine is mostly called to set cfs_rq->curr field when a task | |
976 | * migrates between groups/classes. | |
977 | */ | |
978 | static void set_curr_task_fair(struct rq *rq) | |
979 | { | |
7c6c16f3 | 980 | struct sched_entity *se = &rq->curr->se; |
a8e504d2 | 981 | |
c3b64f1e IM |
982 | for_each_sched_entity(se) |
983 | set_next_entity(cfs_rq_of(se), se); | |
bf0f6f24 IM |
984 | } |
985 | #else | |
986 | static void set_curr_task_fair(struct rq *rq) | |
987 | { | |
988 | } | |
989 | #endif | |
990 | ||
991 | /* | |
992 | * All the scheduling class methods: | |
993 | */ | |
994 | struct sched_class fair_sched_class __read_mostly = { | |
995 | .enqueue_task = enqueue_task_fair, | |
996 | .dequeue_task = dequeue_task_fair, | |
997 | .yield_task = yield_task_fair, | |
998 | ||
2e09bf55 | 999 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1000 | |
1001 | .pick_next_task = pick_next_task_fair, | |
1002 | .put_prev_task = put_prev_task_fair, | |
1003 | ||
1004 | .load_balance = load_balance_fair, | |
1005 | ||
1006 | .set_curr_task = set_curr_task_fair, | |
1007 | .task_tick = task_tick_fair, | |
1008 | .task_new = task_new_fair, | |
1009 | }; | |
1010 | ||
1011 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1012 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1013 | { |
bf0f6f24 IM |
1014 | struct cfs_rq *cfs_rq; |
1015 | ||
c3b64f1e | 1016 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1017 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1018 | } |
1019 | #endif |