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 | 141 | { |
30cfdcfc | 142 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
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); | |
bf0f6f24 IM |
184 | } |
185 | ||
19ccd97a | 186 | static void |
bf0f6f24 IM |
187 | __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
188 | { | |
189 | if (cfs_rq->rb_leftmost == &se->run_node) | |
e9acbff6 IM |
190 | set_leftmost(cfs_rq, rb_next(&se->run_node)); |
191 | ||
bf0f6f24 | 192 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
193 | } |
194 | ||
195 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
196 | { | |
197 | return cfs_rq->rb_leftmost; | |
198 | } | |
199 | ||
200 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
201 | { | |
202 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
203 | } | |
204 | ||
aeb73b04 PZ |
205 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
206 | { | |
207 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
208 | struct sched_entity *se = NULL; | |
209 | struct rb_node *parent; | |
210 | ||
211 | while (*link) { | |
212 | parent = *link; | |
213 | se = rb_entry(parent, struct sched_entity, run_node); | |
214 | link = &parent->rb_right; | |
215 | } | |
216 | ||
217 | return se; | |
218 | } | |
219 | ||
bf0f6f24 IM |
220 | /************************************************************** |
221 | * Scheduling class statistics methods: | |
222 | */ | |
223 | ||
4d78e7b6 PZ |
224 | static u64 __sched_period(unsigned long nr_running) |
225 | { | |
226 | u64 period = sysctl_sched_latency; | |
227 | unsigned long nr_latency = | |
228 | sysctl_sched_latency / sysctl_sched_min_granularity; | |
229 | ||
230 | if (unlikely(nr_running > nr_latency)) { | |
231 | period *= nr_running; | |
232 | do_div(period, nr_latency); | |
233 | } | |
234 | ||
235 | return period; | |
236 | } | |
237 | ||
6d0f0ebd | 238 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 239 | { |
6d0f0ebd | 240 | u64 period = __sched_period(cfs_rq->nr_running); |
21805085 | 241 | |
6d0f0ebd PZ |
242 | period *= se->load.weight; |
243 | do_div(period, cfs_rq->load.weight); | |
21805085 | 244 | |
6d0f0ebd | 245 | return period; |
bf0f6f24 IM |
246 | } |
247 | ||
bf0f6f24 IM |
248 | /* |
249 | * Update the current task's runtime statistics. Skip current tasks that | |
250 | * are not in our scheduling class. | |
251 | */ | |
252 | static inline void | |
8ebc91d9 IM |
253 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
254 | unsigned long delta_exec) | |
bf0f6f24 | 255 | { |
bbdba7c0 | 256 | unsigned long delta_exec_weighted; |
02e0431a | 257 | u64 next_vruntime, min_vruntime; |
bf0f6f24 | 258 | |
8179ca23 | 259 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
260 | |
261 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 262 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
e9acbff6 IM |
263 | delta_exec_weighted = delta_exec; |
264 | if (unlikely(curr->load.weight != NICE_0_LOAD)) { | |
265 | delta_exec_weighted = calc_delta_fair(delta_exec_weighted, | |
266 | &curr->load); | |
267 | } | |
268 | curr->vruntime += delta_exec_weighted; | |
02e0431a PZ |
269 | |
270 | /* | |
271 | * maintain cfs_rq->min_vruntime to be a monotonic increasing | |
272 | * value tracking the leftmost vruntime in the tree. | |
273 | */ | |
274 | if (first_fair(cfs_rq)) { | |
275 | next_vruntime = __pick_next_entity(cfs_rq)->vruntime; | |
276 | ||
277 | /* min_vruntime() := !max_vruntime() */ | |
278 | min_vruntime = max_vruntime(curr->vruntime, next_vruntime); | |
279 | if (min_vruntime == next_vruntime) | |
280 | min_vruntime = curr->vruntime; | |
281 | else | |
282 | min_vruntime = next_vruntime; | |
283 | } else | |
284 | min_vruntime = curr->vruntime; | |
285 | ||
286 | cfs_rq->min_vruntime = | |
287 | max_vruntime(cfs_rq->min_vruntime, min_vruntime); | |
bf0f6f24 IM |
288 | } |
289 | ||
b7cc0896 | 290 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 291 | { |
429d43bc | 292 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 293 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
294 | unsigned long delta_exec; |
295 | ||
296 | if (unlikely(!curr)) | |
297 | return; | |
298 | ||
299 | /* | |
300 | * Get the amount of time the current task was running | |
301 | * since the last time we changed load (this cannot | |
302 | * overflow on 32 bits): | |
303 | */ | |
8ebc91d9 | 304 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 305 | |
8ebc91d9 IM |
306 | __update_curr(cfs_rq, curr, delta_exec); |
307 | curr->exec_start = now; | |
bf0f6f24 IM |
308 | } |
309 | ||
310 | static inline void | |
5870db5b | 311 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 312 | { |
d281918d | 313 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
314 | } |
315 | ||
bf0f6f24 | 316 | static inline unsigned long |
08e2388a | 317 | calc_weighted(unsigned long delta, struct sched_entity *se) |
bf0f6f24 | 318 | { |
08e2388a | 319 | unsigned long weight = se->load.weight; |
bf0f6f24 | 320 | |
08e2388a IM |
321 | if (unlikely(weight != NICE_0_LOAD)) |
322 | return (u64)delta * se->load.weight >> NICE_0_SHIFT; | |
323 | else | |
324 | return delta; | |
bf0f6f24 | 325 | } |
bf0f6f24 IM |
326 | |
327 | /* | |
328 | * Task is being enqueued - update stats: | |
329 | */ | |
d2417e5a | 330 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 331 | { |
bf0f6f24 IM |
332 | /* |
333 | * Are we enqueueing a waiting task? (for current tasks | |
334 | * a dequeue/enqueue event is a NOP) | |
335 | */ | |
429d43bc | 336 | if (se != cfs_rq->curr) |
5870db5b | 337 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
338 | } |
339 | ||
bf0f6f24 | 340 | static void |
9ef0a961 | 341 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 342 | { |
bbdba7c0 IM |
343 | schedstat_set(se->wait_max, max(se->wait_max, |
344 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6cfb0d5d | 345 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
346 | } |
347 | ||
348 | static inline void | |
19b6a2e3 | 349 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 350 | { |
b7cc0896 | 351 | update_curr(cfs_rq); |
bf0f6f24 IM |
352 | /* |
353 | * Mark the end of the wait period if dequeueing a | |
354 | * waiting task: | |
355 | */ | |
429d43bc | 356 | if (se != cfs_rq->curr) |
9ef0a961 | 357 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
358 | } |
359 | ||
360 | /* | |
361 | * We are picking a new current task - update its stats: | |
362 | */ | |
363 | static inline void | |
79303e9e | 364 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
365 | { |
366 | /* | |
367 | * We are starting a new run period: | |
368 | */ | |
d281918d | 369 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
370 | } |
371 | ||
372 | /* | |
373 | * We are descheduling a task - update its stats: | |
374 | */ | |
375 | static inline void | |
c7e9b5b2 | 376 | update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
377 | { |
378 | se->exec_start = 0; | |
379 | } | |
380 | ||
381 | /************************************************** | |
382 | * Scheduling class queueing methods: | |
383 | */ | |
384 | ||
30cfdcfc DA |
385 | static void |
386 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
387 | { | |
388 | update_load_add(&cfs_rq->load, se->load.weight); | |
389 | cfs_rq->nr_running++; | |
390 | se->on_rq = 1; | |
391 | } | |
392 | ||
393 | static void | |
394 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
395 | { | |
396 | update_load_sub(&cfs_rq->load, se->load.weight); | |
397 | cfs_rq->nr_running--; | |
398 | se->on_rq = 0; | |
399 | } | |
400 | ||
2396af69 | 401 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 402 | { |
bf0f6f24 IM |
403 | #ifdef CONFIG_SCHEDSTATS |
404 | if (se->sleep_start) { | |
d281918d | 405 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
406 | |
407 | if ((s64)delta < 0) | |
408 | delta = 0; | |
409 | ||
410 | if (unlikely(delta > se->sleep_max)) | |
411 | se->sleep_max = delta; | |
412 | ||
413 | se->sleep_start = 0; | |
414 | se->sum_sleep_runtime += delta; | |
415 | } | |
416 | if (se->block_start) { | |
d281918d | 417 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
418 | |
419 | if ((s64)delta < 0) | |
420 | delta = 0; | |
421 | ||
422 | if (unlikely(delta > se->block_max)) | |
423 | se->block_max = delta; | |
424 | ||
425 | se->block_start = 0; | |
426 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
427 | |
428 | /* | |
429 | * Blocking time is in units of nanosecs, so shift by 20 to | |
430 | * get a milliseconds-range estimation of the amount of | |
431 | * time that the task spent sleeping: | |
432 | */ | |
433 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf IM |
434 | struct task_struct *tsk = task_of(se); |
435 | ||
30084fbd IM |
436 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
437 | delta >> 20); | |
438 | } | |
bf0f6f24 IM |
439 | } |
440 | #endif | |
441 | } | |
442 | ||
aeb73b04 PZ |
443 | static void |
444 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
445 | { | |
aeb73b04 PZ |
446 | u64 min_runtime, latency; |
447 | ||
448 | min_runtime = cfs_rq->min_vruntime; | |
94dfb5e7 PZ |
449 | |
450 | if (sched_feat(USE_TREE_AVG)) { | |
451 | struct sched_entity *last = __pick_last_entity(cfs_rq); | |
452 | if (last) { | |
453 | min_runtime = __pick_next_entity(cfs_rq)->vruntime; | |
454 | min_runtime += last->vruntime; | |
455 | min_runtime >>= 1; | |
456 | } | |
457 | } else if (sched_feat(APPROX_AVG)) | |
458 | min_runtime += sysctl_sched_latency/2; | |
459 | ||
460 | if (initial && sched_feat(START_DEBIT)) | |
461 | min_runtime += sched_slice(cfs_rq, se); | |
aeb73b04 PZ |
462 | |
463 | if (!initial && sched_feat(NEW_FAIR_SLEEPERS)) { | |
464 | latency = sysctl_sched_latency; | |
465 | if (min_runtime > latency) | |
466 | min_runtime -= latency; | |
467 | else | |
468 | min_runtime = 0; | |
469 | } | |
470 | ||
471 | se->vruntime = max(se->vruntime, min_runtime); | |
472 | } | |
473 | ||
bf0f6f24 | 474 | static void |
668031ca | 475 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
476 | { |
477 | /* | |
478 | * Update the fair clock. | |
479 | */ | |
b7cc0896 | 480 | update_curr(cfs_rq); |
bf0f6f24 | 481 | |
e9acbff6 | 482 | if (wakeup) { |
aeb73b04 | 483 | place_entity(cfs_rq, se, 0); |
2396af69 | 484 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 485 | } |
bf0f6f24 | 486 | |
d2417e5a | 487 | update_stats_enqueue(cfs_rq, se); |
30cfdcfc DA |
488 | if (se != cfs_rq->curr) |
489 | __enqueue_entity(cfs_rq, se); | |
490 | account_entity_enqueue(cfs_rq, se); | |
bf0f6f24 IM |
491 | } |
492 | ||
493 | static void | |
525c2716 | 494 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 495 | { |
19b6a2e3 | 496 | update_stats_dequeue(cfs_rq, se); |
bf0f6f24 | 497 | #ifdef CONFIG_SCHEDSTATS |
db36cc7d | 498 | if (sleep) { |
bf0f6f24 IM |
499 | if (entity_is_task(se)) { |
500 | struct task_struct *tsk = task_of(se); | |
501 | ||
502 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 503 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 504 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 505 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 506 | } |
bf0f6f24 | 507 | } |
db36cc7d | 508 | #endif |
30cfdcfc DA |
509 | if (se != cfs_rq->curr) |
510 | __dequeue_entity(cfs_rq, se); | |
511 | account_entity_dequeue(cfs_rq, se); | |
bf0f6f24 IM |
512 | } |
513 | ||
514 | /* | |
515 | * Preempt the current task with a newly woken task if needed: | |
516 | */ | |
7c92e54f | 517 | static void |
2e09bf55 | 518 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 519 | { |
11697830 PZ |
520 | unsigned long ideal_runtime, delta_exec; |
521 | ||
6d0f0ebd | 522 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 PZ |
523 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
524 | if (delta_exec > ideal_runtime) | |
bf0f6f24 IM |
525 | resched_task(rq_of(cfs_rq)->curr); |
526 | } | |
527 | ||
528 | static inline void | |
8494f412 | 529 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
530 | { |
531 | /* | |
532 | * Any task has to be enqueued before it get to execute on | |
533 | * a CPU. So account for the time it spent waiting on the | |
bbdba7c0 | 534 | * runqueue. |
bf0f6f24 | 535 | */ |
9ef0a961 | 536 | update_stats_wait_end(cfs_rq, se); |
79303e9e | 537 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 538 | cfs_rq->curr = se; |
eba1ed4b IM |
539 | #ifdef CONFIG_SCHEDSTATS |
540 | /* | |
541 | * Track our maximum slice length, if the CPU's load is at | |
542 | * least twice that of our own weight (i.e. dont track it | |
543 | * when there are only lesser-weight tasks around): | |
544 | */ | |
495eca49 | 545 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
546 | se->slice_max = max(se->slice_max, |
547 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
548 | } | |
549 | #endif | |
4a55b450 | 550 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
551 | } |
552 | ||
9948f4b2 | 553 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 IM |
554 | { |
555 | struct sched_entity *se = __pick_next_entity(cfs_rq); | |
556 | ||
30cfdcfc DA |
557 | /* 'current' is not kept within the tree. */ |
558 | if (se) | |
559 | __dequeue_entity(cfs_rq, se); | |
560 | ||
8494f412 | 561 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
562 | |
563 | return se; | |
564 | } | |
565 | ||
ab6cde26 | 566 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
567 | { |
568 | /* | |
569 | * If still on the runqueue then deactivate_task() | |
570 | * was not called and update_curr() has to be done: | |
571 | */ | |
572 | if (prev->on_rq) | |
b7cc0896 | 573 | update_curr(cfs_rq); |
bf0f6f24 | 574 | |
c7e9b5b2 | 575 | update_stats_curr_end(cfs_rq, prev); |
bf0f6f24 | 576 | |
30cfdcfc | 577 | if (prev->on_rq) { |
5870db5b | 578 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
579 | /* Put 'current' back into the tree. */ |
580 | __enqueue_entity(cfs_rq, prev); | |
581 | } | |
429d43bc | 582 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
583 | } |
584 | ||
585 | static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) | |
586 | { | |
bf0f6f24 | 587 | /* |
30cfdcfc | 588 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 589 | */ |
30cfdcfc | 590 | update_curr(cfs_rq); |
bf0f6f24 | 591 | |
2e09bf55 IM |
592 | if (cfs_rq->nr_running > 1) |
593 | check_preempt_tick(cfs_rq, curr); | |
bf0f6f24 IM |
594 | } |
595 | ||
596 | /************************************************** | |
597 | * CFS operations on tasks: | |
598 | */ | |
599 | ||
600 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
601 | ||
602 | /* Walk up scheduling entities hierarchy */ | |
603 | #define for_each_sched_entity(se) \ | |
604 | for (; se; se = se->parent) | |
605 | ||
606 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
607 | { | |
608 | return p->se.cfs_rq; | |
609 | } | |
610 | ||
611 | /* runqueue on which this entity is (to be) queued */ | |
612 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
613 | { | |
614 | return se->cfs_rq; | |
615 | } | |
616 | ||
617 | /* runqueue "owned" by this group */ | |
618 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
619 | { | |
620 | return grp->my_q; | |
621 | } | |
622 | ||
623 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
624 | * another cpu ('this_cpu') | |
625 | */ | |
626 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
627 | { | |
29f59db3 | 628 | return cfs_rq->tg->cfs_rq[this_cpu]; |
bf0f6f24 IM |
629 | } |
630 | ||
631 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
632 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
633 | list_for_each_entry(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
634 | ||
635 | /* Do the two (enqueued) tasks belong to the same group ? */ | |
636 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
637 | { | |
638 | if (curr->se.cfs_rq == p->se.cfs_rq) | |
639 | return 1; | |
640 | ||
641 | return 0; | |
642 | } | |
643 | ||
644 | #else /* CONFIG_FAIR_GROUP_SCHED */ | |
645 | ||
646 | #define for_each_sched_entity(se) \ | |
647 | for (; se; se = NULL) | |
648 | ||
649 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
650 | { | |
651 | return &task_rq(p)->cfs; | |
652 | } | |
653 | ||
654 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
655 | { | |
656 | struct task_struct *p = task_of(se); | |
657 | struct rq *rq = task_rq(p); | |
658 | ||
659 | return &rq->cfs; | |
660 | } | |
661 | ||
662 | /* runqueue "owned" by this group */ | |
663 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
664 | { | |
665 | return NULL; | |
666 | } | |
667 | ||
668 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
669 | { | |
670 | return &cpu_rq(this_cpu)->cfs; | |
671 | } | |
672 | ||
673 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
674 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
675 | ||
676 | static inline int is_same_group(struct task_struct *curr, struct task_struct *p) | |
677 | { | |
678 | return 1; | |
679 | } | |
680 | ||
681 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
682 | ||
683 | /* | |
684 | * The enqueue_task method is called before nr_running is | |
685 | * increased. Here we update the fair scheduling stats and | |
686 | * then put the task into the rbtree: | |
687 | */ | |
fd390f6a | 688 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
689 | { |
690 | struct cfs_rq *cfs_rq; | |
691 | struct sched_entity *se = &p->se; | |
692 | ||
693 | for_each_sched_entity(se) { | |
694 | if (se->on_rq) | |
695 | break; | |
696 | cfs_rq = cfs_rq_of(se); | |
668031ca | 697 | enqueue_entity(cfs_rq, se, wakeup); |
bf0f6f24 IM |
698 | } |
699 | } | |
700 | ||
701 | /* | |
702 | * The dequeue_task method is called before nr_running is | |
703 | * decreased. We remove the task from the rbtree and | |
704 | * update the fair scheduling stats: | |
705 | */ | |
f02231e5 | 706 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
707 | { |
708 | struct cfs_rq *cfs_rq; | |
709 | struct sched_entity *se = &p->se; | |
710 | ||
711 | for_each_sched_entity(se) { | |
712 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 713 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 IM |
714 | /* Don't dequeue parent if it has other entities besides us */ |
715 | if (cfs_rq->load.weight) | |
716 | break; | |
717 | } | |
718 | } | |
719 | ||
720 | /* | |
1799e35d IM |
721 | * sched_yield() support is very simple - we dequeue and enqueue. |
722 | * | |
723 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 IM |
724 | */ |
725 | static void yield_task_fair(struct rq *rq, struct task_struct *p) | |
726 | { | |
727 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1799e35d IM |
728 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; |
729 | struct sched_entity *rightmost, *se = &p->se; | |
730 | struct rb_node *parent; | |
bf0f6f24 IM |
731 | |
732 | /* | |
1799e35d IM |
733 | * Are we the only task in the tree? |
734 | */ | |
735 | if (unlikely(cfs_rq->nr_running == 1)) | |
736 | return; | |
737 | ||
738 | if (likely(!sysctl_sched_compat_yield)) { | |
739 | __update_rq_clock(rq); | |
740 | /* | |
741 | * Dequeue and enqueue the task to update its | |
742 | * position within the tree: | |
743 | */ | |
744 | dequeue_entity(cfs_rq, &p->se, 0); | |
745 | enqueue_entity(cfs_rq, &p->se, 0); | |
746 | ||
747 | return; | |
748 | } | |
749 | /* | |
750 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 751 | */ |
1799e35d IM |
752 | do { |
753 | parent = *link; | |
754 | link = &parent->rb_right; | |
755 | } while (*link); | |
756 | ||
757 | rightmost = rb_entry(parent, struct sched_entity, run_node); | |
758 | /* | |
759 | * Already in the rightmost position? | |
760 | */ | |
761 | if (unlikely(rightmost == se)) | |
762 | return; | |
763 | ||
764 | /* | |
765 | * Minimally necessary key value to be last in the tree: | |
766 | */ | |
30cfdcfc | 767 | se->vruntime = rightmost->vruntime + 1; |
1799e35d IM |
768 | |
769 | if (cfs_rq->rb_leftmost == &se->run_node) | |
770 | cfs_rq->rb_leftmost = rb_next(&se->run_node); | |
771 | /* | |
772 | * Relink the task to the rightmost position: | |
773 | */ | |
774 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); | |
775 | rb_link_node(&se->run_node, parent, link); | |
776 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
777 | } |
778 | ||
779 | /* | |
780 | * Preempt the current task with a newly woken task if needed: | |
781 | */ | |
2e09bf55 | 782 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
783 | { |
784 | struct task_struct *curr = rq->curr; | |
785 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
bf0f6f24 IM |
786 | |
787 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 788 | update_rq_clock(rq); |
b7cc0896 | 789 | update_curr(cfs_rq); |
bf0f6f24 IM |
790 | resched_task(curr); |
791 | return; | |
792 | } | |
2e09bf55 IM |
793 | if (is_same_group(curr, p)) { |
794 | s64 delta = curr->se.vruntime - p->se.vruntime; | |
bf0f6f24 | 795 | |
2e09bf55 IM |
796 | if (delta > (s64)sysctl_sched_wakeup_granularity) |
797 | resched_task(curr); | |
798 | } | |
bf0f6f24 IM |
799 | } |
800 | ||
fb8d4724 | 801 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 IM |
802 | { |
803 | struct cfs_rq *cfs_rq = &rq->cfs; | |
804 | struct sched_entity *se; | |
805 | ||
806 | if (unlikely(!cfs_rq->nr_running)) | |
807 | return NULL; | |
808 | ||
809 | do { | |
9948f4b2 | 810 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
811 | cfs_rq = group_cfs_rq(se); |
812 | } while (cfs_rq); | |
813 | ||
814 | return task_of(se); | |
815 | } | |
816 | ||
817 | /* | |
818 | * Account for a descheduled task: | |
819 | */ | |
31ee529c | 820 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
821 | { |
822 | struct sched_entity *se = &prev->se; | |
823 | struct cfs_rq *cfs_rq; | |
824 | ||
825 | for_each_sched_entity(se) { | |
826 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 827 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
828 | } |
829 | } | |
830 | ||
831 | /************************************************** | |
832 | * Fair scheduling class load-balancing methods: | |
833 | */ | |
834 | ||
835 | /* | |
836 | * Load-balancing iterator. Note: while the runqueue stays locked | |
837 | * during the whole iteration, the current task might be | |
838 | * dequeued so the iterator has to be dequeue-safe. Here we | |
839 | * achieve that by always pre-iterating before returning | |
840 | * the current task: | |
841 | */ | |
842 | static inline struct task_struct * | |
843 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct rb_node *curr) | |
844 | { | |
845 | struct task_struct *p; | |
846 | ||
847 | if (!curr) | |
848 | return NULL; | |
849 | ||
850 | p = rb_entry(curr, struct task_struct, se.run_node); | |
851 | cfs_rq->rb_load_balance_curr = rb_next(curr); | |
852 | ||
853 | return p; | |
854 | } | |
855 | ||
856 | static struct task_struct *load_balance_start_fair(void *arg) | |
857 | { | |
858 | struct cfs_rq *cfs_rq = arg; | |
859 | ||
860 | return __load_balance_iterator(cfs_rq, first_fair(cfs_rq)); | |
861 | } | |
862 | ||
863 | static struct task_struct *load_balance_next_fair(void *arg) | |
864 | { | |
865 | struct cfs_rq *cfs_rq = arg; | |
866 | ||
867 | return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); | |
868 | } | |
869 | ||
a4ac01c3 | 870 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 IM |
871 | static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) |
872 | { | |
873 | struct sched_entity *curr; | |
874 | struct task_struct *p; | |
875 | ||
876 | if (!cfs_rq->nr_running) | |
877 | return MAX_PRIO; | |
878 | ||
879 | curr = __pick_next_entity(cfs_rq); | |
880 | p = task_of(curr); | |
881 | ||
882 | return p->prio; | |
883 | } | |
a4ac01c3 | 884 | #endif |
bf0f6f24 | 885 | |
43010659 | 886 | static unsigned long |
bf0f6f24 | 887 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
a4ac01c3 PW |
888 | unsigned long max_nr_move, unsigned long max_load_move, |
889 | struct sched_domain *sd, enum cpu_idle_type idle, | |
890 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 IM |
891 | { |
892 | struct cfs_rq *busy_cfs_rq; | |
893 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
894 | long rem_load_move = max_load_move; | |
895 | struct rq_iterator cfs_rq_iterator; | |
896 | ||
897 | cfs_rq_iterator.start = load_balance_start_fair; | |
898 | cfs_rq_iterator.next = load_balance_next_fair; | |
899 | ||
900 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
a4ac01c3 | 901 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 902 | struct cfs_rq *this_cfs_rq; |
e56f31aa | 903 | long imbalance; |
bf0f6f24 | 904 | unsigned long maxload; |
bf0f6f24 IM |
905 | |
906 | this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); | |
907 | ||
e56f31aa | 908 | imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight; |
bf0f6f24 IM |
909 | /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */ |
910 | if (imbalance <= 0) | |
911 | continue; | |
912 | ||
913 | /* Don't pull more than imbalance/2 */ | |
914 | imbalance /= 2; | |
915 | maxload = min(rem_load_move, imbalance); | |
916 | ||
a4ac01c3 PW |
917 | *this_best_prio = cfs_rq_best_prio(this_cfs_rq); |
918 | #else | |
e56f31aa | 919 | # define maxload rem_load_move |
a4ac01c3 | 920 | #endif |
bf0f6f24 IM |
921 | /* pass busy_cfs_rq argument into |
922 | * load_balance_[start|next]_fair iterators | |
923 | */ | |
924 | cfs_rq_iterator.arg = busy_cfs_rq; | |
925 | nr_moved = balance_tasks(this_rq, this_cpu, busiest, | |
926 | max_nr_move, maxload, sd, idle, all_pinned, | |
a4ac01c3 | 927 | &load_moved, this_best_prio, &cfs_rq_iterator); |
bf0f6f24 IM |
928 | |
929 | total_nr_moved += nr_moved; | |
930 | max_nr_move -= nr_moved; | |
931 | rem_load_move -= load_moved; | |
932 | ||
933 | if (max_nr_move <= 0 || rem_load_move <= 0) | |
934 | break; | |
935 | } | |
936 | ||
43010659 | 937 | return max_load_move - rem_load_move; |
bf0f6f24 IM |
938 | } |
939 | ||
940 | /* | |
941 | * scheduler tick hitting a task of our scheduling class: | |
942 | */ | |
943 | static void task_tick_fair(struct rq *rq, struct task_struct *curr) | |
944 | { | |
945 | struct cfs_rq *cfs_rq; | |
946 | struct sched_entity *se = &curr->se; | |
947 | ||
948 | for_each_sched_entity(se) { | |
949 | cfs_rq = cfs_rq_of(se); | |
950 | entity_tick(cfs_rq, se); | |
951 | } | |
952 | } | |
953 | ||
4d78e7b6 PZ |
954 | #define swap(a,b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
955 | ||
bf0f6f24 IM |
956 | /* |
957 | * Share the fairness runtime between parent and child, thus the | |
958 | * total amount of pressure for CPU stays equal - new tasks | |
959 | * get a chance to run but frequent forkers are not allowed to | |
960 | * monopolize the CPU. Note: the parent runqueue is locked, | |
961 | * the child is not running yet. | |
962 | */ | |
ee0827d8 | 963 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
964 | { |
965 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 966 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
bf0f6f24 IM |
967 | |
968 | sched_info_queued(p); | |
969 | ||
7109c442 | 970 | update_curr(cfs_rq); |
aeb73b04 | 971 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 972 | |
4d78e7b6 PZ |
973 | if (sysctl_sched_child_runs_first && |
974 | curr->vruntime < se->vruntime) { | |
975 | ||
976 | dequeue_entity(cfs_rq, curr, 0); | |
977 | swap(curr->vruntime, se->vruntime); | |
978 | enqueue_entity(cfs_rq, curr, 0); | |
979 | } | |
bf0f6f24 | 980 | |
e9acbff6 | 981 | update_stats_enqueue(cfs_rq, se); |
bf0f6f24 | 982 | __enqueue_entity(cfs_rq, se); |
30cfdcfc | 983 | account_entity_enqueue(cfs_rq, se); |
bb61c210 | 984 | resched_task(rq->curr); |
bf0f6f24 IM |
985 | } |
986 | ||
987 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
988 | /* Account for a task changing its policy or group. | |
989 | * | |
990 | * This routine is mostly called to set cfs_rq->curr field when a task | |
991 | * migrates between groups/classes. | |
992 | */ | |
993 | static void set_curr_task_fair(struct rq *rq) | |
994 | { | |
7c6c16f3 | 995 | struct sched_entity *se = &rq->curr->se; |
a8e504d2 | 996 | |
c3b64f1e IM |
997 | for_each_sched_entity(se) |
998 | set_next_entity(cfs_rq_of(se), se); | |
bf0f6f24 IM |
999 | } |
1000 | #else | |
1001 | static void set_curr_task_fair(struct rq *rq) | |
1002 | { | |
d02e5ed8 DA |
1003 | struct sched_entity *se = &rq->curr->se; |
1004 | struct cfs_rq *cfs_rq = cfs_rq_of(se); | |
1005 | ||
1006 | cfs_rq->curr = se; | |
bf0f6f24 IM |
1007 | } |
1008 | #endif | |
1009 | ||
1010 | /* | |
1011 | * All the scheduling class methods: | |
1012 | */ | |
1013 | struct sched_class fair_sched_class __read_mostly = { | |
1014 | .enqueue_task = enqueue_task_fair, | |
1015 | .dequeue_task = dequeue_task_fair, | |
1016 | .yield_task = yield_task_fair, | |
1017 | ||
2e09bf55 | 1018 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1019 | |
1020 | .pick_next_task = pick_next_task_fair, | |
1021 | .put_prev_task = put_prev_task_fair, | |
1022 | ||
1023 | .load_balance = load_balance_fair, | |
1024 | ||
1025 | .set_curr_task = set_curr_task_fair, | |
1026 | .task_tick = task_tick_fair, | |
1027 | .task_new = task_new_fair, | |
1028 | }; | |
1029 | ||
1030 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1031 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1032 | { |
bf0f6f24 IM |
1033 | struct cfs_rq *cfs_rq; |
1034 | ||
c3b64f1e | 1035 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1036 | print_cfs_rq(m, cpu, cfs_rq); |
bf0f6f24 IM |
1037 | } |
1038 | #endif |