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