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 | ||
9745512c | 23 | #include <linux/latencytop.h> |
1983a922 | 24 | #include <linux/sched.h> |
9745512c | 25 | |
bf0f6f24 | 26 | /* |
21805085 | 27 | * Targeted preemption latency for CPU-bound tasks: |
172e082a | 28 | * (default: 5ms * (1 + ilog(ncpus)), units: nanoseconds) |
bf0f6f24 | 29 | * |
21805085 | 30 | * NOTE: this latency value is not the same as the concept of |
d274a4ce IM |
31 | * 'timeslice length' - timeslices in CFS are of variable length |
32 | * and have no persistent notion like in traditional, time-slice | |
33 | * based scheduling concepts. | |
bf0f6f24 | 34 | * |
d274a4ce IM |
35 | * (to see the precise effective timeslice length of your workload, |
36 | * run vmstat and monitor the context-switches (cs) field) | |
bf0f6f24 | 37 | */ |
172e082a | 38 | unsigned int sysctl_sched_latency = 5000000ULL; |
0bcdcf28 | 39 | unsigned int normalized_sysctl_sched_latency = 5000000ULL; |
2bd8e6d4 | 40 | |
1983a922 CE |
41 | /* |
42 | * The initial- and re-scaling of tunables is configurable | |
43 | * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus)) | |
44 | * | |
45 | * Options are: | |
46 | * SCHED_TUNABLESCALING_NONE - unscaled, always *1 | |
47 | * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus) | |
48 | * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus | |
49 | */ | |
50 | enum sched_tunable_scaling sysctl_sched_tunable_scaling | |
51 | = SCHED_TUNABLESCALING_LOG; | |
52 | ||
2bd8e6d4 | 53 | /* |
b2be5e96 | 54 | * Minimal preemption granularity for CPU-bound tasks: |
172e082a | 55 | * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds) |
2bd8e6d4 | 56 | */ |
172e082a | 57 | unsigned int sysctl_sched_min_granularity = 1000000ULL; |
0bcdcf28 | 58 | unsigned int normalized_sysctl_sched_min_granularity = 1000000ULL; |
21805085 PZ |
59 | |
60 | /* | |
b2be5e96 PZ |
61 | * is kept at sysctl_sched_latency / sysctl_sched_min_granularity |
62 | */ | |
722aab0c | 63 | static unsigned int sched_nr_latency = 5; |
b2be5e96 PZ |
64 | |
65 | /* | |
2bba22c5 | 66 | * After fork, child runs first. If set to 0 (default) then |
b2be5e96 | 67 | * parent will (try to) run first. |
21805085 | 68 | */ |
2bba22c5 | 69 | unsigned int sysctl_sched_child_runs_first __read_mostly; |
bf0f6f24 | 70 | |
1799e35d IM |
71 | /* |
72 | * sys_sched_yield() compat mode | |
73 | * | |
74 | * This option switches the agressive yield implementation of the | |
75 | * old scheduler back on. | |
76 | */ | |
77 | unsigned int __read_mostly sysctl_sched_compat_yield; | |
78 | ||
bf0f6f24 IM |
79 | /* |
80 | * SCHED_OTHER wake-up granularity. | |
172e082a | 81 | * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds) |
bf0f6f24 IM |
82 | * |
83 | * This option delays the preemption effects of decoupled workloads | |
84 | * and reduces their over-scheduling. Synchronous workloads will still | |
85 | * have immediate wakeup/sleep latencies. | |
86 | */ | |
172e082a | 87 | unsigned int sysctl_sched_wakeup_granularity = 1000000UL; |
0bcdcf28 | 88 | unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL; |
bf0f6f24 | 89 | |
da84d961 IM |
90 | const_debug unsigned int sysctl_sched_migration_cost = 500000UL; |
91 | ||
a4c2f00f PZ |
92 | static const struct sched_class fair_sched_class; |
93 | ||
bf0f6f24 IM |
94 | /************************************************************** |
95 | * CFS operations on generic schedulable entities: | |
96 | */ | |
97 | ||
62160e3f | 98 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 99 | |
62160e3f | 100 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
101 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
102 | { | |
62160e3f | 103 | return cfs_rq->rq; |
bf0f6f24 IM |
104 | } |
105 | ||
62160e3f IM |
106 | /* An entity is a task if it doesn't "own" a runqueue */ |
107 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 108 | |
8f48894f PZ |
109 | static inline struct task_struct *task_of(struct sched_entity *se) |
110 | { | |
111 | #ifdef CONFIG_SCHED_DEBUG | |
112 | WARN_ON_ONCE(!entity_is_task(se)); | |
113 | #endif | |
114 | return container_of(se, struct task_struct, se); | |
115 | } | |
116 | ||
b758149c PZ |
117 | /* Walk up scheduling entities hierarchy */ |
118 | #define for_each_sched_entity(se) \ | |
119 | for (; se; se = se->parent) | |
120 | ||
121 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
122 | { | |
123 | return p->se.cfs_rq; | |
124 | } | |
125 | ||
126 | /* runqueue on which this entity is (to be) queued */ | |
127 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
128 | { | |
129 | return se->cfs_rq; | |
130 | } | |
131 | ||
132 | /* runqueue "owned" by this group */ | |
133 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
134 | { | |
135 | return grp->my_q; | |
136 | } | |
137 | ||
138 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
139 | * another cpu ('this_cpu') | |
140 | */ | |
141 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
142 | { | |
143 | return cfs_rq->tg->cfs_rq[this_cpu]; | |
144 | } | |
145 | ||
146 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
147 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
148 | list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
149 | ||
150 | /* Do the two (enqueued) entities belong to the same group ? */ | |
151 | static inline int | |
152 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
153 | { | |
154 | if (se->cfs_rq == pse->cfs_rq) | |
155 | return 1; | |
156 | ||
157 | return 0; | |
158 | } | |
159 | ||
160 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
161 | { | |
162 | return se->parent; | |
163 | } | |
164 | ||
464b7527 PZ |
165 | /* return depth at which a sched entity is present in the hierarchy */ |
166 | static inline int depth_se(struct sched_entity *se) | |
167 | { | |
168 | int depth = 0; | |
169 | ||
170 | for_each_sched_entity(se) | |
171 | depth++; | |
172 | ||
173 | return depth; | |
174 | } | |
175 | ||
176 | static void | |
177 | find_matching_se(struct sched_entity **se, struct sched_entity **pse) | |
178 | { | |
179 | int se_depth, pse_depth; | |
180 | ||
181 | /* | |
182 | * preemption test can be made between sibling entities who are in the | |
183 | * same cfs_rq i.e who have a common parent. Walk up the hierarchy of | |
184 | * both tasks until we find their ancestors who are siblings of common | |
185 | * parent. | |
186 | */ | |
187 | ||
188 | /* First walk up until both entities are at same depth */ | |
189 | se_depth = depth_se(*se); | |
190 | pse_depth = depth_se(*pse); | |
191 | ||
192 | while (se_depth > pse_depth) { | |
193 | se_depth--; | |
194 | *se = parent_entity(*se); | |
195 | } | |
196 | ||
197 | while (pse_depth > se_depth) { | |
198 | pse_depth--; | |
199 | *pse = parent_entity(*pse); | |
200 | } | |
201 | ||
202 | while (!is_same_group(*se, *pse)) { | |
203 | *se = parent_entity(*se); | |
204 | *pse = parent_entity(*pse); | |
205 | } | |
206 | } | |
207 | ||
8f48894f PZ |
208 | #else /* !CONFIG_FAIR_GROUP_SCHED */ |
209 | ||
210 | static inline struct task_struct *task_of(struct sched_entity *se) | |
211 | { | |
212 | return container_of(se, struct task_struct, se); | |
213 | } | |
bf0f6f24 | 214 | |
62160e3f IM |
215 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
216 | { | |
217 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
218 | } |
219 | ||
220 | #define entity_is_task(se) 1 | |
221 | ||
b758149c PZ |
222 | #define for_each_sched_entity(se) \ |
223 | for (; se; se = NULL) | |
bf0f6f24 | 224 | |
b758149c | 225 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) |
bf0f6f24 | 226 | { |
b758149c | 227 | return &task_rq(p)->cfs; |
bf0f6f24 IM |
228 | } |
229 | ||
b758149c PZ |
230 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) |
231 | { | |
232 | struct task_struct *p = task_of(se); | |
233 | struct rq *rq = task_rq(p); | |
234 | ||
235 | return &rq->cfs; | |
236 | } | |
237 | ||
238 | /* runqueue "owned" by this group */ | |
239 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
240 | { | |
241 | return NULL; | |
242 | } | |
243 | ||
244 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
245 | { | |
246 | return &cpu_rq(this_cpu)->cfs; | |
247 | } | |
248 | ||
249 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
250 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
251 | ||
252 | static inline int | |
253 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
254 | { | |
255 | return 1; | |
256 | } | |
257 | ||
258 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
259 | { | |
260 | return NULL; | |
261 | } | |
262 | ||
464b7527 PZ |
263 | static inline void |
264 | find_matching_se(struct sched_entity **se, struct sched_entity **pse) | |
265 | { | |
266 | } | |
267 | ||
b758149c PZ |
268 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
269 | ||
bf0f6f24 IM |
270 | |
271 | /************************************************************** | |
272 | * Scheduling class tree data structure manipulation methods: | |
273 | */ | |
274 | ||
0702e3eb | 275 | static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime) |
02e0431a | 276 | { |
368059a9 PZ |
277 | s64 delta = (s64)(vruntime - min_vruntime); |
278 | if (delta > 0) | |
02e0431a PZ |
279 | min_vruntime = vruntime; |
280 | ||
281 | return min_vruntime; | |
282 | } | |
283 | ||
0702e3eb | 284 | static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime) |
b0ffd246 PZ |
285 | { |
286 | s64 delta = (s64)(vruntime - min_vruntime); | |
287 | if (delta < 0) | |
288 | min_vruntime = vruntime; | |
289 | ||
290 | return min_vruntime; | |
291 | } | |
292 | ||
54fdc581 FC |
293 | static inline int entity_before(struct sched_entity *a, |
294 | struct sched_entity *b) | |
295 | { | |
296 | return (s64)(a->vruntime - b->vruntime) < 0; | |
297 | } | |
298 | ||
0702e3eb | 299 | static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
9014623c | 300 | { |
30cfdcfc | 301 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
302 | } |
303 | ||
1af5f730 PZ |
304 | static void update_min_vruntime(struct cfs_rq *cfs_rq) |
305 | { | |
306 | u64 vruntime = cfs_rq->min_vruntime; | |
307 | ||
308 | if (cfs_rq->curr) | |
309 | vruntime = cfs_rq->curr->vruntime; | |
310 | ||
311 | if (cfs_rq->rb_leftmost) { | |
312 | struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost, | |
313 | struct sched_entity, | |
314 | run_node); | |
315 | ||
e17036da | 316 | if (!cfs_rq->curr) |
1af5f730 PZ |
317 | vruntime = se->vruntime; |
318 | else | |
319 | vruntime = min_vruntime(vruntime, se->vruntime); | |
320 | } | |
321 | ||
322 | cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime); | |
323 | } | |
324 | ||
bf0f6f24 IM |
325 | /* |
326 | * Enqueue an entity into the rb-tree: | |
327 | */ | |
0702e3eb | 328 | static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
329 | { |
330 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
331 | struct rb_node *parent = NULL; | |
332 | struct sched_entity *entry; | |
9014623c | 333 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
334 | int leftmost = 1; |
335 | ||
336 | /* | |
337 | * Find the right place in the rbtree: | |
338 | */ | |
339 | while (*link) { | |
340 | parent = *link; | |
341 | entry = rb_entry(parent, struct sched_entity, run_node); | |
342 | /* | |
343 | * We dont care about collisions. Nodes with | |
344 | * the same key stay together. | |
345 | */ | |
9014623c | 346 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
347 | link = &parent->rb_left; |
348 | } else { | |
349 | link = &parent->rb_right; | |
350 | leftmost = 0; | |
351 | } | |
352 | } | |
353 | ||
354 | /* | |
355 | * Maintain a cache of leftmost tree entries (it is frequently | |
356 | * used): | |
357 | */ | |
1af5f730 | 358 | if (leftmost) |
57cb499d | 359 | cfs_rq->rb_leftmost = &se->run_node; |
bf0f6f24 IM |
360 | |
361 | rb_link_node(&se->run_node, parent, link); | |
362 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
363 | } |
364 | ||
0702e3eb | 365 | static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 366 | { |
3fe69747 PZ |
367 | if (cfs_rq->rb_leftmost == &se->run_node) { |
368 | struct rb_node *next_node; | |
3fe69747 PZ |
369 | |
370 | next_node = rb_next(&se->run_node); | |
371 | cfs_rq->rb_leftmost = next_node; | |
3fe69747 | 372 | } |
e9acbff6 | 373 | |
bf0f6f24 | 374 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
375 | } |
376 | ||
bf0f6f24 IM |
377 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) |
378 | { | |
f4b6755f PZ |
379 | struct rb_node *left = cfs_rq->rb_leftmost; |
380 | ||
381 | if (!left) | |
382 | return NULL; | |
383 | ||
384 | return rb_entry(left, struct sched_entity, run_node); | |
bf0f6f24 IM |
385 | } |
386 | ||
f4b6755f | 387 | static struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
aeb73b04 | 388 | { |
7eee3e67 | 389 | struct rb_node *last = rb_last(&cfs_rq->tasks_timeline); |
aeb73b04 | 390 | |
70eee74b BS |
391 | if (!last) |
392 | return NULL; | |
7eee3e67 IM |
393 | |
394 | return rb_entry(last, struct sched_entity, run_node); | |
aeb73b04 PZ |
395 | } |
396 | ||
bf0f6f24 IM |
397 | /************************************************************** |
398 | * Scheduling class statistics methods: | |
399 | */ | |
400 | ||
b2be5e96 | 401 | #ifdef CONFIG_SCHED_DEBUG |
acb4a848 | 402 | int sched_proc_update_handler(struct ctl_table *table, int write, |
8d65af78 | 403 | void __user *buffer, size_t *lenp, |
b2be5e96 PZ |
404 | loff_t *ppos) |
405 | { | |
8d65af78 | 406 | int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
acb4a848 | 407 | int factor = get_update_sysctl_factor(); |
b2be5e96 PZ |
408 | |
409 | if (ret || !write) | |
410 | return ret; | |
411 | ||
412 | sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency, | |
413 | sysctl_sched_min_granularity); | |
414 | ||
acb4a848 CE |
415 | #define WRT_SYSCTL(name) \ |
416 | (normalized_sysctl_##name = sysctl_##name / (factor)) | |
417 | WRT_SYSCTL(sched_min_granularity); | |
418 | WRT_SYSCTL(sched_latency); | |
419 | WRT_SYSCTL(sched_wakeup_granularity); | |
420 | WRT_SYSCTL(sched_shares_ratelimit); | |
421 | #undef WRT_SYSCTL | |
422 | ||
b2be5e96 PZ |
423 | return 0; |
424 | } | |
425 | #endif | |
647e7cac | 426 | |
a7be37ac | 427 | /* |
f9c0b095 | 428 | * delta /= w |
a7be37ac PZ |
429 | */ |
430 | static inline unsigned long | |
431 | calc_delta_fair(unsigned long delta, struct sched_entity *se) | |
432 | { | |
f9c0b095 PZ |
433 | if (unlikely(se->load.weight != NICE_0_LOAD)) |
434 | delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load); | |
a7be37ac PZ |
435 | |
436 | return delta; | |
437 | } | |
438 | ||
647e7cac IM |
439 | /* |
440 | * The idea is to set a period in which each task runs once. | |
441 | * | |
442 | * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch | |
443 | * this period because otherwise the slices get too small. | |
444 | * | |
445 | * p = (nr <= nl) ? l : l*nr/nl | |
446 | */ | |
4d78e7b6 PZ |
447 | static u64 __sched_period(unsigned long nr_running) |
448 | { | |
449 | u64 period = sysctl_sched_latency; | |
b2be5e96 | 450 | unsigned long nr_latency = sched_nr_latency; |
4d78e7b6 PZ |
451 | |
452 | if (unlikely(nr_running > nr_latency)) { | |
4bf0b771 | 453 | period = sysctl_sched_min_granularity; |
4d78e7b6 | 454 | period *= nr_running; |
4d78e7b6 PZ |
455 | } |
456 | ||
457 | return period; | |
458 | } | |
459 | ||
647e7cac IM |
460 | /* |
461 | * We calculate the wall-time slice from the period by taking a part | |
462 | * proportional to the weight. | |
463 | * | |
f9c0b095 | 464 | * s = p*P[w/rw] |
647e7cac | 465 | */ |
6d0f0ebd | 466 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 467 | { |
0a582440 | 468 | u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq); |
f9c0b095 | 469 | |
0a582440 | 470 | for_each_sched_entity(se) { |
6272d68c | 471 | struct load_weight *load; |
3104bf03 | 472 | struct load_weight lw; |
6272d68c LM |
473 | |
474 | cfs_rq = cfs_rq_of(se); | |
475 | load = &cfs_rq->load; | |
f9c0b095 | 476 | |
0a582440 | 477 | if (unlikely(!se->on_rq)) { |
3104bf03 | 478 | lw = cfs_rq->load; |
0a582440 MG |
479 | |
480 | update_load_add(&lw, se->load.weight); | |
481 | load = &lw; | |
482 | } | |
483 | slice = calc_delta_mine(slice, se->load.weight, load); | |
484 | } | |
485 | return slice; | |
bf0f6f24 IM |
486 | } |
487 | ||
647e7cac | 488 | /* |
ac884dec | 489 | * We calculate the vruntime slice of a to be inserted task |
647e7cac | 490 | * |
f9c0b095 | 491 | * vs = s/w |
647e7cac | 492 | */ |
f9c0b095 | 493 | static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
67e9fb2a | 494 | { |
f9c0b095 | 495 | return calc_delta_fair(sched_slice(cfs_rq, se), se); |
a7be37ac PZ |
496 | } |
497 | ||
bf0f6f24 IM |
498 | /* |
499 | * Update the current task's runtime statistics. Skip current tasks that | |
500 | * are not in our scheduling class. | |
501 | */ | |
502 | static inline void | |
8ebc91d9 IM |
503 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
504 | unsigned long delta_exec) | |
bf0f6f24 | 505 | { |
bbdba7c0 | 506 | unsigned long delta_exec_weighted; |
bf0f6f24 | 507 | |
8179ca23 | 508 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
509 | |
510 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 511 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
a7be37ac | 512 | delta_exec_weighted = calc_delta_fair(delta_exec, curr); |
e9acbff6 | 513 | curr->vruntime += delta_exec_weighted; |
1af5f730 | 514 | update_min_vruntime(cfs_rq); |
bf0f6f24 IM |
515 | } |
516 | ||
b7cc0896 | 517 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 518 | { |
429d43bc | 519 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 520 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
521 | unsigned long delta_exec; |
522 | ||
523 | if (unlikely(!curr)) | |
524 | return; | |
525 | ||
526 | /* | |
527 | * Get the amount of time the current task was running | |
528 | * since the last time we changed load (this cannot | |
529 | * overflow on 32 bits): | |
530 | */ | |
8ebc91d9 | 531 | delta_exec = (unsigned long)(now - curr->exec_start); |
34f28ecd PZ |
532 | if (!delta_exec) |
533 | return; | |
bf0f6f24 | 534 | |
8ebc91d9 IM |
535 | __update_curr(cfs_rq, curr, delta_exec); |
536 | curr->exec_start = now; | |
d842de87 SV |
537 | |
538 | if (entity_is_task(curr)) { | |
539 | struct task_struct *curtask = task_of(curr); | |
540 | ||
f977bb49 | 541 | trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime); |
d842de87 | 542 | cpuacct_charge(curtask, delta_exec); |
f06febc9 | 543 | account_group_exec_runtime(curtask, delta_exec); |
d842de87 | 544 | } |
bf0f6f24 IM |
545 | } |
546 | ||
547 | static inline void | |
5870db5b | 548 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 549 | { |
d281918d | 550 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
551 | } |
552 | ||
bf0f6f24 IM |
553 | /* |
554 | * Task is being enqueued - update stats: | |
555 | */ | |
d2417e5a | 556 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 557 | { |
bf0f6f24 IM |
558 | /* |
559 | * Are we enqueueing a waiting task? (for current tasks | |
560 | * a dequeue/enqueue event is a NOP) | |
561 | */ | |
429d43bc | 562 | if (se != cfs_rq->curr) |
5870db5b | 563 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
564 | } |
565 | ||
bf0f6f24 | 566 | static void |
9ef0a961 | 567 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 568 | { |
bbdba7c0 IM |
569 | schedstat_set(se->wait_max, max(se->wait_max, |
570 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6d082592 AV |
571 | schedstat_set(se->wait_count, se->wait_count + 1); |
572 | schedstat_set(se->wait_sum, se->wait_sum + | |
573 | rq_of(cfs_rq)->clock - se->wait_start); | |
768d0c27 PZ |
574 | #ifdef CONFIG_SCHEDSTATS |
575 | if (entity_is_task(se)) { | |
576 | trace_sched_stat_wait(task_of(se), | |
577 | rq_of(cfs_rq)->clock - se->wait_start); | |
578 | } | |
579 | #endif | |
e1f84508 | 580 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
581 | } |
582 | ||
583 | static inline void | |
19b6a2e3 | 584 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 585 | { |
bf0f6f24 IM |
586 | /* |
587 | * Mark the end of the wait period if dequeueing a | |
588 | * waiting task: | |
589 | */ | |
429d43bc | 590 | if (se != cfs_rq->curr) |
9ef0a961 | 591 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
592 | } |
593 | ||
594 | /* | |
595 | * We are picking a new current task - update its stats: | |
596 | */ | |
597 | static inline void | |
79303e9e | 598 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
599 | { |
600 | /* | |
601 | * We are starting a new run period: | |
602 | */ | |
d281918d | 603 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
604 | } |
605 | ||
bf0f6f24 IM |
606 | /************************************************** |
607 | * Scheduling class queueing methods: | |
608 | */ | |
609 | ||
c09595f6 PZ |
610 | #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED |
611 | static void | |
612 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
613 | { | |
614 | cfs_rq->task_weight += weight; | |
615 | } | |
616 | #else | |
617 | static inline void | |
618 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
619 | { | |
620 | } | |
621 | #endif | |
622 | ||
30cfdcfc DA |
623 | static void |
624 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
625 | { | |
626 | update_load_add(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
627 | if (!parent_entity(se)) |
628 | inc_cpu_load(rq_of(cfs_rq), se->load.weight); | |
b87f1724 | 629 | if (entity_is_task(se)) { |
c09595f6 | 630 | add_cfs_task_weight(cfs_rq, se->load.weight); |
b87f1724 BR |
631 | list_add(&se->group_node, &cfs_rq->tasks); |
632 | } | |
30cfdcfc DA |
633 | cfs_rq->nr_running++; |
634 | se->on_rq = 1; | |
635 | } | |
636 | ||
637 | static void | |
638 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
639 | { | |
640 | update_load_sub(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
641 | if (!parent_entity(se)) |
642 | dec_cpu_load(rq_of(cfs_rq), se->load.weight); | |
b87f1724 | 643 | if (entity_is_task(se)) { |
c09595f6 | 644 | add_cfs_task_weight(cfs_rq, -se->load.weight); |
b87f1724 BR |
645 | list_del_init(&se->group_node); |
646 | } | |
30cfdcfc DA |
647 | cfs_rq->nr_running--; |
648 | se->on_rq = 0; | |
649 | } | |
650 | ||
2396af69 | 651 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 652 | { |
bf0f6f24 | 653 | #ifdef CONFIG_SCHEDSTATS |
e414314c PZ |
654 | struct task_struct *tsk = NULL; |
655 | ||
656 | if (entity_is_task(se)) | |
657 | tsk = task_of(se); | |
658 | ||
bf0f6f24 | 659 | if (se->sleep_start) { |
d281918d | 660 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
661 | |
662 | if ((s64)delta < 0) | |
663 | delta = 0; | |
664 | ||
665 | if (unlikely(delta > se->sleep_max)) | |
666 | se->sleep_max = delta; | |
667 | ||
668 | se->sleep_start = 0; | |
669 | se->sum_sleep_runtime += delta; | |
9745512c | 670 | |
768d0c27 | 671 | if (tsk) { |
e414314c | 672 | account_scheduler_latency(tsk, delta >> 10, 1); |
768d0c27 PZ |
673 | trace_sched_stat_sleep(tsk, delta); |
674 | } | |
bf0f6f24 IM |
675 | } |
676 | if (se->block_start) { | |
d281918d | 677 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
678 | |
679 | if ((s64)delta < 0) | |
680 | delta = 0; | |
681 | ||
682 | if (unlikely(delta > se->block_max)) | |
683 | se->block_max = delta; | |
684 | ||
685 | se->block_start = 0; | |
686 | se->sum_sleep_runtime += delta; | |
30084fbd | 687 | |
e414314c | 688 | if (tsk) { |
8f0dfc34 AV |
689 | if (tsk->in_iowait) { |
690 | se->iowait_sum += delta; | |
691 | se->iowait_count++; | |
768d0c27 | 692 | trace_sched_stat_iowait(tsk, delta); |
8f0dfc34 AV |
693 | } |
694 | ||
e414314c PZ |
695 | /* |
696 | * Blocking time is in units of nanosecs, so shift by | |
697 | * 20 to get a milliseconds-range estimation of the | |
698 | * amount of time that the task spent sleeping: | |
699 | */ | |
700 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
701 | profile_hits(SLEEP_PROFILING, | |
702 | (void *)get_wchan(tsk), | |
703 | delta >> 20); | |
704 | } | |
705 | account_scheduler_latency(tsk, delta >> 10, 0); | |
30084fbd | 706 | } |
bf0f6f24 IM |
707 | } |
708 | #endif | |
709 | } | |
710 | ||
ddc97297 PZ |
711 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
712 | { | |
713 | #ifdef CONFIG_SCHED_DEBUG | |
714 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
715 | ||
716 | if (d < 0) | |
717 | d = -d; | |
718 | ||
719 | if (d > 3*sysctl_sched_latency) | |
720 | schedstat_inc(cfs_rq, nr_spread_over); | |
721 | #endif | |
722 | } | |
723 | ||
aeb73b04 PZ |
724 | static void |
725 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
726 | { | |
1af5f730 | 727 | u64 vruntime = cfs_rq->min_vruntime; |
94dfb5e7 | 728 | |
2cb8600e PZ |
729 | /* |
730 | * The 'current' period is already promised to the current tasks, | |
731 | * however the extra weight of the new task will slow them down a | |
732 | * little, place the new task so that it fits in the slot that | |
733 | * stays open at the end. | |
734 | */ | |
94dfb5e7 | 735 | if (initial && sched_feat(START_DEBIT)) |
f9c0b095 | 736 | vruntime += sched_vslice(cfs_rq, se); |
aeb73b04 | 737 | |
a2e7a7eb MG |
738 | /* sleeps up to a single latency don't count. */ |
739 | if (!initial && sched_feat(FAIR_SLEEPERS)) { | |
740 | unsigned long thresh = sysctl_sched_latency; | |
a7be37ac | 741 | |
a2e7a7eb MG |
742 | /* |
743 | * Convert the sleeper threshold into virtual time. | |
744 | * SCHED_IDLE is a special sub-class. We care about | |
745 | * fairness only relative to other SCHED_IDLE tasks, | |
746 | * all of which have the same weight. | |
747 | */ | |
748 | if (sched_feat(NORMALIZED_SLEEPER) && (!entity_is_task(se) || | |
749 | task_of(se)->policy != SCHED_IDLE)) | |
750 | thresh = calc_delta_fair(thresh, se); | |
a7be37ac | 751 | |
a2e7a7eb MG |
752 | /* |
753 | * Halve their sleep time's effect, to allow | |
754 | * for a gentler effect of sleepers: | |
755 | */ | |
756 | if (sched_feat(GENTLE_FAIR_SLEEPERS)) | |
757 | thresh >>= 1; | |
51e0304c | 758 | |
a2e7a7eb | 759 | vruntime -= thresh; |
aeb73b04 PZ |
760 | } |
761 | ||
b5d9d734 MG |
762 | /* ensure we never gain time by being placed backwards. */ |
763 | vruntime = max_vruntime(se->vruntime, vruntime); | |
764 | ||
67e9fb2a | 765 | se->vruntime = vruntime; |
aeb73b04 PZ |
766 | } |
767 | ||
bf0f6f24 | 768 | static void |
83b699ed | 769 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
770 | { |
771 | /* | |
a2a2d680 | 772 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 773 | */ |
b7cc0896 | 774 | update_curr(cfs_rq); |
a992241d | 775 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 | 776 | |
e9acbff6 | 777 | if (wakeup) { |
aeb73b04 | 778 | place_entity(cfs_rq, se, 0); |
2396af69 | 779 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 780 | } |
bf0f6f24 | 781 | |
d2417e5a | 782 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 783 | check_spread(cfs_rq, se); |
83b699ed SV |
784 | if (se != cfs_rq->curr) |
785 | __enqueue_entity(cfs_rq, se); | |
bf0f6f24 IM |
786 | } |
787 | ||
a571bbea | 788 | static void __clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se) |
2002c695 | 789 | { |
de69a80b | 790 | if (!se || cfs_rq->last == se) |
2002c695 PZ |
791 | cfs_rq->last = NULL; |
792 | ||
de69a80b | 793 | if (!se || cfs_rq->next == se) |
2002c695 PZ |
794 | cfs_rq->next = NULL; |
795 | } | |
796 | ||
a571bbea PZ |
797 | static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se) |
798 | { | |
799 | for_each_sched_entity(se) | |
800 | __clear_buddies(cfs_rq_of(se), se); | |
801 | } | |
802 | ||
bf0f6f24 | 803 | static void |
525c2716 | 804 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 805 | { |
a2a2d680 DA |
806 | /* |
807 | * Update run-time statistics of the 'current'. | |
808 | */ | |
809 | update_curr(cfs_rq); | |
810 | ||
19b6a2e3 | 811 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 812 | if (sleep) { |
67e9fb2a | 813 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
814 | if (entity_is_task(se)) { |
815 | struct task_struct *tsk = task_of(se); | |
816 | ||
817 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 818 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 819 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 820 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 821 | } |
db36cc7d | 822 | #endif |
67e9fb2a PZ |
823 | } |
824 | ||
2002c695 | 825 | clear_buddies(cfs_rq, se); |
4793241b | 826 | |
83b699ed | 827 | if (se != cfs_rq->curr) |
30cfdcfc DA |
828 | __dequeue_entity(cfs_rq, se); |
829 | account_entity_dequeue(cfs_rq, se); | |
1af5f730 | 830 | update_min_vruntime(cfs_rq); |
bf0f6f24 IM |
831 | } |
832 | ||
833 | /* | |
834 | * Preempt the current task with a newly woken task if needed: | |
835 | */ | |
7c92e54f | 836 | static void |
2e09bf55 | 837 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 838 | { |
11697830 PZ |
839 | unsigned long ideal_runtime, delta_exec; |
840 | ||
6d0f0ebd | 841 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 | 842 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
a9f3e2b5 | 843 | if (delta_exec > ideal_runtime) { |
bf0f6f24 | 844 | resched_task(rq_of(cfs_rq)->curr); |
a9f3e2b5 MG |
845 | /* |
846 | * The current task ran long enough, ensure it doesn't get | |
847 | * re-elected due to buddy favours. | |
848 | */ | |
849 | clear_buddies(cfs_rq, curr); | |
f685ceac MG |
850 | return; |
851 | } | |
852 | ||
853 | /* | |
854 | * Ensure that a task that missed wakeup preemption by a | |
855 | * narrow margin doesn't have to wait for a full slice. | |
856 | * This also mitigates buddy induced latencies under load. | |
857 | */ | |
858 | if (!sched_feat(WAKEUP_PREEMPT)) | |
859 | return; | |
860 | ||
861 | if (delta_exec < sysctl_sched_min_granularity) | |
862 | return; | |
863 | ||
864 | if (cfs_rq->nr_running > 1) { | |
865 | struct sched_entity *se = __pick_next_entity(cfs_rq); | |
866 | s64 delta = curr->vruntime - se->vruntime; | |
867 | ||
868 | if (delta > ideal_runtime) | |
869 | resched_task(rq_of(cfs_rq)->curr); | |
a9f3e2b5 | 870 | } |
bf0f6f24 IM |
871 | } |
872 | ||
83b699ed | 873 | static void |
8494f412 | 874 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 875 | { |
83b699ed SV |
876 | /* 'current' is not kept within the tree. */ |
877 | if (se->on_rq) { | |
878 | /* | |
879 | * Any task has to be enqueued before it get to execute on | |
880 | * a CPU. So account for the time it spent waiting on the | |
881 | * runqueue. | |
882 | */ | |
883 | update_stats_wait_end(cfs_rq, se); | |
884 | __dequeue_entity(cfs_rq, se); | |
885 | } | |
886 | ||
79303e9e | 887 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 888 | cfs_rq->curr = se; |
eba1ed4b IM |
889 | #ifdef CONFIG_SCHEDSTATS |
890 | /* | |
891 | * Track our maximum slice length, if the CPU's load is at | |
892 | * least twice that of our own weight (i.e. dont track it | |
893 | * when there are only lesser-weight tasks around): | |
894 | */ | |
495eca49 | 895 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
896 | se->slice_max = max(se->slice_max, |
897 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
898 | } | |
899 | #endif | |
4a55b450 | 900 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
901 | } |
902 | ||
3f3a4904 PZ |
903 | static int |
904 | wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se); | |
905 | ||
f4b6755f | 906 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
aa2ac252 | 907 | { |
f4b6755f | 908 | struct sched_entity *se = __pick_next_entity(cfs_rq); |
f685ceac | 909 | struct sched_entity *left = se; |
f4b6755f | 910 | |
f685ceac MG |
911 | if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, left) < 1) |
912 | se = cfs_rq->next; | |
aa2ac252 | 913 | |
f685ceac MG |
914 | /* |
915 | * Prefer last buddy, try to return the CPU to a preempted task. | |
916 | */ | |
917 | if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, left) < 1) | |
918 | se = cfs_rq->last; | |
919 | ||
920 | clear_buddies(cfs_rq, se); | |
4793241b PZ |
921 | |
922 | return se; | |
aa2ac252 PZ |
923 | } |
924 | ||
ab6cde26 | 925 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
926 | { |
927 | /* | |
928 | * If still on the runqueue then deactivate_task() | |
929 | * was not called and update_curr() has to be done: | |
930 | */ | |
931 | if (prev->on_rq) | |
b7cc0896 | 932 | update_curr(cfs_rq); |
bf0f6f24 | 933 | |
ddc97297 | 934 | check_spread(cfs_rq, prev); |
30cfdcfc | 935 | if (prev->on_rq) { |
5870db5b | 936 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
937 | /* Put 'current' back into the tree. */ |
938 | __enqueue_entity(cfs_rq, prev); | |
939 | } | |
429d43bc | 940 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
941 | } |
942 | ||
8f4d37ec PZ |
943 | static void |
944 | entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) | |
bf0f6f24 | 945 | { |
bf0f6f24 | 946 | /* |
30cfdcfc | 947 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 948 | */ |
30cfdcfc | 949 | update_curr(cfs_rq); |
bf0f6f24 | 950 | |
8f4d37ec PZ |
951 | #ifdef CONFIG_SCHED_HRTICK |
952 | /* | |
953 | * queued ticks are scheduled to match the slice, so don't bother | |
954 | * validating it and just reschedule. | |
955 | */ | |
983ed7a6 HH |
956 | if (queued) { |
957 | resched_task(rq_of(cfs_rq)->curr); | |
958 | return; | |
959 | } | |
8f4d37ec PZ |
960 | /* |
961 | * don't let the period tick interfere with the hrtick preemption | |
962 | */ | |
963 | if (!sched_feat(DOUBLE_TICK) && | |
964 | hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) | |
965 | return; | |
966 | #endif | |
967 | ||
ce6c1311 | 968 | if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT)) |
2e09bf55 | 969 | check_preempt_tick(cfs_rq, curr); |
bf0f6f24 IM |
970 | } |
971 | ||
972 | /************************************************** | |
973 | * CFS operations on tasks: | |
974 | */ | |
975 | ||
8f4d37ec PZ |
976 | #ifdef CONFIG_SCHED_HRTICK |
977 | static void hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
978 | { | |
8f4d37ec PZ |
979 | struct sched_entity *se = &p->se; |
980 | struct cfs_rq *cfs_rq = cfs_rq_of(se); | |
981 | ||
982 | WARN_ON(task_rq(p) != rq); | |
983 | ||
984 | if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) { | |
985 | u64 slice = sched_slice(cfs_rq, se); | |
986 | u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime; | |
987 | s64 delta = slice - ran; | |
988 | ||
989 | if (delta < 0) { | |
990 | if (rq->curr == p) | |
991 | resched_task(p); | |
992 | return; | |
993 | } | |
994 | ||
995 | /* | |
996 | * Don't schedule slices shorter than 10000ns, that just | |
997 | * doesn't make sense. Rely on vruntime for fairness. | |
998 | */ | |
31656519 | 999 | if (rq->curr != p) |
157124c1 | 1000 | delta = max_t(s64, 10000LL, delta); |
8f4d37ec | 1001 | |
31656519 | 1002 | hrtick_start(rq, delta); |
8f4d37ec PZ |
1003 | } |
1004 | } | |
a4c2f00f PZ |
1005 | |
1006 | /* | |
1007 | * called from enqueue/dequeue and updates the hrtick when the | |
1008 | * current task is from our class and nr_running is low enough | |
1009 | * to matter. | |
1010 | */ | |
1011 | static void hrtick_update(struct rq *rq) | |
1012 | { | |
1013 | struct task_struct *curr = rq->curr; | |
1014 | ||
1015 | if (curr->sched_class != &fair_sched_class) | |
1016 | return; | |
1017 | ||
1018 | if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency) | |
1019 | hrtick_start_fair(rq, curr); | |
1020 | } | |
55e12e5e | 1021 | #else /* !CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1022 | static inline void |
1023 | hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
1024 | { | |
1025 | } | |
a4c2f00f PZ |
1026 | |
1027 | static inline void hrtick_update(struct rq *rq) | |
1028 | { | |
1029 | } | |
8f4d37ec PZ |
1030 | #endif |
1031 | ||
bf0f6f24 IM |
1032 | /* |
1033 | * The enqueue_task method is called before nr_running is | |
1034 | * increased. Here we update the fair scheduling stats and | |
1035 | * then put the task into the rbtree: | |
1036 | */ | |
fd390f6a | 1037 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
1038 | { |
1039 | struct cfs_rq *cfs_rq; | |
62fb1851 | 1040 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
1041 | |
1042 | for_each_sched_entity(se) { | |
62fb1851 | 1043 | if (se->on_rq) |
bf0f6f24 IM |
1044 | break; |
1045 | cfs_rq = cfs_rq_of(se); | |
83b699ed | 1046 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 1047 | wakeup = 1; |
bf0f6f24 | 1048 | } |
8f4d37ec | 1049 | |
a4c2f00f | 1050 | hrtick_update(rq); |
bf0f6f24 IM |
1051 | } |
1052 | ||
1053 | /* | |
1054 | * The dequeue_task method is called before nr_running is | |
1055 | * decreased. We remove the task from the rbtree and | |
1056 | * update the fair scheduling stats: | |
1057 | */ | |
f02231e5 | 1058 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
1059 | { |
1060 | struct cfs_rq *cfs_rq; | |
62fb1851 | 1061 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
1062 | |
1063 | for_each_sched_entity(se) { | |
1064 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 1065 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 | 1066 | /* Don't dequeue parent if it has other entities besides us */ |
62fb1851 | 1067 | if (cfs_rq->load.weight) |
bf0f6f24 | 1068 | break; |
b9fa3df3 | 1069 | sleep = 1; |
bf0f6f24 | 1070 | } |
8f4d37ec | 1071 | |
a4c2f00f | 1072 | hrtick_update(rq); |
bf0f6f24 IM |
1073 | } |
1074 | ||
1075 | /* | |
1799e35d IM |
1076 | * sched_yield() support is very simple - we dequeue and enqueue. |
1077 | * | |
1078 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 1079 | */ |
4530d7ab | 1080 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 1081 | { |
db292ca3 IM |
1082 | struct task_struct *curr = rq->curr; |
1083 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
1084 | struct sched_entity *rightmost, *se = &curr->se; | |
bf0f6f24 IM |
1085 | |
1086 | /* | |
1799e35d IM |
1087 | * Are we the only task in the tree? |
1088 | */ | |
1089 | if (unlikely(cfs_rq->nr_running == 1)) | |
1090 | return; | |
1091 | ||
2002c695 PZ |
1092 | clear_buddies(cfs_rq, se); |
1093 | ||
db292ca3 | 1094 | if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) { |
3e51f33f | 1095 | update_rq_clock(rq); |
1799e35d | 1096 | /* |
a2a2d680 | 1097 | * Update run-time statistics of the 'current'. |
1799e35d | 1098 | */ |
2b1e315d | 1099 | update_curr(cfs_rq); |
1799e35d IM |
1100 | |
1101 | return; | |
1102 | } | |
1103 | /* | |
1104 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 1105 | */ |
2b1e315d | 1106 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
1107 | /* |
1108 | * Already in the rightmost position? | |
1109 | */ | |
54fdc581 | 1110 | if (unlikely(!rightmost || entity_before(rightmost, se))) |
1799e35d IM |
1111 | return; |
1112 | ||
1113 | /* | |
1114 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
1115 | * Upon rescheduling, sched_class::put_prev_task() will place |
1116 | * 'current' within the tree based on its new key value. | |
1799e35d | 1117 | */ |
30cfdcfc | 1118 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
1119 | } |
1120 | ||
e7693a36 | 1121 | #ifdef CONFIG_SMP |
098fb9db | 1122 | |
bb3469ac | 1123 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f5bfb7d9 PZ |
1124 | /* |
1125 | * effective_load() calculates the load change as seen from the root_task_group | |
1126 | * | |
1127 | * Adding load to a group doesn't make a group heavier, but can cause movement | |
1128 | * of group shares between cpus. Assuming the shares were perfectly aligned one | |
1129 | * can calculate the shift in shares. | |
1130 | * | |
1131 | * The problem is that perfectly aligning the shares is rather expensive, hence | |
1132 | * we try to avoid doing that too often - see update_shares(), which ratelimits | |
1133 | * this change. | |
1134 | * | |
1135 | * We compensate this by not only taking the current delta into account, but | |
1136 | * also considering the delta between when the shares were last adjusted and | |
1137 | * now. | |
1138 | * | |
1139 | * We still saw a performance dip, some tracing learned us that between | |
1140 | * cgroup:/ and cgroup:/foo balancing the number of affine wakeups increased | |
1141 | * significantly. Therefore try to bias the error in direction of failing | |
1142 | * the affine wakeup. | |
1143 | * | |
1144 | */ | |
f1d239f7 PZ |
1145 | static long effective_load(struct task_group *tg, int cpu, |
1146 | long wl, long wg) | |
bb3469ac | 1147 | { |
4be9daaa | 1148 | struct sched_entity *se = tg->se[cpu]; |
f1d239f7 PZ |
1149 | |
1150 | if (!tg->parent) | |
1151 | return wl; | |
1152 | ||
f5bfb7d9 PZ |
1153 | /* |
1154 | * By not taking the decrease of shares on the other cpu into | |
1155 | * account our error leans towards reducing the affine wakeups. | |
1156 | */ | |
1157 | if (!wl && sched_feat(ASYM_EFF_LOAD)) | |
1158 | return wl; | |
1159 | ||
4be9daaa | 1160 | for_each_sched_entity(se) { |
cb5ef42a | 1161 | long S, rw, s, a, b; |
940959e9 PZ |
1162 | long more_w; |
1163 | ||
1164 | /* | |
1165 | * Instead of using this increment, also add the difference | |
1166 | * between when the shares were last updated and now. | |
1167 | */ | |
1168 | more_w = se->my_q->load.weight - se->my_q->rq_weight; | |
1169 | wl += more_w; | |
1170 | wg += more_w; | |
4be9daaa PZ |
1171 | |
1172 | S = se->my_q->tg->shares; | |
1173 | s = se->my_q->shares; | |
f1d239f7 | 1174 | rw = se->my_q->rq_weight; |
bb3469ac | 1175 | |
cb5ef42a PZ |
1176 | a = S*(rw + wl); |
1177 | b = S*rw + s*wg; | |
4be9daaa | 1178 | |
940959e9 PZ |
1179 | wl = s*(a-b); |
1180 | ||
1181 | if (likely(b)) | |
1182 | wl /= b; | |
1183 | ||
83378269 PZ |
1184 | /* |
1185 | * Assume the group is already running and will | |
1186 | * thus already be accounted for in the weight. | |
1187 | * | |
1188 | * That is, moving shares between CPUs, does not | |
1189 | * alter the group weight. | |
1190 | */ | |
4be9daaa | 1191 | wg = 0; |
4be9daaa | 1192 | } |
bb3469ac | 1193 | |
4be9daaa | 1194 | return wl; |
bb3469ac | 1195 | } |
4be9daaa | 1196 | |
bb3469ac | 1197 | #else |
4be9daaa | 1198 | |
83378269 PZ |
1199 | static inline unsigned long effective_load(struct task_group *tg, int cpu, |
1200 | unsigned long wl, unsigned long wg) | |
4be9daaa | 1201 | { |
83378269 | 1202 | return wl; |
bb3469ac | 1203 | } |
4be9daaa | 1204 | |
bb3469ac PZ |
1205 | #endif |
1206 | ||
c88d5910 | 1207 | static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync) |
098fb9db | 1208 | { |
c88d5910 PZ |
1209 | struct task_struct *curr = current; |
1210 | unsigned long this_load, load; | |
1211 | int idx, this_cpu, prev_cpu; | |
098fb9db | 1212 | unsigned long tl_per_task; |
c88d5910 PZ |
1213 | unsigned int imbalance; |
1214 | struct task_group *tg; | |
83378269 | 1215 | unsigned long weight; |
b3137bc8 | 1216 | int balanced; |
098fb9db | 1217 | |
c88d5910 PZ |
1218 | idx = sd->wake_idx; |
1219 | this_cpu = smp_processor_id(); | |
1220 | prev_cpu = task_cpu(p); | |
1221 | load = source_load(prev_cpu, idx); | |
1222 | this_load = target_load(this_cpu, idx); | |
098fb9db | 1223 | |
e69b0f1b PZ |
1224 | if (sync) { |
1225 | if (sched_feat(SYNC_LESS) && | |
1226 | (curr->se.avg_overlap > sysctl_sched_migration_cost || | |
1227 | p->se.avg_overlap > sysctl_sched_migration_cost)) | |
1228 | sync = 0; | |
1229 | } else { | |
1230 | if (sched_feat(SYNC_MORE) && | |
1231 | (curr->se.avg_overlap < sysctl_sched_migration_cost && | |
1232 | p->se.avg_overlap < sysctl_sched_migration_cost)) | |
1233 | sync = 1; | |
1234 | } | |
fc631c82 | 1235 | |
b3137bc8 MG |
1236 | /* |
1237 | * If sync wakeup then subtract the (maximum possible) | |
1238 | * effect of the currently running task from the load | |
1239 | * of the current CPU: | |
1240 | */ | |
83378269 PZ |
1241 | if (sync) { |
1242 | tg = task_group(current); | |
1243 | weight = current->se.load.weight; | |
1244 | ||
c88d5910 | 1245 | this_load += effective_load(tg, this_cpu, -weight, -weight); |
83378269 PZ |
1246 | load += effective_load(tg, prev_cpu, 0, -weight); |
1247 | } | |
b3137bc8 | 1248 | |
83378269 PZ |
1249 | tg = task_group(p); |
1250 | weight = p->se.load.weight; | |
b3137bc8 | 1251 | |
c88d5910 PZ |
1252 | imbalance = 100 + (sd->imbalance_pct - 100) / 2; |
1253 | ||
71a29aa7 PZ |
1254 | /* |
1255 | * In low-load situations, where prev_cpu is idle and this_cpu is idle | |
c88d5910 PZ |
1256 | * due to the sync cause above having dropped this_load to 0, we'll |
1257 | * always have an imbalance, but there's really nothing you can do | |
1258 | * about that, so that's good too. | |
71a29aa7 PZ |
1259 | * |
1260 | * Otherwise check if either cpus are near enough in load to allow this | |
1261 | * task to be woken on this_cpu. | |
1262 | */ | |
c88d5910 PZ |
1263 | balanced = !this_load || |
1264 | 100*(this_load + effective_load(tg, this_cpu, weight, weight)) <= | |
83378269 | 1265 | imbalance*(load + effective_load(tg, prev_cpu, 0, weight)); |
b3137bc8 | 1266 | |
098fb9db | 1267 | /* |
4ae7d5ce IM |
1268 | * If the currently running task will sleep within |
1269 | * a reasonable amount of time then attract this newly | |
1270 | * woken task: | |
098fb9db | 1271 | */ |
2fb7635c PZ |
1272 | if (sync && balanced) |
1273 | return 1; | |
098fb9db IM |
1274 | |
1275 | schedstat_inc(p, se.nr_wakeups_affine_attempts); | |
1276 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
1277 | ||
c88d5910 PZ |
1278 | if (balanced || |
1279 | (this_load <= load && | |
1280 | this_load + target_load(prev_cpu, idx) <= tl_per_task)) { | |
098fb9db IM |
1281 | /* |
1282 | * This domain has SD_WAKE_AFFINE and | |
1283 | * p is cache cold in this domain, and | |
1284 | * there is no bad imbalance. | |
1285 | */ | |
c88d5910 | 1286 | schedstat_inc(sd, ttwu_move_affine); |
098fb9db IM |
1287 | schedstat_inc(p, se.nr_wakeups_affine); |
1288 | ||
1289 | return 1; | |
1290 | } | |
1291 | return 0; | |
1292 | } | |
1293 | ||
aaee1203 PZ |
1294 | /* |
1295 | * find_idlest_group finds and returns the least busy CPU group within the | |
1296 | * domain. | |
1297 | */ | |
1298 | static struct sched_group * | |
78e7ed53 | 1299 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, |
5158f4e4 | 1300 | int this_cpu, int load_idx) |
e7693a36 | 1301 | { |
aaee1203 PZ |
1302 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; |
1303 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
aaee1203 | 1304 | int imbalance = 100 + (sd->imbalance_pct-100)/2; |
e7693a36 | 1305 | |
aaee1203 PZ |
1306 | do { |
1307 | unsigned long load, avg_load; | |
1308 | int local_group; | |
1309 | int i; | |
e7693a36 | 1310 | |
aaee1203 PZ |
1311 | /* Skip over this group if it has no CPUs allowed */ |
1312 | if (!cpumask_intersects(sched_group_cpus(group), | |
1313 | &p->cpus_allowed)) | |
1314 | continue; | |
1315 | ||
1316 | local_group = cpumask_test_cpu(this_cpu, | |
1317 | sched_group_cpus(group)); | |
1318 | ||
1319 | /* Tally up the load of all CPUs in the group */ | |
1320 | avg_load = 0; | |
1321 | ||
1322 | for_each_cpu(i, sched_group_cpus(group)) { | |
1323 | /* Bias balancing toward cpus of our domain */ | |
1324 | if (local_group) | |
1325 | load = source_load(i, load_idx); | |
1326 | else | |
1327 | load = target_load(i, load_idx); | |
1328 | ||
1329 | avg_load += load; | |
1330 | } | |
1331 | ||
1332 | /* Adjust by relative CPU power of the group */ | |
1333 | avg_load = (avg_load * SCHED_LOAD_SCALE) / group->cpu_power; | |
1334 | ||
1335 | if (local_group) { | |
1336 | this_load = avg_load; | |
1337 | this = group; | |
1338 | } else if (avg_load < min_load) { | |
1339 | min_load = avg_load; | |
1340 | idlest = group; | |
1341 | } | |
1342 | } while (group = group->next, group != sd->groups); | |
1343 | ||
1344 | if (!idlest || 100*this_load < imbalance*min_load) | |
1345 | return NULL; | |
1346 | return idlest; | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * find_idlest_cpu - find the idlest cpu among the cpus in group. | |
1351 | */ | |
1352 | static int | |
1353 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
1354 | { | |
1355 | unsigned long load, min_load = ULONG_MAX; | |
1356 | int idlest = -1; | |
1357 | int i; | |
1358 | ||
1359 | /* Traverse only the allowed CPUs */ | |
1360 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { | |
1361 | load = weighted_cpuload(i); | |
1362 | ||
1363 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1364 | min_load = load; | |
1365 | idlest = i; | |
e7693a36 GH |
1366 | } |
1367 | } | |
1368 | ||
aaee1203 PZ |
1369 | return idlest; |
1370 | } | |
e7693a36 | 1371 | |
a50bde51 PZ |
1372 | /* |
1373 | * Try and locate an idle CPU in the sched_domain. | |
1374 | */ | |
1375 | static int | |
1376 | select_idle_sibling(struct task_struct *p, struct sched_domain *sd, int target) | |
1377 | { | |
1378 | int cpu = smp_processor_id(); | |
1379 | int prev_cpu = task_cpu(p); | |
1380 | int i; | |
1381 | ||
1382 | /* | |
1383 | * If this domain spans both cpu and prev_cpu (see the SD_WAKE_AFFINE | |
1384 | * test in select_task_rq_fair) and the prev_cpu is idle then that's | |
1385 | * always a better target than the current cpu. | |
1386 | */ | |
fe3bcfe1 PZ |
1387 | if (target == cpu && !cpu_rq(prev_cpu)->cfs.nr_running) |
1388 | return prev_cpu; | |
a50bde51 PZ |
1389 | |
1390 | /* | |
1391 | * Otherwise, iterate the domain and find an elegible idle cpu. | |
1392 | */ | |
fe3bcfe1 PZ |
1393 | for_each_cpu_and(i, sched_domain_span(sd), &p->cpus_allowed) { |
1394 | if (!cpu_rq(i)->cfs.nr_running) { | |
1395 | target = i; | |
1396 | break; | |
a50bde51 PZ |
1397 | } |
1398 | } | |
1399 | ||
1400 | return target; | |
1401 | } | |
1402 | ||
aaee1203 PZ |
1403 | /* |
1404 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1405 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1406 | * SD_BALANCE_EXEC. | |
1407 | * | |
1408 | * Balance, ie. select the least loaded group. | |
1409 | * | |
1410 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1411 | * | |
1412 | * preempt must be disabled. | |
1413 | */ | |
5158f4e4 | 1414 | static int select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags) |
aaee1203 | 1415 | { |
29cd8bae | 1416 | struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL; |
c88d5910 PZ |
1417 | int cpu = smp_processor_id(); |
1418 | int prev_cpu = task_cpu(p); | |
1419 | int new_cpu = cpu; | |
1420 | int want_affine = 0; | |
29cd8bae | 1421 | int want_sd = 1; |
5158f4e4 | 1422 | int sync = wake_flags & WF_SYNC; |
c88d5910 | 1423 | |
0763a660 | 1424 | if (sd_flag & SD_BALANCE_WAKE) { |
3f04e8cd MG |
1425 | if (sched_feat(AFFINE_WAKEUPS) && |
1426 | cpumask_test_cpu(cpu, &p->cpus_allowed)) | |
c88d5910 PZ |
1427 | want_affine = 1; |
1428 | new_cpu = prev_cpu; | |
1429 | } | |
aaee1203 PZ |
1430 | |
1431 | for_each_domain(cpu, tmp) { | |
1432 | /* | |
ae154be1 PZ |
1433 | * If power savings logic is enabled for a domain, see if we |
1434 | * are not overloaded, if so, don't balance wider. | |
aaee1203 | 1435 | */ |
59abf026 | 1436 | if (tmp->flags & (SD_POWERSAVINGS_BALANCE|SD_PREFER_LOCAL)) { |
ae154be1 PZ |
1437 | unsigned long power = 0; |
1438 | unsigned long nr_running = 0; | |
1439 | unsigned long capacity; | |
1440 | int i; | |
1441 | ||
1442 | for_each_cpu(i, sched_domain_span(tmp)) { | |
1443 | power += power_of(i); | |
1444 | nr_running += cpu_rq(i)->cfs.nr_running; | |
1445 | } | |
1446 | ||
1447 | capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE); | |
1448 | ||
59abf026 PZ |
1449 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1450 | nr_running /= 2; | |
1451 | ||
1452 | if (nr_running < capacity) | |
29cd8bae | 1453 | want_sd = 0; |
ae154be1 | 1454 | } |
aaee1203 | 1455 | |
fe3bcfe1 PZ |
1456 | /* |
1457 | * While iterating the domains looking for a spanning | |
1458 | * WAKE_AFFINE domain, adjust the affine target to any idle cpu | |
1459 | * in cache sharing domains along the way. | |
1460 | */ | |
1461 | if (want_affine) { | |
a50bde51 | 1462 | int target = -1; |
c88d5910 | 1463 | |
a50bde51 PZ |
1464 | /* |
1465 | * If both cpu and prev_cpu are part of this domain, | |
1466 | * cpu is a valid SD_WAKE_AFFINE target. | |
1467 | */ | |
a1f84a3a | 1468 | if (cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) |
a50bde51 | 1469 | target = cpu; |
a1f84a3a MG |
1470 | |
1471 | /* | |
a50bde51 PZ |
1472 | * If there's an idle sibling in this domain, make that |
1473 | * the wake_affine target instead of the current cpu. | |
a1f84a3a | 1474 | */ |
a50bde51 PZ |
1475 | if (tmp->flags & SD_PREFER_SIBLING) |
1476 | target = select_idle_sibling(p, tmp, target); | |
a1f84a3a | 1477 | |
a50bde51 | 1478 | if (target >= 0) { |
fe3bcfe1 PZ |
1479 | if (tmp->flags & SD_WAKE_AFFINE) { |
1480 | affine_sd = tmp; | |
1481 | want_affine = 0; | |
1482 | } | |
a50bde51 | 1483 | cpu = target; |
a1f84a3a | 1484 | } |
c88d5910 PZ |
1485 | } |
1486 | ||
29cd8bae PZ |
1487 | if (!want_sd && !want_affine) |
1488 | break; | |
1489 | ||
0763a660 | 1490 | if (!(tmp->flags & sd_flag)) |
c88d5910 PZ |
1491 | continue; |
1492 | ||
29cd8bae PZ |
1493 | if (want_sd) |
1494 | sd = tmp; | |
1495 | } | |
1496 | ||
1497 | if (sched_feat(LB_SHARES_UPDATE)) { | |
1498 | /* | |
1499 | * Pick the largest domain to update shares over | |
1500 | */ | |
1501 | tmp = sd; | |
1502 | if (affine_sd && (!tmp || | |
1503 | cpumask_weight(sched_domain_span(affine_sd)) > | |
1504 | cpumask_weight(sched_domain_span(sd)))) | |
1505 | tmp = affine_sd; | |
1506 | ||
1507 | if (tmp) | |
1508 | update_shares(tmp); | |
c88d5910 | 1509 | } |
aaee1203 | 1510 | |
fb58bac5 PZ |
1511 | if (affine_sd && wake_affine(affine_sd, p, sync)) |
1512 | return cpu; | |
e7693a36 | 1513 | |
aaee1203 | 1514 | while (sd) { |
5158f4e4 | 1515 | int load_idx = sd->forkexec_idx; |
aaee1203 | 1516 | struct sched_group *group; |
c88d5910 | 1517 | int weight; |
098fb9db | 1518 | |
0763a660 | 1519 | if (!(sd->flags & sd_flag)) { |
aaee1203 PZ |
1520 | sd = sd->child; |
1521 | continue; | |
1522 | } | |
098fb9db | 1523 | |
5158f4e4 PZ |
1524 | if (sd_flag & SD_BALANCE_WAKE) |
1525 | load_idx = sd->wake_idx; | |
098fb9db | 1526 | |
5158f4e4 | 1527 | group = find_idlest_group(sd, p, cpu, load_idx); |
aaee1203 PZ |
1528 | if (!group) { |
1529 | sd = sd->child; | |
1530 | continue; | |
1531 | } | |
4ae7d5ce | 1532 | |
d7c33c49 | 1533 | new_cpu = find_idlest_cpu(group, p, cpu); |
aaee1203 PZ |
1534 | if (new_cpu == -1 || new_cpu == cpu) { |
1535 | /* Now try balancing at a lower domain level of cpu */ | |
1536 | sd = sd->child; | |
1537 | continue; | |
e7693a36 | 1538 | } |
aaee1203 PZ |
1539 | |
1540 | /* Now try balancing at a lower domain level of new_cpu */ | |
1541 | cpu = new_cpu; | |
1542 | weight = cpumask_weight(sched_domain_span(sd)); | |
1543 | sd = NULL; | |
1544 | for_each_domain(cpu, tmp) { | |
1545 | if (weight <= cpumask_weight(sched_domain_span(tmp))) | |
1546 | break; | |
0763a660 | 1547 | if (tmp->flags & sd_flag) |
aaee1203 PZ |
1548 | sd = tmp; |
1549 | } | |
1550 | /* while loop will break here if sd == NULL */ | |
e7693a36 GH |
1551 | } |
1552 | ||
c88d5910 | 1553 | return new_cpu; |
e7693a36 GH |
1554 | } |
1555 | #endif /* CONFIG_SMP */ | |
1556 | ||
e52fb7c0 PZ |
1557 | /* |
1558 | * Adaptive granularity | |
1559 | * | |
1560 | * se->avg_wakeup gives the average time a task runs until it does a wakeup, | |
1561 | * with the limit of wakeup_gran -- when it never does a wakeup. | |
1562 | * | |
1563 | * So the smaller avg_wakeup is the faster we want this task to preempt, | |
1564 | * but we don't want to treat the preemptee unfairly and therefore allow it | |
1565 | * to run for at least the amount of time we'd like to run. | |
1566 | * | |
1567 | * NOTE: we use 2*avg_wakeup to increase the probability of actually doing one | |
1568 | * | |
1569 | * NOTE: we use *nr_running to scale with load, this nicely matches the | |
1570 | * degrading latency on load. | |
1571 | */ | |
1572 | static unsigned long | |
1573 | adaptive_gran(struct sched_entity *curr, struct sched_entity *se) | |
1574 | { | |
1575 | u64 this_run = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; | |
1576 | u64 expected_wakeup = 2*se->avg_wakeup * cfs_rq_of(se)->nr_running; | |
1577 | u64 gran = 0; | |
1578 | ||
1579 | if (this_run < expected_wakeup) | |
1580 | gran = expected_wakeup - this_run; | |
1581 | ||
1582 | return min_t(s64, gran, sysctl_sched_wakeup_granularity); | |
1583 | } | |
1584 | ||
1585 | static unsigned long | |
1586 | wakeup_gran(struct sched_entity *curr, struct sched_entity *se) | |
0bbd3336 PZ |
1587 | { |
1588 | unsigned long gran = sysctl_sched_wakeup_granularity; | |
1589 | ||
e52fb7c0 PZ |
1590 | if (cfs_rq_of(curr)->curr && sched_feat(ADAPTIVE_GRAN)) |
1591 | gran = adaptive_gran(curr, se); | |
1592 | ||
0bbd3336 | 1593 | /* |
e52fb7c0 PZ |
1594 | * Since its curr running now, convert the gran from real-time |
1595 | * to virtual-time in his units. | |
0bbd3336 | 1596 | */ |
e52fb7c0 PZ |
1597 | if (sched_feat(ASYM_GRAN)) { |
1598 | /* | |
1599 | * By using 'se' instead of 'curr' we penalize light tasks, so | |
1600 | * they get preempted easier. That is, if 'se' < 'curr' then | |
1601 | * the resulting gran will be larger, therefore penalizing the | |
1602 | * lighter, if otoh 'se' > 'curr' then the resulting gran will | |
1603 | * be smaller, again penalizing the lighter task. | |
1604 | * | |
1605 | * This is especially important for buddies when the leftmost | |
1606 | * task is higher priority than the buddy. | |
1607 | */ | |
1608 | if (unlikely(se->load.weight != NICE_0_LOAD)) | |
1609 | gran = calc_delta_fair(gran, se); | |
1610 | } else { | |
1611 | if (unlikely(curr->load.weight != NICE_0_LOAD)) | |
1612 | gran = calc_delta_fair(gran, curr); | |
1613 | } | |
0bbd3336 PZ |
1614 | |
1615 | return gran; | |
1616 | } | |
1617 | ||
464b7527 PZ |
1618 | /* |
1619 | * Should 'se' preempt 'curr'. | |
1620 | * | |
1621 | * |s1 | |
1622 | * |s2 | |
1623 | * |s3 | |
1624 | * g | |
1625 | * |<--->|c | |
1626 | * | |
1627 | * w(c, s1) = -1 | |
1628 | * w(c, s2) = 0 | |
1629 | * w(c, s3) = 1 | |
1630 | * | |
1631 | */ | |
1632 | static int | |
1633 | wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se) | |
1634 | { | |
1635 | s64 gran, vdiff = curr->vruntime - se->vruntime; | |
1636 | ||
1637 | if (vdiff <= 0) | |
1638 | return -1; | |
1639 | ||
e52fb7c0 | 1640 | gran = wakeup_gran(curr, se); |
464b7527 PZ |
1641 | if (vdiff > gran) |
1642 | return 1; | |
1643 | ||
1644 | return 0; | |
1645 | } | |
1646 | ||
02479099 PZ |
1647 | static void set_last_buddy(struct sched_entity *se) |
1648 | { | |
6bc912b7 PZ |
1649 | if (likely(task_of(se)->policy != SCHED_IDLE)) { |
1650 | for_each_sched_entity(se) | |
1651 | cfs_rq_of(se)->last = se; | |
1652 | } | |
02479099 PZ |
1653 | } |
1654 | ||
1655 | static void set_next_buddy(struct sched_entity *se) | |
1656 | { | |
6bc912b7 PZ |
1657 | if (likely(task_of(se)->policy != SCHED_IDLE)) { |
1658 | for_each_sched_entity(se) | |
1659 | cfs_rq_of(se)->next = se; | |
1660 | } | |
02479099 PZ |
1661 | } |
1662 | ||
bf0f6f24 IM |
1663 | /* |
1664 | * Preempt the current task with a newly woken task if needed: | |
1665 | */ | |
5a9b86f6 | 1666 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
bf0f6f24 IM |
1667 | { |
1668 | struct task_struct *curr = rq->curr; | |
8651a86c | 1669 | struct sched_entity *se = &curr->se, *pse = &p->se; |
03e89e45 | 1670 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
5a9b86f6 | 1671 | int sync = wake_flags & WF_SYNC; |
f685ceac | 1672 | int scale = cfs_rq->nr_running >= sched_nr_latency; |
bf0f6f24 | 1673 | |
3a7e73a2 PZ |
1674 | if (unlikely(rt_prio(p->prio))) |
1675 | goto preempt; | |
aa2ac252 | 1676 | |
d95f98d0 PZ |
1677 | if (unlikely(p->sched_class != &fair_sched_class)) |
1678 | return; | |
1679 | ||
4ae7d5ce IM |
1680 | if (unlikely(se == pse)) |
1681 | return; | |
1682 | ||
f685ceac | 1683 | if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) |
3cb63d52 | 1684 | set_next_buddy(pse); |
57fdc26d | 1685 | |
aec0a514 BR |
1686 | /* |
1687 | * We can come here with TIF_NEED_RESCHED already set from new task | |
1688 | * wake up path. | |
1689 | */ | |
1690 | if (test_tsk_need_resched(curr)) | |
1691 | return; | |
1692 | ||
91c234b4 | 1693 | /* |
6bc912b7 | 1694 | * Batch and idle tasks do not preempt (their preemption is driven by |
91c234b4 IM |
1695 | * the tick): |
1696 | */ | |
6bc912b7 | 1697 | if (unlikely(p->policy != SCHED_NORMAL)) |
91c234b4 | 1698 | return; |
bf0f6f24 | 1699 | |
6bc912b7 | 1700 | /* Idle tasks are by definition preempted by everybody. */ |
3a7e73a2 PZ |
1701 | if (unlikely(curr->policy == SCHED_IDLE)) |
1702 | goto preempt; | |
bf0f6f24 | 1703 | |
3a7e73a2 PZ |
1704 | if (sched_feat(WAKEUP_SYNC) && sync) |
1705 | goto preempt; | |
15afe09b | 1706 | |
3a7e73a2 PZ |
1707 | if (sched_feat(WAKEUP_OVERLAP) && |
1708 | se->avg_overlap < sysctl_sched_migration_cost && | |
1709 | pse->avg_overlap < sysctl_sched_migration_cost) | |
1710 | goto preempt; | |
1711 | ||
ad4b78bb PZ |
1712 | if (!sched_feat(WAKEUP_PREEMPT)) |
1713 | return; | |
1714 | ||
3a7e73a2 | 1715 | update_curr(cfs_rq); |
464b7527 | 1716 | find_matching_se(&se, &pse); |
002f128b | 1717 | BUG_ON(!pse); |
3a7e73a2 PZ |
1718 | if (wakeup_preempt_entity(se, pse) == 1) |
1719 | goto preempt; | |
464b7527 | 1720 | |
3a7e73a2 | 1721 | return; |
a65ac745 | 1722 | |
3a7e73a2 PZ |
1723 | preempt: |
1724 | resched_task(curr); | |
1725 | /* | |
1726 | * Only set the backward buddy when the current task is still | |
1727 | * on the rq. This can happen when a wakeup gets interleaved | |
1728 | * with schedule on the ->pre_schedule() or idle_balance() | |
1729 | * point, either of which can * drop the rq lock. | |
1730 | * | |
1731 | * Also, during early boot the idle thread is in the fair class, | |
1732 | * for obvious reasons its a bad idea to schedule back to it. | |
1733 | */ | |
1734 | if (unlikely(!se->on_rq || curr == rq->idle)) | |
1735 | return; | |
1736 | ||
1737 | if (sched_feat(LAST_BUDDY) && scale && entity_is_task(se)) | |
1738 | set_last_buddy(se); | |
bf0f6f24 IM |
1739 | } |
1740 | ||
fb8d4724 | 1741 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 | 1742 | { |
8f4d37ec | 1743 | struct task_struct *p; |
bf0f6f24 IM |
1744 | struct cfs_rq *cfs_rq = &rq->cfs; |
1745 | struct sched_entity *se; | |
1746 | ||
36ace27e | 1747 | if (!cfs_rq->nr_running) |
bf0f6f24 IM |
1748 | return NULL; |
1749 | ||
1750 | do { | |
9948f4b2 | 1751 | se = pick_next_entity(cfs_rq); |
f4b6755f | 1752 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
1753 | cfs_rq = group_cfs_rq(se); |
1754 | } while (cfs_rq); | |
1755 | ||
8f4d37ec PZ |
1756 | p = task_of(se); |
1757 | hrtick_start_fair(rq, p); | |
1758 | ||
1759 | return p; | |
bf0f6f24 IM |
1760 | } |
1761 | ||
1762 | /* | |
1763 | * Account for a descheduled task: | |
1764 | */ | |
31ee529c | 1765 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
1766 | { |
1767 | struct sched_entity *se = &prev->se; | |
1768 | struct cfs_rq *cfs_rq; | |
1769 | ||
1770 | for_each_sched_entity(se) { | |
1771 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 1772 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
1773 | } |
1774 | } | |
1775 | ||
681f3e68 | 1776 | #ifdef CONFIG_SMP |
bf0f6f24 IM |
1777 | /************************************************** |
1778 | * Fair scheduling class load-balancing methods: | |
1779 | */ | |
1780 | ||
1781 | /* | |
1782 | * Load-balancing iterator. Note: while the runqueue stays locked | |
1783 | * during the whole iteration, the current task might be | |
1784 | * dequeued so the iterator has to be dequeue-safe. Here we | |
1785 | * achieve that by always pre-iterating before returning | |
1786 | * the current task: | |
1787 | */ | |
a9957449 | 1788 | static struct task_struct * |
4a55bd5e | 1789 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next) |
bf0f6f24 | 1790 | { |
354d60c2 DG |
1791 | struct task_struct *p = NULL; |
1792 | struct sched_entity *se; | |
bf0f6f24 | 1793 | |
77ae6513 MG |
1794 | if (next == &cfs_rq->tasks) |
1795 | return NULL; | |
1796 | ||
b87f1724 BR |
1797 | se = list_entry(next, struct sched_entity, group_node); |
1798 | p = task_of(se); | |
1799 | cfs_rq->balance_iterator = next->next; | |
77ae6513 | 1800 | |
bf0f6f24 IM |
1801 | return p; |
1802 | } | |
1803 | ||
1804 | static struct task_struct *load_balance_start_fair(void *arg) | |
1805 | { | |
1806 | struct cfs_rq *cfs_rq = arg; | |
1807 | ||
4a55bd5e | 1808 | return __load_balance_iterator(cfs_rq, cfs_rq->tasks.next); |
bf0f6f24 IM |
1809 | } |
1810 | ||
1811 | static struct task_struct *load_balance_next_fair(void *arg) | |
1812 | { | |
1813 | struct cfs_rq *cfs_rq = arg; | |
1814 | ||
4a55bd5e | 1815 | return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator); |
bf0f6f24 IM |
1816 | } |
1817 | ||
c09595f6 PZ |
1818 | static unsigned long |
1819 | __load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1820 | unsigned long max_load_move, struct sched_domain *sd, | |
1821 | enum cpu_idle_type idle, int *all_pinned, int *this_best_prio, | |
1822 | struct cfs_rq *cfs_rq) | |
62fb1851 | 1823 | { |
c09595f6 | 1824 | struct rq_iterator cfs_rq_iterator; |
62fb1851 | 1825 | |
c09595f6 PZ |
1826 | cfs_rq_iterator.start = load_balance_start_fair; |
1827 | cfs_rq_iterator.next = load_balance_next_fair; | |
1828 | cfs_rq_iterator.arg = cfs_rq; | |
62fb1851 | 1829 | |
c09595f6 PZ |
1830 | return balance_tasks(this_rq, this_cpu, busiest, |
1831 | max_load_move, sd, idle, all_pinned, | |
1832 | this_best_prio, &cfs_rq_iterator); | |
62fb1851 | 1833 | } |
62fb1851 | 1834 | |
c09595f6 | 1835 | #ifdef CONFIG_FAIR_GROUP_SCHED |
43010659 | 1836 | static unsigned long |
bf0f6f24 | 1837 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f | 1838 | unsigned long max_load_move, |
a4ac01c3 PW |
1839 | struct sched_domain *sd, enum cpu_idle_type idle, |
1840 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 | 1841 | { |
bf0f6f24 | 1842 | long rem_load_move = max_load_move; |
c09595f6 PZ |
1843 | int busiest_cpu = cpu_of(busiest); |
1844 | struct task_group *tg; | |
18d95a28 | 1845 | |
c09595f6 | 1846 | rcu_read_lock(); |
c8cba857 | 1847 | update_h_load(busiest_cpu); |
18d95a28 | 1848 | |
caea8a03 | 1849 | list_for_each_entry_rcu(tg, &task_groups, list) { |
c8cba857 | 1850 | struct cfs_rq *busiest_cfs_rq = tg->cfs_rq[busiest_cpu]; |
42a3ac7d PZ |
1851 | unsigned long busiest_h_load = busiest_cfs_rq->h_load; |
1852 | unsigned long busiest_weight = busiest_cfs_rq->load.weight; | |
243e0e7b | 1853 | u64 rem_load, moved_load; |
18d95a28 | 1854 | |
c09595f6 PZ |
1855 | /* |
1856 | * empty group | |
1857 | */ | |
c8cba857 | 1858 | if (!busiest_cfs_rq->task_weight) |
bf0f6f24 IM |
1859 | continue; |
1860 | ||
243e0e7b SV |
1861 | rem_load = (u64)rem_load_move * busiest_weight; |
1862 | rem_load = div_u64(rem_load, busiest_h_load + 1); | |
bf0f6f24 | 1863 | |
c09595f6 | 1864 | moved_load = __load_balance_fair(this_rq, this_cpu, busiest, |
53fecd8a | 1865 | rem_load, sd, idle, all_pinned, this_best_prio, |
c09595f6 | 1866 | tg->cfs_rq[busiest_cpu]); |
bf0f6f24 | 1867 | |
c09595f6 | 1868 | if (!moved_load) |
bf0f6f24 IM |
1869 | continue; |
1870 | ||
42a3ac7d | 1871 | moved_load *= busiest_h_load; |
243e0e7b | 1872 | moved_load = div_u64(moved_load, busiest_weight + 1); |
bf0f6f24 | 1873 | |
c09595f6 PZ |
1874 | rem_load_move -= moved_load; |
1875 | if (rem_load_move < 0) | |
bf0f6f24 IM |
1876 | break; |
1877 | } | |
c09595f6 | 1878 | rcu_read_unlock(); |
bf0f6f24 | 1879 | |
43010659 | 1880 | return max_load_move - rem_load_move; |
bf0f6f24 | 1881 | } |
c09595f6 PZ |
1882 | #else |
1883 | static unsigned long | |
1884 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1885 | unsigned long max_load_move, | |
1886 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1887 | int *all_pinned, int *this_best_prio) | |
1888 | { | |
1889 | return __load_balance_fair(this_rq, this_cpu, busiest, | |
1890 | max_load_move, sd, idle, all_pinned, | |
1891 | this_best_prio, &busiest->cfs); | |
1892 | } | |
1893 | #endif | |
bf0f6f24 | 1894 | |
e1d1484f PW |
1895 | static int |
1896 | move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1897 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1898 | { | |
1899 | struct cfs_rq *busy_cfs_rq; | |
1900 | struct rq_iterator cfs_rq_iterator; | |
1901 | ||
1902 | cfs_rq_iterator.start = load_balance_start_fair; | |
1903 | cfs_rq_iterator.next = load_balance_next_fair; | |
1904 | ||
1905 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
1906 | /* | |
1907 | * pass busy_cfs_rq argument into | |
1908 | * load_balance_[start|next]_fair iterators | |
1909 | */ | |
1910 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1911 | if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle, | |
1912 | &cfs_rq_iterator)) | |
1913 | return 1; | |
1914 | } | |
1915 | ||
1916 | return 0; | |
1917 | } | |
0bcdcf28 CE |
1918 | |
1919 | static void rq_online_fair(struct rq *rq) | |
1920 | { | |
1921 | update_sysctl(); | |
1922 | } | |
1923 | ||
1924 | static void rq_offline_fair(struct rq *rq) | |
1925 | { | |
1926 | update_sysctl(); | |
1927 | } | |
1928 | ||
55e12e5e | 1929 | #endif /* CONFIG_SMP */ |
e1d1484f | 1930 | |
bf0f6f24 IM |
1931 | /* |
1932 | * scheduler tick hitting a task of our scheduling class: | |
1933 | */ | |
8f4d37ec | 1934 | static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) |
bf0f6f24 IM |
1935 | { |
1936 | struct cfs_rq *cfs_rq; | |
1937 | struct sched_entity *se = &curr->se; | |
1938 | ||
1939 | for_each_sched_entity(se) { | |
1940 | cfs_rq = cfs_rq_of(se); | |
8f4d37ec | 1941 | entity_tick(cfs_rq, se, queued); |
bf0f6f24 IM |
1942 | } |
1943 | } | |
1944 | ||
1945 | /* | |
cd29fe6f PZ |
1946 | * called on fork with the child task as argument from the parent's context |
1947 | * - child not yet on the tasklist | |
1948 | * - preemption disabled | |
bf0f6f24 | 1949 | */ |
cd29fe6f | 1950 | static void task_fork_fair(struct task_struct *p) |
bf0f6f24 | 1951 | { |
cd29fe6f | 1952 | struct cfs_rq *cfs_rq = task_cfs_rq(current); |
429d43bc | 1953 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
00bf7bfc | 1954 | int this_cpu = smp_processor_id(); |
cd29fe6f PZ |
1955 | struct rq *rq = this_rq(); |
1956 | unsigned long flags; | |
1957 | ||
1958 | spin_lock_irqsave(&rq->lock, flags); | |
bf0f6f24 | 1959 | |
cd29fe6f PZ |
1960 | if (unlikely(task_cpu(p) != this_cpu)) |
1961 | __set_task_cpu(p, this_cpu); | |
bf0f6f24 | 1962 | |
7109c442 | 1963 | update_curr(cfs_rq); |
cd29fe6f | 1964 | |
b5d9d734 MG |
1965 | if (curr) |
1966 | se->vruntime = curr->vruntime; | |
aeb73b04 | 1967 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1968 | |
cd29fe6f | 1969 | if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) { |
87fefa38 | 1970 | /* |
edcb60a3 IM |
1971 | * Upon rescheduling, sched_class::put_prev_task() will place |
1972 | * 'current' within the tree based on its new key value. | |
1973 | */ | |
4d78e7b6 | 1974 | swap(curr->vruntime, se->vruntime); |
aec0a514 | 1975 | resched_task(rq->curr); |
4d78e7b6 | 1976 | } |
bf0f6f24 | 1977 | |
cd29fe6f | 1978 | spin_unlock_irqrestore(&rq->lock, flags); |
bf0f6f24 IM |
1979 | } |
1980 | ||
cb469845 SR |
1981 | /* |
1982 | * Priority of the task has changed. Check to see if we preempt | |
1983 | * the current task. | |
1984 | */ | |
1985 | static void prio_changed_fair(struct rq *rq, struct task_struct *p, | |
1986 | int oldprio, int running) | |
1987 | { | |
1988 | /* | |
1989 | * Reschedule if we are currently running on this runqueue and | |
1990 | * our priority decreased, or if we are not currently running on | |
1991 | * this runqueue and our priority is higher than the current's | |
1992 | */ | |
1993 | if (running) { | |
1994 | if (p->prio > oldprio) | |
1995 | resched_task(rq->curr); | |
1996 | } else | |
15afe09b | 1997 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
1998 | } |
1999 | ||
2000 | /* | |
2001 | * We switched to the sched_fair class. | |
2002 | */ | |
2003 | static void switched_to_fair(struct rq *rq, struct task_struct *p, | |
2004 | int running) | |
2005 | { | |
2006 | /* | |
2007 | * We were most likely switched from sched_rt, so | |
2008 | * kick off the schedule if running, otherwise just see | |
2009 | * if we can still preempt the current task. | |
2010 | */ | |
2011 | if (running) | |
2012 | resched_task(rq->curr); | |
2013 | else | |
15afe09b | 2014 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
2015 | } |
2016 | ||
83b699ed SV |
2017 | /* Account for a task changing its policy or group. |
2018 | * | |
2019 | * This routine is mostly called to set cfs_rq->curr field when a task | |
2020 | * migrates between groups/classes. | |
2021 | */ | |
2022 | static void set_curr_task_fair(struct rq *rq) | |
2023 | { | |
2024 | struct sched_entity *se = &rq->curr->se; | |
2025 | ||
2026 | for_each_sched_entity(se) | |
2027 | set_next_entity(cfs_rq_of(se), se); | |
2028 | } | |
2029 | ||
810b3817 PZ |
2030 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2031 | static void moved_group_fair(struct task_struct *p) | |
2032 | { | |
2033 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
2034 | ||
2035 | update_curr(cfs_rq); | |
2036 | place_entity(cfs_rq, &p->se, 1); | |
2037 | } | |
2038 | #endif | |
2039 | ||
dba091b9 | 2040 | unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task) |
0d721cea PW |
2041 | { |
2042 | struct sched_entity *se = &task->se; | |
0d721cea PW |
2043 | unsigned int rr_interval = 0; |
2044 | ||
2045 | /* | |
2046 | * Time slice is 0 for SCHED_OTHER tasks that are on an otherwise | |
2047 | * idle runqueue: | |
2048 | */ | |
0d721cea PW |
2049 | if (rq->cfs.load.weight) |
2050 | rr_interval = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
0d721cea PW |
2051 | |
2052 | return rr_interval; | |
2053 | } | |
2054 | ||
bf0f6f24 IM |
2055 | /* |
2056 | * All the scheduling class methods: | |
2057 | */ | |
5522d5d5 IM |
2058 | static const struct sched_class fair_sched_class = { |
2059 | .next = &idle_sched_class, | |
bf0f6f24 IM |
2060 | .enqueue_task = enqueue_task_fair, |
2061 | .dequeue_task = dequeue_task_fair, | |
2062 | .yield_task = yield_task_fair, | |
2063 | ||
2e09bf55 | 2064 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
2065 | |
2066 | .pick_next_task = pick_next_task_fair, | |
2067 | .put_prev_task = put_prev_task_fair, | |
2068 | ||
681f3e68 | 2069 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
2070 | .select_task_rq = select_task_rq_fair, |
2071 | ||
bf0f6f24 | 2072 | .load_balance = load_balance_fair, |
e1d1484f | 2073 | .move_one_task = move_one_task_fair, |
0bcdcf28 CE |
2074 | .rq_online = rq_online_fair, |
2075 | .rq_offline = rq_offline_fair, | |
681f3e68 | 2076 | #endif |
bf0f6f24 | 2077 | |
83b699ed | 2078 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 | 2079 | .task_tick = task_tick_fair, |
cd29fe6f | 2080 | .task_fork = task_fork_fair, |
cb469845 SR |
2081 | |
2082 | .prio_changed = prio_changed_fair, | |
2083 | .switched_to = switched_to_fair, | |
810b3817 | 2084 | |
0d721cea PW |
2085 | .get_rr_interval = get_rr_interval_fair, |
2086 | ||
810b3817 PZ |
2087 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2088 | .moved_group = moved_group_fair, | |
2089 | #endif | |
bf0f6f24 IM |
2090 | }; |
2091 | ||
2092 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 2093 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 2094 | { |
bf0f6f24 IM |
2095 | struct cfs_rq *cfs_rq; |
2096 | ||
5973e5b9 | 2097 | rcu_read_lock(); |
c3b64f1e | 2098 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 2099 | print_cfs_rq(m, cpu, cfs_rq); |
5973e5b9 | 2100 | rcu_read_unlock(); |
bf0f6f24 IM |
2101 | } |
2102 | #endif |