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