]>
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 AV |
23 | #include <linux/latencytop.h> |
24 | ||
bf0f6f24 | 25 | /* |
21805085 | 26 | * Targeted preemption latency for CPU-bound tasks: |
722aab0c | 27 | * (default: 20ms * (1 + ilog(ncpus)), units: nanoseconds) |
bf0f6f24 | 28 | * |
21805085 | 29 | * NOTE: this latency value is not the same as the concept of |
d274a4ce IM |
30 | * 'timeslice length' - timeslices in CFS are of variable length |
31 | * and have no persistent notion like in traditional, time-slice | |
32 | * based scheduling concepts. | |
bf0f6f24 | 33 | * |
d274a4ce IM |
34 | * (to see the precise effective timeslice length of your workload, |
35 | * run vmstat and monitor the context-switches (cs) field) | |
bf0f6f24 | 36 | */ |
19978ca6 | 37 | unsigned int sysctl_sched_latency = 20000000ULL; |
2bd8e6d4 IM |
38 | |
39 | /* | |
b2be5e96 | 40 | * Minimal preemption granularity for CPU-bound tasks: |
722aab0c | 41 | * (default: 4 msec * (1 + ilog(ncpus)), units: nanoseconds) |
2bd8e6d4 | 42 | */ |
722aab0c | 43 | unsigned int sysctl_sched_min_granularity = 4000000ULL; |
21805085 PZ |
44 | |
45 | /* | |
b2be5e96 PZ |
46 | * is kept at sysctl_sched_latency / sysctl_sched_min_granularity |
47 | */ | |
722aab0c | 48 | static unsigned int sched_nr_latency = 5; |
b2be5e96 PZ |
49 | |
50 | /* | |
51 | * After fork, child runs first. (default) If set to 0 then | |
52 | * parent will (try to) run first. | |
21805085 | 53 | */ |
b2be5e96 | 54 | const_debug unsigned int sysctl_sched_child_runs_first = 1; |
bf0f6f24 | 55 | |
1799e35d IM |
56 | /* |
57 | * sys_sched_yield() compat mode | |
58 | * | |
59 | * This option switches the agressive yield implementation of the | |
60 | * old scheduler back on. | |
61 | */ | |
62 | unsigned int __read_mostly sysctl_sched_compat_yield; | |
63 | ||
bf0f6f24 IM |
64 | /* |
65 | * SCHED_OTHER wake-up granularity. | |
103638d9 | 66 | * (default: 5 msec * (1 + ilog(ncpus)), units: nanoseconds) |
bf0f6f24 IM |
67 | * |
68 | * This option delays the preemption effects of decoupled workloads | |
69 | * and reduces their over-scheduling. Synchronous workloads will still | |
70 | * have immediate wakeup/sleep latencies. | |
71 | */ | |
103638d9 | 72 | unsigned int sysctl_sched_wakeup_granularity = 5000000UL; |
bf0f6f24 | 73 | |
da84d961 IM |
74 | const_debug unsigned int sysctl_sched_migration_cost = 500000UL; |
75 | ||
a4c2f00f PZ |
76 | static const struct sched_class fair_sched_class; |
77 | ||
bf0f6f24 IM |
78 | /************************************************************** |
79 | * CFS operations on generic schedulable entities: | |
80 | */ | |
81 | ||
62160e3f | 82 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 83 | |
62160e3f | 84 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
85 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
86 | { | |
62160e3f | 87 | return cfs_rq->rq; |
bf0f6f24 IM |
88 | } |
89 | ||
62160e3f IM |
90 | /* An entity is a task if it doesn't "own" a runqueue */ |
91 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 92 | |
8f48894f PZ |
93 | static inline struct task_struct *task_of(struct sched_entity *se) |
94 | { | |
95 | #ifdef CONFIG_SCHED_DEBUG | |
96 | WARN_ON_ONCE(!entity_is_task(se)); | |
97 | #endif | |
98 | return container_of(se, struct task_struct, se); | |
99 | } | |
100 | ||
b758149c PZ |
101 | /* Walk up scheduling entities hierarchy */ |
102 | #define for_each_sched_entity(se) \ | |
103 | for (; se; se = se->parent) | |
104 | ||
105 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
106 | { | |
107 | return p->se.cfs_rq; | |
108 | } | |
109 | ||
110 | /* runqueue on which this entity is (to be) queued */ | |
111 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
112 | { | |
113 | return se->cfs_rq; | |
114 | } | |
115 | ||
116 | /* runqueue "owned" by this group */ | |
117 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
118 | { | |
119 | return grp->my_q; | |
120 | } | |
121 | ||
122 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
123 | * another cpu ('this_cpu') | |
124 | */ | |
125 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
126 | { | |
127 | return cfs_rq->tg->cfs_rq[this_cpu]; | |
128 | } | |
129 | ||
130 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
131 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
132 | list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
133 | ||
134 | /* Do the two (enqueued) entities belong to the same group ? */ | |
135 | static inline int | |
136 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
137 | { | |
138 | if (se->cfs_rq == pse->cfs_rq) | |
139 | return 1; | |
140 | ||
141 | return 0; | |
142 | } | |
143 | ||
144 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
145 | { | |
146 | return se->parent; | |
147 | } | |
148 | ||
464b7527 PZ |
149 | /* return depth at which a sched entity is present in the hierarchy */ |
150 | static inline int depth_se(struct sched_entity *se) | |
151 | { | |
152 | int depth = 0; | |
153 | ||
154 | for_each_sched_entity(se) | |
155 | depth++; | |
156 | ||
157 | return depth; | |
158 | } | |
159 | ||
160 | static void | |
161 | find_matching_se(struct sched_entity **se, struct sched_entity **pse) | |
162 | { | |
163 | int se_depth, pse_depth; | |
164 | ||
165 | /* | |
166 | * preemption test can be made between sibling entities who are in the | |
167 | * same cfs_rq i.e who have a common parent. Walk up the hierarchy of | |
168 | * both tasks until we find their ancestors who are siblings of common | |
169 | * parent. | |
170 | */ | |
171 | ||
172 | /* First walk up until both entities are at same depth */ | |
173 | se_depth = depth_se(*se); | |
174 | pse_depth = depth_se(*pse); | |
175 | ||
176 | while (se_depth > pse_depth) { | |
177 | se_depth--; | |
178 | *se = parent_entity(*se); | |
179 | } | |
180 | ||
181 | while (pse_depth > se_depth) { | |
182 | pse_depth--; | |
183 | *pse = parent_entity(*pse); | |
184 | } | |
185 | ||
186 | while (!is_same_group(*se, *pse)) { | |
187 | *se = parent_entity(*se); | |
188 | *pse = parent_entity(*pse); | |
189 | } | |
190 | } | |
191 | ||
8f48894f PZ |
192 | #else /* !CONFIG_FAIR_GROUP_SCHED */ |
193 | ||
194 | static inline struct task_struct *task_of(struct sched_entity *se) | |
195 | { | |
196 | return container_of(se, struct task_struct, se); | |
197 | } | |
bf0f6f24 | 198 | |
62160e3f IM |
199 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
200 | { | |
201 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
202 | } |
203 | ||
204 | #define entity_is_task(se) 1 | |
205 | ||
b758149c PZ |
206 | #define for_each_sched_entity(se) \ |
207 | for (; se; se = NULL) | |
bf0f6f24 | 208 | |
b758149c | 209 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) |
bf0f6f24 | 210 | { |
b758149c | 211 | return &task_rq(p)->cfs; |
bf0f6f24 IM |
212 | } |
213 | ||
b758149c PZ |
214 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) |
215 | { | |
216 | struct task_struct *p = task_of(se); | |
217 | struct rq *rq = task_rq(p); | |
218 | ||
219 | return &rq->cfs; | |
220 | } | |
221 | ||
222 | /* runqueue "owned" by this group */ | |
223 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
224 | { | |
225 | return NULL; | |
226 | } | |
227 | ||
228 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
229 | { | |
230 | return &cpu_rq(this_cpu)->cfs; | |
231 | } | |
232 | ||
233 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
234 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
235 | ||
236 | static inline int | |
237 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
238 | { | |
239 | return 1; | |
240 | } | |
241 | ||
242 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
243 | { | |
244 | return NULL; | |
245 | } | |
246 | ||
464b7527 PZ |
247 | static inline void |
248 | find_matching_se(struct sched_entity **se, struct sched_entity **pse) | |
249 | { | |
250 | } | |
251 | ||
b758149c PZ |
252 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
253 | ||
bf0f6f24 IM |
254 | |
255 | /************************************************************** | |
256 | * Scheduling class tree data structure manipulation methods: | |
257 | */ | |
258 | ||
0702e3eb | 259 | static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime) |
02e0431a | 260 | { |
368059a9 PZ |
261 | s64 delta = (s64)(vruntime - min_vruntime); |
262 | if (delta > 0) | |
02e0431a PZ |
263 | min_vruntime = vruntime; |
264 | ||
265 | return min_vruntime; | |
266 | } | |
267 | ||
0702e3eb | 268 | static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime) |
b0ffd246 PZ |
269 | { |
270 | s64 delta = (s64)(vruntime - min_vruntime); | |
271 | if (delta < 0) | |
272 | min_vruntime = vruntime; | |
273 | ||
274 | return min_vruntime; | |
275 | } | |
276 | ||
54fdc581 FC |
277 | static inline int entity_before(struct sched_entity *a, |
278 | struct sched_entity *b) | |
279 | { | |
280 | return (s64)(a->vruntime - b->vruntime) < 0; | |
281 | } | |
282 | ||
0702e3eb | 283 | static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
9014623c | 284 | { |
30cfdcfc | 285 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
286 | } |
287 | ||
1af5f730 PZ |
288 | static void update_min_vruntime(struct cfs_rq *cfs_rq) |
289 | { | |
290 | u64 vruntime = cfs_rq->min_vruntime; | |
291 | ||
292 | if (cfs_rq->curr) | |
293 | vruntime = cfs_rq->curr->vruntime; | |
294 | ||
295 | if (cfs_rq->rb_leftmost) { | |
296 | struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost, | |
297 | struct sched_entity, | |
298 | run_node); | |
299 | ||
e17036da | 300 | if (!cfs_rq->curr) |
1af5f730 PZ |
301 | vruntime = se->vruntime; |
302 | else | |
303 | vruntime = min_vruntime(vruntime, se->vruntime); | |
304 | } | |
305 | ||
306 | cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime); | |
307 | } | |
308 | ||
bf0f6f24 IM |
309 | /* |
310 | * Enqueue an entity into the rb-tree: | |
311 | */ | |
0702e3eb | 312 | static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
313 | { |
314 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
315 | struct rb_node *parent = NULL; | |
316 | struct sched_entity *entry; | |
9014623c | 317 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
318 | int leftmost = 1; |
319 | ||
320 | /* | |
321 | * Find the right place in the rbtree: | |
322 | */ | |
323 | while (*link) { | |
324 | parent = *link; | |
325 | entry = rb_entry(parent, struct sched_entity, run_node); | |
326 | /* | |
327 | * We dont care about collisions. Nodes with | |
328 | * the same key stay together. | |
329 | */ | |
9014623c | 330 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
331 | link = &parent->rb_left; |
332 | } else { | |
333 | link = &parent->rb_right; | |
334 | leftmost = 0; | |
335 | } | |
336 | } | |
337 | ||
338 | /* | |
339 | * Maintain a cache of leftmost tree entries (it is frequently | |
340 | * used): | |
341 | */ | |
1af5f730 | 342 | if (leftmost) |
57cb499d | 343 | cfs_rq->rb_leftmost = &se->run_node; |
bf0f6f24 IM |
344 | |
345 | rb_link_node(&se->run_node, parent, link); | |
346 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
347 | } |
348 | ||
0702e3eb | 349 | static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 350 | { |
3fe69747 PZ |
351 | if (cfs_rq->rb_leftmost == &se->run_node) { |
352 | struct rb_node *next_node; | |
3fe69747 PZ |
353 | |
354 | next_node = rb_next(&se->run_node); | |
355 | cfs_rq->rb_leftmost = next_node; | |
3fe69747 | 356 | } |
e9acbff6 | 357 | |
bf0f6f24 | 358 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
359 | } |
360 | ||
bf0f6f24 IM |
361 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) |
362 | { | |
f4b6755f PZ |
363 | struct rb_node *left = cfs_rq->rb_leftmost; |
364 | ||
365 | if (!left) | |
366 | return NULL; | |
367 | ||
368 | return rb_entry(left, struct sched_entity, run_node); | |
bf0f6f24 IM |
369 | } |
370 | ||
f4b6755f | 371 | static struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
aeb73b04 | 372 | { |
7eee3e67 | 373 | struct rb_node *last = rb_last(&cfs_rq->tasks_timeline); |
aeb73b04 | 374 | |
70eee74b BS |
375 | if (!last) |
376 | return NULL; | |
7eee3e67 IM |
377 | |
378 | return rb_entry(last, struct sched_entity, run_node); | |
aeb73b04 PZ |
379 | } |
380 | ||
bf0f6f24 IM |
381 | /************************************************************** |
382 | * Scheduling class statistics methods: | |
383 | */ | |
384 | ||
b2be5e96 PZ |
385 | #ifdef CONFIG_SCHED_DEBUG |
386 | int sched_nr_latency_handler(struct ctl_table *table, int write, | |
387 | struct file *filp, void __user *buffer, size_t *lenp, | |
388 | loff_t *ppos) | |
389 | { | |
390 | int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); | |
391 | ||
392 | if (ret || !write) | |
393 | return ret; | |
394 | ||
395 | sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency, | |
396 | sysctl_sched_min_granularity); | |
397 | ||
398 | return 0; | |
399 | } | |
400 | #endif | |
647e7cac | 401 | |
a7be37ac | 402 | /* |
f9c0b095 | 403 | * delta /= w |
a7be37ac PZ |
404 | */ |
405 | static inline unsigned long | |
406 | calc_delta_fair(unsigned long delta, struct sched_entity *se) | |
407 | { | |
f9c0b095 PZ |
408 | if (unlikely(se->load.weight != NICE_0_LOAD)) |
409 | delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load); | |
a7be37ac PZ |
410 | |
411 | return delta; | |
412 | } | |
413 | ||
647e7cac IM |
414 | /* |
415 | * The idea is to set a period in which each task runs once. | |
416 | * | |
417 | * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch | |
418 | * this period because otherwise the slices get too small. | |
419 | * | |
420 | * p = (nr <= nl) ? l : l*nr/nl | |
421 | */ | |
4d78e7b6 PZ |
422 | static u64 __sched_period(unsigned long nr_running) |
423 | { | |
424 | u64 period = sysctl_sched_latency; | |
b2be5e96 | 425 | unsigned long nr_latency = sched_nr_latency; |
4d78e7b6 PZ |
426 | |
427 | if (unlikely(nr_running > nr_latency)) { | |
4bf0b771 | 428 | period = sysctl_sched_min_granularity; |
4d78e7b6 | 429 | period *= nr_running; |
4d78e7b6 PZ |
430 | } |
431 | ||
432 | return period; | |
433 | } | |
434 | ||
647e7cac IM |
435 | /* |
436 | * We calculate the wall-time slice from the period by taking a part | |
437 | * proportional to the weight. | |
438 | * | |
f9c0b095 | 439 | * s = p*P[w/rw] |
647e7cac | 440 | */ |
6d0f0ebd | 441 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 442 | { |
0a582440 | 443 | u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq); |
f9c0b095 | 444 | |
0a582440 | 445 | for_each_sched_entity(se) { |
6272d68c | 446 | struct load_weight *load; |
3104bf03 | 447 | struct load_weight lw; |
6272d68c LM |
448 | |
449 | cfs_rq = cfs_rq_of(se); | |
450 | load = &cfs_rq->load; | |
f9c0b095 | 451 | |
0a582440 | 452 | if (unlikely(!se->on_rq)) { |
3104bf03 | 453 | lw = cfs_rq->load; |
0a582440 MG |
454 | |
455 | update_load_add(&lw, se->load.weight); | |
456 | load = &lw; | |
457 | } | |
458 | slice = calc_delta_mine(slice, se->load.weight, load); | |
459 | } | |
460 | return slice; | |
bf0f6f24 IM |
461 | } |
462 | ||
647e7cac | 463 | /* |
ac884dec | 464 | * We calculate the vruntime slice of a to be inserted task |
647e7cac | 465 | * |
f9c0b095 | 466 | * vs = s/w |
647e7cac | 467 | */ |
f9c0b095 | 468 | static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
67e9fb2a | 469 | { |
f9c0b095 | 470 | return calc_delta_fair(sched_slice(cfs_rq, se), se); |
a7be37ac PZ |
471 | } |
472 | ||
bf0f6f24 IM |
473 | /* |
474 | * Update the current task's runtime statistics. Skip current tasks that | |
475 | * are not in our scheduling class. | |
476 | */ | |
477 | static inline void | |
8ebc91d9 IM |
478 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
479 | unsigned long delta_exec) | |
bf0f6f24 | 480 | { |
bbdba7c0 | 481 | unsigned long delta_exec_weighted; |
bf0f6f24 | 482 | |
8179ca23 | 483 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
484 | |
485 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 486 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
a7be37ac | 487 | delta_exec_weighted = calc_delta_fair(delta_exec, curr); |
e9acbff6 | 488 | curr->vruntime += delta_exec_weighted; |
1af5f730 | 489 | update_min_vruntime(cfs_rq); |
bf0f6f24 IM |
490 | } |
491 | ||
b7cc0896 | 492 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 493 | { |
429d43bc | 494 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 495 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
496 | unsigned long delta_exec; |
497 | ||
498 | if (unlikely(!curr)) | |
499 | return; | |
500 | ||
501 | /* | |
502 | * Get the amount of time the current task was running | |
503 | * since the last time we changed load (this cannot | |
504 | * overflow on 32 bits): | |
505 | */ | |
8ebc91d9 | 506 | delta_exec = (unsigned long)(now - curr->exec_start); |
34f28ecd PZ |
507 | if (!delta_exec) |
508 | return; | |
bf0f6f24 | 509 | |
8ebc91d9 IM |
510 | __update_curr(cfs_rq, curr, delta_exec); |
511 | curr->exec_start = now; | |
d842de87 SV |
512 | |
513 | if (entity_is_task(curr)) { | |
514 | struct task_struct *curtask = task_of(curr); | |
515 | ||
516 | cpuacct_charge(curtask, delta_exec); | |
f06febc9 | 517 | account_group_exec_runtime(curtask, delta_exec); |
d842de87 | 518 | } |
bf0f6f24 IM |
519 | } |
520 | ||
521 | static inline void | |
5870db5b | 522 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 523 | { |
d281918d | 524 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
525 | } |
526 | ||
bf0f6f24 IM |
527 | /* |
528 | * Task is being enqueued - update stats: | |
529 | */ | |
d2417e5a | 530 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 531 | { |
bf0f6f24 IM |
532 | /* |
533 | * Are we enqueueing a waiting task? (for current tasks | |
534 | * a dequeue/enqueue event is a NOP) | |
535 | */ | |
429d43bc | 536 | if (se != cfs_rq->curr) |
5870db5b | 537 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
538 | } |
539 | ||
bf0f6f24 | 540 | static void |
9ef0a961 | 541 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 542 | { |
bbdba7c0 IM |
543 | schedstat_set(se->wait_max, max(se->wait_max, |
544 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6d082592 AV |
545 | schedstat_set(se->wait_count, se->wait_count + 1); |
546 | schedstat_set(se->wait_sum, se->wait_sum + | |
547 | rq_of(cfs_rq)->clock - se->wait_start); | |
6cfb0d5d | 548 | schedstat_set(se->wait_start, 0); |
768d0c27 PZ |
549 | |
550 | #ifdef CONFIG_SCHEDSTATS | |
551 | if (entity_is_task(se)) { | |
552 | trace_sched_stat_wait(task_of(se), | |
553 | rq_of(cfs_rq)->clock - se->wait_start); | |
554 | } | |
555 | #endif | |
bf0f6f24 IM |
556 | } |
557 | ||
558 | static inline void | |
19b6a2e3 | 559 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 560 | { |
bf0f6f24 IM |
561 | /* |
562 | * Mark the end of the wait period if dequeueing a | |
563 | * waiting task: | |
564 | */ | |
429d43bc | 565 | if (se != cfs_rq->curr) |
9ef0a961 | 566 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
567 | } |
568 | ||
569 | /* | |
570 | * We are picking a new current task - update its stats: | |
571 | */ | |
572 | static inline void | |
79303e9e | 573 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
574 | { |
575 | /* | |
576 | * We are starting a new run period: | |
577 | */ | |
d281918d | 578 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
579 | } |
580 | ||
bf0f6f24 IM |
581 | /************************************************** |
582 | * Scheduling class queueing methods: | |
583 | */ | |
584 | ||
c09595f6 PZ |
585 | #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED |
586 | static void | |
587 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
588 | { | |
589 | cfs_rq->task_weight += weight; | |
590 | } | |
591 | #else | |
592 | static inline void | |
593 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
594 | { | |
595 | } | |
596 | #endif | |
597 | ||
30cfdcfc DA |
598 | static void |
599 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
600 | { | |
601 | update_load_add(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
602 | if (!parent_entity(se)) |
603 | inc_cpu_load(rq_of(cfs_rq), se->load.weight); | |
b87f1724 | 604 | if (entity_is_task(se)) { |
c09595f6 | 605 | add_cfs_task_weight(cfs_rq, se->load.weight); |
b87f1724 BR |
606 | list_add(&se->group_node, &cfs_rq->tasks); |
607 | } | |
30cfdcfc DA |
608 | cfs_rq->nr_running++; |
609 | se->on_rq = 1; | |
610 | } | |
611 | ||
612 | static void | |
613 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
614 | { | |
615 | update_load_sub(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
616 | if (!parent_entity(se)) |
617 | dec_cpu_load(rq_of(cfs_rq), se->load.weight); | |
b87f1724 | 618 | if (entity_is_task(se)) { |
c09595f6 | 619 | add_cfs_task_weight(cfs_rq, -se->load.weight); |
b87f1724 BR |
620 | list_del_init(&se->group_node); |
621 | } | |
30cfdcfc DA |
622 | cfs_rq->nr_running--; |
623 | se->on_rq = 0; | |
624 | } | |
625 | ||
2396af69 | 626 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 627 | { |
bf0f6f24 | 628 | #ifdef CONFIG_SCHEDSTATS |
e414314c PZ |
629 | struct task_struct *tsk = NULL; |
630 | ||
631 | if (entity_is_task(se)) | |
632 | tsk = task_of(se); | |
633 | ||
bf0f6f24 | 634 | if (se->sleep_start) { |
d281918d | 635 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
bf0f6f24 IM |
636 | |
637 | if ((s64)delta < 0) | |
638 | delta = 0; | |
639 | ||
640 | if (unlikely(delta > se->sleep_max)) | |
641 | se->sleep_max = delta; | |
642 | ||
643 | se->sleep_start = 0; | |
644 | se->sum_sleep_runtime += delta; | |
9745512c | 645 | |
768d0c27 | 646 | if (tsk) { |
e414314c | 647 | account_scheduler_latency(tsk, delta >> 10, 1); |
768d0c27 PZ |
648 | trace_sched_stat_sleep(tsk, delta); |
649 | } | |
bf0f6f24 IM |
650 | } |
651 | if (se->block_start) { | |
d281918d | 652 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
bf0f6f24 IM |
653 | |
654 | if ((s64)delta < 0) | |
655 | delta = 0; | |
656 | ||
657 | if (unlikely(delta > se->block_max)) | |
658 | se->block_max = delta; | |
659 | ||
660 | se->block_start = 0; | |
661 | se->sum_sleep_runtime += delta; | |
30084fbd | 662 | |
e414314c | 663 | if (tsk) { |
8f0dfc34 AV |
664 | if (tsk->in_iowait) { |
665 | se->iowait_sum += delta; | |
666 | se->iowait_count++; | |
768d0c27 | 667 | trace_sched_stat_iowait(tsk, delta); |
8f0dfc34 AV |
668 | } |
669 | ||
e414314c PZ |
670 | /* |
671 | * Blocking time is in units of nanosecs, so shift by | |
672 | * 20 to get a milliseconds-range estimation of the | |
673 | * amount of time that the task spent sleeping: | |
674 | */ | |
675 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
676 | profile_hits(SLEEP_PROFILING, | |
677 | (void *)get_wchan(tsk), | |
678 | delta >> 20); | |
679 | } | |
680 | account_scheduler_latency(tsk, delta >> 10, 0); | |
30084fbd | 681 | } |
bf0f6f24 IM |
682 | } |
683 | #endif | |
684 | } | |
685 | ||
ddc97297 PZ |
686 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
687 | { | |
688 | #ifdef CONFIG_SCHED_DEBUG | |
689 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
690 | ||
691 | if (d < 0) | |
692 | d = -d; | |
693 | ||
694 | if (d > 3*sysctl_sched_latency) | |
695 | schedstat_inc(cfs_rq, nr_spread_over); | |
696 | #endif | |
697 | } | |
698 | ||
aeb73b04 PZ |
699 | static void |
700 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
701 | { | |
1af5f730 | 702 | u64 vruntime = cfs_rq->min_vruntime; |
94dfb5e7 | 703 | |
2cb8600e PZ |
704 | /* |
705 | * The 'current' period is already promised to the current tasks, | |
706 | * however the extra weight of the new task will slow them down a | |
707 | * little, place the new task so that it fits in the slot that | |
708 | * stays open at the end. | |
709 | */ | |
94dfb5e7 | 710 | if (initial && sched_feat(START_DEBIT)) |
f9c0b095 | 711 | vruntime += sched_vslice(cfs_rq, se); |
aeb73b04 | 712 | |
8465e792 | 713 | if (!initial) { |
2cb8600e | 714 | /* sleeps upto a single latency don't count. */ |
a7be37ac PZ |
715 | if (sched_feat(NEW_FAIR_SLEEPERS)) { |
716 | unsigned long thresh = sysctl_sched_latency; | |
717 | ||
718 | /* | |
6bc912b7 PZ |
719 | * Convert the sleeper threshold into virtual time. |
720 | * SCHED_IDLE is a special sub-class. We care about | |
721 | * fairness only relative to other SCHED_IDLE tasks, | |
722 | * all of which have the same weight. | |
a7be37ac | 723 | */ |
6bc912b7 | 724 | if (sched_feat(NORMALIZED_SLEEPER) && |
d07387b4 PT |
725 | (!entity_is_task(se) || |
726 | task_of(se)->policy != SCHED_IDLE)) | |
a7be37ac PZ |
727 | thresh = calc_delta_fair(thresh, se); |
728 | ||
729 | vruntime -= thresh; | |
730 | } | |
94359f05 | 731 | |
2cb8600e PZ |
732 | /* ensure we never gain time by being placed backwards. */ |
733 | vruntime = max_vruntime(se->vruntime, vruntime); | |
aeb73b04 PZ |
734 | } |
735 | ||
67e9fb2a | 736 | se->vruntime = vruntime; |
aeb73b04 PZ |
737 | } |
738 | ||
bf0f6f24 | 739 | static void |
83b699ed | 740 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
741 | { |
742 | /* | |
a2a2d680 | 743 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 744 | */ |
b7cc0896 | 745 | update_curr(cfs_rq); |
a992241d | 746 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 | 747 | |
e9acbff6 | 748 | if (wakeup) { |
aeb73b04 | 749 | place_entity(cfs_rq, se, 0); |
2396af69 | 750 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 751 | } |
bf0f6f24 | 752 | |
d2417e5a | 753 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 754 | check_spread(cfs_rq, se); |
83b699ed SV |
755 | if (se != cfs_rq->curr) |
756 | __enqueue_entity(cfs_rq, se); | |
bf0f6f24 IM |
757 | } |
758 | ||
a571bbea | 759 | static void __clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se) |
2002c695 PZ |
760 | { |
761 | if (cfs_rq->last == se) | |
762 | cfs_rq->last = NULL; | |
763 | ||
764 | if (cfs_rq->next == se) | |
765 | cfs_rq->next = NULL; | |
766 | } | |
767 | ||
a571bbea PZ |
768 | static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se) |
769 | { | |
770 | for_each_sched_entity(se) | |
771 | __clear_buddies(cfs_rq_of(se), se); | |
772 | } | |
773 | ||
bf0f6f24 | 774 | static void |
525c2716 | 775 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 776 | { |
a2a2d680 DA |
777 | /* |
778 | * Update run-time statistics of the 'current'. | |
779 | */ | |
780 | update_curr(cfs_rq); | |
781 | ||
19b6a2e3 | 782 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 783 | if (sleep) { |
67e9fb2a | 784 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
785 | if (entity_is_task(se)) { |
786 | struct task_struct *tsk = task_of(se); | |
787 | ||
788 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 789 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 790 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 791 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 792 | } |
db36cc7d | 793 | #endif |
67e9fb2a PZ |
794 | } |
795 | ||
2002c695 | 796 | clear_buddies(cfs_rq, se); |
4793241b | 797 | |
83b699ed | 798 | if (se != cfs_rq->curr) |
30cfdcfc DA |
799 | __dequeue_entity(cfs_rq, se); |
800 | account_entity_dequeue(cfs_rq, se); | |
1af5f730 | 801 | update_min_vruntime(cfs_rq); |
bf0f6f24 IM |
802 | } |
803 | ||
804 | /* | |
805 | * Preempt the current task with a newly woken task if needed: | |
806 | */ | |
7c92e54f | 807 | static void |
2e09bf55 | 808 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 809 | { |
11697830 PZ |
810 | unsigned long ideal_runtime, delta_exec; |
811 | ||
6d0f0ebd | 812 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 | 813 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
a9f3e2b5 | 814 | if (delta_exec > ideal_runtime) { |
bf0f6f24 | 815 | resched_task(rq_of(cfs_rq)->curr); |
a9f3e2b5 MG |
816 | /* |
817 | * The current task ran long enough, ensure it doesn't get | |
818 | * re-elected due to buddy favours. | |
819 | */ | |
820 | clear_buddies(cfs_rq, curr); | |
821 | } | |
bf0f6f24 IM |
822 | } |
823 | ||
83b699ed | 824 | static void |
8494f412 | 825 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 826 | { |
83b699ed SV |
827 | /* 'current' is not kept within the tree. */ |
828 | if (se->on_rq) { | |
829 | /* | |
830 | * Any task has to be enqueued before it get to execute on | |
831 | * a CPU. So account for the time it spent waiting on the | |
832 | * runqueue. | |
833 | */ | |
834 | update_stats_wait_end(cfs_rq, se); | |
835 | __dequeue_entity(cfs_rq, se); | |
836 | } | |
837 | ||
79303e9e | 838 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 839 | cfs_rq->curr = se; |
eba1ed4b IM |
840 | #ifdef CONFIG_SCHEDSTATS |
841 | /* | |
842 | * Track our maximum slice length, if the CPU's load is at | |
843 | * least twice that of our own weight (i.e. dont track it | |
844 | * when there are only lesser-weight tasks around): | |
845 | */ | |
495eca49 | 846 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
847 | se->slice_max = max(se->slice_max, |
848 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
849 | } | |
850 | #endif | |
4a55b450 | 851 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
852 | } |
853 | ||
3f3a4904 PZ |
854 | static int |
855 | wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se); | |
856 | ||
f4b6755f | 857 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
aa2ac252 | 858 | { |
f4b6755f PZ |
859 | struct sched_entity *se = __pick_next_entity(cfs_rq); |
860 | ||
4793241b PZ |
861 | if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, se) < 1) |
862 | return cfs_rq->next; | |
aa2ac252 | 863 | |
4793241b PZ |
864 | if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, se) < 1) |
865 | return cfs_rq->last; | |
866 | ||
867 | return se; | |
aa2ac252 PZ |
868 | } |
869 | ||
ab6cde26 | 870 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
871 | { |
872 | /* | |
873 | * If still on the runqueue then deactivate_task() | |
874 | * was not called and update_curr() has to be done: | |
875 | */ | |
876 | if (prev->on_rq) | |
b7cc0896 | 877 | update_curr(cfs_rq); |
bf0f6f24 | 878 | |
ddc97297 | 879 | check_spread(cfs_rq, prev); |
30cfdcfc | 880 | if (prev->on_rq) { |
5870db5b | 881 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
882 | /* Put 'current' back into the tree. */ |
883 | __enqueue_entity(cfs_rq, prev); | |
884 | } | |
429d43bc | 885 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
886 | } |
887 | ||
8f4d37ec PZ |
888 | static void |
889 | entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) | |
bf0f6f24 | 890 | { |
bf0f6f24 | 891 | /* |
30cfdcfc | 892 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 893 | */ |
30cfdcfc | 894 | update_curr(cfs_rq); |
bf0f6f24 | 895 | |
8f4d37ec PZ |
896 | #ifdef CONFIG_SCHED_HRTICK |
897 | /* | |
898 | * queued ticks are scheduled to match the slice, so don't bother | |
899 | * validating it and just reschedule. | |
900 | */ | |
983ed7a6 HH |
901 | if (queued) { |
902 | resched_task(rq_of(cfs_rq)->curr); | |
903 | return; | |
904 | } | |
8f4d37ec PZ |
905 | /* |
906 | * don't let the period tick interfere with the hrtick preemption | |
907 | */ | |
908 | if (!sched_feat(DOUBLE_TICK) && | |
909 | hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) | |
910 | return; | |
911 | #endif | |
912 | ||
ce6c1311 | 913 | if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT)) |
2e09bf55 | 914 | check_preempt_tick(cfs_rq, curr); |
bf0f6f24 IM |
915 | } |
916 | ||
917 | /************************************************** | |
918 | * CFS operations on tasks: | |
919 | */ | |
920 | ||
8f4d37ec PZ |
921 | #ifdef CONFIG_SCHED_HRTICK |
922 | static void hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
923 | { | |
8f4d37ec PZ |
924 | struct sched_entity *se = &p->se; |
925 | struct cfs_rq *cfs_rq = cfs_rq_of(se); | |
926 | ||
927 | WARN_ON(task_rq(p) != rq); | |
928 | ||
929 | if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) { | |
930 | u64 slice = sched_slice(cfs_rq, se); | |
931 | u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime; | |
932 | s64 delta = slice - ran; | |
933 | ||
934 | if (delta < 0) { | |
935 | if (rq->curr == p) | |
936 | resched_task(p); | |
937 | return; | |
938 | } | |
939 | ||
940 | /* | |
941 | * Don't schedule slices shorter than 10000ns, that just | |
942 | * doesn't make sense. Rely on vruntime for fairness. | |
943 | */ | |
31656519 | 944 | if (rq->curr != p) |
157124c1 | 945 | delta = max_t(s64, 10000LL, delta); |
8f4d37ec | 946 | |
31656519 | 947 | hrtick_start(rq, delta); |
8f4d37ec PZ |
948 | } |
949 | } | |
a4c2f00f PZ |
950 | |
951 | /* | |
952 | * called from enqueue/dequeue and updates the hrtick when the | |
953 | * current task is from our class and nr_running is low enough | |
954 | * to matter. | |
955 | */ | |
956 | static void hrtick_update(struct rq *rq) | |
957 | { | |
958 | struct task_struct *curr = rq->curr; | |
959 | ||
960 | if (curr->sched_class != &fair_sched_class) | |
961 | return; | |
962 | ||
963 | if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency) | |
964 | hrtick_start_fair(rq, curr); | |
965 | } | |
55e12e5e | 966 | #else /* !CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
967 | static inline void |
968 | hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
969 | { | |
970 | } | |
a4c2f00f PZ |
971 | |
972 | static inline void hrtick_update(struct rq *rq) | |
973 | { | |
974 | } | |
8f4d37ec PZ |
975 | #endif |
976 | ||
bf0f6f24 IM |
977 | /* |
978 | * The enqueue_task method is called before nr_running is | |
979 | * increased. Here we update the fair scheduling stats and | |
980 | * then put the task into the rbtree: | |
981 | */ | |
fd390f6a | 982 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
983 | { |
984 | struct cfs_rq *cfs_rq; | |
62fb1851 | 985 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
986 | |
987 | for_each_sched_entity(se) { | |
62fb1851 | 988 | if (se->on_rq) |
bf0f6f24 IM |
989 | break; |
990 | cfs_rq = cfs_rq_of(se); | |
83b699ed | 991 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 992 | wakeup = 1; |
bf0f6f24 | 993 | } |
8f4d37ec | 994 | |
a4c2f00f | 995 | hrtick_update(rq); |
bf0f6f24 IM |
996 | } |
997 | ||
998 | /* | |
999 | * The dequeue_task method is called before nr_running is | |
1000 | * decreased. We remove the task from the rbtree and | |
1001 | * update the fair scheduling stats: | |
1002 | */ | |
f02231e5 | 1003 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
1004 | { |
1005 | struct cfs_rq *cfs_rq; | |
62fb1851 | 1006 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
1007 | |
1008 | for_each_sched_entity(se) { | |
1009 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 1010 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 | 1011 | /* Don't dequeue parent if it has other entities besides us */ |
62fb1851 | 1012 | if (cfs_rq->load.weight) |
bf0f6f24 | 1013 | break; |
b9fa3df3 | 1014 | sleep = 1; |
bf0f6f24 | 1015 | } |
8f4d37ec | 1016 | |
a4c2f00f | 1017 | hrtick_update(rq); |
bf0f6f24 IM |
1018 | } |
1019 | ||
1020 | /* | |
1799e35d IM |
1021 | * sched_yield() support is very simple - we dequeue and enqueue. |
1022 | * | |
1023 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 1024 | */ |
4530d7ab | 1025 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 1026 | { |
db292ca3 IM |
1027 | struct task_struct *curr = rq->curr; |
1028 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
1029 | struct sched_entity *rightmost, *se = &curr->se; | |
bf0f6f24 IM |
1030 | |
1031 | /* | |
1799e35d IM |
1032 | * Are we the only task in the tree? |
1033 | */ | |
1034 | if (unlikely(cfs_rq->nr_running == 1)) | |
1035 | return; | |
1036 | ||
2002c695 PZ |
1037 | clear_buddies(cfs_rq, se); |
1038 | ||
db292ca3 | 1039 | if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) { |
3e51f33f | 1040 | update_rq_clock(rq); |
1799e35d | 1041 | /* |
a2a2d680 | 1042 | * Update run-time statistics of the 'current'. |
1799e35d | 1043 | */ |
2b1e315d | 1044 | update_curr(cfs_rq); |
1799e35d IM |
1045 | |
1046 | return; | |
1047 | } | |
1048 | /* | |
1049 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 1050 | */ |
2b1e315d | 1051 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
1052 | /* |
1053 | * Already in the rightmost position? | |
1054 | */ | |
54fdc581 | 1055 | if (unlikely(!rightmost || entity_before(rightmost, se))) |
1799e35d IM |
1056 | return; |
1057 | ||
1058 | /* | |
1059 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
1060 | * Upon rescheduling, sched_class::put_prev_task() will place |
1061 | * 'current' within the tree based on its new key value. | |
1799e35d | 1062 | */ |
30cfdcfc | 1063 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
1064 | } |
1065 | ||
e7693a36 GH |
1066 | /* |
1067 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1068 | * not idle and an idle cpu is available. The span of cpus to | |
1069 | * search starts with cpus closest then further out as needed, | |
1070 | * so we always favor a closer, idle cpu. | |
e761b772 | 1071 | * Domains may include CPUs that are not usable for migration, |
00aec93d | 1072 | * hence we need to mask them out (rq->rd->online) |
e7693a36 GH |
1073 | * |
1074 | * Returns the CPU we should wake onto. | |
1075 | */ | |
1076 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
00aec93d GH |
1077 | |
1078 | #define cpu_rd_active(cpu, rq) cpumask_test_cpu(cpu, rq->rd->online) | |
1079 | ||
e7693a36 GH |
1080 | static int wake_idle(int cpu, struct task_struct *p) |
1081 | { | |
e7693a36 GH |
1082 | struct sched_domain *sd; |
1083 | int i; | |
7eb52dfa VS |
1084 | unsigned int chosen_wakeup_cpu; |
1085 | int this_cpu; | |
00aec93d | 1086 | struct rq *task_rq = task_rq(p); |
7eb52dfa VS |
1087 | |
1088 | /* | |
1089 | * At POWERSAVINGS_BALANCE_WAKEUP level, if both this_cpu and prev_cpu | |
1090 | * are idle and this is not a kernel thread and this task's affinity | |
1091 | * allows it to be moved to preferred cpu, then just move! | |
1092 | */ | |
1093 | ||
1094 | this_cpu = smp_processor_id(); | |
1095 | chosen_wakeup_cpu = | |
1096 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu; | |
1097 | ||
1098 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP && | |
1099 | idle_cpu(cpu) && idle_cpu(this_cpu) && | |
1100 | p->mm && !(p->flags & PF_KTHREAD) && | |
1101 | cpu_isset(chosen_wakeup_cpu, p->cpus_allowed)) | |
1102 | return chosen_wakeup_cpu; | |
e7693a36 GH |
1103 | |
1104 | /* | |
1105 | * If it is idle, then it is the best cpu to run this task. | |
1106 | * | |
1107 | * This cpu is also the best, if it has more than one task already. | |
1108 | * Siblings must be also busy(in most cases) as they didn't already | |
1109 | * pickup the extra load from this cpu and hence we need not check | |
1110 | * sibling runqueue info. This will avoid the checks and cache miss | |
1111 | * penalities associated with that. | |
1112 | */ | |
104f6454 | 1113 | if (idle_cpu(cpu) || cpu_rq(cpu)->cfs.nr_running > 1) |
e7693a36 GH |
1114 | return cpu; |
1115 | ||
1116 | for_each_domain(cpu, sd) { | |
1d3504fc HS |
1117 | if ((sd->flags & SD_WAKE_IDLE) |
1118 | || ((sd->flags & SD_WAKE_IDLE_FAR) | |
00aec93d | 1119 | && !task_hot(p, task_rq->clock, sd))) { |
758b2cdc RR |
1120 | for_each_cpu_and(i, sched_domain_span(sd), |
1121 | &p->cpus_allowed) { | |
00aec93d | 1122 | if (cpu_rd_active(i, task_rq) && idle_cpu(i)) { |
e7693a36 GH |
1123 | if (i != task_cpu(p)) { |
1124 | schedstat_inc(p, | |
1125 | se.nr_wakeups_idle); | |
1126 | } | |
1127 | return i; | |
1128 | } | |
1129 | } | |
1130 | } else { | |
1131 | break; | |
1132 | } | |
1133 | } | |
1134 | return cpu; | |
1135 | } | |
55e12e5e | 1136 | #else /* !ARCH_HAS_SCHED_WAKE_IDLE*/ |
e7693a36 GH |
1137 | static inline int wake_idle(int cpu, struct task_struct *p) |
1138 | { | |
1139 | return cpu; | |
1140 | } | |
1141 | #endif | |
1142 | ||
1143 | #ifdef CONFIG_SMP | |
098fb9db | 1144 | |
bb3469ac | 1145 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f5bfb7d9 PZ |
1146 | /* |
1147 | * effective_load() calculates the load change as seen from the root_task_group | |
1148 | * | |
1149 | * Adding load to a group doesn't make a group heavier, but can cause movement | |
1150 | * of group shares between cpus. Assuming the shares were perfectly aligned one | |
1151 | * can calculate the shift in shares. | |
1152 | * | |
1153 | * The problem is that perfectly aligning the shares is rather expensive, hence | |
1154 | * we try to avoid doing that too often - see update_shares(), which ratelimits | |
1155 | * this change. | |
1156 | * | |
1157 | * We compensate this by not only taking the current delta into account, but | |
1158 | * also considering the delta between when the shares were last adjusted and | |
1159 | * now. | |
1160 | * | |
1161 | * We still saw a performance dip, some tracing learned us that between | |
1162 | * cgroup:/ and cgroup:/foo balancing the number of affine wakeups increased | |
1163 | * significantly. Therefore try to bias the error in direction of failing | |
1164 | * the affine wakeup. | |
1165 | * | |
1166 | */ | |
f1d239f7 PZ |
1167 | static long effective_load(struct task_group *tg, int cpu, |
1168 | long wl, long wg) | |
bb3469ac | 1169 | { |
4be9daaa | 1170 | struct sched_entity *se = tg->se[cpu]; |
f1d239f7 PZ |
1171 | |
1172 | if (!tg->parent) | |
1173 | return wl; | |
1174 | ||
f5bfb7d9 PZ |
1175 | /* |
1176 | * By not taking the decrease of shares on the other cpu into | |
1177 | * account our error leans towards reducing the affine wakeups. | |
1178 | */ | |
1179 | if (!wl && sched_feat(ASYM_EFF_LOAD)) | |
1180 | return wl; | |
1181 | ||
4be9daaa | 1182 | for_each_sched_entity(se) { |
cb5ef42a | 1183 | long S, rw, s, a, b; |
940959e9 PZ |
1184 | long more_w; |
1185 | ||
1186 | /* | |
1187 | * Instead of using this increment, also add the difference | |
1188 | * between when the shares were last updated and now. | |
1189 | */ | |
1190 | more_w = se->my_q->load.weight - se->my_q->rq_weight; | |
1191 | wl += more_w; | |
1192 | wg += more_w; | |
4be9daaa PZ |
1193 | |
1194 | S = se->my_q->tg->shares; | |
1195 | s = se->my_q->shares; | |
f1d239f7 | 1196 | rw = se->my_q->rq_weight; |
bb3469ac | 1197 | |
cb5ef42a PZ |
1198 | a = S*(rw + wl); |
1199 | b = S*rw + s*wg; | |
4be9daaa | 1200 | |
940959e9 PZ |
1201 | wl = s*(a-b); |
1202 | ||
1203 | if (likely(b)) | |
1204 | wl /= b; | |
1205 | ||
83378269 PZ |
1206 | /* |
1207 | * Assume the group is already running and will | |
1208 | * thus already be accounted for in the weight. | |
1209 | * | |
1210 | * That is, moving shares between CPUs, does not | |
1211 | * alter the group weight. | |
1212 | */ | |
4be9daaa | 1213 | wg = 0; |
4be9daaa | 1214 | } |
bb3469ac | 1215 | |
4be9daaa | 1216 | return wl; |
bb3469ac | 1217 | } |
4be9daaa | 1218 | |
bb3469ac | 1219 | #else |
4be9daaa | 1220 | |
83378269 PZ |
1221 | static inline unsigned long effective_load(struct task_group *tg, int cpu, |
1222 | unsigned long wl, unsigned long wg) | |
4be9daaa | 1223 | { |
83378269 | 1224 | return wl; |
bb3469ac | 1225 | } |
4be9daaa | 1226 | |
bb3469ac PZ |
1227 | #endif |
1228 | ||
098fb9db | 1229 | static int |
64b9e029 | 1230 | wake_affine(struct sched_domain *this_sd, struct rq *this_rq, |
4ae7d5ce IM |
1231 | struct task_struct *p, int prev_cpu, int this_cpu, int sync, |
1232 | int idx, unsigned long load, unsigned long this_load, | |
098fb9db IM |
1233 | unsigned int imbalance) |
1234 | { | |
fc631c82 PZ |
1235 | struct task_struct *curr = this_rq->curr; |
1236 | struct task_group *tg; | |
098fb9db IM |
1237 | unsigned long tl = this_load; |
1238 | unsigned long tl_per_task; | |
83378269 | 1239 | unsigned long weight; |
b3137bc8 | 1240 | int balanced; |
098fb9db | 1241 | |
b3137bc8 | 1242 | if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS)) |
098fb9db IM |
1243 | return 0; |
1244 | ||
fc631c82 PZ |
1245 | if (sync && (curr->se.avg_overlap > sysctl_sched_migration_cost || |
1246 | p->se.avg_overlap > sysctl_sched_migration_cost)) | |
1247 | sync = 0; | |
1248 | ||
b3137bc8 MG |
1249 | /* |
1250 | * If sync wakeup then subtract the (maximum possible) | |
1251 | * effect of the currently running task from the load | |
1252 | * of the current CPU: | |
1253 | */ | |
83378269 PZ |
1254 | if (sync) { |
1255 | tg = task_group(current); | |
1256 | weight = current->se.load.weight; | |
1257 | ||
1258 | tl += effective_load(tg, this_cpu, -weight, -weight); | |
1259 | load += effective_load(tg, prev_cpu, 0, -weight); | |
1260 | } | |
b3137bc8 | 1261 | |
83378269 PZ |
1262 | tg = task_group(p); |
1263 | weight = p->se.load.weight; | |
b3137bc8 | 1264 | |
83378269 PZ |
1265 | balanced = 100*(tl + effective_load(tg, this_cpu, weight, weight)) <= |
1266 | imbalance*(load + effective_load(tg, prev_cpu, 0, weight)); | |
b3137bc8 | 1267 | |
098fb9db | 1268 | /* |
4ae7d5ce IM |
1269 | * If the currently running task will sleep within |
1270 | * a reasonable amount of time then attract this newly | |
1271 | * woken task: | |
098fb9db | 1272 | */ |
2fb7635c PZ |
1273 | if (sync && balanced) |
1274 | return 1; | |
098fb9db IM |
1275 | |
1276 | schedstat_inc(p, se.nr_wakeups_affine_attempts); | |
1277 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
1278 | ||
64b9e029 AA |
1279 | if (balanced || (tl <= load && tl + target_load(prev_cpu, idx) <= |
1280 | 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 | */ | |
1286 | schedstat_inc(this_sd, ttwu_move_affine); | |
1287 | schedstat_inc(p, se.nr_wakeups_affine); | |
1288 | ||
1289 | return 1; | |
1290 | } | |
1291 | return 0; | |
1292 | } | |
1293 | ||
e7693a36 GH |
1294 | static int select_task_rq_fair(struct task_struct *p, int sync) |
1295 | { | |
e7693a36 | 1296 | struct sched_domain *sd, *this_sd = NULL; |
ac192d39 | 1297 | int prev_cpu, this_cpu, new_cpu; |
098fb9db | 1298 | unsigned long load, this_load; |
64b9e029 | 1299 | struct rq *this_rq; |
098fb9db | 1300 | unsigned int imbalance; |
098fb9db | 1301 | int idx; |
e7693a36 | 1302 | |
ac192d39 | 1303 | prev_cpu = task_cpu(p); |
ac192d39 | 1304 | this_cpu = smp_processor_id(); |
4ae7d5ce | 1305 | this_rq = cpu_rq(this_cpu); |
ac192d39 | 1306 | new_cpu = prev_cpu; |
e7693a36 | 1307 | |
ac192d39 IM |
1308 | /* |
1309 | * 'this_sd' is the first domain that both | |
1310 | * this_cpu and prev_cpu are present in: | |
1311 | */ | |
e7693a36 | 1312 | for_each_domain(this_cpu, sd) { |
758b2cdc | 1313 | if (cpumask_test_cpu(prev_cpu, sched_domain_span(sd))) { |
e7693a36 GH |
1314 | this_sd = sd; |
1315 | break; | |
1316 | } | |
1317 | } | |
1318 | ||
96f874e2 | 1319 | if (unlikely(!cpumask_test_cpu(this_cpu, &p->cpus_allowed))) |
f4827386 | 1320 | goto out; |
e7693a36 GH |
1321 | |
1322 | /* | |
1323 | * Check for affine wakeup and passive balancing possibilities. | |
1324 | */ | |
098fb9db | 1325 | if (!this_sd) |
f4827386 | 1326 | goto out; |
e7693a36 | 1327 | |
098fb9db IM |
1328 | idx = this_sd->wake_idx; |
1329 | ||
1330 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; | |
1331 | ||
ac192d39 | 1332 | load = source_load(prev_cpu, idx); |
098fb9db IM |
1333 | this_load = target_load(this_cpu, idx); |
1334 | ||
64b9e029 | 1335 | if (wake_affine(this_sd, this_rq, p, prev_cpu, this_cpu, sync, idx, |
4ae7d5ce IM |
1336 | load, this_load, imbalance)) |
1337 | return this_cpu; | |
1338 | ||
098fb9db IM |
1339 | /* |
1340 | * Start passive balancing when half the imbalance_pct | |
1341 | * limit is reached. | |
1342 | */ | |
1343 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1344 | if (imbalance*this_load <= 100*load) { | |
1345 | schedstat_inc(this_sd, ttwu_move_balance); | |
1346 | schedstat_inc(p, se.nr_wakeups_passive); | |
4ae7d5ce | 1347 | return this_cpu; |
e7693a36 GH |
1348 | } |
1349 | } | |
1350 | ||
f4827386 | 1351 | out: |
e7693a36 GH |
1352 | return wake_idle(new_cpu, p); |
1353 | } | |
1354 | #endif /* CONFIG_SMP */ | |
1355 | ||
e52fb7c0 PZ |
1356 | /* |
1357 | * Adaptive granularity | |
1358 | * | |
1359 | * se->avg_wakeup gives the average time a task runs until it does a wakeup, | |
1360 | * with the limit of wakeup_gran -- when it never does a wakeup. | |
1361 | * | |
1362 | * So the smaller avg_wakeup is the faster we want this task to preempt, | |
1363 | * but we don't want to treat the preemptee unfairly and therefore allow it | |
1364 | * to run for at least the amount of time we'd like to run. | |
1365 | * | |
1366 | * NOTE: we use 2*avg_wakeup to increase the probability of actually doing one | |
1367 | * | |
1368 | * NOTE: we use *nr_running to scale with load, this nicely matches the | |
1369 | * degrading latency on load. | |
1370 | */ | |
1371 | static unsigned long | |
1372 | adaptive_gran(struct sched_entity *curr, struct sched_entity *se) | |
1373 | { | |
1374 | u64 this_run = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; | |
1375 | u64 expected_wakeup = 2*se->avg_wakeup * cfs_rq_of(se)->nr_running; | |
1376 | u64 gran = 0; | |
1377 | ||
1378 | if (this_run < expected_wakeup) | |
1379 | gran = expected_wakeup - this_run; | |
1380 | ||
1381 | return min_t(s64, gran, sysctl_sched_wakeup_granularity); | |
1382 | } | |
1383 | ||
1384 | static unsigned long | |
1385 | wakeup_gran(struct sched_entity *curr, struct sched_entity *se) | |
0bbd3336 PZ |
1386 | { |
1387 | unsigned long gran = sysctl_sched_wakeup_granularity; | |
1388 | ||
e52fb7c0 PZ |
1389 | if (cfs_rq_of(curr)->curr && sched_feat(ADAPTIVE_GRAN)) |
1390 | gran = adaptive_gran(curr, se); | |
1391 | ||
0bbd3336 | 1392 | /* |
e52fb7c0 PZ |
1393 | * Since its curr running now, convert the gran from real-time |
1394 | * to virtual-time in his units. | |
0bbd3336 | 1395 | */ |
e52fb7c0 PZ |
1396 | if (sched_feat(ASYM_GRAN)) { |
1397 | /* | |
1398 | * By using 'se' instead of 'curr' we penalize light tasks, so | |
1399 | * they get preempted easier. That is, if 'se' < 'curr' then | |
1400 | * the resulting gran will be larger, therefore penalizing the | |
1401 | * lighter, if otoh 'se' > 'curr' then the resulting gran will | |
1402 | * be smaller, again penalizing the lighter task. | |
1403 | * | |
1404 | * This is especially important for buddies when the leftmost | |
1405 | * task is higher priority than the buddy. | |
1406 | */ | |
1407 | if (unlikely(se->load.weight != NICE_0_LOAD)) | |
1408 | gran = calc_delta_fair(gran, se); | |
1409 | } else { | |
1410 | if (unlikely(curr->load.weight != NICE_0_LOAD)) | |
1411 | gran = calc_delta_fair(gran, curr); | |
1412 | } | |
0bbd3336 PZ |
1413 | |
1414 | return gran; | |
1415 | } | |
1416 | ||
464b7527 PZ |
1417 | /* |
1418 | * Should 'se' preempt 'curr'. | |
1419 | * | |
1420 | * |s1 | |
1421 | * |s2 | |
1422 | * |s3 | |
1423 | * g | |
1424 | * |<--->|c | |
1425 | * | |
1426 | * w(c, s1) = -1 | |
1427 | * w(c, s2) = 0 | |
1428 | * w(c, s3) = 1 | |
1429 | * | |
1430 | */ | |
1431 | static int | |
1432 | wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se) | |
1433 | { | |
1434 | s64 gran, vdiff = curr->vruntime - se->vruntime; | |
1435 | ||
1436 | if (vdiff <= 0) | |
1437 | return -1; | |
1438 | ||
e52fb7c0 | 1439 | gran = wakeup_gran(curr, se); |
464b7527 PZ |
1440 | if (vdiff > gran) |
1441 | return 1; | |
1442 | ||
1443 | return 0; | |
1444 | } | |
1445 | ||
02479099 PZ |
1446 | static void set_last_buddy(struct sched_entity *se) |
1447 | { | |
6bc912b7 PZ |
1448 | if (likely(task_of(se)->policy != SCHED_IDLE)) { |
1449 | for_each_sched_entity(se) | |
1450 | cfs_rq_of(se)->last = se; | |
1451 | } | |
02479099 PZ |
1452 | } |
1453 | ||
1454 | static void set_next_buddy(struct sched_entity *se) | |
1455 | { | |
6bc912b7 PZ |
1456 | if (likely(task_of(se)->policy != SCHED_IDLE)) { |
1457 | for_each_sched_entity(se) | |
1458 | cfs_rq_of(se)->next = se; | |
1459 | } | |
02479099 PZ |
1460 | } |
1461 | ||
bf0f6f24 IM |
1462 | /* |
1463 | * Preempt the current task with a newly woken task if needed: | |
1464 | */ | |
15afe09b | 1465 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync) |
bf0f6f24 IM |
1466 | { |
1467 | struct task_struct *curr = rq->curr; | |
8651a86c | 1468 | struct sched_entity *se = &curr->se, *pse = &p->se; |
03e89e45 | 1469 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
bf0f6f24 | 1470 | |
03e89e45 | 1471 | update_curr(cfs_rq); |
4793241b | 1472 | |
03e89e45 | 1473 | if (unlikely(rt_prio(p->prio))) { |
bf0f6f24 IM |
1474 | resched_task(curr); |
1475 | return; | |
1476 | } | |
aa2ac252 | 1477 | |
d95f98d0 PZ |
1478 | if (unlikely(p->sched_class != &fair_sched_class)) |
1479 | return; | |
1480 | ||
4ae7d5ce IM |
1481 | if (unlikely(se == pse)) |
1482 | return; | |
1483 | ||
4793241b PZ |
1484 | /* |
1485 | * Only set the backward buddy when the current task is still on the | |
1486 | * rq. This can happen when a wakeup gets interleaved with schedule on | |
1487 | * the ->pre_schedule() or idle_balance() point, either of which can | |
1488 | * drop the rq lock. | |
1489 | * | |
1490 | * Also, during early boot the idle thread is in the fair class, for | |
1491 | * obvious reasons its a bad idea to schedule back to the idle thread. | |
1492 | */ | |
1493 | if (sched_feat(LAST_BUDDY) && likely(se->on_rq && curr != rq->idle)) | |
02479099 PZ |
1494 | set_last_buddy(se); |
1495 | set_next_buddy(pse); | |
57fdc26d | 1496 | |
aec0a514 BR |
1497 | /* |
1498 | * We can come here with TIF_NEED_RESCHED already set from new task | |
1499 | * wake up path. | |
1500 | */ | |
1501 | if (test_tsk_need_resched(curr)) | |
1502 | return; | |
1503 | ||
91c234b4 | 1504 | /* |
6bc912b7 | 1505 | * Batch and idle tasks do not preempt (their preemption is driven by |
91c234b4 IM |
1506 | * the tick): |
1507 | */ | |
6bc912b7 | 1508 | if (unlikely(p->policy != SCHED_NORMAL)) |
91c234b4 | 1509 | return; |
bf0f6f24 | 1510 | |
6bc912b7 PZ |
1511 | /* Idle tasks are by definition preempted by everybody. */ |
1512 | if (unlikely(curr->policy == SCHED_IDLE)) { | |
1513 | resched_task(curr); | |
91c234b4 | 1514 | return; |
6bc912b7 | 1515 | } |
bf0f6f24 | 1516 | |
77d9cc44 IM |
1517 | if (!sched_feat(WAKEUP_PREEMPT)) |
1518 | return; | |
8651a86c | 1519 | |
fc631c82 PZ |
1520 | if (sched_feat(WAKEUP_OVERLAP) && (sync || |
1521 | (se->avg_overlap < sysctl_sched_migration_cost && | |
1522 | pse->avg_overlap < sysctl_sched_migration_cost))) { | |
15afe09b PZ |
1523 | resched_task(curr); |
1524 | return; | |
1525 | } | |
1526 | ||
464b7527 PZ |
1527 | find_matching_se(&se, &pse); |
1528 | ||
002f128b | 1529 | BUG_ON(!pse); |
464b7527 | 1530 | |
002f128b PT |
1531 | if (wakeup_preempt_entity(se, pse) == 1) |
1532 | resched_task(curr); | |
bf0f6f24 IM |
1533 | } |
1534 | ||
fb8d4724 | 1535 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 | 1536 | { |
8f4d37ec | 1537 | struct task_struct *p; |
bf0f6f24 IM |
1538 | struct cfs_rq *cfs_rq = &rq->cfs; |
1539 | struct sched_entity *se; | |
1540 | ||
1541 | if (unlikely(!cfs_rq->nr_running)) | |
1542 | return NULL; | |
1543 | ||
1544 | do { | |
9948f4b2 | 1545 | se = pick_next_entity(cfs_rq); |
a9f3e2b5 MG |
1546 | /* |
1547 | * If se was a buddy, clear it so that it will have to earn | |
1548 | * the favour again. | |
1549 | */ | |
a571bbea | 1550 | __clear_buddies(cfs_rq, se); |
f4b6755f | 1551 | set_next_entity(cfs_rq, se); |
bf0f6f24 IM |
1552 | cfs_rq = group_cfs_rq(se); |
1553 | } while (cfs_rq); | |
1554 | ||
8f4d37ec PZ |
1555 | p = task_of(se); |
1556 | hrtick_start_fair(rq, p); | |
1557 | ||
1558 | return p; | |
bf0f6f24 IM |
1559 | } |
1560 | ||
1561 | /* | |
1562 | * Account for a descheduled task: | |
1563 | */ | |
31ee529c | 1564 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
1565 | { |
1566 | struct sched_entity *se = &prev->se; | |
1567 | struct cfs_rq *cfs_rq; | |
1568 | ||
1569 | for_each_sched_entity(se) { | |
1570 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 1571 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
1572 | } |
1573 | } | |
1574 | ||
681f3e68 | 1575 | #ifdef CONFIG_SMP |
bf0f6f24 IM |
1576 | /************************************************** |
1577 | * Fair scheduling class load-balancing methods: | |
1578 | */ | |
1579 | ||
1580 | /* | |
1581 | * Load-balancing iterator. Note: while the runqueue stays locked | |
1582 | * during the whole iteration, the current task might be | |
1583 | * dequeued so the iterator has to be dequeue-safe. Here we | |
1584 | * achieve that by always pre-iterating before returning | |
1585 | * the current task: | |
1586 | */ | |
a9957449 | 1587 | static struct task_struct * |
4a55bd5e | 1588 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next) |
bf0f6f24 | 1589 | { |
354d60c2 DG |
1590 | struct task_struct *p = NULL; |
1591 | struct sched_entity *se; | |
bf0f6f24 | 1592 | |
77ae6513 MG |
1593 | if (next == &cfs_rq->tasks) |
1594 | return NULL; | |
1595 | ||
b87f1724 BR |
1596 | se = list_entry(next, struct sched_entity, group_node); |
1597 | p = task_of(se); | |
1598 | cfs_rq->balance_iterator = next->next; | |
77ae6513 | 1599 | |
bf0f6f24 IM |
1600 | return p; |
1601 | } | |
1602 | ||
1603 | static struct task_struct *load_balance_start_fair(void *arg) | |
1604 | { | |
1605 | struct cfs_rq *cfs_rq = arg; | |
1606 | ||
4a55bd5e | 1607 | return __load_balance_iterator(cfs_rq, cfs_rq->tasks.next); |
bf0f6f24 IM |
1608 | } |
1609 | ||
1610 | static struct task_struct *load_balance_next_fair(void *arg) | |
1611 | { | |
1612 | struct cfs_rq *cfs_rq = arg; | |
1613 | ||
4a55bd5e | 1614 | return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator); |
bf0f6f24 IM |
1615 | } |
1616 | ||
c09595f6 PZ |
1617 | static unsigned long |
1618 | __load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1619 | unsigned long max_load_move, struct sched_domain *sd, | |
1620 | enum cpu_idle_type idle, int *all_pinned, int *this_best_prio, | |
1621 | struct cfs_rq *cfs_rq) | |
62fb1851 | 1622 | { |
c09595f6 | 1623 | struct rq_iterator cfs_rq_iterator; |
62fb1851 | 1624 | |
c09595f6 PZ |
1625 | cfs_rq_iterator.start = load_balance_start_fair; |
1626 | cfs_rq_iterator.next = load_balance_next_fair; | |
1627 | cfs_rq_iterator.arg = cfs_rq; | |
62fb1851 | 1628 | |
c09595f6 PZ |
1629 | return balance_tasks(this_rq, this_cpu, busiest, |
1630 | max_load_move, sd, idle, all_pinned, | |
1631 | this_best_prio, &cfs_rq_iterator); | |
62fb1851 | 1632 | } |
62fb1851 | 1633 | |
c09595f6 | 1634 | #ifdef CONFIG_FAIR_GROUP_SCHED |
43010659 | 1635 | static unsigned long |
bf0f6f24 | 1636 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f | 1637 | unsigned long max_load_move, |
a4ac01c3 PW |
1638 | struct sched_domain *sd, enum cpu_idle_type idle, |
1639 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 | 1640 | { |
bf0f6f24 | 1641 | long rem_load_move = max_load_move; |
c09595f6 PZ |
1642 | int busiest_cpu = cpu_of(busiest); |
1643 | struct task_group *tg; | |
18d95a28 | 1644 | |
c09595f6 | 1645 | rcu_read_lock(); |
c8cba857 | 1646 | update_h_load(busiest_cpu); |
18d95a28 | 1647 | |
caea8a03 | 1648 | list_for_each_entry_rcu(tg, &task_groups, list) { |
c8cba857 | 1649 | struct cfs_rq *busiest_cfs_rq = tg->cfs_rq[busiest_cpu]; |
42a3ac7d PZ |
1650 | unsigned long busiest_h_load = busiest_cfs_rq->h_load; |
1651 | unsigned long busiest_weight = busiest_cfs_rq->load.weight; | |
243e0e7b | 1652 | u64 rem_load, moved_load; |
18d95a28 | 1653 | |
c09595f6 PZ |
1654 | /* |
1655 | * empty group | |
1656 | */ | |
c8cba857 | 1657 | if (!busiest_cfs_rq->task_weight) |
bf0f6f24 IM |
1658 | continue; |
1659 | ||
243e0e7b SV |
1660 | rem_load = (u64)rem_load_move * busiest_weight; |
1661 | rem_load = div_u64(rem_load, busiest_h_load + 1); | |
bf0f6f24 | 1662 | |
c09595f6 | 1663 | moved_load = __load_balance_fair(this_rq, this_cpu, busiest, |
53fecd8a | 1664 | rem_load, sd, idle, all_pinned, this_best_prio, |
c09595f6 | 1665 | tg->cfs_rq[busiest_cpu]); |
bf0f6f24 | 1666 | |
c09595f6 | 1667 | if (!moved_load) |
bf0f6f24 IM |
1668 | continue; |
1669 | ||
42a3ac7d | 1670 | moved_load *= busiest_h_load; |
243e0e7b | 1671 | moved_load = div_u64(moved_load, busiest_weight + 1); |
bf0f6f24 | 1672 | |
c09595f6 PZ |
1673 | rem_load_move -= moved_load; |
1674 | if (rem_load_move < 0) | |
bf0f6f24 IM |
1675 | break; |
1676 | } | |
c09595f6 | 1677 | rcu_read_unlock(); |
bf0f6f24 | 1678 | |
43010659 | 1679 | return max_load_move - rem_load_move; |
bf0f6f24 | 1680 | } |
c09595f6 PZ |
1681 | #else |
1682 | static unsigned long | |
1683 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1684 | unsigned long max_load_move, | |
1685 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1686 | int *all_pinned, int *this_best_prio) | |
1687 | { | |
1688 | return __load_balance_fair(this_rq, this_cpu, busiest, | |
1689 | max_load_move, sd, idle, all_pinned, | |
1690 | this_best_prio, &busiest->cfs); | |
1691 | } | |
1692 | #endif | |
bf0f6f24 | 1693 | |
e1d1484f PW |
1694 | static int |
1695 | move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1696 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1697 | { | |
1698 | struct cfs_rq *busy_cfs_rq; | |
1699 | struct rq_iterator cfs_rq_iterator; | |
1700 | ||
1701 | cfs_rq_iterator.start = load_balance_start_fair; | |
1702 | cfs_rq_iterator.next = load_balance_next_fair; | |
1703 | ||
1704 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
1705 | /* | |
1706 | * pass busy_cfs_rq argument into | |
1707 | * load_balance_[start|next]_fair iterators | |
1708 | */ | |
1709 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1710 | if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle, | |
1711 | &cfs_rq_iterator)) | |
1712 | return 1; | |
1713 | } | |
1714 | ||
1715 | return 0; | |
1716 | } | |
55e12e5e | 1717 | #endif /* CONFIG_SMP */ |
e1d1484f | 1718 | |
bf0f6f24 IM |
1719 | /* |
1720 | * scheduler tick hitting a task of our scheduling class: | |
1721 | */ | |
8f4d37ec | 1722 | static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) |
bf0f6f24 IM |
1723 | { |
1724 | struct cfs_rq *cfs_rq; | |
1725 | struct sched_entity *se = &curr->se; | |
1726 | ||
1727 | for_each_sched_entity(se) { | |
1728 | cfs_rq = cfs_rq_of(se); | |
8f4d37ec | 1729 | entity_tick(cfs_rq, se, queued); |
bf0f6f24 IM |
1730 | } |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * Share the fairness runtime between parent and child, thus the | |
1735 | * total amount of pressure for CPU stays equal - new tasks | |
1736 | * get a chance to run but frequent forkers are not allowed to | |
1737 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1738 | * the child is not running yet. | |
1739 | */ | |
ee0827d8 | 1740 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1741 | { |
1742 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1743 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
00bf7bfc | 1744 | int this_cpu = smp_processor_id(); |
bf0f6f24 IM |
1745 | |
1746 | sched_info_queued(p); | |
1747 | ||
7109c442 | 1748 | update_curr(cfs_rq); |
aeb73b04 | 1749 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1750 | |
3c90e6e9 | 1751 | /* 'curr' will be NULL if the child belongs to a different group */ |
00bf7bfc | 1752 | if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) && |
54fdc581 | 1753 | curr && entity_before(curr, se)) { |
87fefa38 | 1754 | /* |
edcb60a3 IM |
1755 | * Upon rescheduling, sched_class::put_prev_task() will place |
1756 | * 'current' within the tree based on its new key value. | |
1757 | */ | |
4d78e7b6 | 1758 | swap(curr->vruntime, se->vruntime); |
aec0a514 | 1759 | resched_task(rq->curr); |
4d78e7b6 | 1760 | } |
bf0f6f24 | 1761 | |
b9dca1e0 | 1762 | enqueue_task_fair(rq, p, 0); |
bf0f6f24 IM |
1763 | } |
1764 | ||
cb469845 SR |
1765 | /* |
1766 | * Priority of the task has changed. Check to see if we preempt | |
1767 | * the current task. | |
1768 | */ | |
1769 | static void prio_changed_fair(struct rq *rq, struct task_struct *p, | |
1770 | int oldprio, int running) | |
1771 | { | |
1772 | /* | |
1773 | * Reschedule if we are currently running on this runqueue and | |
1774 | * our priority decreased, or if we are not currently running on | |
1775 | * this runqueue and our priority is higher than the current's | |
1776 | */ | |
1777 | if (running) { | |
1778 | if (p->prio > oldprio) | |
1779 | resched_task(rq->curr); | |
1780 | } else | |
15afe09b | 1781 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
1782 | } |
1783 | ||
1784 | /* | |
1785 | * We switched to the sched_fair class. | |
1786 | */ | |
1787 | static void switched_to_fair(struct rq *rq, struct task_struct *p, | |
1788 | int running) | |
1789 | { | |
1790 | /* | |
1791 | * We were most likely switched from sched_rt, so | |
1792 | * kick off the schedule if running, otherwise just see | |
1793 | * if we can still preempt the current task. | |
1794 | */ | |
1795 | if (running) | |
1796 | resched_task(rq->curr); | |
1797 | else | |
15afe09b | 1798 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
1799 | } |
1800 | ||
83b699ed SV |
1801 | /* Account for a task changing its policy or group. |
1802 | * | |
1803 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1804 | * migrates between groups/classes. | |
1805 | */ | |
1806 | static void set_curr_task_fair(struct rq *rq) | |
1807 | { | |
1808 | struct sched_entity *se = &rq->curr->se; | |
1809 | ||
1810 | for_each_sched_entity(se) | |
1811 | set_next_entity(cfs_rq_of(se), se); | |
1812 | } | |
1813 | ||
810b3817 PZ |
1814 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1815 | static void moved_group_fair(struct task_struct *p) | |
1816 | { | |
1817 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1818 | ||
1819 | update_curr(cfs_rq); | |
1820 | place_entity(cfs_rq, &p->se, 1); | |
1821 | } | |
1822 | #endif | |
1823 | ||
bf0f6f24 IM |
1824 | /* |
1825 | * All the scheduling class methods: | |
1826 | */ | |
5522d5d5 IM |
1827 | static const struct sched_class fair_sched_class = { |
1828 | .next = &idle_sched_class, | |
bf0f6f24 IM |
1829 | .enqueue_task = enqueue_task_fair, |
1830 | .dequeue_task = dequeue_task_fair, | |
1831 | .yield_task = yield_task_fair, | |
1832 | ||
2e09bf55 | 1833 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1834 | |
1835 | .pick_next_task = pick_next_task_fair, | |
1836 | .put_prev_task = put_prev_task_fair, | |
1837 | ||
681f3e68 | 1838 | #ifdef CONFIG_SMP |
4ce72a2c LZ |
1839 | .select_task_rq = select_task_rq_fair, |
1840 | ||
bf0f6f24 | 1841 | .load_balance = load_balance_fair, |
e1d1484f | 1842 | .move_one_task = move_one_task_fair, |
681f3e68 | 1843 | #endif |
bf0f6f24 | 1844 | |
83b699ed | 1845 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 IM |
1846 | .task_tick = task_tick_fair, |
1847 | .task_new = task_new_fair, | |
cb469845 SR |
1848 | |
1849 | .prio_changed = prio_changed_fair, | |
1850 | .switched_to = switched_to_fair, | |
810b3817 PZ |
1851 | |
1852 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1853 | .moved_group = moved_group_fair, | |
1854 | #endif | |
bf0f6f24 IM |
1855 | }; |
1856 | ||
1857 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1858 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1859 | { |
bf0f6f24 IM |
1860 | struct cfs_rq *cfs_rq; |
1861 | ||
5973e5b9 | 1862 | rcu_read_lock(); |
c3b64f1e | 1863 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1864 | print_cfs_rq(m, cpu, cfs_rq); |
5973e5b9 | 1865 | rcu_read_unlock(); |
bf0f6f24 IM |
1866 | } |
1867 | #endif |