]>
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 | ||
b758149c PZ |
82 | static inline struct task_struct *task_of(struct sched_entity *se) |
83 | { | |
84 | return container_of(se, struct task_struct, se); | |
85 | } | |
86 | ||
62160e3f | 87 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 88 | |
62160e3f | 89 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
90 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
91 | { | |
62160e3f | 92 | return cfs_rq->rq; |
bf0f6f24 IM |
93 | } |
94 | ||
62160e3f IM |
95 | /* An entity is a task if it doesn't "own" a runqueue */ |
96 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 97 | |
b758149c PZ |
98 | /* Walk up scheduling entities hierarchy */ |
99 | #define for_each_sched_entity(se) \ | |
100 | for (; se; se = se->parent) | |
101 | ||
102 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
103 | { | |
104 | return p->se.cfs_rq; | |
105 | } | |
106 | ||
107 | /* runqueue on which this entity is (to be) queued */ | |
108 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
109 | { | |
110 | return se->cfs_rq; | |
111 | } | |
112 | ||
113 | /* runqueue "owned" by this group */ | |
114 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
115 | { | |
116 | return grp->my_q; | |
117 | } | |
118 | ||
119 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
120 | * another cpu ('this_cpu') | |
121 | */ | |
122 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
123 | { | |
124 | return cfs_rq->tg->cfs_rq[this_cpu]; | |
125 | } | |
126 | ||
127 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
128 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
129 | list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
130 | ||
131 | /* Do the two (enqueued) entities belong to the same group ? */ | |
132 | static inline int | |
133 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
134 | { | |
135 | if (se->cfs_rq == pse->cfs_rq) | |
136 | return 1; | |
137 | ||
138 | return 0; | |
139 | } | |
140 | ||
141 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
142 | { | |
143 | return se->parent; | |
144 | } | |
145 | ||
62160e3f | 146 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
bf0f6f24 | 147 | |
62160e3f IM |
148 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
149 | { | |
150 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
151 | } |
152 | ||
153 | #define entity_is_task(se) 1 | |
154 | ||
b758149c PZ |
155 | #define for_each_sched_entity(se) \ |
156 | for (; se; se = NULL) | |
bf0f6f24 | 157 | |
b758149c | 158 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) |
bf0f6f24 | 159 | { |
b758149c | 160 | return &task_rq(p)->cfs; |
bf0f6f24 IM |
161 | } |
162 | ||
b758149c PZ |
163 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) |
164 | { | |
165 | struct task_struct *p = task_of(se); | |
166 | struct rq *rq = task_rq(p); | |
167 | ||
168 | return &rq->cfs; | |
169 | } | |
170 | ||
171 | /* runqueue "owned" by this group */ | |
172 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
173 | { | |
174 | return NULL; | |
175 | } | |
176 | ||
177 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
178 | { | |
179 | return &cpu_rq(this_cpu)->cfs; | |
180 | } | |
181 | ||
182 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
183 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
184 | ||
185 | static inline int | |
186 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
187 | { | |
188 | return 1; | |
189 | } | |
190 | ||
191 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
192 | { | |
193 | return NULL; | |
194 | } | |
195 | ||
196 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
197 | ||
bf0f6f24 IM |
198 | |
199 | /************************************************************** | |
200 | * Scheduling class tree data structure manipulation methods: | |
201 | */ | |
202 | ||
0702e3eb | 203 | static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime) |
02e0431a | 204 | { |
368059a9 PZ |
205 | s64 delta = (s64)(vruntime - min_vruntime); |
206 | if (delta > 0) | |
02e0431a PZ |
207 | min_vruntime = vruntime; |
208 | ||
209 | return min_vruntime; | |
210 | } | |
211 | ||
0702e3eb | 212 | static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime) |
b0ffd246 PZ |
213 | { |
214 | s64 delta = (s64)(vruntime - min_vruntime); | |
215 | if (delta < 0) | |
216 | min_vruntime = vruntime; | |
217 | ||
218 | return min_vruntime; | |
219 | } | |
220 | ||
0702e3eb | 221 | static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
9014623c | 222 | { |
30cfdcfc | 223 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
224 | } |
225 | ||
bf0f6f24 IM |
226 | /* |
227 | * Enqueue an entity into the rb-tree: | |
228 | */ | |
0702e3eb | 229 | static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
230 | { |
231 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
232 | struct rb_node *parent = NULL; | |
233 | struct sched_entity *entry; | |
9014623c | 234 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
235 | int leftmost = 1; |
236 | ||
237 | /* | |
238 | * Find the right place in the rbtree: | |
239 | */ | |
240 | while (*link) { | |
241 | parent = *link; | |
242 | entry = rb_entry(parent, struct sched_entity, run_node); | |
243 | /* | |
244 | * We dont care about collisions. Nodes with | |
245 | * the same key stay together. | |
246 | */ | |
9014623c | 247 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
248 | link = &parent->rb_left; |
249 | } else { | |
250 | link = &parent->rb_right; | |
251 | leftmost = 0; | |
252 | } | |
253 | } | |
254 | ||
255 | /* | |
256 | * Maintain a cache of leftmost tree entries (it is frequently | |
257 | * used): | |
258 | */ | |
3fe69747 | 259 | if (leftmost) { |
57cb499d | 260 | cfs_rq->rb_leftmost = &se->run_node; |
3fe69747 PZ |
261 | /* |
262 | * maintain cfs_rq->min_vruntime to be a monotonic increasing | |
263 | * value tracking the leftmost vruntime in the tree. | |
264 | */ | |
265 | cfs_rq->min_vruntime = | |
266 | max_vruntime(cfs_rq->min_vruntime, se->vruntime); | |
267 | } | |
bf0f6f24 IM |
268 | |
269 | rb_link_node(&se->run_node, parent, link); | |
270 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
271 | } |
272 | ||
0702e3eb | 273 | static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 274 | { |
3fe69747 PZ |
275 | if (cfs_rq->rb_leftmost == &se->run_node) { |
276 | struct rb_node *next_node; | |
277 | struct sched_entity *next; | |
278 | ||
279 | next_node = rb_next(&se->run_node); | |
280 | cfs_rq->rb_leftmost = next_node; | |
281 | ||
282 | if (next_node) { | |
283 | next = rb_entry(next_node, | |
284 | struct sched_entity, run_node); | |
285 | cfs_rq->min_vruntime = | |
286 | max_vruntime(cfs_rq->min_vruntime, | |
287 | next->vruntime); | |
288 | } | |
289 | } | |
e9acbff6 | 290 | |
aa2ac252 PZ |
291 | if (cfs_rq->next == se) |
292 | cfs_rq->next = NULL; | |
293 | ||
bf0f6f24 | 294 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
295 | } |
296 | ||
297 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
298 | { | |
299 | return cfs_rq->rb_leftmost; | |
300 | } | |
301 | ||
302 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
303 | { | |
304 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
305 | } | |
306 | ||
aeb73b04 PZ |
307 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
308 | { | |
7eee3e67 | 309 | struct rb_node *last = rb_last(&cfs_rq->tasks_timeline); |
aeb73b04 | 310 | |
70eee74b BS |
311 | if (!last) |
312 | return NULL; | |
7eee3e67 IM |
313 | |
314 | return rb_entry(last, struct sched_entity, run_node); | |
aeb73b04 PZ |
315 | } |
316 | ||
bf0f6f24 IM |
317 | /************************************************************** |
318 | * Scheduling class statistics methods: | |
319 | */ | |
320 | ||
b2be5e96 PZ |
321 | #ifdef CONFIG_SCHED_DEBUG |
322 | int sched_nr_latency_handler(struct ctl_table *table, int write, | |
323 | struct file *filp, void __user *buffer, size_t *lenp, | |
324 | loff_t *ppos) | |
325 | { | |
326 | int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); | |
327 | ||
328 | if (ret || !write) | |
329 | return ret; | |
330 | ||
331 | sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency, | |
332 | sysctl_sched_min_granularity); | |
333 | ||
334 | return 0; | |
335 | } | |
336 | #endif | |
647e7cac | 337 | |
a7be37ac PZ |
338 | /* |
339 | * delta *= w / rw | |
340 | */ | |
341 | static inline unsigned long | |
342 | calc_delta_weight(unsigned long delta, struct sched_entity *se) | |
343 | { | |
344 | for_each_sched_entity(se) { | |
345 | delta = calc_delta_mine(delta, | |
346 | se->load.weight, &cfs_rq_of(se)->load); | |
347 | } | |
348 | ||
349 | return delta; | |
350 | } | |
351 | ||
352 | /* | |
353 | * delta *= rw / w | |
354 | */ | |
355 | static inline unsigned long | |
356 | calc_delta_fair(unsigned long delta, struct sched_entity *se) | |
357 | { | |
358 | for_each_sched_entity(se) { | |
359 | delta = calc_delta_mine(delta, | |
360 | cfs_rq_of(se)->load.weight, &se->load); | |
361 | } | |
362 | ||
363 | return delta; | |
364 | } | |
365 | ||
647e7cac IM |
366 | /* |
367 | * The idea is to set a period in which each task runs once. | |
368 | * | |
369 | * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch | |
370 | * this period because otherwise the slices get too small. | |
371 | * | |
372 | * p = (nr <= nl) ? l : l*nr/nl | |
373 | */ | |
4d78e7b6 PZ |
374 | static u64 __sched_period(unsigned long nr_running) |
375 | { | |
376 | u64 period = sysctl_sched_latency; | |
b2be5e96 | 377 | unsigned long nr_latency = sched_nr_latency; |
4d78e7b6 PZ |
378 | |
379 | if (unlikely(nr_running > nr_latency)) { | |
4bf0b771 | 380 | period = sysctl_sched_min_granularity; |
4d78e7b6 | 381 | period *= nr_running; |
4d78e7b6 PZ |
382 | } |
383 | ||
384 | return period; | |
385 | } | |
386 | ||
647e7cac IM |
387 | /* |
388 | * We calculate the wall-time slice from the period by taking a part | |
389 | * proportional to the weight. | |
390 | * | |
391 | * s = p*w/rw | |
392 | */ | |
6d0f0ebd | 393 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 394 | { |
a7be37ac | 395 | return calc_delta_weight(__sched_period(cfs_rq->nr_running), se); |
bf0f6f24 IM |
396 | } |
397 | ||
647e7cac | 398 | /* |
ac884dec | 399 | * We calculate the vruntime slice of a to be inserted task |
647e7cac | 400 | * |
a7be37ac | 401 | * vs = s*rw/w = p |
647e7cac | 402 | */ |
ac884dec | 403 | static u64 sched_vslice_add(struct cfs_rq *cfs_rq, struct sched_entity *se) |
67e9fb2a | 404 | { |
ac884dec | 405 | unsigned long nr_running = cfs_rq->nr_running; |
67e9fb2a | 406 | |
ac884dec PZ |
407 | if (!se->on_rq) |
408 | nr_running++; | |
67e9fb2a | 409 | |
a7be37ac PZ |
410 | return __sched_period(nr_running); |
411 | } | |
412 | ||
bf0f6f24 IM |
413 | /* |
414 | * Update the current task's runtime statistics. Skip current tasks that | |
415 | * are not in our scheduling class. | |
416 | */ | |
417 | static inline void | |
8ebc91d9 IM |
418 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
419 | unsigned long delta_exec) | |
bf0f6f24 | 420 | { |
bbdba7c0 | 421 | unsigned long delta_exec_weighted; |
bf0f6f24 | 422 | |
8179ca23 | 423 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
424 | |
425 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 426 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
a7be37ac | 427 | delta_exec_weighted = calc_delta_fair(delta_exec, curr); |
e9acbff6 | 428 | curr->vruntime += delta_exec_weighted; |
bf0f6f24 IM |
429 | } |
430 | ||
b7cc0896 | 431 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 432 | { |
429d43bc | 433 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 434 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
435 | unsigned long delta_exec; |
436 | ||
437 | if (unlikely(!curr)) | |
438 | return; | |
439 | ||
440 | /* | |
441 | * Get the amount of time the current task was running | |
442 | * since the last time we changed load (this cannot | |
443 | * overflow on 32 bits): | |
444 | */ | |
8ebc91d9 | 445 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 446 | |
8ebc91d9 IM |
447 | __update_curr(cfs_rq, curr, delta_exec); |
448 | curr->exec_start = now; | |
d842de87 SV |
449 | |
450 | if (entity_is_task(curr)) { | |
451 | struct task_struct *curtask = task_of(curr); | |
452 | ||
453 | cpuacct_charge(curtask, delta_exec); | |
454 | } | |
bf0f6f24 IM |
455 | } |
456 | ||
457 | static inline void | |
5870db5b | 458 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 459 | { |
d281918d | 460 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
461 | } |
462 | ||
bf0f6f24 IM |
463 | /* |
464 | * Task is being enqueued - update stats: | |
465 | */ | |
d2417e5a | 466 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 467 | { |
bf0f6f24 IM |
468 | /* |
469 | * Are we enqueueing a waiting task? (for current tasks | |
470 | * a dequeue/enqueue event is a NOP) | |
471 | */ | |
429d43bc | 472 | if (se != cfs_rq->curr) |
5870db5b | 473 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
474 | } |
475 | ||
bf0f6f24 | 476 | static void |
9ef0a961 | 477 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 478 | { |
bbdba7c0 IM |
479 | schedstat_set(se->wait_max, max(se->wait_max, |
480 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6d082592 AV |
481 | schedstat_set(se->wait_count, se->wait_count + 1); |
482 | schedstat_set(se->wait_sum, se->wait_sum + | |
483 | rq_of(cfs_rq)->clock - se->wait_start); | |
6cfb0d5d | 484 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
485 | } |
486 | ||
487 | static inline void | |
19b6a2e3 | 488 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 489 | { |
bf0f6f24 IM |
490 | /* |
491 | * Mark the end of the wait period if dequeueing a | |
492 | * waiting task: | |
493 | */ | |
429d43bc | 494 | if (se != cfs_rq->curr) |
9ef0a961 | 495 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
496 | } |
497 | ||
498 | /* | |
499 | * We are picking a new current task - update its stats: | |
500 | */ | |
501 | static inline void | |
79303e9e | 502 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
503 | { |
504 | /* | |
505 | * We are starting a new run period: | |
506 | */ | |
d281918d | 507 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
508 | } |
509 | ||
bf0f6f24 IM |
510 | /************************************************** |
511 | * Scheduling class queueing methods: | |
512 | */ | |
513 | ||
c09595f6 PZ |
514 | #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED |
515 | static void | |
516 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
517 | { | |
518 | cfs_rq->task_weight += weight; | |
519 | } | |
520 | #else | |
521 | static inline void | |
522 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
523 | { | |
524 | } | |
525 | #endif | |
526 | ||
30cfdcfc DA |
527 | static void |
528 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
529 | { | |
530 | update_load_add(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
531 | if (!parent_entity(se)) |
532 | inc_cpu_load(rq_of(cfs_rq), se->load.weight); | |
b87f1724 | 533 | if (entity_is_task(se)) { |
c09595f6 | 534 | add_cfs_task_weight(cfs_rq, se->load.weight); |
b87f1724 BR |
535 | list_add(&se->group_node, &cfs_rq->tasks); |
536 | } | |
30cfdcfc DA |
537 | cfs_rq->nr_running++; |
538 | se->on_rq = 1; | |
539 | } | |
540 | ||
541 | static void | |
542 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
543 | { | |
544 | update_load_sub(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
545 | if (!parent_entity(se)) |
546 | dec_cpu_load(rq_of(cfs_rq), se->load.weight); | |
b87f1724 | 547 | if (entity_is_task(se)) { |
c09595f6 | 548 | add_cfs_task_weight(cfs_rq, -se->load.weight); |
b87f1724 BR |
549 | list_del_init(&se->group_node); |
550 | } | |
30cfdcfc DA |
551 | cfs_rq->nr_running--; |
552 | se->on_rq = 0; | |
553 | } | |
554 | ||
2396af69 | 555 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 556 | { |
bf0f6f24 IM |
557 | #ifdef CONFIG_SCHEDSTATS |
558 | if (se->sleep_start) { | |
d281918d | 559 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
9745512c | 560 | struct task_struct *tsk = task_of(se); |
bf0f6f24 IM |
561 | |
562 | if ((s64)delta < 0) | |
563 | delta = 0; | |
564 | ||
565 | if (unlikely(delta > se->sleep_max)) | |
566 | se->sleep_max = delta; | |
567 | ||
568 | se->sleep_start = 0; | |
569 | se->sum_sleep_runtime += delta; | |
9745512c AV |
570 | |
571 | account_scheduler_latency(tsk, delta >> 10, 1); | |
bf0f6f24 IM |
572 | } |
573 | if (se->block_start) { | |
d281918d | 574 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
9745512c | 575 | struct task_struct *tsk = task_of(se); |
bf0f6f24 IM |
576 | |
577 | if ((s64)delta < 0) | |
578 | delta = 0; | |
579 | ||
580 | if (unlikely(delta > se->block_max)) | |
581 | se->block_max = delta; | |
582 | ||
583 | se->block_start = 0; | |
584 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
585 | |
586 | /* | |
587 | * Blocking time is in units of nanosecs, so shift by 20 to | |
588 | * get a milliseconds-range estimation of the amount of | |
589 | * time that the task spent sleeping: | |
590 | */ | |
591 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf | 592 | |
30084fbd IM |
593 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
594 | delta >> 20); | |
595 | } | |
9745512c | 596 | account_scheduler_latency(tsk, delta >> 10, 0); |
bf0f6f24 IM |
597 | } |
598 | #endif | |
599 | } | |
600 | ||
ddc97297 PZ |
601 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
602 | { | |
603 | #ifdef CONFIG_SCHED_DEBUG | |
604 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
605 | ||
606 | if (d < 0) | |
607 | d = -d; | |
608 | ||
609 | if (d > 3*sysctl_sched_latency) | |
610 | schedstat_inc(cfs_rq, nr_spread_over); | |
611 | #endif | |
612 | } | |
613 | ||
aeb73b04 PZ |
614 | static void |
615 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
616 | { | |
67e9fb2a | 617 | u64 vruntime; |
aeb73b04 | 618 | |
3fe69747 PZ |
619 | if (first_fair(cfs_rq)) { |
620 | vruntime = min_vruntime(cfs_rq->min_vruntime, | |
621 | __pick_next_entity(cfs_rq)->vruntime); | |
622 | } else | |
623 | vruntime = cfs_rq->min_vruntime; | |
94dfb5e7 | 624 | |
2cb8600e PZ |
625 | /* |
626 | * The 'current' period is already promised to the current tasks, | |
627 | * however the extra weight of the new task will slow them down a | |
628 | * little, place the new task so that it fits in the slot that | |
629 | * stays open at the end. | |
630 | */ | |
94dfb5e7 | 631 | if (initial && sched_feat(START_DEBIT)) |
647e7cac | 632 | vruntime += sched_vslice_add(cfs_rq, se); |
aeb73b04 | 633 | |
8465e792 | 634 | if (!initial) { |
2cb8600e | 635 | /* sleeps upto a single latency don't count. */ |
a7be37ac PZ |
636 | if (sched_feat(NEW_FAIR_SLEEPERS)) { |
637 | unsigned long thresh = sysctl_sched_latency; | |
638 | ||
639 | /* | |
640 | * convert the sleeper threshold into virtual time | |
641 | */ | |
642 | if (sched_feat(NORMALIZED_SLEEPER)) | |
643 | thresh = calc_delta_fair(thresh, se); | |
644 | ||
645 | vruntime -= thresh; | |
646 | } | |
94359f05 | 647 | |
2cb8600e PZ |
648 | /* ensure we never gain time by being placed backwards. */ |
649 | vruntime = max_vruntime(se->vruntime, vruntime); | |
aeb73b04 PZ |
650 | } |
651 | ||
67e9fb2a | 652 | se->vruntime = vruntime; |
aeb73b04 PZ |
653 | } |
654 | ||
bf0f6f24 | 655 | static void |
83b699ed | 656 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
657 | { |
658 | /* | |
a2a2d680 | 659 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 660 | */ |
b7cc0896 | 661 | update_curr(cfs_rq); |
a992241d | 662 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 | 663 | |
e9acbff6 | 664 | if (wakeup) { |
aeb73b04 | 665 | place_entity(cfs_rq, se, 0); |
2396af69 | 666 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 667 | } |
bf0f6f24 | 668 | |
d2417e5a | 669 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 670 | check_spread(cfs_rq, se); |
83b699ed SV |
671 | if (se != cfs_rq->curr) |
672 | __enqueue_entity(cfs_rq, se); | |
bf0f6f24 IM |
673 | } |
674 | ||
675 | static void | |
525c2716 | 676 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 677 | { |
a2a2d680 DA |
678 | /* |
679 | * Update run-time statistics of the 'current'. | |
680 | */ | |
681 | update_curr(cfs_rq); | |
682 | ||
19b6a2e3 | 683 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 684 | if (sleep) { |
67e9fb2a | 685 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
686 | if (entity_is_task(se)) { |
687 | struct task_struct *tsk = task_of(se); | |
688 | ||
689 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 690 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 691 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 692 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 693 | } |
db36cc7d | 694 | #endif |
67e9fb2a PZ |
695 | } |
696 | ||
83b699ed | 697 | if (se != cfs_rq->curr) |
30cfdcfc DA |
698 | __dequeue_entity(cfs_rq, se); |
699 | account_entity_dequeue(cfs_rq, se); | |
bf0f6f24 IM |
700 | } |
701 | ||
702 | /* | |
703 | * Preempt the current task with a newly woken task if needed: | |
704 | */ | |
7c92e54f | 705 | static void |
2e09bf55 | 706 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 707 | { |
11697830 PZ |
708 | unsigned long ideal_runtime, delta_exec; |
709 | ||
6d0f0ebd | 710 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 | 711 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
3e3e13f3 | 712 | if (delta_exec > ideal_runtime) |
bf0f6f24 IM |
713 | resched_task(rq_of(cfs_rq)->curr); |
714 | } | |
715 | ||
83b699ed | 716 | static void |
8494f412 | 717 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 718 | { |
83b699ed SV |
719 | /* 'current' is not kept within the tree. */ |
720 | if (se->on_rq) { | |
721 | /* | |
722 | * Any task has to be enqueued before it get to execute on | |
723 | * a CPU. So account for the time it spent waiting on the | |
724 | * runqueue. | |
725 | */ | |
726 | update_stats_wait_end(cfs_rq, se); | |
727 | __dequeue_entity(cfs_rq, se); | |
728 | } | |
729 | ||
79303e9e | 730 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 731 | cfs_rq->curr = se; |
eba1ed4b IM |
732 | #ifdef CONFIG_SCHEDSTATS |
733 | /* | |
734 | * Track our maximum slice length, if the CPU's load is at | |
735 | * least twice that of our own weight (i.e. dont track it | |
736 | * when there are only lesser-weight tasks around): | |
737 | */ | |
495eca49 | 738 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
739 | se->slice_max = max(se->slice_max, |
740 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
741 | } | |
742 | #endif | |
4a55b450 | 743 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
744 | } |
745 | ||
aa2ac252 PZ |
746 | static struct sched_entity * |
747 | pick_next(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
748 | { | |
103638d9 PZ |
749 | struct rq *rq = rq_of(cfs_rq); |
750 | u64 pair_slice = rq->clock - cfs_rq->pair_start; | |
aa2ac252 | 751 | |
b0aa51b9 | 752 | if (!cfs_rq->next || pair_slice > sysctl_sched_min_granularity) { |
103638d9 | 753 | cfs_rq->pair_start = rq->clock; |
aa2ac252 | 754 | return se; |
103638d9 | 755 | } |
aa2ac252 PZ |
756 | |
757 | return cfs_rq->next; | |
758 | } | |
759 | ||
9948f4b2 | 760 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 | 761 | { |
08ec3df5 | 762 | struct sched_entity *se = NULL; |
bf0f6f24 | 763 | |
08ec3df5 DA |
764 | if (first_fair(cfs_rq)) { |
765 | se = __pick_next_entity(cfs_rq); | |
aa2ac252 | 766 | se = pick_next(cfs_rq, se); |
08ec3df5 DA |
767 | set_next_entity(cfs_rq, se); |
768 | } | |
bf0f6f24 IM |
769 | |
770 | return se; | |
771 | } | |
772 | ||
ab6cde26 | 773 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
774 | { |
775 | /* | |
776 | * If still on the runqueue then deactivate_task() | |
777 | * was not called and update_curr() has to be done: | |
778 | */ | |
779 | if (prev->on_rq) | |
b7cc0896 | 780 | update_curr(cfs_rq); |
bf0f6f24 | 781 | |
ddc97297 | 782 | check_spread(cfs_rq, prev); |
30cfdcfc | 783 | if (prev->on_rq) { |
5870db5b | 784 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
785 | /* Put 'current' back into the tree. */ |
786 | __enqueue_entity(cfs_rq, prev); | |
787 | } | |
429d43bc | 788 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
789 | } |
790 | ||
8f4d37ec PZ |
791 | static void |
792 | entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) | |
bf0f6f24 | 793 | { |
bf0f6f24 | 794 | /* |
30cfdcfc | 795 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 796 | */ |
30cfdcfc | 797 | update_curr(cfs_rq); |
bf0f6f24 | 798 | |
8f4d37ec PZ |
799 | #ifdef CONFIG_SCHED_HRTICK |
800 | /* | |
801 | * queued ticks are scheduled to match the slice, so don't bother | |
802 | * validating it and just reschedule. | |
803 | */ | |
983ed7a6 HH |
804 | if (queued) { |
805 | resched_task(rq_of(cfs_rq)->curr); | |
806 | return; | |
807 | } | |
8f4d37ec PZ |
808 | /* |
809 | * don't let the period tick interfere with the hrtick preemption | |
810 | */ | |
811 | if (!sched_feat(DOUBLE_TICK) && | |
812 | hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) | |
813 | return; | |
814 | #endif | |
815 | ||
ce6c1311 | 816 | if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT)) |
2e09bf55 | 817 | check_preempt_tick(cfs_rq, curr); |
bf0f6f24 IM |
818 | } |
819 | ||
820 | /************************************************** | |
821 | * CFS operations on tasks: | |
822 | */ | |
823 | ||
8f4d37ec PZ |
824 | #ifdef CONFIG_SCHED_HRTICK |
825 | static void hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
826 | { | |
8f4d37ec PZ |
827 | struct sched_entity *se = &p->se; |
828 | struct cfs_rq *cfs_rq = cfs_rq_of(se); | |
829 | ||
830 | WARN_ON(task_rq(p) != rq); | |
831 | ||
832 | if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) { | |
833 | u64 slice = sched_slice(cfs_rq, se); | |
834 | u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime; | |
835 | s64 delta = slice - ran; | |
836 | ||
837 | if (delta < 0) { | |
838 | if (rq->curr == p) | |
839 | resched_task(p); | |
840 | return; | |
841 | } | |
842 | ||
843 | /* | |
844 | * Don't schedule slices shorter than 10000ns, that just | |
845 | * doesn't make sense. Rely on vruntime for fairness. | |
846 | */ | |
31656519 | 847 | if (rq->curr != p) |
157124c1 | 848 | delta = max_t(s64, 10000LL, delta); |
8f4d37ec | 849 | |
31656519 | 850 | hrtick_start(rq, delta); |
8f4d37ec PZ |
851 | } |
852 | } | |
a4c2f00f PZ |
853 | |
854 | /* | |
855 | * called from enqueue/dequeue and updates the hrtick when the | |
856 | * current task is from our class and nr_running is low enough | |
857 | * to matter. | |
858 | */ | |
859 | static void hrtick_update(struct rq *rq) | |
860 | { | |
861 | struct task_struct *curr = rq->curr; | |
862 | ||
863 | if (curr->sched_class != &fair_sched_class) | |
864 | return; | |
865 | ||
866 | if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency) | |
867 | hrtick_start_fair(rq, curr); | |
868 | } | |
55e12e5e | 869 | #else /* !CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
870 | static inline void |
871 | hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
872 | { | |
873 | } | |
a4c2f00f PZ |
874 | |
875 | static inline void hrtick_update(struct rq *rq) | |
876 | { | |
877 | } | |
8f4d37ec PZ |
878 | #endif |
879 | ||
bf0f6f24 IM |
880 | /* |
881 | * The enqueue_task method is called before nr_running is | |
882 | * increased. Here we update the fair scheduling stats and | |
883 | * then put the task into the rbtree: | |
884 | */ | |
fd390f6a | 885 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
886 | { |
887 | struct cfs_rq *cfs_rq; | |
62fb1851 | 888 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
889 | |
890 | for_each_sched_entity(se) { | |
62fb1851 | 891 | if (se->on_rq) |
bf0f6f24 IM |
892 | break; |
893 | cfs_rq = cfs_rq_of(se); | |
83b699ed | 894 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 895 | wakeup = 1; |
bf0f6f24 | 896 | } |
8f4d37ec | 897 | |
a4c2f00f | 898 | hrtick_update(rq); |
bf0f6f24 IM |
899 | } |
900 | ||
901 | /* | |
902 | * The dequeue_task method is called before nr_running is | |
903 | * decreased. We remove the task from the rbtree and | |
904 | * update the fair scheduling stats: | |
905 | */ | |
f02231e5 | 906 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
907 | { |
908 | struct cfs_rq *cfs_rq; | |
62fb1851 | 909 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
910 | |
911 | for_each_sched_entity(se) { | |
912 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 913 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 | 914 | /* Don't dequeue parent if it has other entities besides us */ |
62fb1851 | 915 | if (cfs_rq->load.weight) |
bf0f6f24 | 916 | break; |
b9fa3df3 | 917 | sleep = 1; |
bf0f6f24 | 918 | } |
8f4d37ec | 919 | |
a4c2f00f | 920 | hrtick_update(rq); |
bf0f6f24 IM |
921 | } |
922 | ||
923 | /* | |
1799e35d IM |
924 | * sched_yield() support is very simple - we dequeue and enqueue. |
925 | * | |
926 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 927 | */ |
4530d7ab | 928 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 929 | { |
db292ca3 IM |
930 | struct task_struct *curr = rq->curr; |
931 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
932 | struct sched_entity *rightmost, *se = &curr->se; | |
bf0f6f24 IM |
933 | |
934 | /* | |
1799e35d IM |
935 | * Are we the only task in the tree? |
936 | */ | |
937 | if (unlikely(cfs_rq->nr_running == 1)) | |
938 | return; | |
939 | ||
db292ca3 | 940 | if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) { |
3e51f33f | 941 | update_rq_clock(rq); |
1799e35d | 942 | /* |
a2a2d680 | 943 | * Update run-time statistics of the 'current'. |
1799e35d | 944 | */ |
2b1e315d | 945 | update_curr(cfs_rq); |
1799e35d IM |
946 | |
947 | return; | |
948 | } | |
949 | /* | |
950 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 951 | */ |
2b1e315d | 952 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
953 | /* |
954 | * Already in the rightmost position? | |
955 | */ | |
79b3feff | 956 | if (unlikely(!rightmost || rightmost->vruntime < se->vruntime)) |
1799e35d IM |
957 | return; |
958 | ||
959 | /* | |
960 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
961 | * Upon rescheduling, sched_class::put_prev_task() will place |
962 | * 'current' within the tree based on its new key value. | |
1799e35d | 963 | */ |
30cfdcfc | 964 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
965 | } |
966 | ||
e7693a36 GH |
967 | /* |
968 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
969 | * not idle and an idle cpu is available. The span of cpus to | |
970 | * search starts with cpus closest then further out as needed, | |
971 | * so we always favor a closer, idle cpu. | |
e761b772 MK |
972 | * Domains may include CPUs that are not usable for migration, |
973 | * hence we need to mask them out (cpu_active_map) | |
e7693a36 GH |
974 | * |
975 | * Returns the CPU we should wake onto. | |
976 | */ | |
977 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
978 | static int wake_idle(int cpu, struct task_struct *p) | |
979 | { | |
980 | cpumask_t tmp; | |
981 | struct sched_domain *sd; | |
982 | int i; | |
983 | ||
984 | /* | |
985 | * If it is idle, then it is the best cpu to run this task. | |
986 | * | |
987 | * This cpu is also the best, if it has more than one task already. | |
988 | * Siblings must be also busy(in most cases) as they didn't already | |
989 | * pickup the extra load from this cpu and hence we need not check | |
990 | * sibling runqueue info. This will avoid the checks and cache miss | |
991 | * penalities associated with that. | |
992 | */ | |
104f6454 | 993 | if (idle_cpu(cpu) || cpu_rq(cpu)->cfs.nr_running > 1) |
e7693a36 GH |
994 | return cpu; |
995 | ||
996 | for_each_domain(cpu, sd) { | |
1d3504fc HS |
997 | if ((sd->flags & SD_WAKE_IDLE) |
998 | || ((sd->flags & SD_WAKE_IDLE_FAR) | |
999 | && !task_hot(p, task_rq(p)->clock, sd))) { | |
e7693a36 | 1000 | cpus_and(tmp, sd->span, p->cpus_allowed); |
e761b772 | 1001 | cpus_and(tmp, tmp, cpu_active_map); |
363ab6f1 | 1002 | for_each_cpu_mask_nr(i, tmp) { |
e7693a36 GH |
1003 | if (idle_cpu(i)) { |
1004 | if (i != task_cpu(p)) { | |
1005 | schedstat_inc(p, | |
1006 | se.nr_wakeups_idle); | |
1007 | } | |
1008 | return i; | |
1009 | } | |
1010 | } | |
1011 | } else { | |
1012 | break; | |
1013 | } | |
1014 | } | |
1015 | return cpu; | |
1016 | } | |
55e12e5e | 1017 | #else /* !ARCH_HAS_SCHED_WAKE_IDLE*/ |
e7693a36 GH |
1018 | static inline int wake_idle(int cpu, struct task_struct *p) |
1019 | { | |
1020 | return cpu; | |
1021 | } | |
1022 | #endif | |
1023 | ||
1024 | #ifdef CONFIG_SMP | |
098fb9db | 1025 | |
bb3469ac | 1026 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f5bfb7d9 PZ |
1027 | /* |
1028 | * effective_load() calculates the load change as seen from the root_task_group | |
1029 | * | |
1030 | * Adding load to a group doesn't make a group heavier, but can cause movement | |
1031 | * of group shares between cpus. Assuming the shares were perfectly aligned one | |
1032 | * can calculate the shift in shares. | |
1033 | * | |
1034 | * The problem is that perfectly aligning the shares is rather expensive, hence | |
1035 | * we try to avoid doing that too often - see update_shares(), which ratelimits | |
1036 | * this change. | |
1037 | * | |
1038 | * We compensate this by not only taking the current delta into account, but | |
1039 | * also considering the delta between when the shares were last adjusted and | |
1040 | * now. | |
1041 | * | |
1042 | * We still saw a performance dip, some tracing learned us that between | |
1043 | * cgroup:/ and cgroup:/foo balancing the number of affine wakeups increased | |
1044 | * significantly. Therefore try to bias the error in direction of failing | |
1045 | * the affine wakeup. | |
1046 | * | |
1047 | */ | |
f1d239f7 PZ |
1048 | static long effective_load(struct task_group *tg, int cpu, |
1049 | long wl, long wg) | |
bb3469ac | 1050 | { |
4be9daaa | 1051 | struct sched_entity *se = tg->se[cpu]; |
f1d239f7 PZ |
1052 | |
1053 | if (!tg->parent) | |
1054 | return wl; | |
1055 | ||
f5bfb7d9 PZ |
1056 | /* |
1057 | * By not taking the decrease of shares on the other cpu into | |
1058 | * account our error leans towards reducing the affine wakeups. | |
1059 | */ | |
1060 | if (!wl && sched_feat(ASYM_EFF_LOAD)) | |
1061 | return wl; | |
1062 | ||
4be9daaa | 1063 | for_each_sched_entity(se) { |
cb5ef42a | 1064 | long S, rw, s, a, b; |
940959e9 PZ |
1065 | long more_w; |
1066 | ||
1067 | /* | |
1068 | * Instead of using this increment, also add the difference | |
1069 | * between when the shares were last updated and now. | |
1070 | */ | |
1071 | more_w = se->my_q->load.weight - se->my_q->rq_weight; | |
1072 | wl += more_w; | |
1073 | wg += more_w; | |
4be9daaa PZ |
1074 | |
1075 | S = se->my_q->tg->shares; | |
1076 | s = se->my_q->shares; | |
f1d239f7 | 1077 | rw = se->my_q->rq_weight; |
bb3469ac | 1078 | |
cb5ef42a PZ |
1079 | a = S*(rw + wl); |
1080 | b = S*rw + s*wg; | |
4be9daaa | 1081 | |
940959e9 PZ |
1082 | wl = s*(a-b); |
1083 | ||
1084 | if (likely(b)) | |
1085 | wl /= b; | |
1086 | ||
83378269 PZ |
1087 | /* |
1088 | * Assume the group is already running and will | |
1089 | * thus already be accounted for in the weight. | |
1090 | * | |
1091 | * That is, moving shares between CPUs, does not | |
1092 | * alter the group weight. | |
1093 | */ | |
4be9daaa | 1094 | wg = 0; |
4be9daaa | 1095 | } |
bb3469ac | 1096 | |
4be9daaa | 1097 | return wl; |
bb3469ac | 1098 | } |
4be9daaa | 1099 | |
bb3469ac | 1100 | #else |
4be9daaa | 1101 | |
83378269 PZ |
1102 | static inline unsigned long effective_load(struct task_group *tg, int cpu, |
1103 | unsigned long wl, unsigned long wg) | |
4be9daaa | 1104 | { |
83378269 | 1105 | return wl; |
bb3469ac | 1106 | } |
4be9daaa | 1107 | |
bb3469ac PZ |
1108 | #endif |
1109 | ||
098fb9db | 1110 | static int |
64b9e029 | 1111 | wake_affine(struct sched_domain *this_sd, struct rq *this_rq, |
4ae7d5ce IM |
1112 | struct task_struct *p, int prev_cpu, int this_cpu, int sync, |
1113 | int idx, unsigned long load, unsigned long this_load, | |
098fb9db IM |
1114 | unsigned int imbalance) |
1115 | { | |
4ae7d5ce | 1116 | struct task_struct *curr = this_rq->curr; |
83378269 | 1117 | struct task_group *tg; |
098fb9db IM |
1118 | unsigned long tl = this_load; |
1119 | unsigned long tl_per_task; | |
83378269 | 1120 | unsigned long weight; |
b3137bc8 | 1121 | int balanced; |
098fb9db | 1122 | |
b3137bc8 | 1123 | if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS)) |
098fb9db IM |
1124 | return 0; |
1125 | ||
2fb7635c PZ |
1126 | if (!sync && sched_feat(SYNC_WAKEUPS) && |
1127 | curr->se.avg_overlap < sysctl_sched_migration_cost && | |
1128 | p->se.avg_overlap < sysctl_sched_migration_cost) | |
1129 | sync = 1; | |
1130 | ||
b3137bc8 MG |
1131 | /* |
1132 | * If sync wakeup then subtract the (maximum possible) | |
1133 | * effect of the currently running task from the load | |
1134 | * of the current CPU: | |
1135 | */ | |
83378269 PZ |
1136 | if (sync) { |
1137 | tg = task_group(current); | |
1138 | weight = current->se.load.weight; | |
1139 | ||
1140 | tl += effective_load(tg, this_cpu, -weight, -weight); | |
1141 | load += effective_load(tg, prev_cpu, 0, -weight); | |
1142 | } | |
b3137bc8 | 1143 | |
83378269 PZ |
1144 | tg = task_group(p); |
1145 | weight = p->se.load.weight; | |
b3137bc8 | 1146 | |
83378269 PZ |
1147 | balanced = 100*(tl + effective_load(tg, this_cpu, weight, weight)) <= |
1148 | imbalance*(load + effective_load(tg, prev_cpu, 0, weight)); | |
b3137bc8 | 1149 | |
098fb9db | 1150 | /* |
4ae7d5ce IM |
1151 | * If the currently running task will sleep within |
1152 | * a reasonable amount of time then attract this newly | |
1153 | * woken task: | |
098fb9db | 1154 | */ |
2fb7635c PZ |
1155 | if (sync && balanced) |
1156 | return 1; | |
098fb9db IM |
1157 | |
1158 | schedstat_inc(p, se.nr_wakeups_affine_attempts); | |
1159 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
1160 | ||
64b9e029 AA |
1161 | if (balanced || (tl <= load && tl + target_load(prev_cpu, idx) <= |
1162 | tl_per_task)) { | |
098fb9db IM |
1163 | /* |
1164 | * This domain has SD_WAKE_AFFINE and | |
1165 | * p is cache cold in this domain, and | |
1166 | * there is no bad imbalance. | |
1167 | */ | |
1168 | schedstat_inc(this_sd, ttwu_move_affine); | |
1169 | schedstat_inc(p, se.nr_wakeups_affine); | |
1170 | ||
1171 | return 1; | |
1172 | } | |
1173 | return 0; | |
1174 | } | |
1175 | ||
e7693a36 GH |
1176 | static int select_task_rq_fair(struct task_struct *p, int sync) |
1177 | { | |
e7693a36 | 1178 | struct sched_domain *sd, *this_sd = NULL; |
ac192d39 | 1179 | int prev_cpu, this_cpu, new_cpu; |
098fb9db | 1180 | unsigned long load, this_load; |
64b9e029 | 1181 | struct rq *this_rq; |
098fb9db | 1182 | unsigned int imbalance; |
098fb9db | 1183 | int idx; |
e7693a36 | 1184 | |
ac192d39 | 1185 | prev_cpu = task_cpu(p); |
ac192d39 | 1186 | this_cpu = smp_processor_id(); |
4ae7d5ce | 1187 | this_rq = cpu_rq(this_cpu); |
ac192d39 | 1188 | new_cpu = prev_cpu; |
e7693a36 | 1189 | |
64b9e029 AA |
1190 | if (prev_cpu == this_cpu) |
1191 | goto out; | |
ac192d39 IM |
1192 | /* |
1193 | * 'this_sd' is the first domain that both | |
1194 | * this_cpu and prev_cpu are present in: | |
1195 | */ | |
e7693a36 | 1196 | for_each_domain(this_cpu, sd) { |
ac192d39 | 1197 | if (cpu_isset(prev_cpu, sd->span)) { |
e7693a36 GH |
1198 | this_sd = sd; |
1199 | break; | |
1200 | } | |
1201 | } | |
1202 | ||
1203 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) | |
f4827386 | 1204 | goto out; |
e7693a36 GH |
1205 | |
1206 | /* | |
1207 | * Check for affine wakeup and passive balancing possibilities. | |
1208 | */ | |
098fb9db | 1209 | if (!this_sd) |
f4827386 | 1210 | goto out; |
e7693a36 | 1211 | |
098fb9db IM |
1212 | idx = this_sd->wake_idx; |
1213 | ||
1214 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; | |
1215 | ||
ac192d39 | 1216 | load = source_load(prev_cpu, idx); |
098fb9db IM |
1217 | this_load = target_load(this_cpu, idx); |
1218 | ||
64b9e029 | 1219 | if (wake_affine(this_sd, this_rq, p, prev_cpu, this_cpu, sync, idx, |
4ae7d5ce IM |
1220 | load, this_load, imbalance)) |
1221 | return this_cpu; | |
1222 | ||
098fb9db IM |
1223 | /* |
1224 | * Start passive balancing when half the imbalance_pct | |
1225 | * limit is reached. | |
1226 | */ | |
1227 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1228 | if (imbalance*this_load <= 100*load) { | |
1229 | schedstat_inc(this_sd, ttwu_move_balance); | |
1230 | schedstat_inc(p, se.nr_wakeups_passive); | |
4ae7d5ce | 1231 | return this_cpu; |
e7693a36 GH |
1232 | } |
1233 | } | |
1234 | ||
f4827386 | 1235 | out: |
e7693a36 GH |
1236 | return wake_idle(new_cpu, p); |
1237 | } | |
1238 | #endif /* CONFIG_SMP */ | |
1239 | ||
0bbd3336 PZ |
1240 | static unsigned long wakeup_gran(struct sched_entity *se) |
1241 | { | |
1242 | unsigned long gran = sysctl_sched_wakeup_granularity; | |
1243 | ||
1244 | /* | |
a7be37ac PZ |
1245 | * More easily preempt - nice tasks, while not making it harder for |
1246 | * + nice tasks. | |
0bbd3336 | 1247 | */ |
c9c294a6 | 1248 | if (sched_feat(ASYM_GRAN)) |
69569850 | 1249 | gran = calc_delta_mine(gran, NICE_0_LOAD, &se->load); |
0bbd3336 PZ |
1250 | |
1251 | return gran; | |
1252 | } | |
1253 | ||
bf0f6f24 IM |
1254 | /* |
1255 | * Preempt the current task with a newly woken task if needed: | |
1256 | */ | |
15afe09b | 1257 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync) |
bf0f6f24 IM |
1258 | { |
1259 | struct task_struct *curr = rq->curr; | |
fad095a7 | 1260 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
8651a86c | 1261 | struct sched_entity *se = &curr->se, *pse = &p->se; |
69569850 | 1262 | s64 delta_exec; |
bf0f6f24 IM |
1263 | |
1264 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 1265 | update_rq_clock(rq); |
b7cc0896 | 1266 | update_curr(cfs_rq); |
bf0f6f24 IM |
1267 | resched_task(curr); |
1268 | return; | |
1269 | } | |
aa2ac252 | 1270 | |
4ae7d5ce IM |
1271 | if (unlikely(se == pse)) |
1272 | return; | |
1273 | ||
57fdc26d PZ |
1274 | cfs_rq_of(pse)->next = pse; |
1275 | ||
aec0a514 BR |
1276 | /* |
1277 | * We can come here with TIF_NEED_RESCHED already set from new task | |
1278 | * wake up path. | |
1279 | */ | |
1280 | if (test_tsk_need_resched(curr)) | |
1281 | return; | |
1282 | ||
91c234b4 IM |
1283 | /* |
1284 | * Batch tasks do not preempt (their preemption is driven by | |
1285 | * the tick): | |
1286 | */ | |
1287 | if (unlikely(p->policy == SCHED_BATCH)) | |
1288 | return; | |
bf0f6f24 | 1289 | |
77d9cc44 IM |
1290 | if (!sched_feat(WAKEUP_PREEMPT)) |
1291 | return; | |
8651a86c | 1292 | |
2fb7635c PZ |
1293 | if (sched_feat(WAKEUP_OVERLAP) && (sync || |
1294 | (se->avg_overlap < sysctl_sched_migration_cost && | |
1295 | pse->avg_overlap < sysctl_sched_migration_cost))) { | |
15afe09b PZ |
1296 | resched_task(curr); |
1297 | return; | |
1298 | } | |
1299 | ||
69569850 PZ |
1300 | delta_exec = se->sum_exec_runtime - se->prev_sum_exec_runtime; |
1301 | if (delta_exec > wakeup_gran(pse)) | |
77d9cc44 | 1302 | resched_task(curr); |
bf0f6f24 IM |
1303 | } |
1304 | ||
fb8d4724 | 1305 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 | 1306 | { |
8f4d37ec | 1307 | struct task_struct *p; |
bf0f6f24 IM |
1308 | struct cfs_rq *cfs_rq = &rq->cfs; |
1309 | struct sched_entity *se; | |
1310 | ||
1311 | if (unlikely(!cfs_rq->nr_running)) | |
1312 | return NULL; | |
1313 | ||
1314 | do { | |
9948f4b2 | 1315 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
1316 | cfs_rq = group_cfs_rq(se); |
1317 | } while (cfs_rq); | |
1318 | ||
8f4d37ec PZ |
1319 | p = task_of(se); |
1320 | hrtick_start_fair(rq, p); | |
1321 | ||
1322 | return p; | |
bf0f6f24 IM |
1323 | } |
1324 | ||
1325 | /* | |
1326 | * Account for a descheduled task: | |
1327 | */ | |
31ee529c | 1328 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
1329 | { |
1330 | struct sched_entity *se = &prev->se; | |
1331 | struct cfs_rq *cfs_rq; | |
1332 | ||
1333 | for_each_sched_entity(se) { | |
1334 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 1335 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
1336 | } |
1337 | } | |
1338 | ||
681f3e68 | 1339 | #ifdef CONFIG_SMP |
bf0f6f24 IM |
1340 | /************************************************** |
1341 | * Fair scheduling class load-balancing methods: | |
1342 | */ | |
1343 | ||
1344 | /* | |
1345 | * Load-balancing iterator. Note: while the runqueue stays locked | |
1346 | * during the whole iteration, the current task might be | |
1347 | * dequeued so the iterator has to be dequeue-safe. Here we | |
1348 | * achieve that by always pre-iterating before returning | |
1349 | * the current task: | |
1350 | */ | |
a9957449 | 1351 | static struct task_struct * |
4a55bd5e | 1352 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next) |
bf0f6f24 | 1353 | { |
354d60c2 DG |
1354 | struct task_struct *p = NULL; |
1355 | struct sched_entity *se; | |
bf0f6f24 | 1356 | |
77ae6513 MG |
1357 | if (next == &cfs_rq->tasks) |
1358 | return NULL; | |
1359 | ||
b87f1724 BR |
1360 | se = list_entry(next, struct sched_entity, group_node); |
1361 | p = task_of(se); | |
1362 | cfs_rq->balance_iterator = next->next; | |
77ae6513 | 1363 | |
bf0f6f24 IM |
1364 | return p; |
1365 | } | |
1366 | ||
1367 | static struct task_struct *load_balance_start_fair(void *arg) | |
1368 | { | |
1369 | struct cfs_rq *cfs_rq = arg; | |
1370 | ||
4a55bd5e | 1371 | return __load_balance_iterator(cfs_rq, cfs_rq->tasks.next); |
bf0f6f24 IM |
1372 | } |
1373 | ||
1374 | static struct task_struct *load_balance_next_fair(void *arg) | |
1375 | { | |
1376 | struct cfs_rq *cfs_rq = arg; | |
1377 | ||
4a55bd5e | 1378 | return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator); |
bf0f6f24 IM |
1379 | } |
1380 | ||
c09595f6 PZ |
1381 | static unsigned long |
1382 | __load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1383 | unsigned long max_load_move, struct sched_domain *sd, | |
1384 | enum cpu_idle_type idle, int *all_pinned, int *this_best_prio, | |
1385 | struct cfs_rq *cfs_rq) | |
62fb1851 | 1386 | { |
c09595f6 | 1387 | struct rq_iterator cfs_rq_iterator; |
62fb1851 | 1388 | |
c09595f6 PZ |
1389 | cfs_rq_iterator.start = load_balance_start_fair; |
1390 | cfs_rq_iterator.next = load_balance_next_fair; | |
1391 | cfs_rq_iterator.arg = cfs_rq; | |
62fb1851 | 1392 | |
c09595f6 PZ |
1393 | return balance_tasks(this_rq, this_cpu, busiest, |
1394 | max_load_move, sd, idle, all_pinned, | |
1395 | this_best_prio, &cfs_rq_iterator); | |
62fb1851 | 1396 | } |
62fb1851 | 1397 | |
c09595f6 | 1398 | #ifdef CONFIG_FAIR_GROUP_SCHED |
43010659 | 1399 | static unsigned long |
bf0f6f24 | 1400 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f | 1401 | unsigned long max_load_move, |
a4ac01c3 PW |
1402 | struct sched_domain *sd, enum cpu_idle_type idle, |
1403 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 | 1404 | { |
bf0f6f24 | 1405 | long rem_load_move = max_load_move; |
c09595f6 PZ |
1406 | int busiest_cpu = cpu_of(busiest); |
1407 | struct task_group *tg; | |
18d95a28 | 1408 | |
c09595f6 | 1409 | rcu_read_lock(); |
c8cba857 | 1410 | update_h_load(busiest_cpu); |
18d95a28 | 1411 | |
caea8a03 | 1412 | list_for_each_entry_rcu(tg, &task_groups, list) { |
c8cba857 | 1413 | struct cfs_rq *busiest_cfs_rq = tg->cfs_rq[busiest_cpu]; |
42a3ac7d PZ |
1414 | unsigned long busiest_h_load = busiest_cfs_rq->h_load; |
1415 | unsigned long busiest_weight = busiest_cfs_rq->load.weight; | |
243e0e7b | 1416 | u64 rem_load, moved_load; |
18d95a28 | 1417 | |
c09595f6 PZ |
1418 | /* |
1419 | * empty group | |
1420 | */ | |
c8cba857 | 1421 | if (!busiest_cfs_rq->task_weight) |
bf0f6f24 IM |
1422 | continue; |
1423 | ||
243e0e7b SV |
1424 | rem_load = (u64)rem_load_move * busiest_weight; |
1425 | rem_load = div_u64(rem_load, busiest_h_load + 1); | |
bf0f6f24 | 1426 | |
c09595f6 | 1427 | moved_load = __load_balance_fair(this_rq, this_cpu, busiest, |
53fecd8a | 1428 | rem_load, sd, idle, all_pinned, this_best_prio, |
c09595f6 | 1429 | tg->cfs_rq[busiest_cpu]); |
bf0f6f24 | 1430 | |
c09595f6 | 1431 | if (!moved_load) |
bf0f6f24 IM |
1432 | continue; |
1433 | ||
42a3ac7d | 1434 | moved_load *= busiest_h_load; |
243e0e7b | 1435 | moved_load = div_u64(moved_load, busiest_weight + 1); |
bf0f6f24 | 1436 | |
c09595f6 PZ |
1437 | rem_load_move -= moved_load; |
1438 | if (rem_load_move < 0) | |
bf0f6f24 IM |
1439 | break; |
1440 | } | |
c09595f6 | 1441 | rcu_read_unlock(); |
bf0f6f24 | 1442 | |
43010659 | 1443 | return max_load_move - rem_load_move; |
bf0f6f24 | 1444 | } |
c09595f6 PZ |
1445 | #else |
1446 | static unsigned long | |
1447 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1448 | unsigned long max_load_move, | |
1449 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1450 | int *all_pinned, int *this_best_prio) | |
1451 | { | |
1452 | return __load_balance_fair(this_rq, this_cpu, busiest, | |
1453 | max_load_move, sd, idle, all_pinned, | |
1454 | this_best_prio, &busiest->cfs); | |
1455 | } | |
1456 | #endif | |
bf0f6f24 | 1457 | |
e1d1484f PW |
1458 | static int |
1459 | move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1460 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1461 | { | |
1462 | struct cfs_rq *busy_cfs_rq; | |
1463 | struct rq_iterator cfs_rq_iterator; | |
1464 | ||
1465 | cfs_rq_iterator.start = load_balance_start_fair; | |
1466 | cfs_rq_iterator.next = load_balance_next_fair; | |
1467 | ||
1468 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
1469 | /* | |
1470 | * pass busy_cfs_rq argument into | |
1471 | * load_balance_[start|next]_fair iterators | |
1472 | */ | |
1473 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1474 | if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle, | |
1475 | &cfs_rq_iterator)) | |
1476 | return 1; | |
1477 | } | |
1478 | ||
1479 | return 0; | |
1480 | } | |
55e12e5e | 1481 | #endif /* CONFIG_SMP */ |
e1d1484f | 1482 | |
bf0f6f24 IM |
1483 | /* |
1484 | * scheduler tick hitting a task of our scheduling class: | |
1485 | */ | |
8f4d37ec | 1486 | static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) |
bf0f6f24 IM |
1487 | { |
1488 | struct cfs_rq *cfs_rq; | |
1489 | struct sched_entity *se = &curr->se; | |
1490 | ||
1491 | for_each_sched_entity(se) { | |
1492 | cfs_rq = cfs_rq_of(se); | |
8f4d37ec | 1493 | entity_tick(cfs_rq, se, queued); |
bf0f6f24 IM |
1494 | } |
1495 | } | |
1496 | ||
8eb172d9 | 1497 | #define swap(a, b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
4d78e7b6 | 1498 | |
bf0f6f24 IM |
1499 | /* |
1500 | * Share the fairness runtime between parent and child, thus the | |
1501 | * total amount of pressure for CPU stays equal - new tasks | |
1502 | * get a chance to run but frequent forkers are not allowed to | |
1503 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1504 | * the child is not running yet. | |
1505 | */ | |
ee0827d8 | 1506 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1507 | { |
1508 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1509 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
00bf7bfc | 1510 | int this_cpu = smp_processor_id(); |
bf0f6f24 IM |
1511 | |
1512 | sched_info_queued(p); | |
1513 | ||
7109c442 | 1514 | update_curr(cfs_rq); |
aeb73b04 | 1515 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1516 | |
3c90e6e9 | 1517 | /* 'curr' will be NULL if the child belongs to a different group */ |
00bf7bfc | 1518 | if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) && |
3c90e6e9 | 1519 | curr && curr->vruntime < se->vruntime) { |
87fefa38 | 1520 | /* |
edcb60a3 IM |
1521 | * Upon rescheduling, sched_class::put_prev_task() will place |
1522 | * 'current' within the tree based on its new key value. | |
1523 | */ | |
4d78e7b6 | 1524 | swap(curr->vruntime, se->vruntime); |
aec0a514 | 1525 | resched_task(rq->curr); |
4d78e7b6 | 1526 | } |
bf0f6f24 | 1527 | |
b9dca1e0 | 1528 | enqueue_task_fair(rq, p, 0); |
bf0f6f24 IM |
1529 | } |
1530 | ||
cb469845 SR |
1531 | /* |
1532 | * Priority of the task has changed. Check to see if we preempt | |
1533 | * the current task. | |
1534 | */ | |
1535 | static void prio_changed_fair(struct rq *rq, struct task_struct *p, | |
1536 | int oldprio, int running) | |
1537 | { | |
1538 | /* | |
1539 | * Reschedule if we are currently running on this runqueue and | |
1540 | * our priority decreased, or if we are not currently running on | |
1541 | * this runqueue and our priority is higher than the current's | |
1542 | */ | |
1543 | if (running) { | |
1544 | if (p->prio > oldprio) | |
1545 | resched_task(rq->curr); | |
1546 | } else | |
15afe09b | 1547 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
1548 | } |
1549 | ||
1550 | /* | |
1551 | * We switched to the sched_fair class. | |
1552 | */ | |
1553 | static void switched_to_fair(struct rq *rq, struct task_struct *p, | |
1554 | int running) | |
1555 | { | |
1556 | /* | |
1557 | * We were most likely switched from sched_rt, so | |
1558 | * kick off the schedule if running, otherwise just see | |
1559 | * if we can still preempt the current task. | |
1560 | */ | |
1561 | if (running) | |
1562 | resched_task(rq->curr); | |
1563 | else | |
15afe09b | 1564 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
1565 | } |
1566 | ||
83b699ed SV |
1567 | /* Account for a task changing its policy or group. |
1568 | * | |
1569 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1570 | * migrates between groups/classes. | |
1571 | */ | |
1572 | static void set_curr_task_fair(struct rq *rq) | |
1573 | { | |
1574 | struct sched_entity *se = &rq->curr->se; | |
1575 | ||
1576 | for_each_sched_entity(se) | |
1577 | set_next_entity(cfs_rq_of(se), se); | |
1578 | } | |
1579 | ||
810b3817 PZ |
1580 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1581 | static void moved_group_fair(struct task_struct *p) | |
1582 | { | |
1583 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1584 | ||
1585 | update_curr(cfs_rq); | |
1586 | place_entity(cfs_rq, &p->se, 1); | |
1587 | } | |
1588 | #endif | |
1589 | ||
bf0f6f24 IM |
1590 | /* |
1591 | * All the scheduling class methods: | |
1592 | */ | |
5522d5d5 IM |
1593 | static const struct sched_class fair_sched_class = { |
1594 | .next = &idle_sched_class, | |
bf0f6f24 IM |
1595 | .enqueue_task = enqueue_task_fair, |
1596 | .dequeue_task = dequeue_task_fair, | |
1597 | .yield_task = yield_task_fair, | |
e7693a36 GH |
1598 | #ifdef CONFIG_SMP |
1599 | .select_task_rq = select_task_rq_fair, | |
1600 | #endif /* CONFIG_SMP */ | |
bf0f6f24 | 1601 | |
2e09bf55 | 1602 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1603 | |
1604 | .pick_next_task = pick_next_task_fair, | |
1605 | .put_prev_task = put_prev_task_fair, | |
1606 | ||
681f3e68 | 1607 | #ifdef CONFIG_SMP |
bf0f6f24 | 1608 | .load_balance = load_balance_fair, |
e1d1484f | 1609 | .move_one_task = move_one_task_fair, |
681f3e68 | 1610 | #endif |
bf0f6f24 | 1611 | |
83b699ed | 1612 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 IM |
1613 | .task_tick = task_tick_fair, |
1614 | .task_new = task_new_fair, | |
cb469845 SR |
1615 | |
1616 | .prio_changed = prio_changed_fair, | |
1617 | .switched_to = switched_to_fair, | |
810b3817 PZ |
1618 | |
1619 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1620 | .moved_group = moved_group_fair, | |
1621 | #endif | |
bf0f6f24 IM |
1622 | }; |
1623 | ||
1624 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1625 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1626 | { |
bf0f6f24 IM |
1627 | struct cfs_rq *cfs_rq; |
1628 | ||
5973e5b9 | 1629 | rcu_read_lock(); |
c3b64f1e | 1630 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1631 | print_cfs_rq(m, cpu, cfs_rq); |
5973e5b9 | 1632 | rcu_read_unlock(); |
bf0f6f24 IM |
1633 | } |
1634 | #endif |