]>
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 | ||
bf0f6f24 IM |
76 | /************************************************************** |
77 | * CFS operations on generic schedulable entities: | |
78 | */ | |
79 | ||
b758149c PZ |
80 | static inline struct task_struct *task_of(struct sched_entity *se) |
81 | { | |
82 | return container_of(se, struct task_struct, se); | |
83 | } | |
84 | ||
62160e3f | 85 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bf0f6f24 | 86 | |
62160e3f | 87 | /* cpu runqueue to which this cfs_rq is attached */ |
bf0f6f24 IM |
88 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
89 | { | |
62160e3f | 90 | return cfs_rq->rq; |
bf0f6f24 IM |
91 | } |
92 | ||
62160e3f IM |
93 | /* An entity is a task if it doesn't "own" a runqueue */ |
94 | #define entity_is_task(se) (!se->my_q) | |
bf0f6f24 | 95 | |
b758149c PZ |
96 | /* Walk up scheduling entities hierarchy */ |
97 | #define for_each_sched_entity(se) \ | |
98 | for (; se; se = se->parent) | |
99 | ||
100 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) | |
101 | { | |
102 | return p->se.cfs_rq; | |
103 | } | |
104 | ||
105 | /* runqueue on which this entity is (to be) queued */ | |
106 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) | |
107 | { | |
108 | return se->cfs_rq; | |
109 | } | |
110 | ||
111 | /* runqueue "owned" by this group */ | |
112 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
113 | { | |
114 | return grp->my_q; | |
115 | } | |
116 | ||
117 | /* Given a group's cfs_rq on one cpu, return its corresponding cfs_rq on | |
118 | * another cpu ('this_cpu') | |
119 | */ | |
120 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
121 | { | |
122 | return cfs_rq->tg->cfs_rq[this_cpu]; | |
123 | } | |
124 | ||
125 | /* Iterate thr' all leaf cfs_rq's on a runqueue */ | |
126 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
127 | list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list) | |
128 | ||
129 | /* Do the two (enqueued) entities belong to the same group ? */ | |
130 | static inline int | |
131 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
132 | { | |
133 | if (se->cfs_rq == pse->cfs_rq) | |
134 | return 1; | |
135 | ||
136 | return 0; | |
137 | } | |
138 | ||
139 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
140 | { | |
141 | return se->parent; | |
142 | } | |
143 | ||
62160e3f | 144 | #else /* CONFIG_FAIR_GROUP_SCHED */ |
bf0f6f24 | 145 | |
62160e3f IM |
146 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) |
147 | { | |
148 | return container_of(cfs_rq, struct rq, cfs); | |
bf0f6f24 IM |
149 | } |
150 | ||
151 | #define entity_is_task(se) 1 | |
152 | ||
b758149c PZ |
153 | #define for_each_sched_entity(se) \ |
154 | for (; se; se = NULL) | |
bf0f6f24 | 155 | |
b758149c | 156 | static inline struct cfs_rq *task_cfs_rq(struct task_struct *p) |
bf0f6f24 | 157 | { |
b758149c | 158 | return &task_rq(p)->cfs; |
bf0f6f24 IM |
159 | } |
160 | ||
b758149c PZ |
161 | static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se) |
162 | { | |
163 | struct task_struct *p = task_of(se); | |
164 | struct rq *rq = task_rq(p); | |
165 | ||
166 | return &rq->cfs; | |
167 | } | |
168 | ||
169 | /* runqueue "owned" by this group */ | |
170 | static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp) | |
171 | { | |
172 | return NULL; | |
173 | } | |
174 | ||
175 | static inline struct cfs_rq *cpu_cfs_rq(struct cfs_rq *cfs_rq, int this_cpu) | |
176 | { | |
177 | return &cpu_rq(this_cpu)->cfs; | |
178 | } | |
179 | ||
180 | #define for_each_leaf_cfs_rq(rq, cfs_rq) \ | |
181 | for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL) | |
182 | ||
183 | static inline int | |
184 | is_same_group(struct sched_entity *se, struct sched_entity *pse) | |
185 | { | |
186 | return 1; | |
187 | } | |
188 | ||
189 | static inline struct sched_entity *parent_entity(struct sched_entity *se) | |
190 | { | |
191 | return NULL; | |
192 | } | |
193 | ||
194 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
195 | ||
bf0f6f24 IM |
196 | |
197 | /************************************************************** | |
198 | * Scheduling class tree data structure manipulation methods: | |
199 | */ | |
200 | ||
0702e3eb | 201 | static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime) |
02e0431a | 202 | { |
368059a9 PZ |
203 | s64 delta = (s64)(vruntime - min_vruntime); |
204 | if (delta > 0) | |
02e0431a PZ |
205 | min_vruntime = vruntime; |
206 | ||
207 | return min_vruntime; | |
208 | } | |
209 | ||
0702e3eb | 210 | static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime) |
b0ffd246 PZ |
211 | { |
212 | s64 delta = (s64)(vruntime - min_vruntime); | |
213 | if (delta < 0) | |
214 | min_vruntime = vruntime; | |
215 | ||
216 | return min_vruntime; | |
217 | } | |
218 | ||
0702e3eb | 219 | static inline s64 entity_key(struct cfs_rq *cfs_rq, struct sched_entity *se) |
9014623c | 220 | { |
30cfdcfc | 221 | return se->vruntime - cfs_rq->min_vruntime; |
9014623c PZ |
222 | } |
223 | ||
bf0f6f24 IM |
224 | /* |
225 | * Enqueue an entity into the rb-tree: | |
226 | */ | |
0702e3eb | 227 | static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
228 | { |
229 | struct rb_node **link = &cfs_rq->tasks_timeline.rb_node; | |
230 | struct rb_node *parent = NULL; | |
231 | struct sched_entity *entry; | |
9014623c | 232 | s64 key = entity_key(cfs_rq, se); |
bf0f6f24 IM |
233 | int leftmost = 1; |
234 | ||
235 | /* | |
236 | * Find the right place in the rbtree: | |
237 | */ | |
238 | while (*link) { | |
239 | parent = *link; | |
240 | entry = rb_entry(parent, struct sched_entity, run_node); | |
241 | /* | |
242 | * We dont care about collisions. Nodes with | |
243 | * the same key stay together. | |
244 | */ | |
9014623c | 245 | if (key < entity_key(cfs_rq, entry)) { |
bf0f6f24 IM |
246 | link = &parent->rb_left; |
247 | } else { | |
248 | link = &parent->rb_right; | |
249 | leftmost = 0; | |
250 | } | |
251 | } | |
252 | ||
253 | /* | |
254 | * Maintain a cache of leftmost tree entries (it is frequently | |
255 | * used): | |
256 | */ | |
3fe69747 | 257 | if (leftmost) { |
57cb499d | 258 | cfs_rq->rb_leftmost = &se->run_node; |
3fe69747 PZ |
259 | /* |
260 | * maintain cfs_rq->min_vruntime to be a monotonic increasing | |
261 | * value tracking the leftmost vruntime in the tree. | |
262 | */ | |
263 | cfs_rq->min_vruntime = | |
264 | max_vruntime(cfs_rq->min_vruntime, se->vruntime); | |
265 | } | |
bf0f6f24 IM |
266 | |
267 | rb_link_node(&se->run_node, parent, link); | |
268 | rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline); | |
bf0f6f24 IM |
269 | } |
270 | ||
0702e3eb | 271 | static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 272 | { |
3fe69747 PZ |
273 | if (cfs_rq->rb_leftmost == &se->run_node) { |
274 | struct rb_node *next_node; | |
275 | struct sched_entity *next; | |
276 | ||
277 | next_node = rb_next(&se->run_node); | |
278 | cfs_rq->rb_leftmost = next_node; | |
279 | ||
280 | if (next_node) { | |
281 | next = rb_entry(next_node, | |
282 | struct sched_entity, run_node); | |
283 | cfs_rq->min_vruntime = | |
284 | max_vruntime(cfs_rq->min_vruntime, | |
285 | next->vruntime); | |
286 | } | |
287 | } | |
e9acbff6 | 288 | |
aa2ac252 PZ |
289 | if (cfs_rq->next == se) |
290 | cfs_rq->next = NULL; | |
291 | ||
bf0f6f24 | 292 | rb_erase(&se->run_node, &cfs_rq->tasks_timeline); |
bf0f6f24 IM |
293 | } |
294 | ||
295 | static inline struct rb_node *first_fair(struct cfs_rq *cfs_rq) | |
296 | { | |
297 | return cfs_rq->rb_leftmost; | |
298 | } | |
299 | ||
300 | static struct sched_entity *__pick_next_entity(struct cfs_rq *cfs_rq) | |
301 | { | |
302 | return rb_entry(first_fair(cfs_rq), struct sched_entity, run_node); | |
303 | } | |
304 | ||
aeb73b04 PZ |
305 | static inline struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq) |
306 | { | |
7eee3e67 | 307 | struct rb_node *last = rb_last(&cfs_rq->tasks_timeline); |
aeb73b04 | 308 | |
70eee74b BS |
309 | if (!last) |
310 | return NULL; | |
7eee3e67 IM |
311 | |
312 | return rb_entry(last, struct sched_entity, run_node); | |
aeb73b04 PZ |
313 | } |
314 | ||
bf0f6f24 IM |
315 | /************************************************************** |
316 | * Scheduling class statistics methods: | |
317 | */ | |
318 | ||
b2be5e96 PZ |
319 | #ifdef CONFIG_SCHED_DEBUG |
320 | int sched_nr_latency_handler(struct ctl_table *table, int write, | |
321 | struct file *filp, void __user *buffer, size_t *lenp, | |
322 | loff_t *ppos) | |
323 | { | |
324 | int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); | |
325 | ||
326 | if (ret || !write) | |
327 | return ret; | |
328 | ||
329 | sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency, | |
330 | sysctl_sched_min_granularity); | |
331 | ||
332 | return 0; | |
333 | } | |
334 | #endif | |
647e7cac | 335 | |
a7be37ac PZ |
336 | /* |
337 | * delta *= w / rw | |
338 | */ | |
339 | static inline unsigned long | |
340 | calc_delta_weight(unsigned long delta, struct sched_entity *se) | |
341 | { | |
342 | for_each_sched_entity(se) { | |
343 | delta = calc_delta_mine(delta, | |
344 | se->load.weight, &cfs_rq_of(se)->load); | |
345 | } | |
346 | ||
347 | return delta; | |
348 | } | |
349 | ||
350 | /* | |
351 | * delta *= rw / w | |
352 | */ | |
353 | static inline unsigned long | |
354 | calc_delta_fair(unsigned long delta, struct sched_entity *se) | |
355 | { | |
356 | for_each_sched_entity(se) { | |
357 | delta = calc_delta_mine(delta, | |
358 | cfs_rq_of(se)->load.weight, &se->load); | |
359 | } | |
360 | ||
361 | return delta; | |
362 | } | |
363 | ||
647e7cac IM |
364 | /* |
365 | * The idea is to set a period in which each task runs once. | |
366 | * | |
367 | * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch | |
368 | * this period because otherwise the slices get too small. | |
369 | * | |
370 | * p = (nr <= nl) ? l : l*nr/nl | |
371 | */ | |
4d78e7b6 PZ |
372 | static u64 __sched_period(unsigned long nr_running) |
373 | { | |
374 | u64 period = sysctl_sched_latency; | |
b2be5e96 | 375 | unsigned long nr_latency = sched_nr_latency; |
4d78e7b6 PZ |
376 | |
377 | if (unlikely(nr_running > nr_latency)) { | |
4bf0b771 | 378 | period = sysctl_sched_min_granularity; |
4d78e7b6 | 379 | period *= nr_running; |
4d78e7b6 PZ |
380 | } |
381 | ||
382 | return period; | |
383 | } | |
384 | ||
647e7cac IM |
385 | /* |
386 | * We calculate the wall-time slice from the period by taking a part | |
387 | * proportional to the weight. | |
388 | * | |
389 | * s = p*w/rw | |
390 | */ | |
6d0f0ebd | 391 | static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) |
21805085 | 392 | { |
a7be37ac | 393 | return calc_delta_weight(__sched_period(cfs_rq->nr_running), se); |
bf0f6f24 IM |
394 | } |
395 | ||
647e7cac | 396 | /* |
ac884dec | 397 | * We calculate the vruntime slice of a to be inserted task |
647e7cac | 398 | * |
a7be37ac | 399 | * vs = s*rw/w = p |
647e7cac | 400 | */ |
ac884dec | 401 | static u64 sched_vslice_add(struct cfs_rq *cfs_rq, struct sched_entity *se) |
67e9fb2a | 402 | { |
ac884dec | 403 | unsigned long nr_running = cfs_rq->nr_running; |
67e9fb2a | 404 | |
ac884dec PZ |
405 | if (!se->on_rq) |
406 | nr_running++; | |
67e9fb2a | 407 | |
a7be37ac PZ |
408 | return __sched_period(nr_running); |
409 | } | |
410 | ||
bf0f6f24 IM |
411 | /* |
412 | * Update the current task's runtime statistics. Skip current tasks that | |
413 | * are not in our scheduling class. | |
414 | */ | |
415 | static inline void | |
8ebc91d9 IM |
416 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
417 | unsigned long delta_exec) | |
bf0f6f24 | 418 | { |
bbdba7c0 | 419 | unsigned long delta_exec_weighted; |
bf0f6f24 | 420 | |
8179ca23 | 421 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
422 | |
423 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 424 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
a7be37ac | 425 | delta_exec_weighted = calc_delta_fair(delta_exec, curr); |
e9acbff6 | 426 | curr->vruntime += delta_exec_weighted; |
bf0f6f24 IM |
427 | } |
428 | ||
b7cc0896 | 429 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 430 | { |
429d43bc | 431 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 432 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
433 | unsigned long delta_exec; |
434 | ||
435 | if (unlikely(!curr)) | |
436 | return; | |
437 | ||
438 | /* | |
439 | * Get the amount of time the current task was running | |
440 | * since the last time we changed load (this cannot | |
441 | * overflow on 32 bits): | |
442 | */ | |
8ebc91d9 | 443 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 444 | |
8ebc91d9 IM |
445 | __update_curr(cfs_rq, curr, delta_exec); |
446 | curr->exec_start = now; | |
d842de87 SV |
447 | |
448 | if (entity_is_task(curr)) { | |
449 | struct task_struct *curtask = task_of(curr); | |
450 | ||
451 | cpuacct_charge(curtask, delta_exec); | |
452 | } | |
bf0f6f24 IM |
453 | } |
454 | ||
455 | static inline void | |
5870db5b | 456 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 457 | { |
d281918d | 458 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
459 | } |
460 | ||
bf0f6f24 IM |
461 | /* |
462 | * Task is being enqueued - update stats: | |
463 | */ | |
d2417e5a | 464 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 465 | { |
bf0f6f24 IM |
466 | /* |
467 | * Are we enqueueing a waiting task? (for current tasks | |
468 | * a dequeue/enqueue event is a NOP) | |
469 | */ | |
429d43bc | 470 | if (se != cfs_rq->curr) |
5870db5b | 471 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
472 | } |
473 | ||
bf0f6f24 | 474 | static void |
9ef0a961 | 475 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 476 | { |
bbdba7c0 IM |
477 | schedstat_set(se->wait_max, max(se->wait_max, |
478 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6d082592 AV |
479 | schedstat_set(se->wait_count, se->wait_count + 1); |
480 | schedstat_set(se->wait_sum, se->wait_sum + | |
481 | rq_of(cfs_rq)->clock - se->wait_start); | |
6cfb0d5d | 482 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
483 | } |
484 | ||
485 | static inline void | |
19b6a2e3 | 486 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 487 | { |
bf0f6f24 IM |
488 | /* |
489 | * Mark the end of the wait period if dequeueing a | |
490 | * waiting task: | |
491 | */ | |
429d43bc | 492 | if (se != cfs_rq->curr) |
9ef0a961 | 493 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
494 | } |
495 | ||
496 | /* | |
497 | * We are picking a new current task - update its stats: | |
498 | */ | |
499 | static inline void | |
79303e9e | 500 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
501 | { |
502 | /* | |
503 | * We are starting a new run period: | |
504 | */ | |
d281918d | 505 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
506 | } |
507 | ||
bf0f6f24 IM |
508 | /************************************************** |
509 | * Scheduling class queueing methods: | |
510 | */ | |
511 | ||
c09595f6 PZ |
512 | #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED |
513 | static void | |
514 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
515 | { | |
516 | cfs_rq->task_weight += weight; | |
517 | } | |
518 | #else | |
519 | static inline void | |
520 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
521 | { | |
522 | } | |
523 | #endif | |
524 | ||
30cfdcfc DA |
525 | static void |
526 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
527 | { | |
528 | update_load_add(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
529 | if (!parent_entity(se)) |
530 | inc_cpu_load(rq_of(cfs_rq), se->load.weight); | |
531 | if (entity_is_task(se)) | |
532 | add_cfs_task_weight(cfs_rq, se->load.weight); | |
30cfdcfc DA |
533 | cfs_rq->nr_running++; |
534 | se->on_rq = 1; | |
4a55bd5e | 535 | list_add(&se->group_node, &cfs_rq->tasks); |
30cfdcfc DA |
536 | } |
537 | ||
538 | static void | |
539 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
540 | { | |
541 | update_load_sub(&cfs_rq->load, se->load.weight); | |
c09595f6 PZ |
542 | if (!parent_entity(se)) |
543 | dec_cpu_load(rq_of(cfs_rq), se->load.weight); | |
544 | if (entity_is_task(se)) | |
545 | add_cfs_task_weight(cfs_rq, -se->load.weight); | |
30cfdcfc DA |
546 | cfs_rq->nr_running--; |
547 | se->on_rq = 0; | |
4a55bd5e | 548 | list_del_init(&se->group_node); |
30cfdcfc DA |
549 | } |
550 | ||
2396af69 | 551 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 552 | { |
bf0f6f24 IM |
553 | #ifdef CONFIG_SCHEDSTATS |
554 | if (se->sleep_start) { | |
d281918d | 555 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
9745512c | 556 | struct task_struct *tsk = task_of(se); |
bf0f6f24 IM |
557 | |
558 | if ((s64)delta < 0) | |
559 | delta = 0; | |
560 | ||
561 | if (unlikely(delta > se->sleep_max)) | |
562 | se->sleep_max = delta; | |
563 | ||
564 | se->sleep_start = 0; | |
565 | se->sum_sleep_runtime += delta; | |
9745512c AV |
566 | |
567 | account_scheduler_latency(tsk, delta >> 10, 1); | |
bf0f6f24 IM |
568 | } |
569 | if (se->block_start) { | |
d281918d | 570 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
9745512c | 571 | struct task_struct *tsk = task_of(se); |
bf0f6f24 IM |
572 | |
573 | if ((s64)delta < 0) | |
574 | delta = 0; | |
575 | ||
576 | if (unlikely(delta > se->block_max)) | |
577 | se->block_max = delta; | |
578 | ||
579 | se->block_start = 0; | |
580 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
581 | |
582 | /* | |
583 | * Blocking time is in units of nanosecs, so shift by 20 to | |
584 | * get a milliseconds-range estimation of the amount of | |
585 | * time that the task spent sleeping: | |
586 | */ | |
587 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf | 588 | |
30084fbd IM |
589 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
590 | delta >> 20); | |
591 | } | |
9745512c | 592 | account_scheduler_latency(tsk, delta >> 10, 0); |
bf0f6f24 IM |
593 | } |
594 | #endif | |
595 | } | |
596 | ||
ddc97297 PZ |
597 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
598 | { | |
599 | #ifdef CONFIG_SCHED_DEBUG | |
600 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
601 | ||
602 | if (d < 0) | |
603 | d = -d; | |
604 | ||
605 | if (d > 3*sysctl_sched_latency) | |
606 | schedstat_inc(cfs_rq, nr_spread_over); | |
607 | #endif | |
608 | } | |
609 | ||
aeb73b04 PZ |
610 | static void |
611 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
612 | { | |
67e9fb2a | 613 | u64 vruntime; |
aeb73b04 | 614 | |
3fe69747 PZ |
615 | if (first_fair(cfs_rq)) { |
616 | vruntime = min_vruntime(cfs_rq->min_vruntime, | |
617 | __pick_next_entity(cfs_rq)->vruntime); | |
618 | } else | |
619 | vruntime = cfs_rq->min_vruntime; | |
94dfb5e7 | 620 | |
2cb8600e PZ |
621 | /* |
622 | * The 'current' period is already promised to the current tasks, | |
623 | * however the extra weight of the new task will slow them down a | |
624 | * little, place the new task so that it fits in the slot that | |
625 | * stays open at the end. | |
626 | */ | |
94dfb5e7 | 627 | if (initial && sched_feat(START_DEBIT)) |
647e7cac | 628 | vruntime += sched_vslice_add(cfs_rq, se); |
aeb73b04 | 629 | |
8465e792 | 630 | if (!initial) { |
2cb8600e | 631 | /* sleeps upto a single latency don't count. */ |
a7be37ac PZ |
632 | if (sched_feat(NEW_FAIR_SLEEPERS)) { |
633 | unsigned long thresh = sysctl_sched_latency; | |
634 | ||
635 | /* | |
636 | * convert the sleeper threshold into virtual time | |
637 | */ | |
638 | if (sched_feat(NORMALIZED_SLEEPER)) | |
639 | thresh = calc_delta_fair(thresh, se); | |
640 | ||
641 | vruntime -= thresh; | |
642 | } | |
94359f05 | 643 | |
2cb8600e PZ |
644 | /* ensure we never gain time by being placed backwards. */ |
645 | vruntime = max_vruntime(se->vruntime, vruntime); | |
aeb73b04 PZ |
646 | } |
647 | ||
67e9fb2a | 648 | se->vruntime = vruntime; |
aeb73b04 PZ |
649 | } |
650 | ||
bf0f6f24 | 651 | static void |
83b699ed | 652 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
653 | { |
654 | /* | |
a2a2d680 | 655 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 656 | */ |
b7cc0896 | 657 | update_curr(cfs_rq); |
a992241d | 658 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 | 659 | |
e9acbff6 | 660 | if (wakeup) { |
aeb73b04 | 661 | place_entity(cfs_rq, se, 0); |
2396af69 | 662 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 663 | } |
bf0f6f24 | 664 | |
d2417e5a | 665 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 666 | check_spread(cfs_rq, se); |
83b699ed SV |
667 | if (se != cfs_rq->curr) |
668 | __enqueue_entity(cfs_rq, se); | |
bf0f6f24 IM |
669 | } |
670 | ||
671 | static void | |
525c2716 | 672 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 673 | { |
a2a2d680 DA |
674 | /* |
675 | * Update run-time statistics of the 'current'. | |
676 | */ | |
677 | update_curr(cfs_rq); | |
678 | ||
19b6a2e3 | 679 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 680 | if (sleep) { |
67e9fb2a | 681 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
682 | if (entity_is_task(se)) { |
683 | struct task_struct *tsk = task_of(se); | |
684 | ||
685 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 686 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 687 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 688 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 689 | } |
db36cc7d | 690 | #endif |
67e9fb2a PZ |
691 | } |
692 | ||
83b699ed | 693 | if (se != cfs_rq->curr) |
30cfdcfc DA |
694 | __dequeue_entity(cfs_rq, se); |
695 | account_entity_dequeue(cfs_rq, se); | |
bf0f6f24 IM |
696 | } |
697 | ||
698 | /* | |
699 | * Preempt the current task with a newly woken task if needed: | |
700 | */ | |
7c92e54f | 701 | static void |
2e09bf55 | 702 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 703 | { |
11697830 PZ |
704 | unsigned long ideal_runtime, delta_exec; |
705 | ||
6d0f0ebd | 706 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 | 707 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
3e3e13f3 | 708 | if (delta_exec > ideal_runtime) |
bf0f6f24 IM |
709 | resched_task(rq_of(cfs_rq)->curr); |
710 | } | |
711 | ||
83b699ed | 712 | static void |
8494f412 | 713 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 714 | { |
83b699ed SV |
715 | /* 'current' is not kept within the tree. */ |
716 | if (se->on_rq) { | |
717 | /* | |
718 | * Any task has to be enqueued before it get to execute on | |
719 | * a CPU. So account for the time it spent waiting on the | |
720 | * runqueue. | |
721 | */ | |
722 | update_stats_wait_end(cfs_rq, se); | |
723 | __dequeue_entity(cfs_rq, se); | |
724 | } | |
725 | ||
79303e9e | 726 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 727 | cfs_rq->curr = se; |
eba1ed4b IM |
728 | #ifdef CONFIG_SCHEDSTATS |
729 | /* | |
730 | * Track our maximum slice length, if the CPU's load is at | |
731 | * least twice that of our own weight (i.e. dont track it | |
732 | * when there are only lesser-weight tasks around): | |
733 | */ | |
495eca49 | 734 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
735 | se->slice_max = max(se->slice_max, |
736 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
737 | } | |
738 | #endif | |
4a55b450 | 739 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
740 | } |
741 | ||
aa2ac252 PZ |
742 | static struct sched_entity * |
743 | pick_next(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
744 | { | |
103638d9 PZ |
745 | struct rq *rq = rq_of(cfs_rq); |
746 | u64 pair_slice = rq->clock - cfs_rq->pair_start; | |
aa2ac252 | 747 | |
103638d9 PZ |
748 | if (!cfs_rq->next || pair_slice > sched_slice(cfs_rq, cfs_rq->next)) { |
749 | cfs_rq->pair_start = rq->clock; | |
aa2ac252 | 750 | return se; |
103638d9 | 751 | } |
aa2ac252 PZ |
752 | |
753 | return cfs_rq->next; | |
754 | } | |
755 | ||
9948f4b2 | 756 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 | 757 | { |
08ec3df5 | 758 | struct sched_entity *se = NULL; |
bf0f6f24 | 759 | |
08ec3df5 DA |
760 | if (first_fair(cfs_rq)) { |
761 | se = __pick_next_entity(cfs_rq); | |
aa2ac252 | 762 | se = pick_next(cfs_rq, se); |
08ec3df5 DA |
763 | set_next_entity(cfs_rq, se); |
764 | } | |
bf0f6f24 IM |
765 | |
766 | return se; | |
767 | } | |
768 | ||
ab6cde26 | 769 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
770 | { |
771 | /* | |
772 | * If still on the runqueue then deactivate_task() | |
773 | * was not called and update_curr() has to be done: | |
774 | */ | |
775 | if (prev->on_rq) | |
b7cc0896 | 776 | update_curr(cfs_rq); |
bf0f6f24 | 777 | |
ddc97297 | 778 | check_spread(cfs_rq, prev); |
30cfdcfc | 779 | if (prev->on_rq) { |
5870db5b | 780 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
781 | /* Put 'current' back into the tree. */ |
782 | __enqueue_entity(cfs_rq, prev); | |
783 | } | |
429d43bc | 784 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
785 | } |
786 | ||
8f4d37ec PZ |
787 | static void |
788 | entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) | |
bf0f6f24 | 789 | { |
bf0f6f24 | 790 | /* |
30cfdcfc | 791 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 792 | */ |
30cfdcfc | 793 | update_curr(cfs_rq); |
bf0f6f24 | 794 | |
8f4d37ec PZ |
795 | #ifdef CONFIG_SCHED_HRTICK |
796 | /* | |
797 | * queued ticks are scheduled to match the slice, so don't bother | |
798 | * validating it and just reschedule. | |
799 | */ | |
983ed7a6 HH |
800 | if (queued) { |
801 | resched_task(rq_of(cfs_rq)->curr); | |
802 | return; | |
803 | } | |
8f4d37ec PZ |
804 | /* |
805 | * don't let the period tick interfere with the hrtick preemption | |
806 | */ | |
807 | if (!sched_feat(DOUBLE_TICK) && | |
808 | hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) | |
809 | return; | |
810 | #endif | |
811 | ||
ce6c1311 | 812 | if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT)) |
2e09bf55 | 813 | check_preempt_tick(cfs_rq, curr); |
bf0f6f24 IM |
814 | } |
815 | ||
816 | /************************************************** | |
817 | * CFS operations on tasks: | |
818 | */ | |
819 | ||
8f4d37ec PZ |
820 | #ifdef CONFIG_SCHED_HRTICK |
821 | static void hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
822 | { | |
8f4d37ec PZ |
823 | struct sched_entity *se = &p->se; |
824 | struct cfs_rq *cfs_rq = cfs_rq_of(se); | |
825 | ||
826 | WARN_ON(task_rq(p) != rq); | |
827 | ||
828 | if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) { | |
829 | u64 slice = sched_slice(cfs_rq, se); | |
830 | u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime; | |
831 | s64 delta = slice - ran; | |
832 | ||
833 | if (delta < 0) { | |
834 | if (rq->curr == p) | |
835 | resched_task(p); | |
836 | return; | |
837 | } | |
838 | ||
839 | /* | |
840 | * Don't schedule slices shorter than 10000ns, that just | |
841 | * doesn't make sense. Rely on vruntime for fairness. | |
842 | */ | |
31656519 | 843 | if (rq->curr != p) |
157124c1 | 844 | delta = max_t(s64, 10000LL, delta); |
8f4d37ec | 845 | |
31656519 | 846 | hrtick_start(rq, delta); |
8f4d37ec PZ |
847 | } |
848 | } | |
55e12e5e | 849 | #else /* !CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
850 | static inline void |
851 | hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
852 | { | |
853 | } | |
854 | #endif | |
855 | ||
bf0f6f24 IM |
856 | /* |
857 | * The enqueue_task method is called before nr_running is | |
858 | * increased. Here we update the fair scheduling stats and | |
859 | * then put the task into the rbtree: | |
860 | */ | |
fd390f6a | 861 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
862 | { |
863 | struct cfs_rq *cfs_rq; | |
62fb1851 | 864 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
865 | |
866 | for_each_sched_entity(se) { | |
62fb1851 | 867 | if (se->on_rq) |
bf0f6f24 IM |
868 | break; |
869 | cfs_rq = cfs_rq_of(se); | |
83b699ed | 870 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 871 | wakeup = 1; |
bf0f6f24 | 872 | } |
8f4d37ec PZ |
873 | |
874 | hrtick_start_fair(rq, rq->curr); | |
bf0f6f24 IM |
875 | } |
876 | ||
877 | /* | |
878 | * The dequeue_task method is called before nr_running is | |
879 | * decreased. We remove the task from the rbtree and | |
880 | * update the fair scheduling stats: | |
881 | */ | |
f02231e5 | 882 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
883 | { |
884 | struct cfs_rq *cfs_rq; | |
62fb1851 | 885 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
886 | |
887 | for_each_sched_entity(se) { | |
888 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 889 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 | 890 | /* Don't dequeue parent if it has other entities besides us */ |
62fb1851 | 891 | if (cfs_rq->load.weight) |
bf0f6f24 | 892 | break; |
b9fa3df3 | 893 | sleep = 1; |
bf0f6f24 | 894 | } |
8f4d37ec PZ |
895 | |
896 | hrtick_start_fair(rq, rq->curr); | |
bf0f6f24 IM |
897 | } |
898 | ||
899 | /* | |
1799e35d IM |
900 | * sched_yield() support is very simple - we dequeue and enqueue. |
901 | * | |
902 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 903 | */ |
4530d7ab | 904 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 905 | { |
db292ca3 IM |
906 | struct task_struct *curr = rq->curr; |
907 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
908 | struct sched_entity *rightmost, *se = &curr->se; | |
bf0f6f24 IM |
909 | |
910 | /* | |
1799e35d IM |
911 | * Are we the only task in the tree? |
912 | */ | |
913 | if (unlikely(cfs_rq->nr_running == 1)) | |
914 | return; | |
915 | ||
db292ca3 | 916 | if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) { |
3e51f33f | 917 | update_rq_clock(rq); |
1799e35d | 918 | /* |
a2a2d680 | 919 | * Update run-time statistics of the 'current'. |
1799e35d | 920 | */ |
2b1e315d | 921 | update_curr(cfs_rq); |
1799e35d IM |
922 | |
923 | return; | |
924 | } | |
925 | /* | |
926 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 927 | */ |
2b1e315d | 928 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
929 | /* |
930 | * Already in the rightmost position? | |
931 | */ | |
79b3feff | 932 | if (unlikely(!rightmost || rightmost->vruntime < se->vruntime)) |
1799e35d IM |
933 | return; |
934 | ||
935 | /* | |
936 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
937 | * Upon rescheduling, sched_class::put_prev_task() will place |
938 | * 'current' within the tree based on its new key value. | |
1799e35d | 939 | */ |
30cfdcfc | 940 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
941 | } |
942 | ||
e7693a36 GH |
943 | /* |
944 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
945 | * not idle and an idle cpu is available. The span of cpus to | |
946 | * search starts with cpus closest then further out as needed, | |
947 | * so we always favor a closer, idle cpu. | |
e761b772 MK |
948 | * Domains may include CPUs that are not usable for migration, |
949 | * hence we need to mask them out (cpu_active_map) | |
e7693a36 GH |
950 | * |
951 | * Returns the CPU we should wake onto. | |
952 | */ | |
953 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
954 | static int wake_idle(int cpu, struct task_struct *p) | |
955 | { | |
956 | cpumask_t tmp; | |
957 | struct sched_domain *sd; | |
958 | int i; | |
959 | ||
960 | /* | |
961 | * If it is idle, then it is the best cpu to run this task. | |
962 | * | |
963 | * This cpu is also the best, if it has more than one task already. | |
964 | * Siblings must be also busy(in most cases) as they didn't already | |
965 | * pickup the extra load from this cpu and hence we need not check | |
966 | * sibling runqueue info. This will avoid the checks and cache miss | |
967 | * penalities associated with that. | |
968 | */ | |
104f6454 | 969 | if (idle_cpu(cpu) || cpu_rq(cpu)->cfs.nr_running > 1) |
e7693a36 GH |
970 | return cpu; |
971 | ||
972 | for_each_domain(cpu, sd) { | |
1d3504fc HS |
973 | if ((sd->flags & SD_WAKE_IDLE) |
974 | || ((sd->flags & SD_WAKE_IDLE_FAR) | |
975 | && !task_hot(p, task_rq(p)->clock, sd))) { | |
e7693a36 | 976 | cpus_and(tmp, sd->span, p->cpus_allowed); |
e761b772 | 977 | cpus_and(tmp, tmp, cpu_active_map); |
363ab6f1 | 978 | for_each_cpu_mask_nr(i, tmp) { |
e7693a36 GH |
979 | if (idle_cpu(i)) { |
980 | if (i != task_cpu(p)) { | |
981 | schedstat_inc(p, | |
982 | se.nr_wakeups_idle); | |
983 | } | |
984 | return i; | |
985 | } | |
986 | } | |
987 | } else { | |
988 | break; | |
989 | } | |
990 | } | |
991 | return cpu; | |
992 | } | |
55e12e5e | 993 | #else /* !ARCH_HAS_SCHED_WAKE_IDLE*/ |
e7693a36 GH |
994 | static inline int wake_idle(int cpu, struct task_struct *p) |
995 | { | |
996 | return cpu; | |
997 | } | |
998 | #endif | |
999 | ||
1000 | #ifdef CONFIG_SMP | |
098fb9db | 1001 | |
4ae7d5ce IM |
1002 | static const struct sched_class fair_sched_class; |
1003 | ||
bb3469ac | 1004 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f5bfb7d9 PZ |
1005 | /* |
1006 | * effective_load() calculates the load change as seen from the root_task_group | |
1007 | * | |
1008 | * Adding load to a group doesn't make a group heavier, but can cause movement | |
1009 | * of group shares between cpus. Assuming the shares were perfectly aligned one | |
1010 | * can calculate the shift in shares. | |
1011 | * | |
1012 | * The problem is that perfectly aligning the shares is rather expensive, hence | |
1013 | * we try to avoid doing that too often - see update_shares(), which ratelimits | |
1014 | * this change. | |
1015 | * | |
1016 | * We compensate this by not only taking the current delta into account, but | |
1017 | * also considering the delta between when the shares were last adjusted and | |
1018 | * now. | |
1019 | * | |
1020 | * We still saw a performance dip, some tracing learned us that between | |
1021 | * cgroup:/ and cgroup:/foo balancing the number of affine wakeups increased | |
1022 | * significantly. Therefore try to bias the error in direction of failing | |
1023 | * the affine wakeup. | |
1024 | * | |
1025 | */ | |
f1d239f7 PZ |
1026 | static long effective_load(struct task_group *tg, int cpu, |
1027 | long wl, long wg) | |
bb3469ac | 1028 | { |
4be9daaa | 1029 | struct sched_entity *se = tg->se[cpu]; |
f1d239f7 PZ |
1030 | |
1031 | if (!tg->parent) | |
1032 | return wl; | |
1033 | ||
f5bfb7d9 PZ |
1034 | /* |
1035 | * By not taking the decrease of shares on the other cpu into | |
1036 | * account our error leans towards reducing the affine wakeups. | |
1037 | */ | |
1038 | if (!wl && sched_feat(ASYM_EFF_LOAD)) | |
1039 | return wl; | |
1040 | ||
4be9daaa | 1041 | for_each_sched_entity(se) { |
cb5ef42a | 1042 | long S, rw, s, a, b; |
940959e9 PZ |
1043 | long more_w; |
1044 | ||
1045 | /* | |
1046 | * Instead of using this increment, also add the difference | |
1047 | * between when the shares were last updated and now. | |
1048 | */ | |
1049 | more_w = se->my_q->load.weight - se->my_q->rq_weight; | |
1050 | wl += more_w; | |
1051 | wg += more_w; | |
4be9daaa PZ |
1052 | |
1053 | S = se->my_q->tg->shares; | |
1054 | s = se->my_q->shares; | |
f1d239f7 | 1055 | rw = se->my_q->rq_weight; |
bb3469ac | 1056 | |
cb5ef42a PZ |
1057 | a = S*(rw + wl); |
1058 | b = S*rw + s*wg; | |
4be9daaa | 1059 | |
940959e9 PZ |
1060 | wl = s*(a-b); |
1061 | ||
1062 | if (likely(b)) | |
1063 | wl /= b; | |
1064 | ||
83378269 PZ |
1065 | /* |
1066 | * Assume the group is already running and will | |
1067 | * thus already be accounted for in the weight. | |
1068 | * | |
1069 | * That is, moving shares between CPUs, does not | |
1070 | * alter the group weight. | |
1071 | */ | |
4be9daaa | 1072 | wg = 0; |
4be9daaa | 1073 | } |
bb3469ac | 1074 | |
4be9daaa | 1075 | return wl; |
bb3469ac | 1076 | } |
4be9daaa | 1077 | |
bb3469ac | 1078 | #else |
4be9daaa | 1079 | |
83378269 PZ |
1080 | static inline unsigned long effective_load(struct task_group *tg, int cpu, |
1081 | unsigned long wl, unsigned long wg) | |
4be9daaa | 1082 | { |
83378269 | 1083 | return wl; |
bb3469ac | 1084 | } |
4be9daaa | 1085 | |
bb3469ac PZ |
1086 | #endif |
1087 | ||
098fb9db | 1088 | static int |
4ae7d5ce IM |
1089 | wake_affine(struct rq *rq, struct sched_domain *this_sd, struct rq *this_rq, |
1090 | struct task_struct *p, int prev_cpu, int this_cpu, int sync, | |
1091 | int idx, unsigned long load, unsigned long this_load, | |
098fb9db IM |
1092 | unsigned int imbalance) |
1093 | { | |
4ae7d5ce | 1094 | struct task_struct *curr = this_rq->curr; |
83378269 | 1095 | struct task_group *tg; |
098fb9db IM |
1096 | unsigned long tl = this_load; |
1097 | unsigned long tl_per_task; | |
83378269 | 1098 | unsigned long weight; |
b3137bc8 | 1099 | int balanced; |
098fb9db | 1100 | |
b3137bc8 | 1101 | if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS)) |
098fb9db IM |
1102 | return 0; |
1103 | ||
b3137bc8 MG |
1104 | /* |
1105 | * If sync wakeup then subtract the (maximum possible) | |
1106 | * effect of the currently running task from the load | |
1107 | * of the current CPU: | |
1108 | */ | |
83378269 PZ |
1109 | if (sync) { |
1110 | tg = task_group(current); | |
1111 | weight = current->se.load.weight; | |
1112 | ||
1113 | tl += effective_load(tg, this_cpu, -weight, -weight); | |
1114 | load += effective_load(tg, prev_cpu, 0, -weight); | |
1115 | } | |
b3137bc8 | 1116 | |
83378269 PZ |
1117 | tg = task_group(p); |
1118 | weight = p->se.load.weight; | |
b3137bc8 | 1119 | |
83378269 PZ |
1120 | balanced = 100*(tl + effective_load(tg, this_cpu, weight, weight)) <= |
1121 | imbalance*(load + effective_load(tg, prev_cpu, 0, weight)); | |
b3137bc8 | 1122 | |
098fb9db | 1123 | /* |
4ae7d5ce IM |
1124 | * If the currently running task will sleep within |
1125 | * a reasonable amount of time then attract this newly | |
1126 | * woken task: | |
098fb9db | 1127 | */ |
2087a1ad | 1128 | if (sync && balanced) { |
4ae7d5ce | 1129 | if (curr->se.avg_overlap < sysctl_sched_migration_cost && |
2087a1ad | 1130 | p->se.avg_overlap < sysctl_sched_migration_cost) |
4ae7d5ce IM |
1131 | return 1; |
1132 | } | |
098fb9db IM |
1133 | |
1134 | schedstat_inc(p, se.nr_wakeups_affine_attempts); | |
1135 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
1136 | ||
ac192d39 | 1137 | if ((tl <= load && tl + target_load(prev_cpu, idx) <= tl_per_task) || |
b3137bc8 | 1138 | balanced) { |
098fb9db IM |
1139 | /* |
1140 | * This domain has SD_WAKE_AFFINE and | |
1141 | * p is cache cold in this domain, and | |
1142 | * there is no bad imbalance. | |
1143 | */ | |
1144 | schedstat_inc(this_sd, ttwu_move_affine); | |
1145 | schedstat_inc(p, se.nr_wakeups_affine); | |
1146 | ||
1147 | return 1; | |
1148 | } | |
1149 | return 0; | |
1150 | } | |
1151 | ||
e7693a36 GH |
1152 | static int select_task_rq_fair(struct task_struct *p, int sync) |
1153 | { | |
e7693a36 | 1154 | struct sched_domain *sd, *this_sd = NULL; |
ac192d39 | 1155 | int prev_cpu, this_cpu, new_cpu; |
098fb9db | 1156 | unsigned long load, this_load; |
4ae7d5ce | 1157 | struct rq *rq, *this_rq; |
098fb9db | 1158 | unsigned int imbalance; |
098fb9db | 1159 | int idx; |
e7693a36 | 1160 | |
ac192d39 IM |
1161 | prev_cpu = task_cpu(p); |
1162 | rq = task_rq(p); | |
1163 | this_cpu = smp_processor_id(); | |
4ae7d5ce | 1164 | this_rq = cpu_rq(this_cpu); |
ac192d39 | 1165 | new_cpu = prev_cpu; |
e7693a36 | 1166 | |
ac192d39 IM |
1167 | /* |
1168 | * 'this_sd' is the first domain that both | |
1169 | * this_cpu and prev_cpu are present in: | |
1170 | */ | |
e7693a36 | 1171 | for_each_domain(this_cpu, sd) { |
ac192d39 | 1172 | if (cpu_isset(prev_cpu, sd->span)) { |
e7693a36 GH |
1173 | this_sd = sd; |
1174 | break; | |
1175 | } | |
1176 | } | |
1177 | ||
1178 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) | |
f4827386 | 1179 | goto out; |
e7693a36 GH |
1180 | |
1181 | /* | |
1182 | * Check for affine wakeup and passive balancing possibilities. | |
1183 | */ | |
098fb9db | 1184 | if (!this_sd) |
f4827386 | 1185 | goto out; |
e7693a36 | 1186 | |
098fb9db IM |
1187 | idx = this_sd->wake_idx; |
1188 | ||
1189 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; | |
1190 | ||
ac192d39 | 1191 | load = source_load(prev_cpu, idx); |
098fb9db IM |
1192 | this_load = target_load(this_cpu, idx); |
1193 | ||
4ae7d5ce IM |
1194 | if (wake_affine(rq, this_sd, this_rq, p, prev_cpu, this_cpu, sync, idx, |
1195 | load, this_load, imbalance)) | |
1196 | return this_cpu; | |
1197 | ||
1198 | if (prev_cpu == this_cpu) | |
f4827386 | 1199 | goto out; |
098fb9db IM |
1200 | |
1201 | /* | |
1202 | * Start passive balancing when half the imbalance_pct | |
1203 | * limit is reached. | |
1204 | */ | |
1205 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1206 | if (imbalance*this_load <= 100*load) { | |
1207 | schedstat_inc(this_sd, ttwu_move_balance); | |
1208 | schedstat_inc(p, se.nr_wakeups_passive); | |
4ae7d5ce | 1209 | return this_cpu; |
e7693a36 GH |
1210 | } |
1211 | } | |
1212 | ||
f4827386 | 1213 | out: |
e7693a36 GH |
1214 | return wake_idle(new_cpu, p); |
1215 | } | |
1216 | #endif /* CONFIG_SMP */ | |
1217 | ||
0bbd3336 PZ |
1218 | static unsigned long wakeup_gran(struct sched_entity *se) |
1219 | { | |
1220 | unsigned long gran = sysctl_sched_wakeup_granularity; | |
1221 | ||
1222 | /* | |
a7be37ac PZ |
1223 | * More easily preempt - nice tasks, while not making it harder for |
1224 | * + nice tasks. | |
0bbd3336 | 1225 | */ |
c9c294a6 | 1226 | if (sched_feat(ASYM_GRAN)) |
69569850 | 1227 | gran = calc_delta_mine(gran, NICE_0_LOAD, &se->load); |
0bbd3336 PZ |
1228 | |
1229 | return gran; | |
1230 | } | |
1231 | ||
bf0f6f24 IM |
1232 | /* |
1233 | * Preempt the current task with a newly woken task if needed: | |
1234 | */ | |
15afe09b | 1235 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync) |
bf0f6f24 IM |
1236 | { |
1237 | struct task_struct *curr = rq->curr; | |
fad095a7 | 1238 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
8651a86c | 1239 | struct sched_entity *se = &curr->se, *pse = &p->se; |
69569850 | 1240 | s64 delta_exec; |
bf0f6f24 IM |
1241 | |
1242 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 1243 | update_rq_clock(rq); |
b7cc0896 | 1244 | update_curr(cfs_rq); |
bf0f6f24 IM |
1245 | resched_task(curr); |
1246 | return; | |
1247 | } | |
aa2ac252 | 1248 | |
4ae7d5ce IM |
1249 | if (unlikely(se == pse)) |
1250 | return; | |
1251 | ||
aec0a514 BR |
1252 | /* |
1253 | * We can come here with TIF_NEED_RESCHED already set from new task | |
1254 | * wake up path. | |
1255 | */ | |
1256 | if (test_tsk_need_resched(curr)) | |
1257 | return; | |
1258 | ||
aa2ac252 PZ |
1259 | cfs_rq_of(pse)->next = pse; |
1260 | ||
91c234b4 IM |
1261 | /* |
1262 | * Batch tasks do not preempt (their preemption is driven by | |
1263 | * the tick): | |
1264 | */ | |
1265 | if (unlikely(p->policy == SCHED_BATCH)) | |
1266 | return; | |
bf0f6f24 | 1267 | |
77d9cc44 IM |
1268 | if (!sched_feat(WAKEUP_PREEMPT)) |
1269 | return; | |
8651a86c | 1270 | |
15afe09b PZ |
1271 | if (sched_feat(WAKEUP_OVERLAP) && sync && |
1272 | se->avg_overlap < sysctl_sched_migration_cost && | |
1273 | pse->avg_overlap < sysctl_sched_migration_cost) { | |
1274 | resched_task(curr); | |
1275 | return; | |
1276 | } | |
1277 | ||
69569850 PZ |
1278 | delta_exec = se->sum_exec_runtime - se->prev_sum_exec_runtime; |
1279 | if (delta_exec > wakeup_gran(pse)) | |
77d9cc44 | 1280 | resched_task(curr); |
bf0f6f24 IM |
1281 | } |
1282 | ||
fb8d4724 | 1283 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 | 1284 | { |
8f4d37ec | 1285 | struct task_struct *p; |
bf0f6f24 IM |
1286 | struct cfs_rq *cfs_rq = &rq->cfs; |
1287 | struct sched_entity *se; | |
1288 | ||
1289 | if (unlikely(!cfs_rq->nr_running)) | |
1290 | return NULL; | |
1291 | ||
1292 | do { | |
9948f4b2 | 1293 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
1294 | cfs_rq = group_cfs_rq(se); |
1295 | } while (cfs_rq); | |
1296 | ||
8f4d37ec PZ |
1297 | p = task_of(se); |
1298 | hrtick_start_fair(rq, p); | |
1299 | ||
1300 | return p; | |
bf0f6f24 IM |
1301 | } |
1302 | ||
1303 | /* | |
1304 | * Account for a descheduled task: | |
1305 | */ | |
31ee529c | 1306 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
1307 | { |
1308 | struct sched_entity *se = &prev->se; | |
1309 | struct cfs_rq *cfs_rq; | |
1310 | ||
1311 | for_each_sched_entity(se) { | |
1312 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 1313 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
1314 | } |
1315 | } | |
1316 | ||
681f3e68 | 1317 | #ifdef CONFIG_SMP |
bf0f6f24 IM |
1318 | /************************************************** |
1319 | * Fair scheduling class load-balancing methods: | |
1320 | */ | |
1321 | ||
1322 | /* | |
1323 | * Load-balancing iterator. Note: while the runqueue stays locked | |
1324 | * during the whole iteration, the current task might be | |
1325 | * dequeued so the iterator has to be dequeue-safe. Here we | |
1326 | * achieve that by always pre-iterating before returning | |
1327 | * the current task: | |
1328 | */ | |
a9957449 | 1329 | static struct task_struct * |
4a55bd5e | 1330 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next) |
bf0f6f24 | 1331 | { |
354d60c2 DG |
1332 | struct task_struct *p = NULL; |
1333 | struct sched_entity *se; | |
bf0f6f24 | 1334 | |
77ae6513 MG |
1335 | if (next == &cfs_rq->tasks) |
1336 | return NULL; | |
1337 | ||
1338 | /* Skip over entities that are not tasks */ | |
1339 | do { | |
4a55bd5e PZ |
1340 | se = list_entry(next, struct sched_entity, group_node); |
1341 | next = next->next; | |
77ae6513 | 1342 | } while (next != &cfs_rq->tasks && !entity_is_task(se)); |
354d60c2 | 1343 | |
38736f47 | 1344 | if (next == &cfs_rq->tasks && !entity_is_task(se)) |
77ae6513 | 1345 | return NULL; |
4a55bd5e PZ |
1346 | |
1347 | cfs_rq->balance_iterator = next; | |
77ae6513 MG |
1348 | |
1349 | if (entity_is_task(se)) | |
1350 | p = task_of(se); | |
1351 | ||
bf0f6f24 IM |
1352 | return p; |
1353 | } | |
1354 | ||
1355 | static struct task_struct *load_balance_start_fair(void *arg) | |
1356 | { | |
1357 | struct cfs_rq *cfs_rq = arg; | |
1358 | ||
4a55bd5e | 1359 | return __load_balance_iterator(cfs_rq, cfs_rq->tasks.next); |
bf0f6f24 IM |
1360 | } |
1361 | ||
1362 | static struct task_struct *load_balance_next_fair(void *arg) | |
1363 | { | |
1364 | struct cfs_rq *cfs_rq = arg; | |
1365 | ||
4a55bd5e | 1366 | return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator); |
bf0f6f24 IM |
1367 | } |
1368 | ||
c09595f6 PZ |
1369 | static unsigned long |
1370 | __load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1371 | unsigned long max_load_move, struct sched_domain *sd, | |
1372 | enum cpu_idle_type idle, int *all_pinned, int *this_best_prio, | |
1373 | struct cfs_rq *cfs_rq) | |
62fb1851 | 1374 | { |
c09595f6 | 1375 | struct rq_iterator cfs_rq_iterator; |
62fb1851 | 1376 | |
c09595f6 PZ |
1377 | cfs_rq_iterator.start = load_balance_start_fair; |
1378 | cfs_rq_iterator.next = load_balance_next_fair; | |
1379 | cfs_rq_iterator.arg = cfs_rq; | |
62fb1851 | 1380 | |
c09595f6 PZ |
1381 | return balance_tasks(this_rq, this_cpu, busiest, |
1382 | max_load_move, sd, idle, all_pinned, | |
1383 | this_best_prio, &cfs_rq_iterator); | |
62fb1851 | 1384 | } |
62fb1851 | 1385 | |
c09595f6 | 1386 | #ifdef CONFIG_FAIR_GROUP_SCHED |
43010659 | 1387 | static unsigned long |
bf0f6f24 | 1388 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f | 1389 | unsigned long max_load_move, |
a4ac01c3 PW |
1390 | struct sched_domain *sd, enum cpu_idle_type idle, |
1391 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 | 1392 | { |
bf0f6f24 | 1393 | long rem_load_move = max_load_move; |
c09595f6 PZ |
1394 | int busiest_cpu = cpu_of(busiest); |
1395 | struct task_group *tg; | |
18d95a28 | 1396 | |
c09595f6 | 1397 | rcu_read_lock(); |
c8cba857 | 1398 | update_h_load(busiest_cpu); |
18d95a28 | 1399 | |
caea8a03 | 1400 | list_for_each_entry_rcu(tg, &task_groups, list) { |
c8cba857 | 1401 | struct cfs_rq *busiest_cfs_rq = tg->cfs_rq[busiest_cpu]; |
42a3ac7d PZ |
1402 | unsigned long busiest_h_load = busiest_cfs_rq->h_load; |
1403 | unsigned long busiest_weight = busiest_cfs_rq->load.weight; | |
243e0e7b | 1404 | u64 rem_load, moved_load; |
18d95a28 | 1405 | |
c09595f6 PZ |
1406 | /* |
1407 | * empty group | |
1408 | */ | |
c8cba857 | 1409 | if (!busiest_cfs_rq->task_weight) |
bf0f6f24 IM |
1410 | continue; |
1411 | ||
243e0e7b SV |
1412 | rem_load = (u64)rem_load_move * busiest_weight; |
1413 | rem_load = div_u64(rem_load, busiest_h_load + 1); | |
bf0f6f24 | 1414 | |
c09595f6 | 1415 | moved_load = __load_balance_fair(this_rq, this_cpu, busiest, |
53fecd8a | 1416 | rem_load, sd, idle, all_pinned, this_best_prio, |
c09595f6 | 1417 | tg->cfs_rq[busiest_cpu]); |
bf0f6f24 | 1418 | |
c09595f6 | 1419 | if (!moved_load) |
bf0f6f24 IM |
1420 | continue; |
1421 | ||
42a3ac7d | 1422 | moved_load *= busiest_h_load; |
243e0e7b | 1423 | moved_load = div_u64(moved_load, busiest_weight + 1); |
bf0f6f24 | 1424 | |
c09595f6 PZ |
1425 | rem_load_move -= moved_load; |
1426 | if (rem_load_move < 0) | |
bf0f6f24 IM |
1427 | break; |
1428 | } | |
c09595f6 | 1429 | rcu_read_unlock(); |
bf0f6f24 | 1430 | |
43010659 | 1431 | return max_load_move - rem_load_move; |
bf0f6f24 | 1432 | } |
c09595f6 PZ |
1433 | #else |
1434 | static unsigned long | |
1435 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1436 | unsigned long max_load_move, | |
1437 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1438 | int *all_pinned, int *this_best_prio) | |
1439 | { | |
1440 | return __load_balance_fair(this_rq, this_cpu, busiest, | |
1441 | max_load_move, sd, idle, all_pinned, | |
1442 | this_best_prio, &busiest->cfs); | |
1443 | } | |
1444 | #endif | |
bf0f6f24 | 1445 | |
e1d1484f PW |
1446 | static int |
1447 | move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1448 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1449 | { | |
1450 | struct cfs_rq *busy_cfs_rq; | |
1451 | struct rq_iterator cfs_rq_iterator; | |
1452 | ||
1453 | cfs_rq_iterator.start = load_balance_start_fair; | |
1454 | cfs_rq_iterator.next = load_balance_next_fair; | |
1455 | ||
1456 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
1457 | /* | |
1458 | * pass busy_cfs_rq argument into | |
1459 | * load_balance_[start|next]_fair iterators | |
1460 | */ | |
1461 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1462 | if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle, | |
1463 | &cfs_rq_iterator)) | |
1464 | return 1; | |
1465 | } | |
1466 | ||
1467 | return 0; | |
1468 | } | |
55e12e5e | 1469 | #endif /* CONFIG_SMP */ |
e1d1484f | 1470 | |
bf0f6f24 IM |
1471 | /* |
1472 | * scheduler tick hitting a task of our scheduling class: | |
1473 | */ | |
8f4d37ec | 1474 | static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) |
bf0f6f24 IM |
1475 | { |
1476 | struct cfs_rq *cfs_rq; | |
1477 | struct sched_entity *se = &curr->se; | |
1478 | ||
1479 | for_each_sched_entity(se) { | |
1480 | cfs_rq = cfs_rq_of(se); | |
8f4d37ec | 1481 | entity_tick(cfs_rq, se, queued); |
bf0f6f24 IM |
1482 | } |
1483 | } | |
1484 | ||
8eb172d9 | 1485 | #define swap(a, b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
4d78e7b6 | 1486 | |
bf0f6f24 IM |
1487 | /* |
1488 | * Share the fairness runtime between parent and child, thus the | |
1489 | * total amount of pressure for CPU stays equal - new tasks | |
1490 | * get a chance to run but frequent forkers are not allowed to | |
1491 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1492 | * the child is not running yet. | |
1493 | */ | |
ee0827d8 | 1494 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1495 | { |
1496 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1497 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
00bf7bfc | 1498 | int this_cpu = smp_processor_id(); |
bf0f6f24 IM |
1499 | |
1500 | sched_info_queued(p); | |
1501 | ||
7109c442 | 1502 | update_curr(cfs_rq); |
aeb73b04 | 1503 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1504 | |
3c90e6e9 | 1505 | /* 'curr' will be NULL if the child belongs to a different group */ |
00bf7bfc | 1506 | if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) && |
3c90e6e9 | 1507 | curr && curr->vruntime < se->vruntime) { |
87fefa38 | 1508 | /* |
edcb60a3 IM |
1509 | * Upon rescheduling, sched_class::put_prev_task() will place |
1510 | * 'current' within the tree based on its new key value. | |
1511 | */ | |
4d78e7b6 | 1512 | swap(curr->vruntime, se->vruntime); |
aec0a514 | 1513 | resched_task(rq->curr); |
4d78e7b6 | 1514 | } |
bf0f6f24 | 1515 | |
b9dca1e0 | 1516 | enqueue_task_fair(rq, p, 0); |
bf0f6f24 IM |
1517 | } |
1518 | ||
cb469845 SR |
1519 | /* |
1520 | * Priority of the task has changed. Check to see if we preempt | |
1521 | * the current task. | |
1522 | */ | |
1523 | static void prio_changed_fair(struct rq *rq, struct task_struct *p, | |
1524 | int oldprio, int running) | |
1525 | { | |
1526 | /* | |
1527 | * Reschedule if we are currently running on this runqueue and | |
1528 | * our priority decreased, or if we are not currently running on | |
1529 | * this runqueue and our priority is higher than the current's | |
1530 | */ | |
1531 | if (running) { | |
1532 | if (p->prio > oldprio) | |
1533 | resched_task(rq->curr); | |
1534 | } else | |
15afe09b | 1535 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
1536 | } |
1537 | ||
1538 | /* | |
1539 | * We switched to the sched_fair class. | |
1540 | */ | |
1541 | static void switched_to_fair(struct rq *rq, struct task_struct *p, | |
1542 | int running) | |
1543 | { | |
1544 | /* | |
1545 | * We were most likely switched from sched_rt, so | |
1546 | * kick off the schedule if running, otherwise just see | |
1547 | * if we can still preempt the current task. | |
1548 | */ | |
1549 | if (running) | |
1550 | resched_task(rq->curr); | |
1551 | else | |
15afe09b | 1552 | check_preempt_curr(rq, p, 0); |
cb469845 SR |
1553 | } |
1554 | ||
83b699ed SV |
1555 | /* Account for a task changing its policy or group. |
1556 | * | |
1557 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1558 | * migrates between groups/classes. | |
1559 | */ | |
1560 | static void set_curr_task_fair(struct rq *rq) | |
1561 | { | |
1562 | struct sched_entity *se = &rq->curr->se; | |
1563 | ||
1564 | for_each_sched_entity(se) | |
1565 | set_next_entity(cfs_rq_of(se), se); | |
1566 | } | |
1567 | ||
810b3817 PZ |
1568 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1569 | static void moved_group_fair(struct task_struct *p) | |
1570 | { | |
1571 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1572 | ||
1573 | update_curr(cfs_rq); | |
1574 | place_entity(cfs_rq, &p->se, 1); | |
1575 | } | |
1576 | #endif | |
1577 | ||
bf0f6f24 IM |
1578 | /* |
1579 | * All the scheduling class methods: | |
1580 | */ | |
5522d5d5 IM |
1581 | static const struct sched_class fair_sched_class = { |
1582 | .next = &idle_sched_class, | |
bf0f6f24 IM |
1583 | .enqueue_task = enqueue_task_fair, |
1584 | .dequeue_task = dequeue_task_fair, | |
1585 | .yield_task = yield_task_fair, | |
e7693a36 GH |
1586 | #ifdef CONFIG_SMP |
1587 | .select_task_rq = select_task_rq_fair, | |
1588 | #endif /* CONFIG_SMP */ | |
bf0f6f24 | 1589 | |
2e09bf55 | 1590 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1591 | |
1592 | .pick_next_task = pick_next_task_fair, | |
1593 | .put_prev_task = put_prev_task_fair, | |
1594 | ||
681f3e68 | 1595 | #ifdef CONFIG_SMP |
bf0f6f24 | 1596 | .load_balance = load_balance_fair, |
e1d1484f | 1597 | .move_one_task = move_one_task_fair, |
681f3e68 | 1598 | #endif |
bf0f6f24 | 1599 | |
83b699ed | 1600 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 IM |
1601 | .task_tick = task_tick_fair, |
1602 | .task_new = task_new_fair, | |
cb469845 SR |
1603 | |
1604 | .prio_changed = prio_changed_fair, | |
1605 | .switched_to = switched_to_fair, | |
810b3817 PZ |
1606 | |
1607 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1608 | .moved_group = moved_group_fair, | |
1609 | #endif | |
bf0f6f24 IM |
1610 | }; |
1611 | ||
1612 | #ifdef CONFIG_SCHED_DEBUG | |
5cef9eca | 1613 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1614 | { |
bf0f6f24 IM |
1615 | struct cfs_rq *cfs_rq; |
1616 | ||
5973e5b9 | 1617 | rcu_read_lock(); |
c3b64f1e | 1618 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1619 | print_cfs_rq(m, cpu, cfs_rq); |
5973e5b9 | 1620 | rcu_read_unlock(); |
bf0f6f24 IM |
1621 | } |
1622 | #endif |