]>
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. | |
0bbd3336 | 66 | * (default: 10 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 | */ | |
0bbd3336 | 72 | unsigned int sysctl_sched_wakeup_granularity = 10000000UL; |
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 | |
8f1bc385 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 | { |
8f1bc385 | 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 | * |
8f1bc385 | 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 | |
8f1bc385 PZ |
408 | return __sched_period(nr_running); |
409 | } | |
410 | ||
411 | /* | |
412 | * The goal of calc_delta_asym() is to be asymmetrically around NICE_0_LOAD, in | |
413 | * that it favours >=0 over <0. | |
414 | * | |
415 | * -20 | | |
416 | * | | |
417 | * 0 --------+------- | |
418 | * .' | |
419 | * 19 .' | |
420 | * | |
421 | */ | |
422 | static unsigned long | |
423 | calc_delta_asym(unsigned long delta, struct sched_entity *se) | |
424 | { | |
425 | struct load_weight lw = { | |
426 | .weight = NICE_0_LOAD, | |
427 | .inv_weight = 1UL << (WMULT_SHIFT-NICE_0_SHIFT) | |
428 | }; | |
5f6d858e | 429 | |
ac884dec | 430 | for_each_sched_entity(se) { |
8f1bc385 | 431 | struct load_weight *se_lw = &se->load; |
ac884dec | 432 | |
8f1bc385 PZ |
433 | if (se->load.weight < NICE_0_LOAD) |
434 | se_lw = &lw; | |
ac884dec | 435 | |
8f1bc385 PZ |
436 | delta = calc_delta_mine(delta, |
437 | cfs_rq_of(se)->load.weight, se_lw); | |
ac884dec PZ |
438 | } |
439 | ||
8f1bc385 | 440 | return delta; |
67e9fb2a PZ |
441 | } |
442 | ||
bf0f6f24 IM |
443 | /* |
444 | * Update the current task's runtime statistics. Skip current tasks that | |
445 | * are not in our scheduling class. | |
446 | */ | |
447 | static inline void | |
8ebc91d9 IM |
448 | __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, |
449 | unsigned long delta_exec) | |
bf0f6f24 | 450 | { |
bbdba7c0 | 451 | unsigned long delta_exec_weighted; |
bf0f6f24 | 452 | |
8179ca23 | 453 | schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); |
bf0f6f24 IM |
454 | |
455 | curr->sum_exec_runtime += delta_exec; | |
7a62eabc | 456 | schedstat_add(cfs_rq, exec_clock, delta_exec); |
8f1bc385 | 457 | delta_exec_weighted = calc_delta_fair(delta_exec, curr); |
e9acbff6 | 458 | curr->vruntime += delta_exec_weighted; |
bf0f6f24 IM |
459 | } |
460 | ||
b7cc0896 | 461 | static void update_curr(struct cfs_rq *cfs_rq) |
bf0f6f24 | 462 | { |
429d43bc | 463 | struct sched_entity *curr = cfs_rq->curr; |
8ebc91d9 | 464 | u64 now = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
465 | unsigned long delta_exec; |
466 | ||
467 | if (unlikely(!curr)) | |
468 | return; | |
469 | ||
470 | /* | |
471 | * Get the amount of time the current task was running | |
472 | * since the last time we changed load (this cannot | |
473 | * overflow on 32 bits): | |
474 | */ | |
8ebc91d9 | 475 | delta_exec = (unsigned long)(now - curr->exec_start); |
bf0f6f24 | 476 | |
8ebc91d9 IM |
477 | __update_curr(cfs_rq, curr, delta_exec); |
478 | curr->exec_start = now; | |
d842de87 SV |
479 | |
480 | if (entity_is_task(curr)) { | |
481 | struct task_struct *curtask = task_of(curr); | |
482 | ||
483 | cpuacct_charge(curtask, delta_exec); | |
484 | } | |
bf0f6f24 IM |
485 | } |
486 | ||
487 | static inline void | |
5870db5b | 488 | update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 489 | { |
d281918d | 490 | schedstat_set(se->wait_start, rq_of(cfs_rq)->clock); |
bf0f6f24 IM |
491 | } |
492 | ||
bf0f6f24 IM |
493 | /* |
494 | * Task is being enqueued - update stats: | |
495 | */ | |
d2417e5a | 496 | static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 497 | { |
bf0f6f24 IM |
498 | /* |
499 | * Are we enqueueing a waiting task? (for current tasks | |
500 | * a dequeue/enqueue event is a NOP) | |
501 | */ | |
429d43bc | 502 | if (se != cfs_rq->curr) |
5870db5b | 503 | update_stats_wait_start(cfs_rq, se); |
bf0f6f24 IM |
504 | } |
505 | ||
bf0f6f24 | 506 | static void |
9ef0a961 | 507 | update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 508 | { |
bbdba7c0 IM |
509 | schedstat_set(se->wait_max, max(se->wait_max, |
510 | rq_of(cfs_rq)->clock - se->wait_start)); | |
6d082592 AV |
511 | schedstat_set(se->wait_count, se->wait_count + 1); |
512 | schedstat_set(se->wait_sum, se->wait_sum + | |
513 | rq_of(cfs_rq)->clock - se->wait_start); | |
6cfb0d5d | 514 | schedstat_set(se->wait_start, 0); |
bf0f6f24 IM |
515 | } |
516 | ||
517 | static inline void | |
19b6a2e3 | 518 | update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 519 | { |
bf0f6f24 IM |
520 | /* |
521 | * Mark the end of the wait period if dequeueing a | |
522 | * waiting task: | |
523 | */ | |
429d43bc | 524 | if (se != cfs_rq->curr) |
9ef0a961 | 525 | update_stats_wait_end(cfs_rq, se); |
bf0f6f24 IM |
526 | } |
527 | ||
528 | /* | |
529 | * We are picking a new current task - update its stats: | |
530 | */ | |
531 | static inline void | |
79303e9e | 532 | update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 IM |
533 | { |
534 | /* | |
535 | * We are starting a new run period: | |
536 | */ | |
d281918d | 537 | se->exec_start = rq_of(cfs_rq)->clock; |
bf0f6f24 IM |
538 | } |
539 | ||
bf0f6f24 IM |
540 | /************************************************** |
541 | * Scheduling class queueing methods: | |
542 | */ | |
543 | ||
18d95a28 PZ |
544 | #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED |
545 | static void | |
546 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
547 | { | |
548 | cfs_rq->task_weight += weight; | |
549 | } | |
550 | #else | |
551 | static inline void | |
552 | add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight) | |
553 | { | |
554 | } | |
555 | #endif | |
556 | ||
30cfdcfc DA |
557 | static void |
558 | account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
559 | { | |
560 | update_load_add(&cfs_rq->load, se->load.weight); | |
18d95a28 PZ |
561 | if (!parent_entity(se)) |
562 | inc_cpu_load(rq_of(cfs_rq), se->load.weight); | |
563 | if (entity_is_task(se)) | |
564 | add_cfs_task_weight(cfs_rq, se->load.weight); | |
30cfdcfc DA |
565 | cfs_rq->nr_running++; |
566 | se->on_rq = 1; | |
4a55bd5e | 567 | list_add(&se->group_node, &cfs_rq->tasks); |
30cfdcfc DA |
568 | } |
569 | ||
570 | static void | |
571 | account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
572 | { | |
573 | update_load_sub(&cfs_rq->load, se->load.weight); | |
18d95a28 PZ |
574 | if (!parent_entity(se)) |
575 | dec_cpu_load(rq_of(cfs_rq), se->load.weight); | |
576 | if (entity_is_task(se)) | |
577 | add_cfs_task_weight(cfs_rq, -se->load.weight); | |
30cfdcfc DA |
578 | cfs_rq->nr_running--; |
579 | se->on_rq = 0; | |
4a55bd5e | 580 | list_del_init(&se->group_node); |
30cfdcfc DA |
581 | } |
582 | ||
2396af69 | 583 | static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 584 | { |
bf0f6f24 IM |
585 | #ifdef CONFIG_SCHEDSTATS |
586 | if (se->sleep_start) { | |
d281918d | 587 | u64 delta = rq_of(cfs_rq)->clock - se->sleep_start; |
9745512c | 588 | struct task_struct *tsk = task_of(se); |
bf0f6f24 IM |
589 | |
590 | if ((s64)delta < 0) | |
591 | delta = 0; | |
592 | ||
593 | if (unlikely(delta > se->sleep_max)) | |
594 | se->sleep_max = delta; | |
595 | ||
596 | se->sleep_start = 0; | |
597 | se->sum_sleep_runtime += delta; | |
9745512c AV |
598 | |
599 | account_scheduler_latency(tsk, delta >> 10, 1); | |
bf0f6f24 IM |
600 | } |
601 | if (se->block_start) { | |
d281918d | 602 | u64 delta = rq_of(cfs_rq)->clock - se->block_start; |
9745512c | 603 | struct task_struct *tsk = task_of(se); |
bf0f6f24 IM |
604 | |
605 | if ((s64)delta < 0) | |
606 | delta = 0; | |
607 | ||
608 | if (unlikely(delta > se->block_max)) | |
609 | se->block_max = delta; | |
610 | ||
611 | se->block_start = 0; | |
612 | se->sum_sleep_runtime += delta; | |
30084fbd IM |
613 | |
614 | /* | |
615 | * Blocking time is in units of nanosecs, so shift by 20 to | |
616 | * get a milliseconds-range estimation of the amount of | |
617 | * time that the task spent sleeping: | |
618 | */ | |
619 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
e22f5bbf | 620 | |
30084fbd IM |
621 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(tsk), |
622 | delta >> 20); | |
623 | } | |
9745512c | 624 | account_scheduler_latency(tsk, delta >> 10, 0); |
bf0f6f24 IM |
625 | } |
626 | #endif | |
627 | } | |
628 | ||
ddc97297 PZ |
629 | static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se) |
630 | { | |
631 | #ifdef CONFIG_SCHED_DEBUG | |
632 | s64 d = se->vruntime - cfs_rq->min_vruntime; | |
633 | ||
634 | if (d < 0) | |
635 | d = -d; | |
636 | ||
637 | if (d > 3*sysctl_sched_latency) | |
638 | schedstat_inc(cfs_rq, nr_spread_over); | |
639 | #endif | |
640 | } | |
641 | ||
aeb73b04 PZ |
642 | static void |
643 | place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial) | |
644 | { | |
67e9fb2a | 645 | u64 vruntime; |
aeb73b04 | 646 | |
3fe69747 PZ |
647 | if (first_fair(cfs_rq)) { |
648 | vruntime = min_vruntime(cfs_rq->min_vruntime, | |
649 | __pick_next_entity(cfs_rq)->vruntime); | |
650 | } else | |
651 | vruntime = cfs_rq->min_vruntime; | |
94dfb5e7 | 652 | |
2cb8600e PZ |
653 | /* |
654 | * The 'current' period is already promised to the current tasks, | |
655 | * however the extra weight of the new task will slow them down a | |
656 | * little, place the new task so that it fits in the slot that | |
657 | * stays open at the end. | |
658 | */ | |
94dfb5e7 | 659 | if (initial && sched_feat(START_DEBIT)) |
647e7cac | 660 | vruntime += sched_vslice_add(cfs_rq, se); |
aeb73b04 | 661 | |
8465e792 | 662 | if (!initial) { |
2cb8600e | 663 | /* sleeps upto a single latency don't count. */ |
018d6db4 | 664 | if (sched_feat(NEW_FAIR_SLEEPERS)) { |
112f53f5 | 665 | if (sched_feat(NORMALIZED_SLEEPER)) |
8f1bc385 | 666 | vruntime -= calc_delta_weight(sysctl_sched_latency, se); |
112f53f5 PZ |
667 | else |
668 | vruntime -= sysctl_sched_latency; | |
018d6db4 | 669 | } |
94359f05 | 670 | |
2cb8600e PZ |
671 | /* ensure we never gain time by being placed backwards. */ |
672 | vruntime = max_vruntime(se->vruntime, vruntime); | |
aeb73b04 PZ |
673 | } |
674 | ||
67e9fb2a | 675 | se->vruntime = vruntime; |
aeb73b04 PZ |
676 | } |
677 | ||
bf0f6f24 | 678 | static void |
83b699ed | 679 | enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup) |
bf0f6f24 IM |
680 | { |
681 | /* | |
a2a2d680 | 682 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 683 | */ |
b7cc0896 | 684 | update_curr(cfs_rq); |
a992241d | 685 | account_entity_enqueue(cfs_rq, se); |
bf0f6f24 | 686 | |
e9acbff6 | 687 | if (wakeup) { |
aeb73b04 | 688 | place_entity(cfs_rq, se, 0); |
2396af69 | 689 | enqueue_sleeper(cfs_rq, se); |
e9acbff6 | 690 | } |
bf0f6f24 | 691 | |
d2417e5a | 692 | update_stats_enqueue(cfs_rq, se); |
ddc97297 | 693 | check_spread(cfs_rq, se); |
83b699ed SV |
694 | if (se != cfs_rq->curr) |
695 | __enqueue_entity(cfs_rq, se); | |
bf0f6f24 IM |
696 | } |
697 | ||
4ae7d5ce IM |
698 | static void update_avg(u64 *avg, u64 sample) |
699 | { | |
700 | s64 diff = sample - *avg; | |
701 | *avg += diff >> 3; | |
702 | } | |
703 | ||
704 | static void update_avg_stats(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
705 | { | |
706 | if (!se->last_wakeup) | |
707 | return; | |
708 | ||
709 | update_avg(&se->avg_overlap, se->sum_exec_runtime - se->last_wakeup); | |
710 | se->last_wakeup = 0; | |
711 | } | |
712 | ||
bf0f6f24 | 713 | static void |
525c2716 | 714 | dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep) |
bf0f6f24 | 715 | { |
a2a2d680 DA |
716 | /* |
717 | * Update run-time statistics of the 'current'. | |
718 | */ | |
719 | update_curr(cfs_rq); | |
720 | ||
19b6a2e3 | 721 | update_stats_dequeue(cfs_rq, se); |
db36cc7d | 722 | if (sleep) { |
4ae7d5ce | 723 | update_avg_stats(cfs_rq, se); |
67e9fb2a | 724 | #ifdef CONFIG_SCHEDSTATS |
bf0f6f24 IM |
725 | if (entity_is_task(se)) { |
726 | struct task_struct *tsk = task_of(se); | |
727 | ||
728 | if (tsk->state & TASK_INTERRUPTIBLE) | |
d281918d | 729 | se->sleep_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 730 | if (tsk->state & TASK_UNINTERRUPTIBLE) |
d281918d | 731 | se->block_start = rq_of(cfs_rq)->clock; |
bf0f6f24 | 732 | } |
db36cc7d | 733 | #endif |
67e9fb2a PZ |
734 | } |
735 | ||
83b699ed | 736 | if (se != cfs_rq->curr) |
30cfdcfc DA |
737 | __dequeue_entity(cfs_rq, se); |
738 | account_entity_dequeue(cfs_rq, se); | |
bf0f6f24 IM |
739 | } |
740 | ||
741 | /* | |
742 | * Preempt the current task with a newly woken task if needed: | |
743 | */ | |
7c92e54f | 744 | static void |
2e09bf55 | 745 | check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) |
bf0f6f24 | 746 | { |
11697830 PZ |
747 | unsigned long ideal_runtime, delta_exec; |
748 | ||
6d0f0ebd | 749 | ideal_runtime = sched_slice(cfs_rq, curr); |
11697830 | 750 | delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime; |
3e3e13f3 | 751 | if (delta_exec > ideal_runtime) |
bf0f6f24 IM |
752 | resched_task(rq_of(cfs_rq)->curr); |
753 | } | |
754 | ||
83b699ed | 755 | static void |
8494f412 | 756 | set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) |
bf0f6f24 | 757 | { |
83b699ed SV |
758 | /* 'current' is not kept within the tree. */ |
759 | if (se->on_rq) { | |
760 | /* | |
761 | * Any task has to be enqueued before it get to execute on | |
762 | * a CPU. So account for the time it spent waiting on the | |
763 | * runqueue. | |
764 | */ | |
765 | update_stats_wait_end(cfs_rq, se); | |
766 | __dequeue_entity(cfs_rq, se); | |
767 | } | |
768 | ||
79303e9e | 769 | update_stats_curr_start(cfs_rq, se); |
429d43bc | 770 | cfs_rq->curr = se; |
eba1ed4b IM |
771 | #ifdef CONFIG_SCHEDSTATS |
772 | /* | |
773 | * Track our maximum slice length, if the CPU's load is at | |
774 | * least twice that of our own weight (i.e. dont track it | |
775 | * when there are only lesser-weight tasks around): | |
776 | */ | |
495eca49 | 777 | if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) { |
eba1ed4b IM |
778 | se->slice_max = max(se->slice_max, |
779 | se->sum_exec_runtime - se->prev_sum_exec_runtime); | |
780 | } | |
781 | #endif | |
4a55b450 | 782 | se->prev_sum_exec_runtime = se->sum_exec_runtime; |
bf0f6f24 IM |
783 | } |
784 | ||
0bbd3336 PZ |
785 | static int |
786 | wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se); | |
787 | ||
aa2ac252 PZ |
788 | static struct sched_entity * |
789 | pick_next(struct cfs_rq *cfs_rq, struct sched_entity *se) | |
790 | { | |
aa2ac252 PZ |
791 | if (!cfs_rq->next) |
792 | return se; | |
793 | ||
0bbd3336 | 794 | if (wakeup_preempt_entity(cfs_rq->next, se) != 0) |
aa2ac252 PZ |
795 | return se; |
796 | ||
797 | return cfs_rq->next; | |
798 | } | |
799 | ||
9948f4b2 | 800 | static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq) |
bf0f6f24 | 801 | { |
08ec3df5 | 802 | struct sched_entity *se = NULL; |
bf0f6f24 | 803 | |
08ec3df5 DA |
804 | if (first_fair(cfs_rq)) { |
805 | se = __pick_next_entity(cfs_rq); | |
aa2ac252 | 806 | se = pick_next(cfs_rq, se); |
08ec3df5 DA |
807 | set_next_entity(cfs_rq, se); |
808 | } | |
bf0f6f24 IM |
809 | |
810 | return se; | |
811 | } | |
812 | ||
ab6cde26 | 813 | static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev) |
bf0f6f24 IM |
814 | { |
815 | /* | |
816 | * If still on the runqueue then deactivate_task() | |
817 | * was not called and update_curr() has to be done: | |
818 | */ | |
819 | if (prev->on_rq) | |
b7cc0896 | 820 | update_curr(cfs_rq); |
bf0f6f24 | 821 | |
ddc97297 | 822 | check_spread(cfs_rq, prev); |
30cfdcfc | 823 | if (prev->on_rq) { |
5870db5b | 824 | update_stats_wait_start(cfs_rq, prev); |
30cfdcfc DA |
825 | /* Put 'current' back into the tree. */ |
826 | __enqueue_entity(cfs_rq, prev); | |
827 | } | |
429d43bc | 828 | cfs_rq->curr = NULL; |
bf0f6f24 IM |
829 | } |
830 | ||
8f4d37ec PZ |
831 | static void |
832 | entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) | |
bf0f6f24 | 833 | { |
bf0f6f24 | 834 | /* |
30cfdcfc | 835 | * Update run-time statistics of the 'current'. |
bf0f6f24 | 836 | */ |
30cfdcfc | 837 | update_curr(cfs_rq); |
bf0f6f24 | 838 | |
8f4d37ec PZ |
839 | #ifdef CONFIG_SCHED_HRTICK |
840 | /* | |
841 | * queued ticks are scheduled to match the slice, so don't bother | |
842 | * validating it and just reschedule. | |
843 | */ | |
844 | if (queued) | |
845 | return resched_task(rq_of(cfs_rq)->curr); | |
846 | /* | |
847 | * don't let the period tick interfere with the hrtick preemption | |
848 | */ | |
849 | if (!sched_feat(DOUBLE_TICK) && | |
850 | hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) | |
851 | return; | |
852 | #endif | |
853 | ||
ce6c1311 | 854 | if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT)) |
2e09bf55 | 855 | check_preempt_tick(cfs_rq, curr); |
bf0f6f24 IM |
856 | } |
857 | ||
858 | /************************************************** | |
859 | * CFS operations on tasks: | |
860 | */ | |
861 | ||
8f4d37ec PZ |
862 | #ifdef CONFIG_SCHED_HRTICK |
863 | static void hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
864 | { | |
865 | int requeue = rq->curr == p; | |
866 | struct sched_entity *se = &p->se; | |
867 | struct cfs_rq *cfs_rq = cfs_rq_of(se); | |
868 | ||
869 | WARN_ON(task_rq(p) != rq); | |
870 | ||
871 | if (hrtick_enabled(rq) && cfs_rq->nr_running > 1) { | |
872 | u64 slice = sched_slice(cfs_rq, se); | |
873 | u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime; | |
874 | s64 delta = slice - ran; | |
875 | ||
876 | if (delta < 0) { | |
877 | if (rq->curr == p) | |
878 | resched_task(p); | |
879 | return; | |
880 | } | |
881 | ||
882 | /* | |
883 | * Don't schedule slices shorter than 10000ns, that just | |
884 | * doesn't make sense. Rely on vruntime for fairness. | |
885 | */ | |
886 | if (!requeue) | |
887 | delta = max(10000LL, delta); | |
888 | ||
889 | hrtick_start(rq, delta, requeue); | |
890 | } | |
891 | } | |
892 | #else | |
893 | static inline void | |
894 | hrtick_start_fair(struct rq *rq, struct task_struct *p) | |
895 | { | |
896 | } | |
897 | #endif | |
898 | ||
bf0f6f24 IM |
899 | /* |
900 | * The enqueue_task method is called before nr_running is | |
901 | * increased. Here we update the fair scheduling stats and | |
902 | * then put the task into the rbtree: | |
903 | */ | |
fd390f6a | 904 | static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup) |
bf0f6f24 IM |
905 | { |
906 | struct cfs_rq *cfs_rq; | |
62fb1851 | 907 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
908 | |
909 | for_each_sched_entity(se) { | |
62fb1851 | 910 | if (se->on_rq) |
bf0f6f24 IM |
911 | break; |
912 | cfs_rq = cfs_rq_of(se); | |
83b699ed | 913 | enqueue_entity(cfs_rq, se, wakeup); |
b9fa3df3 | 914 | wakeup = 1; |
bf0f6f24 | 915 | } |
8f4d37ec PZ |
916 | |
917 | hrtick_start_fair(rq, rq->curr); | |
bf0f6f24 IM |
918 | } |
919 | ||
920 | /* | |
921 | * The dequeue_task method is called before nr_running is | |
922 | * decreased. We remove the task from the rbtree and | |
923 | * update the fair scheduling stats: | |
924 | */ | |
f02231e5 | 925 | static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep) |
bf0f6f24 IM |
926 | { |
927 | struct cfs_rq *cfs_rq; | |
62fb1851 | 928 | struct sched_entity *se = &p->se; |
bf0f6f24 IM |
929 | |
930 | for_each_sched_entity(se) { | |
931 | cfs_rq = cfs_rq_of(se); | |
525c2716 | 932 | dequeue_entity(cfs_rq, se, sleep); |
bf0f6f24 | 933 | /* Don't dequeue parent if it has other entities besides us */ |
62fb1851 | 934 | if (cfs_rq->load.weight) |
bf0f6f24 | 935 | break; |
b9fa3df3 | 936 | sleep = 1; |
bf0f6f24 | 937 | } |
8f4d37ec PZ |
938 | |
939 | hrtick_start_fair(rq, rq->curr); | |
bf0f6f24 IM |
940 | } |
941 | ||
942 | /* | |
1799e35d IM |
943 | * sched_yield() support is very simple - we dequeue and enqueue. |
944 | * | |
945 | * If compat_yield is turned on then we requeue to the end of the tree. | |
bf0f6f24 | 946 | */ |
4530d7ab | 947 | static void yield_task_fair(struct rq *rq) |
bf0f6f24 | 948 | { |
db292ca3 IM |
949 | struct task_struct *curr = rq->curr; |
950 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); | |
951 | struct sched_entity *rightmost, *se = &curr->se; | |
bf0f6f24 IM |
952 | |
953 | /* | |
1799e35d IM |
954 | * Are we the only task in the tree? |
955 | */ | |
956 | if (unlikely(cfs_rq->nr_running == 1)) | |
957 | return; | |
958 | ||
db292ca3 | 959 | if (likely(!sysctl_sched_compat_yield) && curr->policy != SCHED_BATCH) { |
1799e35d IM |
960 | __update_rq_clock(rq); |
961 | /* | |
a2a2d680 | 962 | * Update run-time statistics of the 'current'. |
1799e35d | 963 | */ |
2b1e315d | 964 | update_curr(cfs_rq); |
1799e35d IM |
965 | |
966 | return; | |
967 | } | |
968 | /* | |
969 | * Find the rightmost entry in the rbtree: | |
bf0f6f24 | 970 | */ |
2b1e315d | 971 | rightmost = __pick_last_entity(cfs_rq); |
1799e35d IM |
972 | /* |
973 | * Already in the rightmost position? | |
974 | */ | |
79b3feff | 975 | if (unlikely(!rightmost || rightmost->vruntime < se->vruntime)) |
1799e35d IM |
976 | return; |
977 | ||
978 | /* | |
979 | * Minimally necessary key value to be last in the tree: | |
2b1e315d DA |
980 | * Upon rescheduling, sched_class::put_prev_task() will place |
981 | * 'current' within the tree based on its new key value. | |
1799e35d | 982 | */ |
30cfdcfc | 983 | se->vruntime = rightmost->vruntime + 1; |
bf0f6f24 IM |
984 | } |
985 | ||
e7693a36 GH |
986 | /* |
987 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
988 | * not idle and an idle cpu is available. The span of cpus to | |
989 | * search starts with cpus closest then further out as needed, | |
990 | * so we always favor a closer, idle cpu. | |
991 | * | |
992 | * Returns the CPU we should wake onto. | |
993 | */ | |
994 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
995 | static int wake_idle(int cpu, struct task_struct *p) | |
996 | { | |
997 | cpumask_t tmp; | |
998 | struct sched_domain *sd; | |
999 | int i; | |
1000 | ||
1001 | /* | |
1002 | * If it is idle, then it is the best cpu to run this task. | |
1003 | * | |
1004 | * This cpu is also the best, if it has more than one task already. | |
1005 | * Siblings must be also busy(in most cases) as they didn't already | |
1006 | * pickup the extra load from this cpu and hence we need not check | |
1007 | * sibling runqueue info. This will avoid the checks and cache miss | |
1008 | * penalities associated with that. | |
1009 | */ | |
1010 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1011 | return cpu; | |
1012 | ||
1013 | for_each_domain(cpu, sd) { | |
1d3504fc HS |
1014 | if ((sd->flags & SD_WAKE_IDLE) |
1015 | || ((sd->flags & SD_WAKE_IDLE_FAR) | |
1016 | && !task_hot(p, task_rq(p)->clock, sd))) { | |
e7693a36 GH |
1017 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1018 | for_each_cpu_mask(i, tmp) { | |
1019 | if (idle_cpu(i)) { | |
1020 | if (i != task_cpu(p)) { | |
1021 | schedstat_inc(p, | |
1022 | se.nr_wakeups_idle); | |
1023 | } | |
1024 | return i; | |
1025 | } | |
1026 | } | |
1027 | } else { | |
1028 | break; | |
1029 | } | |
1030 | } | |
1031 | return cpu; | |
1032 | } | |
1033 | #else | |
1034 | static inline int wake_idle(int cpu, struct task_struct *p) | |
1035 | { | |
1036 | return cpu; | |
1037 | } | |
1038 | #endif | |
1039 | ||
1040 | #ifdef CONFIG_SMP | |
098fb9db | 1041 | |
4ae7d5ce IM |
1042 | static const struct sched_class fair_sched_class; |
1043 | ||
098fb9db | 1044 | static int |
4ae7d5ce IM |
1045 | wake_affine(struct rq *rq, struct sched_domain *this_sd, struct rq *this_rq, |
1046 | struct task_struct *p, int prev_cpu, int this_cpu, int sync, | |
1047 | int idx, unsigned long load, unsigned long this_load, | |
098fb9db IM |
1048 | unsigned int imbalance) |
1049 | { | |
4ae7d5ce | 1050 | struct task_struct *curr = this_rq->curr; |
098fb9db IM |
1051 | unsigned long tl = this_load; |
1052 | unsigned long tl_per_task; | |
1053 | ||
1054 | if (!(this_sd->flags & SD_WAKE_AFFINE)) | |
1055 | return 0; | |
1056 | ||
1057 | /* | |
4ae7d5ce IM |
1058 | * If the currently running task will sleep within |
1059 | * a reasonable amount of time then attract this newly | |
1060 | * woken task: | |
098fb9db | 1061 | */ |
4ae7d5ce IM |
1062 | if (sync && curr->sched_class == &fair_sched_class) { |
1063 | if (curr->se.avg_overlap < sysctl_sched_migration_cost && | |
1064 | p->se.avg_overlap < sysctl_sched_migration_cost) | |
1065 | return 1; | |
1066 | } | |
098fb9db IM |
1067 | |
1068 | schedstat_inc(p, se.nr_wakeups_affine_attempts); | |
1069 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
1070 | ||
1071 | /* | |
1072 | * If sync wakeup then subtract the (maximum possible) | |
1073 | * effect of the currently running task from the load | |
1074 | * of the current CPU: | |
1075 | */ | |
1076 | if (sync) | |
1077 | tl -= current->se.load.weight; | |
1078 | ||
ac192d39 | 1079 | if ((tl <= load && tl + target_load(prev_cpu, idx) <= tl_per_task) || |
098fb9db IM |
1080 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
1081 | /* | |
1082 | * This domain has SD_WAKE_AFFINE and | |
1083 | * p is cache cold in this domain, and | |
1084 | * there is no bad imbalance. | |
1085 | */ | |
1086 | schedstat_inc(this_sd, ttwu_move_affine); | |
1087 | schedstat_inc(p, se.nr_wakeups_affine); | |
1088 | ||
1089 | return 1; | |
1090 | } | |
1091 | return 0; | |
1092 | } | |
1093 | ||
e7693a36 GH |
1094 | static int select_task_rq_fair(struct task_struct *p, int sync) |
1095 | { | |
e7693a36 | 1096 | struct sched_domain *sd, *this_sd = NULL; |
ac192d39 | 1097 | int prev_cpu, this_cpu, new_cpu; |
098fb9db | 1098 | unsigned long load, this_load; |
4ae7d5ce | 1099 | struct rq *rq, *this_rq; |
098fb9db | 1100 | unsigned int imbalance; |
098fb9db | 1101 | int idx; |
e7693a36 | 1102 | |
ac192d39 IM |
1103 | prev_cpu = task_cpu(p); |
1104 | rq = task_rq(p); | |
1105 | this_cpu = smp_processor_id(); | |
4ae7d5ce | 1106 | this_rq = cpu_rq(this_cpu); |
ac192d39 | 1107 | new_cpu = prev_cpu; |
e7693a36 | 1108 | |
ac192d39 IM |
1109 | /* |
1110 | * 'this_sd' is the first domain that both | |
1111 | * this_cpu and prev_cpu are present in: | |
1112 | */ | |
e7693a36 | 1113 | for_each_domain(this_cpu, sd) { |
ac192d39 | 1114 | if (cpu_isset(prev_cpu, sd->span)) { |
e7693a36 GH |
1115 | this_sd = sd; |
1116 | break; | |
1117 | } | |
1118 | } | |
1119 | ||
1120 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) | |
f4827386 | 1121 | goto out; |
e7693a36 GH |
1122 | |
1123 | /* | |
1124 | * Check for affine wakeup and passive balancing possibilities. | |
1125 | */ | |
098fb9db | 1126 | if (!this_sd) |
f4827386 | 1127 | goto out; |
e7693a36 | 1128 | |
098fb9db IM |
1129 | idx = this_sd->wake_idx; |
1130 | ||
1131 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; | |
1132 | ||
ac192d39 | 1133 | load = source_load(prev_cpu, idx); |
098fb9db IM |
1134 | this_load = target_load(this_cpu, idx); |
1135 | ||
4ae7d5ce IM |
1136 | if (wake_affine(rq, this_sd, this_rq, p, prev_cpu, this_cpu, sync, idx, |
1137 | load, this_load, imbalance)) | |
1138 | return this_cpu; | |
1139 | ||
1140 | if (prev_cpu == this_cpu) | |
f4827386 | 1141 | goto out; |
098fb9db IM |
1142 | |
1143 | /* | |
1144 | * Start passive balancing when half the imbalance_pct | |
1145 | * limit is reached. | |
1146 | */ | |
1147 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1148 | if (imbalance*this_load <= 100*load) { | |
1149 | schedstat_inc(this_sd, ttwu_move_balance); | |
1150 | schedstat_inc(p, se.nr_wakeups_passive); | |
4ae7d5ce | 1151 | return this_cpu; |
e7693a36 GH |
1152 | } |
1153 | } | |
1154 | ||
f4827386 | 1155 | out: |
e7693a36 GH |
1156 | return wake_idle(new_cpu, p); |
1157 | } | |
1158 | #endif /* CONFIG_SMP */ | |
1159 | ||
0bbd3336 PZ |
1160 | static unsigned long wakeup_gran(struct sched_entity *se) |
1161 | { | |
1162 | unsigned long gran = sysctl_sched_wakeup_granularity; | |
1163 | ||
1164 | /* | |
8f1bc385 PZ |
1165 | * More easily preempt - nice tasks, while not making it harder for |
1166 | * + nice tasks. | |
0bbd3336 | 1167 | */ |
8f1bc385 | 1168 | gran = calc_delta_asym(sysctl_sched_wakeup_granularity, se); |
0bbd3336 PZ |
1169 | |
1170 | return gran; | |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * Should 'se' preempt 'curr'. | |
1175 | * | |
1176 | * |s1 | |
1177 | * |s2 | |
1178 | * |s3 | |
1179 | * g | |
1180 | * |<--->|c | |
1181 | * | |
1182 | * w(c, s1) = -1 | |
1183 | * w(c, s2) = 0 | |
1184 | * w(c, s3) = 1 | |
1185 | * | |
1186 | */ | |
1187 | static int | |
1188 | wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se) | |
1189 | { | |
1190 | s64 gran, vdiff = curr->vruntime - se->vruntime; | |
1191 | ||
1192 | if (vdiff < 0) | |
1193 | return -1; | |
1194 | ||
1195 | gran = wakeup_gran(curr); | |
1196 | if (vdiff > gran) | |
1197 | return 1; | |
1198 | ||
1199 | return 0; | |
1200 | } | |
e7693a36 | 1201 | |
354d60c2 DG |
1202 | /* return depth at which a sched entity is present in the hierarchy */ |
1203 | static inline int depth_se(struct sched_entity *se) | |
1204 | { | |
1205 | int depth = 0; | |
1206 | ||
1207 | for_each_sched_entity(se) | |
1208 | depth++; | |
1209 | ||
1210 | return depth; | |
1211 | } | |
1212 | ||
bf0f6f24 IM |
1213 | /* |
1214 | * Preempt the current task with a newly woken task if needed: | |
1215 | */ | |
2e09bf55 | 1216 | static void check_preempt_wakeup(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1217 | { |
1218 | struct task_struct *curr = rq->curr; | |
fad095a7 | 1219 | struct cfs_rq *cfs_rq = task_cfs_rq(curr); |
8651a86c | 1220 | struct sched_entity *se = &curr->se, *pse = &p->se; |
354d60c2 | 1221 | int se_depth, pse_depth; |
bf0f6f24 IM |
1222 | |
1223 | if (unlikely(rt_prio(p->prio))) { | |
a8e504d2 | 1224 | update_rq_clock(rq); |
b7cc0896 | 1225 | update_curr(cfs_rq); |
bf0f6f24 IM |
1226 | resched_task(curr); |
1227 | return; | |
1228 | } | |
aa2ac252 | 1229 | |
4ae7d5ce IM |
1230 | se->last_wakeup = se->sum_exec_runtime; |
1231 | if (unlikely(se == pse)) | |
1232 | return; | |
1233 | ||
aa2ac252 PZ |
1234 | cfs_rq_of(pse)->next = pse; |
1235 | ||
91c234b4 IM |
1236 | /* |
1237 | * Batch tasks do not preempt (their preemption is driven by | |
1238 | * the tick): | |
1239 | */ | |
1240 | if (unlikely(p->policy == SCHED_BATCH)) | |
1241 | return; | |
bf0f6f24 | 1242 | |
77d9cc44 IM |
1243 | if (!sched_feat(WAKEUP_PREEMPT)) |
1244 | return; | |
8651a86c | 1245 | |
354d60c2 DG |
1246 | /* |
1247 | * preemption test can be made between sibling entities who are in the | |
1248 | * same cfs_rq i.e who have a common parent. Walk up the hierarchy of | |
1249 | * both tasks until we find their ancestors who are siblings of common | |
1250 | * parent. | |
1251 | */ | |
1252 | ||
1253 | /* First walk up until both entities are at same depth */ | |
1254 | se_depth = depth_se(se); | |
1255 | pse_depth = depth_se(pse); | |
1256 | ||
1257 | while (se_depth > pse_depth) { | |
1258 | se_depth--; | |
1259 | se = parent_entity(se); | |
1260 | } | |
1261 | ||
1262 | while (pse_depth > se_depth) { | |
1263 | pse_depth--; | |
1264 | pse = parent_entity(pse); | |
1265 | } | |
1266 | ||
77d9cc44 IM |
1267 | while (!is_same_group(se, pse)) { |
1268 | se = parent_entity(se); | |
1269 | pse = parent_entity(pse); | |
ce6c1311 | 1270 | } |
77d9cc44 | 1271 | |
0bbd3336 | 1272 | if (wakeup_preempt_entity(se, pse) == 1) |
77d9cc44 | 1273 | resched_task(curr); |
bf0f6f24 IM |
1274 | } |
1275 | ||
fb8d4724 | 1276 | static struct task_struct *pick_next_task_fair(struct rq *rq) |
bf0f6f24 | 1277 | { |
8f4d37ec | 1278 | struct task_struct *p; |
bf0f6f24 IM |
1279 | struct cfs_rq *cfs_rq = &rq->cfs; |
1280 | struct sched_entity *se; | |
1281 | ||
1282 | if (unlikely(!cfs_rq->nr_running)) | |
1283 | return NULL; | |
1284 | ||
1285 | do { | |
9948f4b2 | 1286 | se = pick_next_entity(cfs_rq); |
bf0f6f24 IM |
1287 | cfs_rq = group_cfs_rq(se); |
1288 | } while (cfs_rq); | |
1289 | ||
8f4d37ec PZ |
1290 | p = task_of(se); |
1291 | hrtick_start_fair(rq, p); | |
1292 | ||
1293 | return p; | |
bf0f6f24 IM |
1294 | } |
1295 | ||
1296 | /* | |
1297 | * Account for a descheduled task: | |
1298 | */ | |
31ee529c | 1299 | static void put_prev_task_fair(struct rq *rq, struct task_struct *prev) |
bf0f6f24 IM |
1300 | { |
1301 | struct sched_entity *se = &prev->se; | |
1302 | struct cfs_rq *cfs_rq; | |
1303 | ||
1304 | for_each_sched_entity(se) { | |
1305 | cfs_rq = cfs_rq_of(se); | |
ab6cde26 | 1306 | put_prev_entity(cfs_rq, se); |
bf0f6f24 IM |
1307 | } |
1308 | } | |
1309 | ||
681f3e68 | 1310 | #ifdef CONFIG_SMP |
bf0f6f24 IM |
1311 | /************************************************** |
1312 | * Fair scheduling class load-balancing methods: | |
1313 | */ | |
1314 | ||
1315 | /* | |
1316 | * Load-balancing iterator. Note: while the runqueue stays locked | |
1317 | * during the whole iteration, the current task might be | |
1318 | * dequeued so the iterator has to be dequeue-safe. Here we | |
1319 | * achieve that by always pre-iterating before returning | |
1320 | * the current task: | |
1321 | */ | |
a9957449 | 1322 | static struct task_struct * |
4a55bd5e | 1323 | __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next) |
bf0f6f24 | 1324 | { |
354d60c2 DG |
1325 | struct task_struct *p = NULL; |
1326 | struct sched_entity *se; | |
bf0f6f24 | 1327 | |
4a55bd5e | 1328 | if (next == &cfs_rq->tasks) |
bf0f6f24 IM |
1329 | return NULL; |
1330 | ||
354d60c2 DG |
1331 | /* Skip over entities that are not tasks */ |
1332 | do { | |
4a55bd5e PZ |
1333 | se = list_entry(next, struct sched_entity, group_node); |
1334 | next = next->next; | |
1335 | } while (next != &cfs_rq->tasks && !entity_is_task(se)); | |
354d60c2 | 1336 | |
4a55bd5e PZ |
1337 | if (next == &cfs_rq->tasks) |
1338 | return NULL; | |
1339 | ||
1340 | cfs_rq->balance_iterator = next; | |
354d60c2 DG |
1341 | |
1342 | if (entity_is_task(se)) | |
1343 | p = task_of(se); | |
bf0f6f24 IM |
1344 | |
1345 | return p; | |
1346 | } | |
1347 | ||
1348 | static struct task_struct *load_balance_start_fair(void *arg) | |
1349 | { | |
1350 | struct cfs_rq *cfs_rq = arg; | |
1351 | ||
4a55bd5e | 1352 | return __load_balance_iterator(cfs_rq, cfs_rq->tasks.next); |
bf0f6f24 IM |
1353 | } |
1354 | ||
1355 | static struct task_struct *load_balance_next_fair(void *arg) | |
1356 | { | |
1357 | struct cfs_rq *cfs_rq = arg; | |
1358 | ||
4a55bd5e | 1359 | return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator); |
bf0f6f24 IM |
1360 | } |
1361 | ||
18d95a28 PZ |
1362 | static unsigned long |
1363 | __load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1364 | unsigned long max_load_move, struct sched_domain *sd, | |
1365 | enum cpu_idle_type idle, int *all_pinned, int *this_best_prio, | |
1366 | struct cfs_rq *cfs_rq) | |
62fb1851 | 1367 | { |
18d95a28 | 1368 | struct rq_iterator cfs_rq_iterator; |
62fb1851 | 1369 | |
18d95a28 PZ |
1370 | cfs_rq_iterator.start = load_balance_start_fair; |
1371 | cfs_rq_iterator.next = load_balance_next_fair; | |
1372 | cfs_rq_iterator.arg = cfs_rq; | |
62fb1851 | 1373 | |
18d95a28 PZ |
1374 | return balance_tasks(this_rq, this_cpu, busiest, |
1375 | max_load_move, sd, idle, all_pinned, | |
1376 | this_best_prio, &cfs_rq_iterator); | |
62fb1851 | 1377 | } |
62fb1851 | 1378 | |
18d95a28 | 1379 | #ifdef CONFIG_FAIR_GROUP_SCHED |
43010659 | 1380 | static unsigned long |
bf0f6f24 | 1381 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, |
e1d1484f | 1382 | unsigned long max_load_move, |
a4ac01c3 PW |
1383 | struct sched_domain *sd, enum cpu_idle_type idle, |
1384 | int *all_pinned, int *this_best_prio) | |
bf0f6f24 | 1385 | { |
bf0f6f24 | 1386 | long rem_load_move = max_load_move; |
18d95a28 PZ |
1387 | int busiest_cpu = cpu_of(busiest); |
1388 | struct task_group *tg; | |
bf0f6f24 | 1389 | |
18d95a28 PZ |
1390 | rcu_read_lock(); |
1391 | list_for_each_entry(tg, &task_groups, list) { | |
62fb1851 | 1392 | long imbalance; |
18d95a28 PZ |
1393 | unsigned long this_weight, busiest_weight; |
1394 | long rem_load, max_load, moved_load; | |
bf0f6f24 | 1395 | |
18d95a28 PZ |
1396 | /* |
1397 | * empty group | |
1398 | */ | |
1399 | if (!aggregate(tg, sd)->task_weight) | |
1400 | continue; | |
6b2d7700 | 1401 | |
18d95a28 PZ |
1402 | rem_load = rem_load_move * aggregate(tg, sd)->rq_weight; |
1403 | rem_load /= aggregate(tg, sd)->load + 1; | |
1404 | ||
1405 | this_weight = tg->cfs_rq[this_cpu]->task_weight; | |
1406 | busiest_weight = tg->cfs_rq[busiest_cpu]->task_weight; | |
1407 | ||
1408 | imbalance = (busiest_weight - this_weight) / 2; | |
1409 | ||
1410 | if (imbalance < 0) | |
1411 | imbalance = busiest_weight; | |
1412 | ||
1413 | max_load = max(rem_load, imbalance); | |
1414 | moved_load = __load_balance_fair(this_rq, this_cpu, busiest, | |
1415 | max_load, sd, idle, all_pinned, this_best_prio, | |
1416 | tg->cfs_rq[busiest_cpu]); | |
1417 | ||
1418 | if (!moved_load) | |
bf0f6f24 IM |
1419 | continue; |
1420 | ||
18d95a28 | 1421 | move_group_shares(tg, sd, busiest_cpu, this_cpu); |
bf0f6f24 | 1422 | |
18d95a28 PZ |
1423 | moved_load *= aggregate(tg, sd)->load; |
1424 | moved_load /= aggregate(tg, sd)->rq_weight + 1; | |
bf0f6f24 | 1425 | |
18d95a28 PZ |
1426 | rem_load_move -= moved_load; |
1427 | if (rem_load_move < 0) | |
bf0f6f24 IM |
1428 | break; |
1429 | } | |
18d95a28 | 1430 | rcu_read_unlock(); |
bf0f6f24 | 1431 | |
43010659 | 1432 | return max_load_move - rem_load_move; |
bf0f6f24 | 1433 | } |
18d95a28 PZ |
1434 | #else |
1435 | static unsigned long | |
1436 | load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1437 | unsigned long max_load_move, | |
1438 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1439 | int *all_pinned, int *this_best_prio) | |
1440 | { | |
1441 | return __load_balance_fair(this_rq, this_cpu, busiest, | |
1442 | max_load_move, sd, idle, all_pinned, | |
1443 | this_best_prio, &busiest->cfs); | |
1444 | } | |
1445 | #endif | |
bf0f6f24 | 1446 | |
e1d1484f PW |
1447 | static int |
1448 | move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1449 | struct sched_domain *sd, enum cpu_idle_type idle) | |
1450 | { | |
1451 | struct cfs_rq *busy_cfs_rq; | |
1452 | struct rq_iterator cfs_rq_iterator; | |
1453 | ||
1454 | cfs_rq_iterator.start = load_balance_start_fair; | |
1455 | cfs_rq_iterator.next = load_balance_next_fair; | |
1456 | ||
1457 | for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { | |
1458 | /* | |
1459 | * pass busy_cfs_rq argument into | |
1460 | * load_balance_[start|next]_fair iterators | |
1461 | */ | |
1462 | cfs_rq_iterator.arg = busy_cfs_rq; | |
1463 | if (iter_move_one_task(this_rq, this_cpu, busiest, sd, idle, | |
1464 | &cfs_rq_iterator)) | |
1465 | return 1; | |
1466 | } | |
1467 | ||
1468 | return 0; | |
1469 | } | |
681f3e68 | 1470 | #endif |
e1d1484f | 1471 | |
bf0f6f24 IM |
1472 | /* |
1473 | * scheduler tick hitting a task of our scheduling class: | |
1474 | */ | |
8f4d37ec | 1475 | static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued) |
bf0f6f24 IM |
1476 | { |
1477 | struct cfs_rq *cfs_rq; | |
1478 | struct sched_entity *se = &curr->se; | |
1479 | ||
1480 | for_each_sched_entity(se) { | |
1481 | cfs_rq = cfs_rq_of(se); | |
8f4d37ec | 1482 | entity_tick(cfs_rq, se, queued); |
bf0f6f24 IM |
1483 | } |
1484 | } | |
1485 | ||
8eb172d9 | 1486 | #define swap(a, b) do { typeof(a) tmp = (a); (a) = (b); (b) = tmp; } while (0) |
4d78e7b6 | 1487 | |
bf0f6f24 IM |
1488 | /* |
1489 | * Share the fairness runtime between parent and child, thus the | |
1490 | * total amount of pressure for CPU stays equal - new tasks | |
1491 | * get a chance to run but frequent forkers are not allowed to | |
1492 | * monopolize the CPU. Note: the parent runqueue is locked, | |
1493 | * the child is not running yet. | |
1494 | */ | |
ee0827d8 | 1495 | static void task_new_fair(struct rq *rq, struct task_struct *p) |
bf0f6f24 IM |
1496 | { |
1497 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
429d43bc | 1498 | struct sched_entity *se = &p->se, *curr = cfs_rq->curr; |
00bf7bfc | 1499 | int this_cpu = smp_processor_id(); |
bf0f6f24 IM |
1500 | |
1501 | sched_info_queued(p); | |
1502 | ||
7109c442 | 1503 | update_curr(cfs_rq); |
aeb73b04 | 1504 | place_entity(cfs_rq, se, 1); |
4d78e7b6 | 1505 | |
3c90e6e9 | 1506 | /* 'curr' will be NULL if the child belongs to a different group */ |
00bf7bfc | 1507 | if (sysctl_sched_child_runs_first && this_cpu == task_cpu(p) && |
3c90e6e9 | 1508 | curr && curr->vruntime < se->vruntime) { |
87fefa38 | 1509 | /* |
edcb60a3 IM |
1510 | * Upon rescheduling, sched_class::put_prev_task() will place |
1511 | * 'current' within the tree based on its new key value. | |
1512 | */ | |
4d78e7b6 | 1513 | swap(curr->vruntime, se->vruntime); |
4d78e7b6 | 1514 | } |
bf0f6f24 | 1515 | |
b9dca1e0 | 1516 | enqueue_task_fair(rq, p, 0); |
bb61c210 | 1517 | resched_task(rq->curr); |
bf0f6f24 IM |
1518 | } |
1519 | ||
cb469845 SR |
1520 | /* |
1521 | * Priority of the task has changed. Check to see if we preempt | |
1522 | * the current task. | |
1523 | */ | |
1524 | static void prio_changed_fair(struct rq *rq, struct task_struct *p, | |
1525 | int oldprio, int running) | |
1526 | { | |
1527 | /* | |
1528 | * Reschedule if we are currently running on this runqueue and | |
1529 | * our priority decreased, or if we are not currently running on | |
1530 | * this runqueue and our priority is higher than the current's | |
1531 | */ | |
1532 | if (running) { | |
1533 | if (p->prio > oldprio) | |
1534 | resched_task(rq->curr); | |
1535 | } else | |
1536 | check_preempt_curr(rq, p); | |
1537 | } | |
1538 | ||
1539 | /* | |
1540 | * We switched to the sched_fair class. | |
1541 | */ | |
1542 | static void switched_to_fair(struct rq *rq, struct task_struct *p, | |
1543 | int running) | |
1544 | { | |
1545 | /* | |
1546 | * We were most likely switched from sched_rt, so | |
1547 | * kick off the schedule if running, otherwise just see | |
1548 | * if we can still preempt the current task. | |
1549 | */ | |
1550 | if (running) | |
1551 | resched_task(rq->curr); | |
1552 | else | |
1553 | check_preempt_curr(rq, p); | |
1554 | } | |
1555 | ||
83b699ed SV |
1556 | /* Account for a task changing its policy or group. |
1557 | * | |
1558 | * This routine is mostly called to set cfs_rq->curr field when a task | |
1559 | * migrates between groups/classes. | |
1560 | */ | |
1561 | static void set_curr_task_fair(struct rq *rq) | |
1562 | { | |
1563 | struct sched_entity *se = &rq->curr->se; | |
1564 | ||
1565 | for_each_sched_entity(se) | |
1566 | set_next_entity(cfs_rq_of(se), se); | |
1567 | } | |
1568 | ||
810b3817 PZ |
1569 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1570 | static void moved_group_fair(struct task_struct *p) | |
1571 | { | |
1572 | struct cfs_rq *cfs_rq = task_cfs_rq(p); | |
1573 | ||
1574 | update_curr(cfs_rq); | |
1575 | place_entity(cfs_rq, &p->se, 1); | |
1576 | } | |
1577 | #endif | |
1578 | ||
bf0f6f24 IM |
1579 | /* |
1580 | * All the scheduling class methods: | |
1581 | */ | |
5522d5d5 IM |
1582 | static const struct sched_class fair_sched_class = { |
1583 | .next = &idle_sched_class, | |
bf0f6f24 IM |
1584 | .enqueue_task = enqueue_task_fair, |
1585 | .dequeue_task = dequeue_task_fair, | |
1586 | .yield_task = yield_task_fair, | |
e7693a36 GH |
1587 | #ifdef CONFIG_SMP |
1588 | .select_task_rq = select_task_rq_fair, | |
1589 | #endif /* CONFIG_SMP */ | |
bf0f6f24 | 1590 | |
2e09bf55 | 1591 | .check_preempt_curr = check_preempt_wakeup, |
bf0f6f24 IM |
1592 | |
1593 | .pick_next_task = pick_next_task_fair, | |
1594 | .put_prev_task = put_prev_task_fair, | |
1595 | ||
681f3e68 | 1596 | #ifdef CONFIG_SMP |
bf0f6f24 | 1597 | .load_balance = load_balance_fair, |
e1d1484f | 1598 | .move_one_task = move_one_task_fair, |
681f3e68 | 1599 | #endif |
bf0f6f24 | 1600 | |
83b699ed | 1601 | .set_curr_task = set_curr_task_fair, |
bf0f6f24 IM |
1602 | .task_tick = task_tick_fair, |
1603 | .task_new = task_new_fair, | |
cb469845 SR |
1604 | |
1605 | .prio_changed = prio_changed_fair, | |
1606 | .switched_to = switched_to_fair, | |
810b3817 PZ |
1607 | |
1608 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1609 | .moved_group = moved_group_fair, | |
1610 | #endif | |
bf0f6f24 IM |
1611 | }; |
1612 | ||
1613 | #ifdef CONFIG_SCHED_DEBUG | |
7ba2e74a PZ |
1614 | static void |
1615 | print_cfs_rq_tasks(struct seq_file *m, struct cfs_rq *cfs_rq, int depth) | |
1616 | { | |
1617 | struct sched_entity *se; | |
1618 | ||
1619 | if (!cfs_rq) | |
1620 | return; | |
1621 | ||
1622 | list_for_each_entry_rcu(se, &cfs_rq->tasks, group_node) { | |
1623 | int i; | |
1624 | ||
1625 | for (i = depth; i; i--) | |
1626 | seq_puts(m, " "); | |
1627 | ||
1628 | seq_printf(m, "%lu %s %lu\n", | |
1629 | se->load.weight, | |
1630 | entity_is_task(se) ? "T" : "G", | |
1631 | calc_delta_weight(SCHED_LOAD_SCALE, se) | |
1632 | ); | |
1633 | if (!entity_is_task(se)) | |
1634 | print_cfs_rq_tasks(m, group_cfs_rq(se), depth + 1); | |
1635 | } | |
1636 | } | |
1637 | ||
5cef9eca | 1638 | static void print_cfs_stats(struct seq_file *m, int cpu) |
bf0f6f24 | 1639 | { |
bf0f6f24 IM |
1640 | struct cfs_rq *cfs_rq; |
1641 | ||
5973e5b9 | 1642 | rcu_read_lock(); |
c3b64f1e | 1643 | for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq) |
5cef9eca | 1644 | print_cfs_rq(m, cpu, cfs_rq); |
7ba2e74a PZ |
1645 | |
1646 | seq_printf(m, "\nWeight tree:\n"); | |
1647 | print_cfs_rq_tasks(m, &cpu_rq(cpu)->cfs, 1); | |
5973e5b9 | 1648 | rcu_read_unlock(); |
bf0f6f24 IM |
1649 | } |
1650 | #endif |