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>
18 *
19 * Adaptive scheduling granularity, math enhancements by Peter Zijlstra
20 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
21 */
22
23 #include <linux/latencytop.h>
24 #include <linux/sched.h>
25 #include <linux/cpumask.h>
26 #include <linux/slab.h>
27 #include <linux/profile.h>
28 #include <linux/interrupt.h>
29
30 #include <trace/events/sched.h>
31
32 #include "sched.h"
33
34 /*
35 * Targeted preemption latency for CPU-bound tasks:
36 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
37 *
38 * NOTE: this latency value is not the same as the concept of
39 * 'timeslice length' - timeslices in CFS are of variable length
40 * and have no persistent notion like in traditional, time-slice
41 * based scheduling concepts.
42 *
43 * (to see the precise effective timeslice length of your workload,
44 * run vmstat and monitor the context-switches (cs) field)
45 */
46 unsigned int sysctl_sched_latency = 6000000ULL;
47 unsigned int normalized_sysctl_sched_latency = 6000000ULL;
48
49 /*
50 * The initial- and re-scaling of tunables is configurable
51 * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
52 *
53 * Options are:
54 * SCHED_TUNABLESCALING_NONE - unscaled, always *1
55 * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
56 * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
57 */
58 enum sched_tunable_scaling sysctl_sched_tunable_scaling
59 = SCHED_TUNABLESCALING_LOG;
60
61 /*
62 * Minimal preemption granularity for CPU-bound tasks:
63 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
64 */
65 unsigned int sysctl_sched_min_granularity = 750000ULL;
66 unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
67
68 /*
69 * is kept at sysctl_sched_latency / sysctl_sched_min_granularity
70 */
71 static unsigned int sched_nr_latency = 8;
72
73 /*
74 * After fork, child runs first. If set to 0 (default) then
75 * parent will (try to) run first.
76 */
77 unsigned int sysctl_sched_child_runs_first __read_mostly;
78
79 /*
80 * SCHED_OTHER wake-up granularity.
81 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
82 *
83 * This option delays the preemption effects of decoupled workloads
84 * and reduces their over-scheduling. Synchronous workloads will still
85 * have immediate wakeup/sleep latencies.
86 */
87 unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
88 unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL;
89
90 const_debug unsigned int sysctl_sched_migration_cost = 500000UL;
91
92 /*
93 * The exponential sliding window over which load is averaged for shares
94 * distribution.
95 * (default: 10msec)
96 */
97 unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;
98
99 #ifdef CONFIG_CFS_BANDWIDTH
100 /*
101 * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
102 * each time a cfs_rq requests quota.
103 *
104 * Note: in the case that the slice exceeds the runtime remaining (either due
105 * to consumption or the quota being specified to be smaller than the slice)
106 * we will always only issue the remaining available time.
107 *
108 * default: 5 msec, units: microseconds
109 */
110 unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL;
111 #endif
112
113 /*
114 * Increase the granularity value when there are more CPUs,
115 * because with more CPUs the 'effective latency' as visible
116 * to users decreases. But the relationship is not linear,
117 * so pick a second-best guess by going with the log2 of the
118 * number of CPUs.
119 *
120 * This idea comes from the SD scheduler of Con Kolivas:
121 */
get_update_sysctl_factor(void)122 static int get_update_sysctl_factor(void)
123 {
124 unsigned int cpus = min_t(int, num_online_cpus(), 8);
125 unsigned int factor;
126
127 switch (sysctl_sched_tunable_scaling) {
128 case SCHED_TUNABLESCALING_NONE:
129 factor = 1;
130 break;
131 case SCHED_TUNABLESCALING_LINEAR:
132 factor = cpus;
133 break;
134 case SCHED_TUNABLESCALING_LOG:
135 default:
136 factor = 1 + ilog2(cpus);
137 break;
138 }
139
140 return factor;
141 }
142
update_sysctl(void)143 static void update_sysctl(void)
144 {
145 unsigned int factor = get_update_sysctl_factor();
146
147 #define SET_SYSCTL(name) \
148 (sysctl_##name = (factor) * normalized_sysctl_##name)
149 SET_SYSCTL(sched_min_granularity);
150 SET_SYSCTL(sched_latency);
151 SET_SYSCTL(sched_wakeup_granularity);
152 #undef SET_SYSCTL
153 }
154
sched_init_granularity(void)155 void sched_init_granularity(void)
156 {
157 update_sysctl();
158 }
159
160 #if BITS_PER_LONG == 32
161 # define WMULT_CONST (~0UL)
162 #else
163 # define WMULT_CONST (1UL << 32)
164 #endif
165
166 #define WMULT_SHIFT 32
167
168 /*
169 * Shift right and round:
170 */
171 #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
172
173 /*
174 * delta *= weight / lw
175 */
176 static unsigned long
calc_delta_mine(unsigned long delta_exec,unsigned long weight,struct load_weight * lw)177 calc_delta_mine(unsigned long delta_exec, unsigned long weight,
178 struct load_weight *lw)
179 {
180 u64 tmp;
181
182 /*
183 * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched
184 * entities since MIN_SHARES = 2. Treat weight as 1 if less than
185 * 2^SCHED_LOAD_RESOLUTION.
186 */
187 if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION)))
188 tmp = (u64)delta_exec * scale_load_down(weight);
189 else
190 tmp = (u64)delta_exec;
191
192 if (!lw->inv_weight) {
193 unsigned long w = scale_load_down(lw->weight);
194
195 if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
196 lw->inv_weight = 1;
197 else if (unlikely(!w))
198 lw->inv_weight = WMULT_CONST;
199 else
200 lw->inv_weight = WMULT_CONST / w;
201 }
202
203 /*
204 * Check whether we'd overflow the 64-bit multiplication:
205 */
206 if (unlikely(tmp > WMULT_CONST))
207 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
208 WMULT_SHIFT/2);
209 else
210 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
211
212 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
213 }
214
215
216 const struct sched_class fair_sched_class;
217
218 /**************************************************************
219 * CFS operations on generic schedulable entities:
220 */
221
222 #ifdef CONFIG_FAIR_GROUP_SCHED
223
224 /* cpu runqueue to which this cfs_rq is attached */
rq_of(struct cfs_rq * cfs_rq)225 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
226 {
227 return cfs_rq->rq;
228 }
229
230 /* An entity is a task if it doesn't "own" a runqueue */
231 #define entity_is_task(se) (!se->my_q)
232
task_of(struct sched_entity * se)233 static inline struct task_struct *task_of(struct sched_entity *se)
234 {
235 #ifdef CONFIG_SCHED_DEBUG
236 WARN_ON_ONCE(!entity_is_task(se));
237 #endif
238 return container_of(se, struct task_struct, se);
239 }
240
241 /* Walk up scheduling entities hierarchy */
242 #define for_each_sched_entity(se) \
243 for (; se; se = se->parent)
244
task_cfs_rq(struct task_struct * p)245 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
246 {
247 return p->se.cfs_rq;
248 }
249
250 /* runqueue on which this entity is (to be) queued */
cfs_rq_of(struct sched_entity * se)251 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
252 {
253 return se->cfs_rq;
254 }
255
256 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)257 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
258 {
259 return grp->my_q;
260 }
261
list_add_leaf_cfs_rq(struct cfs_rq * cfs_rq)262 static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
263 {
264 if (!cfs_rq->on_list) {
265 /*
266 * Ensure we either appear before our parent (if already
267 * enqueued) or force our parent to appear after us when it is
268 * enqueued. The fact that we always enqueue bottom-up
269 * reduces this to two cases.
270 */
271 if (cfs_rq->tg->parent &&
272 cfs_rq->tg->parent->cfs_rq[cpu_of(rq_of(cfs_rq))]->on_list) {
273 list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
274 &rq_of(cfs_rq)->leaf_cfs_rq_list);
275 } else {
276 list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
277 &rq_of(cfs_rq)->leaf_cfs_rq_list);
278 }
279
280 cfs_rq->on_list = 1;
281 }
282 }
283
list_del_leaf_cfs_rq(struct cfs_rq * cfs_rq)284 static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
285 {
286 if (cfs_rq->on_list) {
287 list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
288 cfs_rq->on_list = 0;
289 }
290 }
291
292 /* Iterate thr' all leaf cfs_rq's on a runqueue */
293 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
294 list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)
295
296 /* Do the two (enqueued) entities belong to the same group ? */
297 static inline int
is_same_group(struct sched_entity * se,struct sched_entity * pse)298 is_same_group(struct sched_entity *se, struct sched_entity *pse)
299 {
300 if (se->cfs_rq == pse->cfs_rq)
301 return 1;
302
303 return 0;
304 }
305
parent_entity(struct sched_entity * se)306 static inline struct sched_entity *parent_entity(struct sched_entity *se)
307 {
308 return se->parent;
309 }
310
311 /* return depth at which a sched entity is present in the hierarchy */
depth_se(struct sched_entity * se)312 static inline int depth_se(struct sched_entity *se)
313 {
314 int depth = 0;
315
316 for_each_sched_entity(se)
317 depth++;
318
319 return depth;
320 }
321
322 static void
find_matching_se(struct sched_entity ** se,struct sched_entity ** pse)323 find_matching_se(struct sched_entity **se, struct sched_entity **pse)
324 {
325 int se_depth, pse_depth;
326
327 /*
328 * preemption test can be made between sibling entities who are in the
329 * same cfs_rq i.e who have a common parent. Walk up the hierarchy of
330 * both tasks until we find their ancestors who are siblings of common
331 * parent.
332 */
333
334 /* First walk up until both entities are at same depth */
335 se_depth = depth_se(*se);
336 pse_depth = depth_se(*pse);
337
338 while (se_depth > pse_depth) {
339 se_depth--;
340 *se = parent_entity(*se);
341 }
342
343 while (pse_depth > se_depth) {
344 pse_depth--;
345 *pse = parent_entity(*pse);
346 }
347
348 while (!is_same_group(*se, *pse)) {
349 *se = parent_entity(*se);
350 *pse = parent_entity(*pse);
351 }
352 }
353
354 #else /* !CONFIG_FAIR_GROUP_SCHED */
355
task_of(struct sched_entity * se)356 static inline struct task_struct *task_of(struct sched_entity *se)
357 {
358 return container_of(se, struct task_struct, se);
359 }
360
rq_of(struct cfs_rq * cfs_rq)361 static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
362 {
363 return container_of(cfs_rq, struct rq, cfs);
364 }
365
366 #define entity_is_task(se) 1
367
368 #define for_each_sched_entity(se) \
369 for (; se; se = NULL)
370
task_cfs_rq(struct task_struct * p)371 static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
372 {
373 return &task_rq(p)->cfs;
374 }
375
cfs_rq_of(struct sched_entity * se)376 static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
377 {
378 struct task_struct *p = task_of(se);
379 struct rq *rq = task_rq(p);
380
381 return &rq->cfs;
382 }
383
384 /* runqueue "owned" by this group */
group_cfs_rq(struct sched_entity * grp)385 static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
386 {
387 return NULL;
388 }
389
list_add_leaf_cfs_rq(struct cfs_rq * cfs_rq)390 static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
391 {
392 }
393
list_del_leaf_cfs_rq(struct cfs_rq * cfs_rq)394 static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
395 {
396 }
397
398 #define for_each_leaf_cfs_rq(rq, cfs_rq) \
399 for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)
400
401 static inline int
is_same_group(struct sched_entity * se,struct sched_entity * pse)402 is_same_group(struct sched_entity *se, struct sched_entity *pse)
403 {
404 return 1;
405 }
406
parent_entity(struct sched_entity * se)407 static inline struct sched_entity *parent_entity(struct sched_entity *se)
408 {
409 return NULL;
410 }
411
412 static inline void
find_matching_se(struct sched_entity ** se,struct sched_entity ** pse)413 find_matching_se(struct sched_entity **se, struct sched_entity **pse)
414 {
415 }
416
417 #endif /* CONFIG_FAIR_GROUP_SCHED */
418
419 static __always_inline
420 void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec);
421
422 /**************************************************************
423 * Scheduling class tree data structure manipulation methods:
424 */
425
max_vruntime(u64 min_vruntime,u64 vruntime)426 static inline u64 max_vruntime(u64 min_vruntime, u64 vruntime)
427 {
428 s64 delta = (s64)(vruntime - min_vruntime);
429 if (delta > 0)
430 min_vruntime = vruntime;
431
432 return min_vruntime;
433 }
434
min_vruntime(u64 min_vruntime,u64 vruntime)435 static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
436 {
437 s64 delta = (s64)(vruntime - min_vruntime);
438 if (delta < 0)
439 min_vruntime = vruntime;
440
441 return min_vruntime;
442 }
443
entity_before(struct sched_entity * a,struct sched_entity * b)444 static inline int entity_before(struct sched_entity *a,
445 struct sched_entity *b)
446 {
447 return (s64)(a->vruntime - b->vruntime) < 0;
448 }
449
update_min_vruntime(struct cfs_rq * cfs_rq)450 static void update_min_vruntime(struct cfs_rq *cfs_rq)
451 {
452 u64 vruntime = cfs_rq->min_vruntime;
453
454 if (cfs_rq->curr)
455 vruntime = cfs_rq->curr->vruntime;
456
457 if (cfs_rq->rb_leftmost) {
458 struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost,
459 struct sched_entity,
460 run_node);
461
462 if (!cfs_rq->curr)
463 vruntime = se->vruntime;
464 else
465 vruntime = min_vruntime(vruntime, se->vruntime);
466 }
467
468 cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
469 #ifndef CONFIG_64BIT
470 smp_wmb();
471 cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
472 #endif
473 }
474
475 /*
476 * Enqueue an entity into the rb-tree:
477 */
__enqueue_entity(struct cfs_rq * cfs_rq,struct sched_entity * se)478 static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
479 {
480 struct rb_node **link = &cfs_rq->tasks_timeline.rb_node;
481 struct rb_node *parent = NULL;
482 struct sched_entity *entry;
483 int leftmost = 1;
484
485 /*
486 * Find the right place in the rbtree:
487 */
488 while (*link) {
489 parent = *link;
490 entry = rb_entry(parent, struct sched_entity, run_node);
491 /*
492 * We dont care about collisions. Nodes with
493 * the same key stay together.
494 */
495 if (entity_before(se, entry)) {
496 link = &parent->rb_left;
497 } else {
498 link = &parent->rb_right;
499 leftmost = 0;
500 }
501 }
502
503 /*
504 * Maintain a cache of leftmost tree entries (it is frequently
505 * used):
506 */
507 if (leftmost)
508 cfs_rq->rb_leftmost = &se->run_node;
509
510 rb_link_node(&se->run_node, parent, link);
511 rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline);
512 }
513
__dequeue_entity(struct cfs_rq * cfs_rq,struct sched_entity * se)514 static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
515 {
516 if (cfs_rq->rb_leftmost == &se->run_node) {
517 struct rb_node *next_node;
518
519 next_node = rb_next(&se->run_node);
520 cfs_rq->rb_leftmost = next_node;
521 }
522
523 rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
524 }
525
__pick_first_entity(struct cfs_rq * cfs_rq)526 struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
527 {
528 struct rb_node *left = cfs_rq->rb_leftmost;
529
530 if (!left)
531 return NULL;
532
533 return rb_entry(left, struct sched_entity, run_node);
534 }
535
__pick_next_entity(struct sched_entity * se)536 static struct sched_entity *__pick_next_entity(struct sched_entity *se)
537 {
538 struct rb_node *next = rb_next(&se->run_node);
539
540 if (!next)
541 return NULL;
542
543 return rb_entry(next, struct sched_entity, run_node);
544 }
545
546 #ifdef CONFIG_SCHED_DEBUG
__pick_last_entity(struct cfs_rq * cfs_rq)547 struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
548 {
549 struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
550
551 if (!last)
552 return NULL;
553
554 return rb_entry(last, struct sched_entity, run_node);
555 }
556
557 /**************************************************************
558 * Scheduling class statistics methods:
559 */
560
sched_proc_update_handler(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)561 int sched_proc_update_handler(struct ctl_table *table, int write,
562 void __user *buffer, size_t *lenp,
563 loff_t *ppos)
564 {
565 int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
566 int factor = get_update_sysctl_factor();
567
568 if (ret || !write)
569 return ret;
570
571 sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
572 sysctl_sched_min_granularity);
573
574 #define WRT_SYSCTL(name) \
575 (normalized_sysctl_##name = sysctl_##name / (factor))
576 WRT_SYSCTL(sched_min_granularity);
577 WRT_SYSCTL(sched_latency);
578 WRT_SYSCTL(sched_wakeup_granularity);
579 #undef WRT_SYSCTL
580
581 return 0;
582 }
583 #endif
584
585 /*
586 * delta /= w
587 */
588 static inline unsigned long
calc_delta_fair(unsigned long delta,struct sched_entity * se)589 calc_delta_fair(unsigned long delta, struct sched_entity *se)
590 {
591 if (unlikely(se->load.weight != NICE_0_LOAD))
592 delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load);
593
594 return delta;
595 }
596
597 /*
598 * The idea is to set a period in which each task runs once.
599 *
600 * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch
601 * this period because otherwise the slices get too small.
602 *
603 * p = (nr <= nl) ? l : l*nr/nl
604 */
__sched_period(unsigned long nr_running)605 static u64 __sched_period(unsigned long nr_running)
606 {
607 u64 period = sysctl_sched_latency;
608 unsigned long nr_latency = sched_nr_latency;
609
610 if (unlikely(nr_running > nr_latency)) {
611 period = sysctl_sched_min_granularity;
612 period *= nr_running;
613 }
614
615 return period;
616 }
617
618 /*
619 * We calculate the wall-time slice from the period by taking a part
620 * proportional to the weight.
621 *
622 * s = p*P[w/rw]
623 */
sched_slice(struct cfs_rq * cfs_rq,struct sched_entity * se)624 static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
625 {
626 u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
627
628 for_each_sched_entity(se) {
629 struct load_weight *load;
630 struct load_weight lw;
631
632 cfs_rq = cfs_rq_of(se);
633 load = &cfs_rq->load;
634
635 if (unlikely(!se->on_rq)) {
636 lw = cfs_rq->load;
637
638 update_load_add(&lw, se->load.weight);
639 load = &lw;
640 }
641 slice = calc_delta_mine(slice, se->load.weight, load);
642 }
643 return slice;
644 }
645
646 /*
647 * We calculate the vruntime slice of a to be inserted task
648 *
649 * vs = s/w
650 */
sched_vslice(struct cfs_rq * cfs_rq,struct sched_entity * se)651 static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
652 {
653 return calc_delta_fair(sched_slice(cfs_rq, se), se);
654 }
655
656 static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update);
657 static void update_cfs_shares(struct cfs_rq *cfs_rq);
658
659 /*
660 * Update the current task's runtime statistics. Skip current tasks that
661 * are not in our scheduling class.
662 */
663 static inline void
__update_curr(struct cfs_rq * cfs_rq,struct sched_entity * curr,unsigned long delta_exec)664 __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
665 unsigned long delta_exec)
666 {
667 unsigned long delta_exec_weighted;
668
669 schedstat_set(curr->statistics.exec_max,
670 max((u64)delta_exec, curr->statistics.exec_max));
671
672 curr->sum_exec_runtime += delta_exec;
673 schedstat_add(cfs_rq, exec_clock, delta_exec);
674 delta_exec_weighted = calc_delta_fair(delta_exec, curr);
675
676 curr->vruntime += delta_exec_weighted;
677 update_min_vruntime(cfs_rq);
678
679 #if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED
680 cfs_rq->load_unacc_exec_time += delta_exec;
681 #endif
682 }
683
update_curr(struct cfs_rq * cfs_rq)684 static void update_curr(struct cfs_rq *cfs_rq)
685 {
686 struct sched_entity *curr = cfs_rq->curr;
687 u64 now = rq_of(cfs_rq)->clock_task;
688 unsigned long delta_exec;
689
690 if (unlikely(!curr))
691 return;
692
693 /*
694 * Get the amount of time the current task was running
695 * since the last time we changed load (this cannot
696 * overflow on 32 bits):
697 */
698 delta_exec = (unsigned long)(now - curr->exec_start);
699 if (!delta_exec)
700 return;
701
702 __update_curr(cfs_rq, curr, delta_exec);
703 curr->exec_start = now;
704
705 if (entity_is_task(curr)) {
706 struct task_struct *curtask = task_of(curr);
707
708 trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
709 cpuacct_charge(curtask, delta_exec);
710 account_group_exec_runtime(curtask, delta_exec);
711 }
712
713 account_cfs_rq_runtime(cfs_rq, delta_exec);
714 }
715
716 static inline void
update_stats_wait_start(struct cfs_rq * cfs_rq,struct sched_entity * se)717 update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
718 {
719 schedstat_set(se->statistics.wait_start, rq_of(cfs_rq)->clock);
720 }
721
722 /*
723 * Task is being enqueued - update stats:
724 */
update_stats_enqueue(struct cfs_rq * cfs_rq,struct sched_entity * se)725 static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
726 {
727 /*
728 * Are we enqueueing a waiting task? (for current tasks
729 * a dequeue/enqueue event is a NOP)
730 */
731 if (se != cfs_rq->curr)
732 update_stats_wait_start(cfs_rq, se);
733 }
734
735 static void
update_stats_wait_end(struct cfs_rq * cfs_rq,struct sched_entity * se)736 update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
737 {
738 schedstat_set(se->statistics.wait_max, max(se->statistics.wait_max,
739 rq_of(cfs_rq)->clock - se->statistics.wait_start));
740 schedstat_set(se->statistics.wait_count, se->statistics.wait_count + 1);
741 schedstat_set(se->statistics.wait_sum, se->statistics.wait_sum +
742 rq_of(cfs_rq)->clock - se->statistics.wait_start);
743 #ifdef CONFIG_SCHEDSTATS
744 if (entity_is_task(se)) {
745 trace_sched_stat_wait(task_of(se),
746 rq_of(cfs_rq)->clock - se->statistics.wait_start);
747 }
748 #endif
749 schedstat_set(se->statistics.wait_start, 0);
750 }
751
752 static inline void
update_stats_dequeue(struct cfs_rq * cfs_rq,struct sched_entity * se)753 update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
754 {
755 /*
756 * Mark the end of the wait period if dequeueing a
757 * waiting task:
758 */
759 if (se != cfs_rq->curr)
760 update_stats_wait_end(cfs_rq, se);
761 }
762
763 /*
764 * We are picking a new current task - update its stats:
765 */
766 static inline void
update_stats_curr_start(struct cfs_rq * cfs_rq,struct sched_entity * se)767 update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
768 {
769 /*
770 * We are starting a new run period:
771 */
772 se->exec_start = rq_of(cfs_rq)->clock_task;
773 }
774
775 /**************************************************
776 * Scheduling class queueing methods:
777 */
778
779 static void
account_entity_enqueue(struct cfs_rq * cfs_rq,struct sched_entity * se)780 account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
781 {
782 update_load_add(&cfs_rq->load, se->load.weight);
783 if (!parent_entity(se))
784 update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
785 #ifdef CONFIG_SMP
786 if (entity_is_task(se))
787 list_add(&se->group_node, &rq_of(cfs_rq)->cfs_tasks);
788 #endif
789 cfs_rq->nr_running++;
790 }
791
792 static void
account_entity_dequeue(struct cfs_rq * cfs_rq,struct sched_entity * se)793 account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
794 {
795 update_load_sub(&cfs_rq->load, se->load.weight);
796 if (!parent_entity(se))
797 update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
798 if (entity_is_task(se))
799 list_del_init(&se->group_node);
800 cfs_rq->nr_running--;
801 }
802
803 #ifdef CONFIG_FAIR_GROUP_SCHED
804 /* we need this in update_cfs_load and load-balance functions below */
805 static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);
806 # ifdef CONFIG_SMP
update_cfs_rq_load_contribution(struct cfs_rq * cfs_rq,int global_update)807 static void update_cfs_rq_load_contribution(struct cfs_rq *cfs_rq,
808 int global_update)
809 {
810 struct task_group *tg = cfs_rq->tg;
811 long load_avg;
812
813 load_avg = div64_u64(cfs_rq->load_avg, cfs_rq->load_period+1);
814 load_avg -= cfs_rq->load_contribution;
815
816 if (global_update || abs(load_avg) > cfs_rq->load_contribution / 8) {
817 atomic_add(load_avg, &tg->load_weight);
818 cfs_rq->load_contribution += load_avg;
819 }
820 }
821
update_cfs_load(struct cfs_rq * cfs_rq,int global_update)822 static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
823 {
824 u64 period = sysctl_sched_shares_window;
825 u64 now, delta;
826 unsigned long load = cfs_rq->load.weight;
827
828 if (cfs_rq->tg == &root_task_group || throttled_hierarchy(cfs_rq))
829 return;
830
831 now = rq_of(cfs_rq)->clock_task;
832 delta = now - cfs_rq->load_stamp;
833
834 /* truncate load history at 4 idle periods */
835 if (cfs_rq->load_stamp > cfs_rq->load_last &&
836 now - cfs_rq->load_last > 4 * period) {
837 cfs_rq->load_period = 0;
838 cfs_rq->load_avg = 0;
839 delta = period - 1;
840 }
841
842 cfs_rq->load_stamp = now;
843 cfs_rq->load_unacc_exec_time = 0;
844 cfs_rq->load_period += delta;
845 if (load) {
846 cfs_rq->load_last = now;
847 cfs_rq->load_avg += delta * load;
848 }
849
850 /* consider updating load contribution on each fold or truncate */
851 if (global_update || cfs_rq->load_period > period
852 || !cfs_rq->load_period)
853 update_cfs_rq_load_contribution(cfs_rq, global_update);
854
855 while (cfs_rq->load_period > period) {
856 /*
857 * Inline assembly required to prevent the compiler
858 * optimising this loop into a divmod call.
859 * See __iter_div_u64_rem() for another example of this.
860 */
861 asm("" : "+rm" (cfs_rq->load_period));
862 cfs_rq->load_period /= 2;
863 cfs_rq->load_avg /= 2;
864 }
865
866 if (!cfs_rq->curr && !cfs_rq->nr_running && !cfs_rq->load_avg)
867 list_del_leaf_cfs_rq(cfs_rq);
868 }
869
calc_tg_weight(struct task_group * tg,struct cfs_rq * cfs_rq)870 static inline long calc_tg_weight(struct task_group *tg, struct cfs_rq *cfs_rq)
871 {
872 long tg_weight;
873
874 /*
875 * Use this CPU's actual weight instead of the last load_contribution
876 * to gain a more accurate current total weight. See
877 * update_cfs_rq_load_contribution().
878 */
879 tg_weight = atomic_read(&tg->load_weight);
880 tg_weight -= cfs_rq->load_contribution;
881 tg_weight += cfs_rq->load.weight;
882
883 return tg_weight;
884 }
885
calc_cfs_shares(struct cfs_rq * cfs_rq,struct task_group * tg)886 static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
887 {
888 long tg_weight, load, shares;
889
890 tg_weight = calc_tg_weight(tg, cfs_rq);
891 load = cfs_rq->load.weight;
892
893 shares = (tg->shares * load);
894 if (tg_weight)
895 shares /= tg_weight;
896
897 if (shares < MIN_SHARES)
898 shares = MIN_SHARES;
899 if (shares > tg->shares)
900 shares = tg->shares;
901
902 return shares;
903 }
904
update_entity_shares_tick(struct cfs_rq * cfs_rq)905 static void update_entity_shares_tick(struct cfs_rq *cfs_rq)
906 {
907 if (cfs_rq->load_unacc_exec_time > sysctl_sched_shares_window) {
908 update_cfs_load(cfs_rq, 0);
909 update_cfs_shares(cfs_rq);
910 }
911 }
912 # else /* CONFIG_SMP */
update_cfs_load(struct cfs_rq * cfs_rq,int global_update)913 static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
914 {
915 }
916
calc_cfs_shares(struct cfs_rq * cfs_rq,struct task_group * tg)917 static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
918 {
919 return tg->shares;
920 }
921
update_entity_shares_tick(struct cfs_rq * cfs_rq)922 static inline void update_entity_shares_tick(struct cfs_rq *cfs_rq)
923 {
924 }
925 # endif /* CONFIG_SMP */
reweight_entity(struct cfs_rq * cfs_rq,struct sched_entity * se,unsigned long weight)926 static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
927 unsigned long weight)
928 {
929 if (se->on_rq) {
930 /* commit outstanding execution time */
931 if (cfs_rq->curr == se)
932 update_curr(cfs_rq);
933 account_entity_dequeue(cfs_rq, se);
934 }
935
936 update_load_set(&se->load, weight);
937
938 if (se->on_rq)
939 account_entity_enqueue(cfs_rq, se);
940 }
941
update_cfs_shares(struct cfs_rq * cfs_rq)942 static void update_cfs_shares(struct cfs_rq *cfs_rq)
943 {
944 struct task_group *tg;
945 struct sched_entity *se;
946 long shares;
947
948 tg = cfs_rq->tg;
949 se = tg->se[cpu_of(rq_of(cfs_rq))];
950 if (!se || throttled_hierarchy(cfs_rq))
951 return;
952 #ifndef CONFIG_SMP
953 if (likely(se->load.weight == tg->shares))
954 return;
955 #endif
956 shares = calc_cfs_shares(cfs_rq, tg);
957
958 reweight_entity(cfs_rq_of(se), se, shares);
959 }
960 #else /* CONFIG_FAIR_GROUP_SCHED */
update_cfs_load(struct cfs_rq * cfs_rq,int global_update)961 static void update_cfs_load(struct cfs_rq *cfs_rq, int global_update)
962 {
963 }
964
update_cfs_shares(struct cfs_rq * cfs_rq)965 static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
966 {
967 }
968
update_entity_shares_tick(struct cfs_rq * cfs_rq)969 static inline void update_entity_shares_tick(struct cfs_rq *cfs_rq)
970 {
971 }
972 #endif /* CONFIG_FAIR_GROUP_SCHED */
973
enqueue_sleeper(struct cfs_rq * cfs_rq,struct sched_entity * se)974 static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
975 {
976 #ifdef CONFIG_SCHEDSTATS
977 struct task_struct *tsk = NULL;
978
979 if (entity_is_task(se))
980 tsk = task_of(se);
981
982 if (se->statistics.sleep_start) {
983 u64 delta = rq_of(cfs_rq)->clock - se->statistics.sleep_start;
984
985 if ((s64)delta < 0)
986 delta = 0;
987
988 if (unlikely(delta > se->statistics.sleep_max))
989 se->statistics.sleep_max = delta;
990
991 se->statistics.sleep_start = 0;
992 se->statistics.sum_sleep_runtime += delta;
993
994 if (tsk) {
995 account_scheduler_latency(tsk, delta >> 10, 1);
996 trace_sched_stat_sleep(tsk, delta);
997 }
998 }
999 if (se->statistics.block_start) {
1000 u64 delta = rq_of(cfs_rq)->clock - se->statistics.block_start;
1001
1002 if ((s64)delta < 0)
1003 delta = 0;
1004
1005 if (unlikely(delta > se->statistics.block_max))
1006 se->statistics.block_max = delta;
1007
1008 se->statistics.block_start = 0;
1009 se->statistics.sum_sleep_runtime += delta;
1010
1011 if (tsk) {
1012 if (tsk->in_iowait) {
1013 se->statistics.iowait_sum += delta;
1014 se->statistics.iowait_count++;
1015 trace_sched_stat_iowait(tsk, delta);
1016 }
1017
1018 trace_sched_stat_blocked(tsk, delta);
1019
1020 /*
1021 * Blocking time is in units of nanosecs, so shift by
1022 * 20 to get a milliseconds-range estimation of the
1023 * amount of time that the task spent sleeping:
1024 */
1025 if (unlikely(prof_on == SLEEP_PROFILING)) {
1026 profile_hits(SLEEP_PROFILING,
1027 (void *)get_wchan(tsk),
1028 delta >> 20);
1029 }
1030 account_scheduler_latency(tsk, delta >> 10, 0);
1031 }
1032 }
1033 #endif
1034 }
1035
check_spread(struct cfs_rq * cfs_rq,struct sched_entity * se)1036 static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se)
1037 {
1038 #ifdef CONFIG_SCHED_DEBUG
1039 s64 d = se->vruntime - cfs_rq->min_vruntime;
1040
1041 if (d < 0)
1042 d = -d;
1043
1044 if (d > 3*sysctl_sched_latency)
1045 schedstat_inc(cfs_rq, nr_spread_over);
1046 #endif
1047 }
1048
1049 static void
place_entity(struct cfs_rq * cfs_rq,struct sched_entity * se,int initial)1050 place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
1051 {
1052 u64 vruntime = cfs_rq->min_vruntime;
1053
1054 /*
1055 * The 'current' period is already promised to the current tasks,
1056 * however the extra weight of the new task will slow them down a
1057 * little, place the new task so that it fits in the slot that
1058 * stays open at the end.
1059 */
1060 if (initial && sched_feat(START_DEBIT))
1061 vruntime += sched_vslice(cfs_rq, se);
1062
1063 /* sleeps up to a single latency don't count. */
1064 if (!initial) {
1065 unsigned long thresh = sysctl_sched_latency;
1066
1067 /*
1068 * Halve their sleep time's effect, to allow
1069 * for a gentler effect of sleepers:
1070 */
1071 if (sched_feat(GENTLE_FAIR_SLEEPERS))
1072 thresh >>= 1;
1073
1074 vruntime -= thresh;
1075 }
1076
1077 /* ensure we never gain time by being placed backwards. */
1078 vruntime = max_vruntime(se->vruntime, vruntime);
1079
1080 se->vruntime = vruntime;
1081 }
1082
1083 static void check_enqueue_throttle(struct cfs_rq *cfs_rq);
1084
1085 static void
enqueue_entity(struct cfs_rq * cfs_rq,struct sched_entity * se,int flags)1086 enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
1087 {
1088 /*
1089 * Update the normalized vruntime before updating min_vruntime
1090 * through callig update_curr().
1091 */
1092 if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
1093 se->vruntime += cfs_rq->min_vruntime;
1094
1095 /*
1096 * Update run-time statistics of the 'current'.
1097 */
1098 update_curr(cfs_rq);
1099 update_cfs_load(cfs_rq, 0);
1100 account_entity_enqueue(cfs_rq, se);
1101 update_cfs_shares(cfs_rq);
1102
1103 if (flags & ENQUEUE_WAKEUP) {
1104 place_entity(cfs_rq, se, 0);
1105 enqueue_sleeper(cfs_rq, se);
1106 }
1107
1108 update_stats_enqueue(cfs_rq, se);
1109 check_spread(cfs_rq, se);
1110 if (se != cfs_rq->curr)
1111 __enqueue_entity(cfs_rq, se);
1112 se->on_rq = 1;
1113
1114 if (cfs_rq->nr_running == 1) {
1115 list_add_leaf_cfs_rq(cfs_rq);
1116 check_enqueue_throttle(cfs_rq);
1117 }
1118 }
1119
__clear_buddies_last(struct sched_entity * se)1120 static void __clear_buddies_last(struct sched_entity *se)
1121 {
1122 for_each_sched_entity(se) {
1123 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1124 if (cfs_rq->last == se)
1125 cfs_rq->last = NULL;
1126 else
1127 break;
1128 }
1129 }
1130
__clear_buddies_next(struct sched_entity * se)1131 static void __clear_buddies_next(struct sched_entity *se)
1132 {
1133 for_each_sched_entity(se) {
1134 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1135 if (cfs_rq->next == se)
1136 cfs_rq->next = NULL;
1137 else
1138 break;
1139 }
1140 }
1141
__clear_buddies_skip(struct sched_entity * se)1142 static void __clear_buddies_skip(struct sched_entity *se)
1143 {
1144 for_each_sched_entity(se) {
1145 struct cfs_rq *cfs_rq = cfs_rq_of(se);
1146 if (cfs_rq->skip == se)
1147 cfs_rq->skip = NULL;
1148 else
1149 break;
1150 }
1151 }
1152
clear_buddies(struct cfs_rq * cfs_rq,struct sched_entity * se)1153 static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
1154 {
1155 if (cfs_rq->last == se)
1156 __clear_buddies_last(se);
1157
1158 if (cfs_rq->next == se)
1159 __clear_buddies_next(se);
1160
1161 if (cfs_rq->skip == se)
1162 __clear_buddies_skip(se);
1163 }
1164
1165 static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
1166
1167 static void
dequeue_entity(struct cfs_rq * cfs_rq,struct sched_entity * se,int flags)1168 dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
1169 {
1170 /*
1171 * Update run-time statistics of the 'current'.
1172 */
1173 update_curr(cfs_rq);
1174
1175 update_stats_dequeue(cfs_rq, se);
1176 if (flags & DEQUEUE_SLEEP) {
1177 #ifdef CONFIG_SCHEDSTATS
1178 if (entity_is_task(se)) {
1179 struct task_struct *tsk = task_of(se);
1180
1181 if (tsk->state & TASK_INTERRUPTIBLE)
1182 se->statistics.sleep_start = rq_of(cfs_rq)->clock;
1183 if (tsk->state & TASK_UNINTERRUPTIBLE)
1184 se->statistics.block_start = rq_of(cfs_rq)->clock;
1185 }
1186 #endif
1187 }
1188
1189 clear_buddies(cfs_rq, se);
1190
1191 if (se != cfs_rq->curr)
1192 __dequeue_entity(cfs_rq, se);
1193 se->on_rq = 0;
1194 update_cfs_load(cfs_rq, 0);
1195 account_entity_dequeue(cfs_rq, se);
1196
1197 /*
1198 * Normalize the entity after updating the min_vruntime because the
1199 * update can refer to the ->curr item and we need to reflect this
1200 * movement in our normalized position.
1201 */
1202 if (!(flags & DEQUEUE_SLEEP))
1203 se->vruntime -= cfs_rq->min_vruntime;
1204
1205 /* return excess runtime on last dequeue */
1206 return_cfs_rq_runtime(cfs_rq);
1207
1208 update_min_vruntime(cfs_rq);
1209 update_cfs_shares(cfs_rq);
1210 }
1211
1212 /*
1213 * Preempt the current task with a newly woken task if needed:
1214 */
1215 static void
check_preempt_tick(struct cfs_rq * cfs_rq,struct sched_entity * curr)1216 check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
1217 {
1218 unsigned long ideal_runtime, delta_exec;
1219 struct sched_entity *se;
1220 s64 delta;
1221
1222 ideal_runtime = sched_slice(cfs_rq, curr);
1223 delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
1224 if (delta_exec > ideal_runtime) {
1225 resched_task(rq_of(cfs_rq)->curr);
1226 /*
1227 * The current task ran long enough, ensure it doesn't get
1228 * re-elected due to buddy favours.
1229 */
1230 clear_buddies(cfs_rq, curr);
1231 return;
1232 }
1233
1234 /*
1235 * Ensure that a task that missed wakeup preemption by a
1236 * narrow margin doesn't have to wait for a full slice.
1237 * This also mitigates buddy induced latencies under load.
1238 */
1239 if (delta_exec < sysctl_sched_min_granularity)
1240 return;
1241
1242 se = __pick_first_entity(cfs_rq);
1243 delta = curr->vruntime - se->vruntime;
1244
1245 if (delta < 0)
1246 return;
1247
1248 if (delta > ideal_runtime)
1249 resched_task(rq_of(cfs_rq)->curr);
1250 }
1251
1252 static void
set_next_entity(struct cfs_rq * cfs_rq,struct sched_entity * se)1253 set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
1254 {
1255 /* 'current' is not kept within the tree. */
1256 if (se->on_rq) {
1257 /*
1258 * Any task has to be enqueued before it get to execute on
1259 * a CPU. So account for the time it spent waiting on the
1260 * runqueue.
1261 */
1262 update_stats_wait_end(cfs_rq, se);
1263 __dequeue_entity(cfs_rq, se);
1264 }
1265
1266 update_stats_curr_start(cfs_rq, se);
1267 cfs_rq->curr = se;
1268 #ifdef CONFIG_SCHEDSTATS
1269 /*
1270 * Track our maximum slice length, if the CPU's load is at
1271 * least twice that of our own weight (i.e. dont track it
1272 * when there are only lesser-weight tasks around):
1273 */
1274 if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
1275 se->statistics.slice_max = max(se->statistics.slice_max,
1276 se->sum_exec_runtime - se->prev_sum_exec_runtime);
1277 }
1278 #endif
1279 se->prev_sum_exec_runtime = se->sum_exec_runtime;
1280 }
1281
1282 static int
1283 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);
1284
1285 /*
1286 * Pick the next process, keeping these things in mind, in this order:
1287 * 1) keep things fair between processes/task groups
1288 * 2) pick the "next" process, since someone really wants that to run
1289 * 3) pick the "last" process, for cache locality
1290 * 4) do not run the "skip" process, if something else is available
1291 */
pick_next_entity(struct cfs_rq * cfs_rq)1292 static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
1293 {
1294 struct sched_entity *se = __pick_first_entity(cfs_rq);
1295 struct sched_entity *left = se;
1296
1297 /*
1298 * Avoid running the skip buddy, if running something else can
1299 * be done without getting too unfair.
1300 */
1301 if (cfs_rq->skip == se) {
1302 struct sched_entity *second = __pick_next_entity(se);
1303 if (second && wakeup_preempt_entity(second, left) < 1)
1304 se = second;
1305 }
1306
1307 /*
1308 * Prefer last buddy, try to return the CPU to a preempted task.
1309 */
1310 if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, left) < 1)
1311 se = cfs_rq->last;
1312
1313 /*
1314 * Someone really wants this to run. If it's not unfair, run it.
1315 */
1316 if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, left) < 1)
1317 se = cfs_rq->next;
1318
1319 clear_buddies(cfs_rq, se);
1320
1321 return se;
1322 }
1323
1324 static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
1325
put_prev_entity(struct cfs_rq * cfs_rq,struct sched_entity * prev)1326 static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
1327 {
1328 /*
1329 * If still on the runqueue then deactivate_task()
1330 * was not called and update_curr() has to be done:
1331 */
1332 if (prev->on_rq)
1333 update_curr(cfs_rq);
1334
1335 /* throttle cfs_rqs exceeding runtime */
1336 check_cfs_rq_runtime(cfs_rq);
1337
1338 check_spread(cfs_rq, prev);
1339 if (prev->on_rq) {
1340 update_stats_wait_start(cfs_rq, prev);
1341 /* Put 'current' back into the tree. */
1342 __enqueue_entity(cfs_rq, prev);
1343 }
1344 cfs_rq->curr = NULL;
1345 }
1346
1347 static void
entity_tick(struct cfs_rq * cfs_rq,struct sched_entity * curr,int queued)1348 entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
1349 {
1350 /*
1351 * Update run-time statistics of the 'current'.
1352 */
1353 update_curr(cfs_rq);
1354
1355 /*
1356 * Update share accounting for long-running entities.
1357 */
1358 update_entity_shares_tick(cfs_rq);
1359
1360 #ifdef CONFIG_SCHED_HRTICK
1361 /*
1362 * queued ticks are scheduled to match the slice, so don't bother
1363 * validating it and just reschedule.
1364 */
1365 if (queued) {
1366 resched_task(rq_of(cfs_rq)->curr);
1367 return;
1368 }
1369 /*
1370 * don't let the period tick interfere with the hrtick preemption
1371 */
1372 if (!sched_feat(DOUBLE_TICK) &&
1373 hrtimer_active(&rq_of(cfs_rq)->hrtick_timer))
1374 return;
1375 #endif
1376
1377 if (cfs_rq->nr_running > 1)
1378 check_preempt_tick(cfs_rq, curr);
1379 }
1380
1381
1382 /**************************************************
1383 * CFS bandwidth control machinery
1384 */
1385
1386 #ifdef CONFIG_CFS_BANDWIDTH
1387
1388 #ifdef HAVE_JUMP_LABEL
1389 static struct static_key __cfs_bandwidth_used;
1390
cfs_bandwidth_used(void)1391 static inline bool cfs_bandwidth_used(void)
1392 {
1393 return static_key_false(&__cfs_bandwidth_used);
1394 }
1395
cfs_bandwidth_usage_inc(void)1396 void cfs_bandwidth_usage_inc(void)
1397 {
1398 static_key_slow_inc(&__cfs_bandwidth_used);
1399 }
1400
cfs_bandwidth_usage_dec(void)1401 void cfs_bandwidth_usage_dec(void)
1402 {
1403 static_key_slow_dec(&__cfs_bandwidth_used);
1404 }
1405 #else /* HAVE_JUMP_LABEL */
cfs_bandwidth_used(void)1406 static bool cfs_bandwidth_used(void)
1407 {
1408 return true;
1409 }
1410
cfs_bandwidth_usage_inc(void)1411 void cfs_bandwidth_usage_inc(void) {}
cfs_bandwidth_usage_dec(void)1412 void cfs_bandwidth_usage_dec(void) {}
1413 #endif /* HAVE_JUMP_LABEL */
1414
1415 /*
1416 * default period for cfs group bandwidth.
1417 * default: 0.1s, units: nanoseconds
1418 */
default_cfs_period(void)1419 static inline u64 default_cfs_period(void)
1420 {
1421 return 100000000ULL;
1422 }
1423
sched_cfs_bandwidth_slice(void)1424 static inline u64 sched_cfs_bandwidth_slice(void)
1425 {
1426 return (u64)sysctl_sched_cfs_bandwidth_slice * NSEC_PER_USEC;
1427 }
1428
1429 /*
1430 * Replenish runtime according to assigned quota and update expiration time.
1431 * We use sched_clock_cpu directly instead of rq->clock to avoid adding
1432 * additional synchronization around rq->lock.
1433 *
1434 * requires cfs_b->lock
1435 */
__refill_cfs_bandwidth_runtime(struct cfs_bandwidth * cfs_b)1436 void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
1437 {
1438 u64 now;
1439
1440 if (cfs_b->quota == RUNTIME_INF)
1441 return;
1442
1443 now = sched_clock_cpu(smp_processor_id());
1444 cfs_b->runtime = cfs_b->quota;
1445 cfs_b->runtime_expires = now + ktime_to_ns(cfs_b->period);
1446 }
1447
tg_cfs_bandwidth(struct task_group * tg)1448 static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
1449 {
1450 return &tg->cfs_bandwidth;
1451 }
1452
1453 /* returns 0 on failure to allocate runtime */
assign_cfs_rq_runtime(struct cfs_rq * cfs_rq)1454 static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
1455 {
1456 struct task_group *tg = cfs_rq->tg;
1457 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
1458 u64 amount = 0, min_amount, expires;
1459
1460 /* note: this is a positive sum as runtime_remaining <= 0 */
1461 min_amount = sched_cfs_bandwidth_slice() - cfs_rq->runtime_remaining;
1462
1463 raw_spin_lock(&cfs_b->lock);
1464 if (cfs_b->quota == RUNTIME_INF)
1465 amount = min_amount;
1466 else {
1467 /*
1468 * If the bandwidth pool has become inactive, then at least one
1469 * period must have elapsed since the last consumption.
1470 * Refresh the global state and ensure bandwidth timer becomes
1471 * active.
1472 */
1473 if (!cfs_b->timer_active) {
1474 __refill_cfs_bandwidth_runtime(cfs_b);
1475 __start_cfs_bandwidth(cfs_b);
1476 }
1477
1478 if (cfs_b->runtime > 0) {
1479 amount = min(cfs_b->runtime, min_amount);
1480 cfs_b->runtime -= amount;
1481 cfs_b->idle = 0;
1482 }
1483 }
1484 expires = cfs_b->runtime_expires;
1485 raw_spin_unlock(&cfs_b->lock);
1486
1487 cfs_rq->runtime_remaining += amount;
1488 /*
1489 * we may have advanced our local expiration to account for allowed
1490 * spread between our sched_clock and the one on which runtime was
1491 * issued.
1492 */
1493 if ((s64)(expires - cfs_rq->runtime_expires) > 0)
1494 cfs_rq->runtime_expires = expires;
1495
1496 return cfs_rq->runtime_remaining > 0;
1497 }
1498
1499 /*
1500 * Note: This depends on the synchronization provided by sched_clock and the
1501 * fact that rq->clock snapshots this value.
1502 */
expire_cfs_rq_runtime(struct cfs_rq * cfs_rq)1503 static void expire_cfs_rq_runtime(struct cfs_rq *cfs_rq)
1504 {
1505 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
1506 struct rq *rq = rq_of(cfs_rq);
1507
1508 /* if the deadline is ahead of our clock, nothing to do */
1509 if (likely((s64)(rq->clock - cfs_rq->runtime_expires) < 0))
1510 return;
1511
1512 if (cfs_rq->runtime_remaining < 0)
1513 return;
1514
1515 /*
1516 * If the local deadline has passed we have to consider the
1517 * possibility that our sched_clock is 'fast' and the global deadline
1518 * has not truly expired.
1519 *
1520 * Fortunately we can check determine whether this the case by checking
1521 * whether the global deadline has advanced.
1522 */
1523
1524 if ((s64)(cfs_rq->runtime_expires - cfs_b->runtime_expires) >= 0) {
1525 /* extend local deadline, drift is bounded above by 2 ticks */
1526 cfs_rq->runtime_expires += TICK_NSEC;
1527 } else {
1528 /* global deadline is ahead, expiration has passed */
1529 cfs_rq->runtime_remaining = 0;
1530 }
1531 }
1532
__account_cfs_rq_runtime(struct cfs_rq * cfs_rq,unsigned long delta_exec)1533 static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
1534 unsigned long delta_exec)
1535 {
1536 /* dock delta_exec before expiring quota (as it could span periods) */
1537 cfs_rq->runtime_remaining -= delta_exec;
1538 expire_cfs_rq_runtime(cfs_rq);
1539
1540 if (likely(cfs_rq->runtime_remaining > 0))
1541 return;
1542
1543 /*
1544 * if we're unable to extend our runtime we resched so that the active
1545 * hierarchy can be throttled
1546 */
1547 if (!assign_cfs_rq_runtime(cfs_rq) && likely(cfs_rq->curr))
1548 resched_task(rq_of(cfs_rq)->curr);
1549 }
1550
1551 static __always_inline
account_cfs_rq_runtime(struct cfs_rq * cfs_rq,unsigned long delta_exec)1552 void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec)
1553 {
1554 if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
1555 return;
1556
1557 __account_cfs_rq_runtime(cfs_rq, delta_exec);
1558 }
1559
cfs_rq_throttled(struct cfs_rq * cfs_rq)1560 static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
1561 {
1562 return cfs_bandwidth_used() && cfs_rq->throttled;
1563 }
1564
1565 /* check whether cfs_rq, or any parent, is throttled */
throttled_hierarchy(struct cfs_rq * cfs_rq)1566 static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
1567 {
1568 return cfs_bandwidth_used() && cfs_rq->throttle_count;
1569 }
1570
1571 /*
1572 * Ensure that neither of the group entities corresponding to src_cpu or
1573 * dest_cpu are members of a throttled hierarchy when performing group
1574 * load-balance operations.
1575 */
throttled_lb_pair(struct task_group * tg,int src_cpu,int dest_cpu)1576 static inline int throttled_lb_pair(struct task_group *tg,
1577 int src_cpu, int dest_cpu)
1578 {
1579 struct cfs_rq *src_cfs_rq, *dest_cfs_rq;
1580
1581 src_cfs_rq = tg->cfs_rq[src_cpu];
1582 dest_cfs_rq = tg->cfs_rq[dest_cpu];
1583
1584 return throttled_hierarchy(src_cfs_rq) ||
1585 throttled_hierarchy(dest_cfs_rq);
1586 }
1587
1588 /* updated child weight may affect parent so we have to do this bottom up */
tg_unthrottle_up(struct task_group * tg,void * data)1589 static int tg_unthrottle_up(struct task_group *tg, void *data)
1590 {
1591 struct rq *rq = data;
1592 struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
1593
1594 cfs_rq->throttle_count--;
1595 #ifdef CONFIG_SMP
1596 if (!cfs_rq->throttle_count) {
1597 u64 delta = rq->clock_task - cfs_rq->load_stamp;
1598
1599 /* leaving throttled state, advance shares averaging windows */
1600 cfs_rq->load_stamp += delta;
1601 cfs_rq->load_last += delta;
1602
1603 /* update entity weight now that we are on_rq again */
1604 update_cfs_shares(cfs_rq);
1605 }
1606 #endif
1607
1608 return 0;
1609 }
1610
tg_throttle_down(struct task_group * tg,void * data)1611 static int tg_throttle_down(struct task_group *tg, void *data)
1612 {
1613 struct rq *rq = data;
1614 struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
1615
1616 /* group is entering throttled state, record last load */
1617 if (!cfs_rq->throttle_count)
1618 update_cfs_load(cfs_rq, 0);
1619 cfs_rq->throttle_count++;
1620
1621 return 0;
1622 }
1623
throttle_cfs_rq(struct cfs_rq * cfs_rq)1624 static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
1625 {
1626 struct rq *rq = rq_of(cfs_rq);
1627 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
1628 struct sched_entity *se;
1629 long task_delta, dequeue = 1;
1630
1631 se = cfs_rq->tg->se[cpu_of(rq_of(cfs_rq))];
1632
1633 /* account load preceding throttle */
1634 rcu_read_lock();
1635 walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
1636 rcu_read_unlock();
1637
1638 task_delta = cfs_rq->h_nr_running;
1639 for_each_sched_entity(se) {
1640 struct cfs_rq *qcfs_rq = cfs_rq_of(se);
1641 /* throttled entity or throttle-on-deactivate */
1642 if (!se->on_rq)
1643 break;
1644
1645 if (dequeue)
1646 dequeue_entity(qcfs_rq, se, DEQUEUE_SLEEP);
1647 qcfs_rq->h_nr_running -= task_delta;
1648
1649 if (qcfs_rq->load.weight)
1650 dequeue = 0;
1651 }
1652
1653 if (!se)
1654 rq->nr_running -= task_delta;
1655
1656 cfs_rq->throttled = 1;
1657 cfs_rq->throttled_timestamp = rq->clock;
1658 raw_spin_lock(&cfs_b->lock);
1659 list_add_tail_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
1660 if (!cfs_b->timer_active)
1661 __start_cfs_bandwidth(cfs_b);
1662 raw_spin_unlock(&cfs_b->lock);
1663 }
1664
unthrottle_cfs_rq(struct cfs_rq * cfs_rq)1665 void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
1666 {
1667 struct rq *rq = rq_of(cfs_rq);
1668 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
1669 struct sched_entity *se;
1670 int enqueue = 1;
1671 long task_delta;
1672
1673 se = cfs_rq->tg->se[cpu_of(rq_of(cfs_rq))];
1674
1675 cfs_rq->throttled = 0;
1676 raw_spin_lock(&cfs_b->lock);
1677 cfs_b->throttled_time += rq->clock - cfs_rq->throttled_timestamp;
1678 list_del_rcu(&cfs_rq->throttled_list);
1679 raw_spin_unlock(&cfs_b->lock);
1680 cfs_rq->throttled_timestamp = 0;
1681
1682 update_rq_clock(rq);
1683 /* update hierarchical throttle state */
1684 walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);
1685
1686 if (!cfs_rq->load.weight)
1687 return;
1688
1689 task_delta = cfs_rq->h_nr_running;
1690 for_each_sched_entity(se) {
1691 if (se->on_rq)
1692 enqueue = 0;
1693
1694 cfs_rq = cfs_rq_of(se);
1695 if (enqueue)
1696 enqueue_entity(cfs_rq, se, ENQUEUE_WAKEUP);
1697 cfs_rq->h_nr_running += task_delta;
1698
1699 if (cfs_rq_throttled(cfs_rq))
1700 break;
1701 }
1702
1703 if (!se)
1704 rq->nr_running += task_delta;
1705
1706 /* determine whether we need to wake up potentially idle cpu */
1707 if (rq->curr == rq->idle && rq->cfs.nr_running)
1708 resched_task(rq->curr);
1709 }
1710
distribute_cfs_runtime(struct cfs_bandwidth * cfs_b,u64 remaining,u64 expires)1711 static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
1712 u64 remaining, u64 expires)
1713 {
1714 struct cfs_rq *cfs_rq;
1715 u64 runtime = remaining;
1716
1717 rcu_read_lock();
1718 list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
1719 throttled_list) {
1720 struct rq *rq = rq_of(cfs_rq);
1721
1722 raw_spin_lock(&rq->lock);
1723 if (!cfs_rq_throttled(cfs_rq))
1724 goto next;
1725
1726 runtime = -cfs_rq->runtime_remaining + 1;
1727 if (runtime > remaining)
1728 runtime = remaining;
1729 remaining -= runtime;
1730
1731 cfs_rq->runtime_remaining += runtime;
1732 cfs_rq->runtime_expires = expires;
1733
1734 /* we check whether we're throttled above */
1735 if (cfs_rq->runtime_remaining > 0)
1736 unthrottle_cfs_rq(cfs_rq);
1737
1738 next:
1739 raw_spin_unlock(&rq->lock);
1740
1741 if (!remaining)
1742 break;
1743 }
1744 rcu_read_unlock();
1745
1746 return remaining;
1747 }
1748
1749 /*
1750 * Responsible for refilling a task_group's bandwidth and unthrottling its
1751 * cfs_rqs as appropriate. If there has been no activity within the last
1752 * period the timer is deactivated until scheduling resumes; cfs_b->idle is
1753 * used to track this state.
1754 */
do_sched_cfs_period_timer(struct cfs_bandwidth * cfs_b,int overrun)1755 static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun)
1756 {
1757 u64 runtime, runtime_expires;
1758 int idle = 1, throttled;
1759
1760 raw_spin_lock(&cfs_b->lock);
1761 /* no need to continue the timer with no bandwidth constraint */
1762 if (cfs_b->quota == RUNTIME_INF)
1763 goto out_unlock;
1764
1765 throttled = !list_empty(&cfs_b->throttled_cfs_rq);
1766 /* idle depends on !throttled (for the case of a large deficit) */
1767 idle = cfs_b->idle && !throttled;
1768 cfs_b->nr_periods += overrun;
1769
1770 /* if we're going inactive then everything else can be deferred */
1771 if (idle)
1772 goto out_unlock;
1773
1774 /*
1775 * if we have relooped after returning idle once, we need to update our
1776 * status as actually running, so that other cpus doing
1777 * __start_cfs_bandwidth will stop trying to cancel us.
1778 */
1779 cfs_b->timer_active = 1;
1780
1781 __refill_cfs_bandwidth_runtime(cfs_b);
1782
1783 if (!throttled) {
1784 /* mark as potentially idle for the upcoming period */
1785 cfs_b->idle = 1;
1786 goto out_unlock;
1787 }
1788
1789 /* account preceding periods in which throttling occurred */
1790 cfs_b->nr_throttled += overrun;
1791
1792 /*
1793 * There are throttled entities so we must first use the new bandwidth
1794 * to unthrottle them before making it generally available. This
1795 * ensures that all existing debts will be paid before a new cfs_rq is
1796 * allowed to run.
1797 */
1798 runtime = cfs_b->runtime;
1799 runtime_expires = cfs_b->runtime_expires;
1800 cfs_b->runtime = 0;
1801
1802 /*
1803 * This check is repeated as we are holding onto the new bandwidth
1804 * while we unthrottle. This can potentially race with an unthrottled
1805 * group trying to acquire new bandwidth from the global pool.
1806 */
1807 while (throttled && runtime > 0) {
1808 raw_spin_unlock(&cfs_b->lock);
1809 /* we can't nest cfs_b->lock while distributing bandwidth */
1810 runtime = distribute_cfs_runtime(cfs_b, runtime,
1811 runtime_expires);
1812 raw_spin_lock(&cfs_b->lock);
1813
1814 throttled = !list_empty(&cfs_b->throttled_cfs_rq);
1815 }
1816
1817 /* return (any) remaining runtime */
1818 cfs_b->runtime = runtime;
1819 /*
1820 * While we are ensured activity in the period following an
1821 * unthrottle, this also covers the case in which the new bandwidth is
1822 * insufficient to cover the existing bandwidth deficit. (Forcing the
1823 * timer to remain active while there are any throttled entities.)
1824 */
1825 cfs_b->idle = 0;
1826 out_unlock:
1827 if (idle)
1828 cfs_b->timer_active = 0;
1829 raw_spin_unlock(&cfs_b->lock);
1830
1831 return idle;
1832 }
1833
1834 /* a cfs_rq won't donate quota below this amount */
1835 static const u64 min_cfs_rq_runtime = 1 * NSEC_PER_MSEC;
1836 /* minimum remaining period time to redistribute slack quota */
1837 static const u64 min_bandwidth_expiration = 2 * NSEC_PER_MSEC;
1838 /* how long we wait to gather additional slack before distributing */
1839 static const u64 cfs_bandwidth_slack_period = 5 * NSEC_PER_MSEC;
1840
1841 /*
1842 * Are we near the end of the current quota period?
1843 *
1844 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
1845 * hrtimer base being cleared by __hrtimer_start_range_ns. In the case of
1846 * migrate_hrtimers, base is never cleared, so we are fine.
1847 */
runtime_refresh_within(struct cfs_bandwidth * cfs_b,u64 min_expire)1848 static int runtime_refresh_within(struct cfs_bandwidth *cfs_b, u64 min_expire)
1849 {
1850 struct hrtimer *refresh_timer = &cfs_b->period_timer;
1851 u64 remaining;
1852
1853 /* if the call-back is running a quota refresh is already occurring */
1854 if (hrtimer_callback_running(refresh_timer))
1855 return 1;
1856
1857 /* is a quota refresh about to occur? */
1858 remaining = ktime_to_ns(hrtimer_expires_remaining(refresh_timer));
1859 if (remaining < min_expire)
1860 return 1;
1861
1862 return 0;
1863 }
1864
start_cfs_slack_bandwidth(struct cfs_bandwidth * cfs_b)1865 static void start_cfs_slack_bandwidth(struct cfs_bandwidth *cfs_b)
1866 {
1867 u64 min_left = cfs_bandwidth_slack_period + min_bandwidth_expiration;
1868
1869 /* if there's a quota refresh soon don't bother with slack */
1870 if (runtime_refresh_within(cfs_b, min_left))
1871 return;
1872
1873 start_bandwidth_timer(&cfs_b->slack_timer,
1874 ns_to_ktime(cfs_bandwidth_slack_period));
1875 }
1876
1877 /* we know any runtime found here is valid as update_curr() precedes return */
__return_cfs_rq_runtime(struct cfs_rq * cfs_rq)1878 static void __return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
1879 {
1880 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
1881 s64 slack_runtime = cfs_rq->runtime_remaining - min_cfs_rq_runtime;
1882
1883 if (slack_runtime <= 0)
1884 return;
1885
1886 raw_spin_lock(&cfs_b->lock);
1887 if (cfs_b->quota != RUNTIME_INF &&
1888 cfs_rq->runtime_expires == cfs_b->runtime_expires) {
1889 cfs_b->runtime += slack_runtime;
1890
1891 /* we are under rq->lock, defer unthrottling using a timer */
1892 if (cfs_b->runtime > sched_cfs_bandwidth_slice() &&
1893 !list_empty(&cfs_b->throttled_cfs_rq))
1894 start_cfs_slack_bandwidth(cfs_b);
1895 }
1896 raw_spin_unlock(&cfs_b->lock);
1897
1898 /* even if it's not valid for return we don't want to try again */
1899 cfs_rq->runtime_remaining -= slack_runtime;
1900 }
1901
return_cfs_rq_runtime(struct cfs_rq * cfs_rq)1902 static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
1903 {
1904 if (!cfs_bandwidth_used())
1905 return;
1906
1907 if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
1908 return;
1909
1910 __return_cfs_rq_runtime(cfs_rq);
1911 }
1912
1913 /*
1914 * This is done with a timer (instead of inline with bandwidth return) since
1915 * it's necessary to juggle rq->locks to unthrottle their respective cfs_rqs.
1916 */
do_sched_cfs_slack_timer(struct cfs_bandwidth * cfs_b)1917 static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b)
1918 {
1919 u64 runtime = 0, slice = sched_cfs_bandwidth_slice();
1920 u64 expires;
1921
1922 /* confirm we're still not at a refresh boundary */
1923 raw_spin_lock(&cfs_b->lock);
1924 if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
1925 raw_spin_unlock(&cfs_b->lock);
1926 return;
1927 }
1928
1929 if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice) {
1930 runtime = cfs_b->runtime;
1931 cfs_b->runtime = 0;
1932 }
1933 expires = cfs_b->runtime_expires;
1934 raw_spin_unlock(&cfs_b->lock);
1935
1936 if (!runtime)
1937 return;
1938
1939 runtime = distribute_cfs_runtime(cfs_b, runtime, expires);
1940
1941 raw_spin_lock(&cfs_b->lock);
1942 if (expires == cfs_b->runtime_expires)
1943 cfs_b->runtime = runtime;
1944 raw_spin_unlock(&cfs_b->lock);
1945 }
1946
1947 /*
1948 * When a group wakes up we want to make sure that its quota is not already
1949 * expired/exceeded, otherwise it may be allowed to steal additional ticks of
1950 * runtime as update_curr() throttling can not not trigger until it's on-rq.
1951 */
check_enqueue_throttle(struct cfs_rq * cfs_rq)1952 static void check_enqueue_throttle(struct cfs_rq *cfs_rq)
1953 {
1954 if (!cfs_bandwidth_used())
1955 return;
1956
1957 /* an active group must be handled by the update_curr()->put() path */
1958 if (!cfs_rq->runtime_enabled || cfs_rq->curr)
1959 return;
1960
1961 /* ensure the group is not already throttled */
1962 if (cfs_rq_throttled(cfs_rq))
1963 return;
1964
1965 /* update runtime allocation */
1966 account_cfs_rq_runtime(cfs_rq, 0);
1967 if (cfs_rq->runtime_remaining <= 0)
1968 throttle_cfs_rq(cfs_rq);
1969 }
1970
1971 /* conditionally throttle active cfs_rq's from put_prev_entity() */
check_cfs_rq_runtime(struct cfs_rq * cfs_rq)1972 static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
1973 {
1974 if (!cfs_bandwidth_used())
1975 return;
1976
1977 if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
1978 return;
1979
1980 /*
1981 * it's possible for a throttled entity to be forced into a running
1982 * state (e.g. set_curr_task), in this case we're finished.
1983 */
1984 if (cfs_rq_throttled(cfs_rq))
1985 return;
1986
1987 throttle_cfs_rq(cfs_rq);
1988 }
1989
1990 static inline u64 default_cfs_period(void);
1991 static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun);
1992 static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b);
1993
sched_cfs_slack_timer(struct hrtimer * timer)1994 static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
1995 {
1996 struct cfs_bandwidth *cfs_b =
1997 container_of(timer, struct cfs_bandwidth, slack_timer);
1998 do_sched_cfs_slack_timer(cfs_b);
1999
2000 return HRTIMER_NORESTART;
2001 }
2002
sched_cfs_period_timer(struct hrtimer * timer)2003 static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
2004 {
2005 struct cfs_bandwidth *cfs_b =
2006 container_of(timer, struct cfs_bandwidth, period_timer);
2007 ktime_t now;
2008 int overrun;
2009 int idle = 0;
2010
2011 for (;;) {
2012 now = hrtimer_cb_get_time(timer);
2013 overrun = hrtimer_forward(timer, now, cfs_b->period);
2014
2015 if (!overrun)
2016 break;
2017
2018 idle = do_sched_cfs_period_timer(cfs_b, overrun);
2019 }
2020
2021 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
2022 }
2023
init_cfs_bandwidth(struct cfs_bandwidth * cfs_b)2024 void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
2025 {
2026 raw_spin_lock_init(&cfs_b->lock);
2027 cfs_b->runtime = 0;
2028 cfs_b->quota = RUNTIME_INF;
2029 cfs_b->period = ns_to_ktime(default_cfs_period());
2030
2031 INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
2032 hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2033 cfs_b->period_timer.function = sched_cfs_period_timer;
2034 hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2035 cfs_b->slack_timer.function = sched_cfs_slack_timer;
2036 }
2037
init_cfs_rq_runtime(struct cfs_rq * cfs_rq)2038 static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2039 {
2040 cfs_rq->runtime_enabled = 0;
2041 INIT_LIST_HEAD(&cfs_rq->throttled_list);
2042 }
2043
2044 /* requires cfs_b->lock, may release to reprogram timer */
__start_cfs_bandwidth(struct cfs_bandwidth * cfs_b)2045 void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
2046 {
2047 /*
2048 * The timer may be active because we're trying to set a new bandwidth
2049 * period or because we're racing with the tear-down path
2050 * (timer_active==0 becomes visible before the hrtimer call-back
2051 * terminates). In either case we ensure that it's re-programmed
2052 */
2053 while (unlikely(hrtimer_active(&cfs_b->period_timer)) &&
2054 hrtimer_try_to_cancel(&cfs_b->period_timer) < 0) {
2055 /* bounce the lock to allow do_sched_cfs_period_timer to run */
2056 raw_spin_unlock(&cfs_b->lock);
2057 cpu_relax();
2058 raw_spin_lock(&cfs_b->lock);
2059 /* if someone else restarted the timer then we're done */
2060 if (cfs_b->timer_active)
2061 return;
2062 }
2063
2064 cfs_b->timer_active = 1;
2065 start_bandwidth_timer(&cfs_b->period_timer, cfs_b->period);
2066 }
2067
destroy_cfs_bandwidth(struct cfs_bandwidth * cfs_b)2068 static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
2069 {
2070 hrtimer_cancel(&cfs_b->period_timer);
2071 hrtimer_cancel(&cfs_b->slack_timer);
2072 }
2073
unthrottle_offline_cfs_rqs(struct rq * rq)2074 static void unthrottle_offline_cfs_rqs(struct rq *rq)
2075 {
2076 struct cfs_rq *cfs_rq;
2077
2078 for_each_leaf_cfs_rq(rq, cfs_rq) {
2079 struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
2080
2081 if (!cfs_rq->runtime_enabled)
2082 continue;
2083
2084 /*
2085 * clock_task is not advancing so we just need to make sure
2086 * there's some valid quota amount
2087 */
2088 cfs_rq->runtime_remaining = cfs_b->quota;
2089 if (cfs_rq_throttled(cfs_rq))
2090 unthrottle_cfs_rq(cfs_rq);
2091 }
2092 }
2093
2094 #else /* CONFIG_CFS_BANDWIDTH */
2095 static __always_inline
account_cfs_rq_runtime(struct cfs_rq * cfs_rq,unsigned long delta_exec)2096 void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec) {}
check_cfs_rq_runtime(struct cfs_rq * cfs_rq)2097 static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
check_enqueue_throttle(struct cfs_rq * cfs_rq)2098 static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
return_cfs_rq_runtime(struct cfs_rq * cfs_rq)2099 static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
2100
cfs_rq_throttled(struct cfs_rq * cfs_rq)2101 static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
2102 {
2103 return 0;
2104 }
2105
throttled_hierarchy(struct cfs_rq * cfs_rq)2106 static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
2107 {
2108 return 0;
2109 }
2110
throttled_lb_pair(struct task_group * tg,int src_cpu,int dest_cpu)2111 static inline int throttled_lb_pair(struct task_group *tg,
2112 int src_cpu, int dest_cpu)
2113 {
2114 return 0;
2115 }
2116
init_cfs_bandwidth(struct cfs_bandwidth * cfs_b)2117 void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}
2118
2119 #ifdef CONFIG_FAIR_GROUP_SCHED
init_cfs_rq_runtime(struct cfs_rq * cfs_rq)2120 static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
2121 #endif
2122
tg_cfs_bandwidth(struct task_group * tg)2123 static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
2124 {
2125 return NULL;
2126 }
destroy_cfs_bandwidth(struct cfs_bandwidth * cfs_b)2127 static inline void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}
unthrottle_offline_cfs_rqs(struct rq * rq)2128 static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
2129
2130 #endif /* CONFIG_CFS_BANDWIDTH */
2131
2132 /**************************************************
2133 * CFS operations on tasks:
2134 */
2135
2136 #ifdef CONFIG_SCHED_HRTICK
hrtick_start_fair(struct rq * rq,struct task_struct * p)2137 static void hrtick_start_fair(struct rq *rq, struct task_struct *p)
2138 {
2139 struct sched_entity *se = &p->se;
2140 struct cfs_rq *cfs_rq = cfs_rq_of(se);
2141
2142 WARN_ON(task_rq(p) != rq);
2143
2144 if (cfs_rq->nr_running > 1) {
2145 u64 slice = sched_slice(cfs_rq, se);
2146 u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime;
2147 s64 delta = slice - ran;
2148
2149 if (delta < 0) {
2150 if (rq->curr == p)
2151 resched_task(p);
2152 return;
2153 }
2154
2155 /*
2156 * Don't schedule slices shorter than 10000ns, that just
2157 * doesn't make sense. Rely on vruntime for fairness.
2158 */
2159 if (rq->curr != p)
2160 delta = max_t(s64, 10000LL, delta);
2161
2162 hrtick_start(rq, delta);
2163 }
2164 }
2165
2166 /*
2167 * called from enqueue/dequeue and updates the hrtick when the
2168 * current task is from our class and nr_running is low enough
2169 * to matter.
2170 */
hrtick_update(struct rq * rq)2171 static void hrtick_update(struct rq *rq)
2172 {
2173 struct task_struct *curr = rq->curr;
2174
2175 if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
2176 return;
2177
2178 if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
2179 hrtick_start_fair(rq, curr);
2180 }
2181 #else /* !CONFIG_SCHED_HRTICK */
2182 static inline void
hrtick_start_fair(struct rq * rq,struct task_struct * p)2183 hrtick_start_fair(struct rq *rq, struct task_struct *p)
2184 {
2185 }
2186
hrtick_update(struct rq * rq)2187 static inline void hrtick_update(struct rq *rq)
2188 {
2189 }
2190 #endif
2191
2192 /*
2193 * The enqueue_task method is called before nr_running is
2194 * increased. Here we update the fair scheduling stats and
2195 * then put the task into the rbtree:
2196 */
2197 static void
enqueue_task_fair(struct rq * rq,struct task_struct * p,int flags)2198 enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
2199 {
2200 struct cfs_rq *cfs_rq;
2201 struct sched_entity *se = &p->se;
2202
2203 for_each_sched_entity(se) {
2204 if (se->on_rq)
2205 break;
2206 cfs_rq = cfs_rq_of(se);
2207 enqueue_entity(cfs_rq, se, flags);
2208
2209 /*
2210 * end evaluation on encountering a throttled cfs_rq
2211 *
2212 * note: in the case of encountering a throttled cfs_rq we will
2213 * post the final h_nr_running increment below.
2214 */
2215 if (cfs_rq_throttled(cfs_rq))
2216 break;
2217 cfs_rq->h_nr_running++;
2218
2219 flags = ENQUEUE_WAKEUP;
2220 }
2221
2222 for_each_sched_entity(se) {
2223 cfs_rq = cfs_rq_of(se);
2224 cfs_rq->h_nr_running++;
2225
2226 if (cfs_rq_throttled(cfs_rq))
2227 break;
2228
2229 update_cfs_load(cfs_rq, 0);
2230 update_cfs_shares(cfs_rq);
2231 }
2232
2233 if (!se)
2234 inc_nr_running(rq);
2235 hrtick_update(rq);
2236 }
2237
2238 static void set_next_buddy(struct sched_entity *se);
2239
2240 /*
2241 * The dequeue_task method is called before nr_running is
2242 * decreased. We remove the task from the rbtree and
2243 * update the fair scheduling stats:
2244 */
dequeue_task_fair(struct rq * rq,struct task_struct * p,int flags)2245 static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
2246 {
2247 struct cfs_rq *cfs_rq;
2248 struct sched_entity *se = &p->se;
2249 int task_sleep = flags & DEQUEUE_SLEEP;
2250
2251 for_each_sched_entity(se) {
2252 cfs_rq = cfs_rq_of(se);
2253 dequeue_entity(cfs_rq, se, flags);
2254
2255 /*
2256 * end evaluation on encountering a throttled cfs_rq
2257 *
2258 * note: in the case of encountering a throttled cfs_rq we will
2259 * post the final h_nr_running decrement below.
2260 */
2261 if (cfs_rq_throttled(cfs_rq))
2262 break;
2263 cfs_rq->h_nr_running--;
2264
2265 /* Don't dequeue parent if it has other entities besides us */
2266 if (cfs_rq->load.weight) {
2267 /*
2268 * Bias pick_next to pick a task from this cfs_rq, as
2269 * p is sleeping when it is within its sched_slice.
2270 */
2271 if (task_sleep && parent_entity(se))
2272 set_next_buddy(parent_entity(se));
2273
2274 /* avoid re-evaluating load for this entity */
2275 se = parent_entity(se);
2276 break;
2277 }
2278 flags |= DEQUEUE_SLEEP;
2279 }
2280
2281 for_each_sched_entity(se) {
2282 cfs_rq = cfs_rq_of(se);
2283 cfs_rq->h_nr_running--;
2284
2285 if (cfs_rq_throttled(cfs_rq))
2286 break;
2287
2288 update_cfs_load(cfs_rq, 0);
2289 update_cfs_shares(cfs_rq);
2290 }
2291
2292 if (!se)
2293 dec_nr_running(rq);
2294 hrtick_update(rq);
2295 }
2296
2297 #ifdef CONFIG_SMP
2298 /* Used instead of source_load when we know the type == 0 */
weighted_cpuload(const int cpu)2299 static unsigned long weighted_cpuload(const int cpu)
2300 {
2301 return cpu_rq(cpu)->load.weight;
2302 }
2303
2304 /*
2305 * Return a low guess at the load of a migration-source cpu weighted
2306 * according to the scheduling class and "nice" value.
2307 *
2308 * We want to under-estimate the load of migration sources, to
2309 * balance conservatively.
2310 */
source_load(int cpu,int type)2311 static unsigned long source_load(int cpu, int type)
2312 {
2313 struct rq *rq = cpu_rq(cpu);
2314 unsigned long total = weighted_cpuload(cpu);
2315
2316 if (type == 0 || !sched_feat(LB_BIAS))
2317 return total;
2318
2319 return min(rq->cpu_load[type-1], total);
2320 }
2321
2322 /*
2323 * Return a high guess at the load of a migration-target cpu weighted
2324 * according to the scheduling class and "nice" value.
2325 */
target_load(int cpu,int type)2326 static unsigned long target_load(int cpu, int type)
2327 {
2328 struct rq *rq = cpu_rq(cpu);
2329 unsigned long total = weighted_cpuload(cpu);
2330
2331 if (type == 0 || !sched_feat(LB_BIAS))
2332 return total;
2333
2334 return max(rq->cpu_load[type-1], total);
2335 }
2336
power_of(int cpu)2337 static unsigned long power_of(int cpu)
2338 {
2339 return cpu_rq(cpu)->cpu_power;
2340 }
2341
cpu_avg_load_per_task(int cpu)2342 static unsigned long cpu_avg_load_per_task(int cpu)
2343 {
2344 struct rq *rq = cpu_rq(cpu);
2345 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
2346
2347 if (nr_running)
2348 return rq->load.weight / nr_running;
2349
2350 return 0;
2351 }
2352
2353
task_waking_fair(struct task_struct * p)2354 static void task_waking_fair(struct task_struct *p)
2355 {
2356 struct sched_entity *se = &p->se;
2357 struct cfs_rq *cfs_rq = cfs_rq_of(se);
2358 u64 min_vruntime;
2359
2360 #ifndef CONFIG_64BIT
2361 u64 min_vruntime_copy;
2362
2363 do {
2364 min_vruntime_copy = cfs_rq->min_vruntime_copy;
2365 smp_rmb();
2366 min_vruntime = cfs_rq->min_vruntime;
2367 } while (min_vruntime != min_vruntime_copy);
2368 #else
2369 min_vruntime = cfs_rq->min_vruntime;
2370 #endif
2371
2372 se->vruntime -= min_vruntime;
2373 }
2374
2375 #ifdef CONFIG_FAIR_GROUP_SCHED
2376 /*
2377 * effective_load() calculates the load change as seen from the root_task_group
2378 *
2379 * Adding load to a group doesn't make a group heavier, but can cause movement
2380 * of group shares between cpus. Assuming the shares were perfectly aligned one
2381 * can calculate the shift in shares.
2382 *
2383 * Calculate the effective load difference if @wl is added (subtracted) to @tg
2384 * on this @cpu and results in a total addition (subtraction) of @wg to the
2385 * total group weight.
2386 *
2387 * Given a runqueue weight distribution (rw_i) we can compute a shares
2388 * distribution (s_i) using:
2389 *
2390 * s_i = rw_i / \Sum rw_j (1)
2391 *
2392 * Suppose we have 4 CPUs and our @tg is a direct child of the root group and
2393 * has 7 equal weight tasks, distributed as below (rw_i), with the resulting
2394 * shares distribution (s_i):
2395 *
2396 * rw_i = { 2, 4, 1, 0 }
2397 * s_i = { 2/7, 4/7, 1/7, 0 }
2398 *
2399 * As per wake_affine() we're interested in the load of two CPUs (the CPU the
2400 * task used to run on and the CPU the waker is running on), we need to
2401 * compute the effect of waking a task on either CPU and, in case of a sync
2402 * wakeup, compute the effect of the current task going to sleep.
2403 *
2404 * So for a change of @wl to the local @cpu with an overall group weight change
2405 * of @wl we can compute the new shares distribution (s'_i) using:
2406 *
2407 * s'_i = (rw_i + @wl) / (@wg + \Sum rw_j) (2)
2408 *
2409 * Suppose we're interested in CPUs 0 and 1, and want to compute the load
2410 * differences in waking a task to CPU 0. The additional task changes the
2411 * weight and shares distributions like:
2412 *
2413 * rw'_i = { 3, 4, 1, 0 }
2414 * s'_i = { 3/8, 4/8, 1/8, 0 }
2415 *
2416 * We can then compute the difference in effective weight by using:
2417 *
2418 * dw_i = S * (s'_i - s_i) (3)
2419 *
2420 * Where 'S' is the group weight as seen by its parent.
2421 *
2422 * Therefore the effective change in loads on CPU 0 would be 5/56 (3/8 - 2/7)
2423 * times the weight of the group. The effect on CPU 1 would be -4/56 (4/8 -
2424 * 4/7) times the weight of the group.
2425 */
effective_load(struct task_group * tg,int cpu,long wl,long wg)2426 static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
2427 {
2428 struct sched_entity *se = tg->se[cpu];
2429
2430 if (!tg->parent) /* the trivial, non-cgroup case */
2431 return wl;
2432
2433 for_each_sched_entity(se) {
2434 long w, W;
2435
2436 tg = se->my_q->tg;
2437
2438 /*
2439 * W = @wg + \Sum rw_j
2440 */
2441 W = wg + calc_tg_weight(tg, se->my_q);
2442
2443 /*
2444 * w = rw_i + @wl
2445 */
2446 w = se->my_q->load.weight + wl;
2447
2448 /*
2449 * wl = S * s'_i; see (2)
2450 */
2451 if (W > 0 && w < W)
2452 wl = (w * tg->shares) / W;
2453 else
2454 wl = tg->shares;
2455
2456 /*
2457 * Per the above, wl is the new se->load.weight value; since
2458 * those are clipped to [MIN_SHARES, ...) do so now. See
2459 * calc_cfs_shares().
2460 */
2461 if (wl < MIN_SHARES)
2462 wl = MIN_SHARES;
2463
2464 /*
2465 * wl = dw_i = S * (s'_i - s_i); see (3)
2466 */
2467 wl -= se->load.weight;
2468
2469 /*
2470 * Recursively apply this logic to all parent groups to compute
2471 * the final effective load change on the root group. Since
2472 * only the @tg group gets extra weight, all parent groups can
2473 * only redistribute existing shares. @wl is the shift in shares
2474 * resulting from this level per the above.
2475 */
2476 wg = 0;
2477 }
2478
2479 return wl;
2480 }
2481 #else
2482
effective_load(struct task_group * tg,int cpu,unsigned long wl,unsigned long wg)2483 static inline unsigned long effective_load(struct task_group *tg, int cpu,
2484 unsigned long wl, unsigned long wg)
2485 {
2486 return wl;
2487 }
2488
2489 #endif
2490
wake_affine(struct sched_domain * sd,struct task_struct * p,int sync)2491 static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
2492 {
2493 s64 this_load, load;
2494 int idx, this_cpu, prev_cpu;
2495 unsigned long tl_per_task;
2496 struct task_group *tg;
2497 unsigned long weight;
2498 int balanced;
2499
2500 idx = sd->wake_idx;
2501 this_cpu = smp_processor_id();
2502 prev_cpu = task_cpu(p);
2503 load = source_load(prev_cpu, idx);
2504 this_load = target_load(this_cpu, idx);
2505
2506 /*
2507 * If sync wakeup then subtract the (maximum possible)
2508 * effect of the currently running task from the load
2509 * of the current CPU:
2510 */
2511 if (sync) {
2512 tg = task_group(current);
2513 weight = current->se.load.weight;
2514
2515 this_load += effective_load(tg, this_cpu, -weight, -weight);
2516 load += effective_load(tg, prev_cpu, 0, -weight);
2517 }
2518
2519 tg = task_group(p);
2520 weight = p->se.load.weight;
2521
2522 /*
2523 * In low-load situations, where prev_cpu is idle and this_cpu is idle
2524 * due to the sync cause above having dropped this_load to 0, we'll
2525 * always have an imbalance, but there's really nothing you can do
2526 * about that, so that's good too.
2527 *
2528 * Otherwise check if either cpus are near enough in load to allow this
2529 * task to be woken on this_cpu.
2530 */
2531 if (this_load > 0) {
2532 s64 this_eff_load, prev_eff_load;
2533
2534 this_eff_load = 100;
2535 this_eff_load *= power_of(prev_cpu);
2536 this_eff_load *= this_load +
2537 effective_load(tg, this_cpu, weight, weight);
2538
2539 prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
2540 prev_eff_load *= power_of(this_cpu);
2541 prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
2542
2543 balanced = this_eff_load <= prev_eff_load;
2544 } else
2545 balanced = true;
2546
2547 /*
2548 * If the currently running task will sleep within
2549 * a reasonable amount of time then attract this newly
2550 * woken task:
2551 */
2552 if (sync && balanced)
2553 return 1;
2554
2555 schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
2556 tl_per_task = cpu_avg_load_per_task(this_cpu);
2557
2558 if (balanced ||
2559 (this_load <= load &&
2560 this_load + target_load(prev_cpu, idx) <= tl_per_task)) {
2561 /*
2562 * This domain has SD_WAKE_AFFINE and
2563 * p is cache cold in this domain, and
2564 * there is no bad imbalance.
2565 */
2566 schedstat_inc(sd, ttwu_move_affine);
2567 schedstat_inc(p, se.statistics.nr_wakeups_affine);
2568
2569 return 1;
2570 }
2571 return 0;
2572 }
2573
2574 /*
2575 * find_idlest_group finds and returns the least busy CPU group within the
2576 * domain.
2577 */
2578 static struct sched_group *
find_idlest_group(struct sched_domain * sd,struct task_struct * p,int this_cpu,int load_idx)2579 find_idlest_group(struct sched_domain *sd, struct task_struct *p,
2580 int this_cpu, int load_idx)
2581 {
2582 struct sched_group *idlest = NULL, *group = sd->groups;
2583 unsigned long min_load = ULONG_MAX, this_load = 0;
2584 int imbalance = 100 + (sd->imbalance_pct-100)/2;
2585
2586 do {
2587 unsigned long load, avg_load;
2588 int local_group;
2589 int i;
2590
2591 /* Skip over this group if it has no CPUs allowed */
2592 if (!cpumask_intersects(sched_group_cpus(group),
2593 tsk_cpus_allowed(p)))
2594 continue;
2595
2596 local_group = cpumask_test_cpu(this_cpu,
2597 sched_group_cpus(group));
2598
2599 /* Tally up the load of all CPUs in the group */
2600 avg_load = 0;
2601
2602 for_each_cpu(i, sched_group_cpus(group)) {
2603 /* Bias balancing toward cpus of our domain */
2604 if (local_group)
2605 load = source_load(i, load_idx);
2606 else
2607 load = target_load(i, load_idx);
2608
2609 avg_load += load;
2610 }
2611
2612 /* Adjust by relative CPU power of the group */
2613 avg_load = (avg_load * SCHED_POWER_SCALE) / group->sgp->power;
2614
2615 if (local_group) {
2616 this_load = avg_load;
2617 } else if (avg_load < min_load) {
2618 min_load = avg_load;
2619 idlest = group;
2620 }
2621 } while (group = group->next, group != sd->groups);
2622
2623 if (!idlest || 100*this_load < imbalance*min_load)
2624 return NULL;
2625 return idlest;
2626 }
2627
2628 /*
2629 * find_idlest_cpu - find the idlest cpu among the cpus in group.
2630 */
2631 static int
find_idlest_cpu(struct sched_group * group,struct task_struct * p,int this_cpu)2632 find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
2633 {
2634 unsigned long load, min_load = ULONG_MAX;
2635 int idlest = -1;
2636 int i;
2637
2638 /* Traverse only the allowed CPUs */
2639 for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
2640 load = weighted_cpuload(i);
2641
2642 if (load < min_load || (load == min_load && i == this_cpu)) {
2643 min_load = load;
2644 idlest = i;
2645 }
2646 }
2647
2648 return idlest;
2649 }
2650
2651 /*
2652 * Try and locate an idle CPU in the sched_domain.
2653 */
select_idle_sibling(struct task_struct * p,int target)2654 static int select_idle_sibling(struct task_struct *p, int target)
2655 {
2656 int cpu = smp_processor_id();
2657 int prev_cpu = task_cpu(p);
2658 struct sched_domain *sd;
2659 struct sched_group *sg;
2660 int i;
2661
2662 /*
2663 * If the task is going to be woken-up on this cpu and if it is
2664 * already idle, then it is the right target.
2665 */
2666 if (target == cpu && idle_cpu(cpu))
2667 return cpu;
2668
2669 /*
2670 * If the task is going to be woken-up on the cpu where it previously
2671 * ran and if it is currently idle, then it the right target.
2672 */
2673 if (target == prev_cpu && idle_cpu(prev_cpu))
2674 return prev_cpu;
2675
2676 /*
2677 * Otherwise, iterate the domains and find an elegible idle cpu.
2678 */
2679 sd = rcu_dereference(per_cpu(sd_llc, target));
2680 for_each_lower_domain(sd) {
2681 sg = sd->groups;
2682 do {
2683 if (!cpumask_intersects(sched_group_cpus(sg),
2684 tsk_cpus_allowed(p)))
2685 goto next;
2686
2687 for_each_cpu(i, sched_group_cpus(sg)) {
2688 if (!idle_cpu(i))
2689 goto next;
2690 }
2691
2692 target = cpumask_first_and(sched_group_cpus(sg),
2693 tsk_cpus_allowed(p));
2694 goto done;
2695 next:
2696 sg = sg->next;
2697 } while (sg != sd->groups);
2698 }
2699 done:
2700 return target;
2701 }
2702
2703 /*
2704 * sched_balance_self: balance the current task (running on cpu) in domains
2705 * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
2706 * SD_BALANCE_EXEC.
2707 *
2708 * Balance, ie. select the least loaded group.
2709 *
2710 * Returns the target CPU number, or the same CPU if no balancing is needed.
2711 *
2712 * preempt must be disabled.
2713 */
2714 static int
select_task_rq_fair(struct task_struct * p,int sd_flag,int wake_flags)2715 select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags)
2716 {
2717 struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
2718 int cpu = smp_processor_id();
2719 int prev_cpu = task_cpu(p);
2720 int new_cpu = cpu;
2721 int want_affine = 0;
2722 int want_sd = 1;
2723 int sync = wake_flags & WF_SYNC;
2724
2725 if (p->rt.nr_cpus_allowed == 1)
2726 return prev_cpu;
2727
2728 if (sd_flag & SD_BALANCE_WAKE) {
2729 if (cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
2730 want_affine = 1;
2731 new_cpu = prev_cpu;
2732 }
2733
2734 rcu_read_lock();
2735 for_each_domain(cpu, tmp) {
2736 if (!(tmp->flags & SD_LOAD_BALANCE))
2737 continue;
2738
2739 /*
2740 * If power savings logic is enabled for a domain, see if we
2741 * are not overloaded, if so, don't balance wider.
2742 */
2743 if (tmp->flags & (SD_POWERSAVINGS_BALANCE|SD_PREFER_LOCAL)) {
2744 unsigned long power = 0;
2745 unsigned long nr_running = 0;
2746 unsigned long capacity;
2747 int i;
2748
2749 for_each_cpu(i, sched_domain_span(tmp)) {
2750 power += power_of(i);
2751 nr_running += cpu_rq(i)->cfs.nr_running;
2752 }
2753
2754 capacity = DIV_ROUND_CLOSEST(power, SCHED_POWER_SCALE);
2755
2756 if (tmp->flags & SD_POWERSAVINGS_BALANCE)
2757 nr_running /= 2;
2758
2759 if (nr_running < capacity)
2760 want_sd = 0;
2761 }
2762
2763 /*
2764 * If both cpu and prev_cpu are part of this domain,
2765 * cpu is a valid SD_WAKE_AFFINE target.
2766 */
2767 if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
2768 cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
2769 affine_sd = tmp;
2770 want_affine = 0;
2771 }
2772
2773 if (!want_sd && !want_affine)
2774 break;
2775
2776 if (!(tmp->flags & sd_flag))
2777 continue;
2778
2779 if (want_sd)
2780 sd = tmp;
2781 }
2782
2783 if (affine_sd) {
2784 if (cpu == prev_cpu || wake_affine(affine_sd, p, sync))
2785 prev_cpu = cpu;
2786
2787 new_cpu = select_idle_sibling(p, prev_cpu);
2788 goto unlock;
2789 }
2790
2791 while (sd) {
2792 int load_idx = sd->forkexec_idx;
2793 struct sched_group *group;
2794 int weight;
2795
2796 if (!(sd->flags & sd_flag)) {
2797 sd = sd->child;
2798 continue;
2799 }
2800
2801 if (sd_flag & SD_BALANCE_WAKE)
2802 load_idx = sd->wake_idx;
2803
2804 group = find_idlest_group(sd, p, cpu, load_idx);
2805 if (!group) {
2806 sd = sd->child;
2807 continue;
2808 }
2809
2810 new_cpu = find_idlest_cpu(group, p, cpu);
2811 if (new_cpu == -1 || new_cpu == cpu) {
2812 /* Now try balancing at a lower domain level of cpu */
2813 sd = sd->child;
2814 continue;
2815 }
2816
2817 /* Now try balancing at a lower domain level of new_cpu */
2818 cpu = new_cpu;
2819 weight = sd->span_weight;
2820 sd = NULL;
2821 for_each_domain(cpu, tmp) {
2822 if (weight <= tmp->span_weight)
2823 break;
2824 if (tmp->flags & sd_flag)
2825 sd = tmp;
2826 }
2827 /* while loop will break here if sd == NULL */
2828 }
2829 unlock:
2830 rcu_read_unlock();
2831
2832 return new_cpu;
2833 }
2834 #endif /* CONFIG_SMP */
2835
2836 static unsigned long
wakeup_gran(struct sched_entity * curr,struct sched_entity * se)2837 wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
2838 {
2839 unsigned long gran = sysctl_sched_wakeup_granularity;
2840
2841 /*
2842 * Since its curr running now, convert the gran from real-time
2843 * to virtual-time in his units.
2844 *
2845 * By using 'se' instead of 'curr' we penalize light tasks, so
2846 * they get preempted easier. That is, if 'se' < 'curr' then
2847 * the resulting gran will be larger, therefore penalizing the
2848 * lighter, if otoh 'se' > 'curr' then the resulting gran will
2849 * be smaller, again penalizing the lighter task.
2850 *
2851 * This is especially important for buddies when the leftmost
2852 * task is higher priority than the buddy.
2853 */
2854 return calc_delta_fair(gran, se);
2855 }
2856
2857 /*
2858 * Should 'se' preempt 'curr'.
2859 *
2860 * |s1
2861 * |s2
2862 * |s3
2863 * g
2864 * |<--->|c
2865 *
2866 * w(c, s1) = -1
2867 * w(c, s2) = 0
2868 * w(c, s3) = 1
2869 *
2870 */
2871 static int
wakeup_preempt_entity(struct sched_entity * curr,struct sched_entity * se)2872 wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
2873 {
2874 s64 gran, vdiff = curr->vruntime - se->vruntime;
2875
2876 if (vdiff <= 0)
2877 return -1;
2878
2879 gran = wakeup_gran(curr, se);
2880 if (vdiff > gran)
2881 return 1;
2882
2883 return 0;
2884 }
2885
set_last_buddy(struct sched_entity * se)2886 static void set_last_buddy(struct sched_entity *se)
2887 {
2888 if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
2889 return;
2890
2891 for_each_sched_entity(se)
2892 cfs_rq_of(se)->last = se;
2893 }
2894
set_next_buddy(struct sched_entity * se)2895 static void set_next_buddy(struct sched_entity *se)
2896 {
2897 if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
2898 return;
2899
2900 for_each_sched_entity(se)
2901 cfs_rq_of(se)->next = se;
2902 }
2903
set_skip_buddy(struct sched_entity * se)2904 static void set_skip_buddy(struct sched_entity *se)
2905 {
2906 for_each_sched_entity(se)
2907 cfs_rq_of(se)->skip = se;
2908 }
2909
2910 /*
2911 * Preempt the current task with a newly woken task if needed:
2912 */
check_preempt_wakeup(struct rq * rq,struct task_struct * p,int wake_flags)2913 static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
2914 {
2915 struct task_struct *curr = rq->curr;
2916 struct sched_entity *se = &curr->se, *pse = &p->se;
2917 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
2918 int scale = cfs_rq->nr_running >= sched_nr_latency;
2919 int next_buddy_marked = 0;
2920
2921 if (unlikely(se == pse))
2922 return;
2923
2924 /*
2925 * This is possible from callers such as move_task(), in which we
2926 * unconditionally check_prempt_curr() after an enqueue (which may have
2927 * lead to a throttle). This both saves work and prevents false
2928 * next-buddy nomination below.
2929 */
2930 if (unlikely(throttled_hierarchy(cfs_rq_of(pse))))
2931 return;
2932
2933 if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
2934 set_next_buddy(pse);
2935 next_buddy_marked = 1;
2936 }
2937
2938 /*
2939 * We can come here with TIF_NEED_RESCHED already set from new task
2940 * wake up path.
2941 *
2942 * Note: this also catches the edge-case of curr being in a throttled
2943 * group (e.g. via set_curr_task), since update_curr() (in the
2944 * enqueue of curr) will have resulted in resched being set. This
2945 * prevents us from potentially nominating it as a false LAST_BUDDY
2946 * below.
2947 */
2948 if (test_tsk_need_resched(curr))
2949 return;
2950
2951 /* Idle tasks are by definition preempted by non-idle tasks. */
2952 if (unlikely(curr->policy == SCHED_IDLE) &&
2953 likely(p->policy != SCHED_IDLE))
2954 goto preempt;
2955
2956 /*
2957 * Batch and idle tasks do not preempt non-idle tasks (their preemption
2958 * is driven by the tick):
2959 */
2960 if (unlikely(p->policy != SCHED_NORMAL))
2961 return;
2962
2963 find_matching_se(&se, &pse);
2964 update_curr(cfs_rq_of(se));
2965 BUG_ON(!pse);
2966 if (wakeup_preempt_entity(se, pse) == 1) {
2967 /*
2968 * Bias pick_next to pick the sched entity that is
2969 * triggering this preemption.
2970 */
2971 if (!next_buddy_marked)
2972 set_next_buddy(pse);
2973 goto preempt;
2974 }
2975
2976 return;
2977
2978 preempt:
2979 resched_task(curr);
2980 /*
2981 * Only set the backward buddy when the current task is still
2982 * on the rq. This can happen when a wakeup gets interleaved
2983 * with schedule on the ->pre_schedule() or idle_balance()
2984 * point, either of which can * drop the rq lock.
2985 *
2986 * Also, during early boot the idle thread is in the fair class,
2987 * for obvious reasons its a bad idea to schedule back to it.
2988 */
2989 if (unlikely(!se->on_rq || curr == rq->idle))
2990 return;
2991
2992 if (sched_feat(LAST_BUDDY) && scale && entity_is_task(se))
2993 set_last_buddy(se);
2994 }
2995
pick_next_task_fair(struct rq * rq)2996 static struct task_struct *pick_next_task_fair(struct rq *rq)
2997 {
2998 struct task_struct *p;
2999 struct cfs_rq *cfs_rq = &rq->cfs;
3000 struct sched_entity *se;
3001
3002 if (!cfs_rq->nr_running)
3003 return NULL;
3004
3005 do {
3006 se = pick_next_entity(cfs_rq);
3007 set_next_entity(cfs_rq, se);
3008 cfs_rq = group_cfs_rq(se);
3009 } while (cfs_rq);
3010
3011 p = task_of(se);
3012 if (hrtick_enabled(rq))
3013 hrtick_start_fair(rq, p);
3014
3015 return p;
3016 }
3017
3018 /*
3019 * Account for a descheduled task:
3020 */
put_prev_task_fair(struct rq * rq,struct task_struct * prev)3021 static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
3022 {
3023 struct sched_entity *se = &prev->se;
3024 struct cfs_rq *cfs_rq;
3025
3026 for_each_sched_entity(se) {
3027 cfs_rq = cfs_rq_of(se);
3028 put_prev_entity(cfs_rq, se);
3029 }
3030 }
3031
3032 /*
3033 * sched_yield() is very simple
3034 *
3035 * The magic of dealing with the ->skip buddy is in pick_next_entity.
3036 */
yield_task_fair(struct rq * rq)3037 static void yield_task_fair(struct rq *rq)
3038 {
3039 struct task_struct *curr = rq->curr;
3040 struct cfs_rq *cfs_rq = task_cfs_rq(curr);
3041 struct sched_entity *se = &curr->se;
3042
3043 /*
3044 * Are we the only task in the tree?
3045 */
3046 if (unlikely(rq->nr_running == 1))
3047 return;
3048
3049 clear_buddies(cfs_rq, se);
3050
3051 if (curr->policy != SCHED_BATCH) {
3052 update_rq_clock(rq);
3053 /*
3054 * Update run-time statistics of the 'current'.
3055 */
3056 update_curr(cfs_rq);
3057 /*
3058 * Tell update_rq_clock() that we've just updated,
3059 * so we don't do microscopic update in schedule()
3060 * and double the fastpath cost.
3061 */
3062 rq->skip_clock_update = 1;
3063 }
3064
3065 set_skip_buddy(se);
3066 }
3067
yield_to_task_fair(struct rq * rq,struct task_struct * p,bool preempt)3068 static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
3069 {
3070 struct sched_entity *se = &p->se;
3071
3072 /* throttled hierarchies are not runnable */
3073 if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
3074 return false;
3075
3076 /* Tell the scheduler that we'd really like pse to run next. */
3077 set_next_buddy(se);
3078
3079 yield_task_fair(rq);
3080
3081 return true;
3082 }
3083
3084 #ifdef CONFIG_SMP
3085 /**************************************************
3086 * Fair scheduling class load-balancing methods:
3087 */
3088
3089 static unsigned long __read_mostly max_load_balance_interval = HZ/10;
3090
3091 #define LBF_ALL_PINNED 0x01
3092 #define LBF_NEED_BREAK 0x02
3093
3094 struct lb_env {
3095 struct sched_domain *sd;
3096
3097 int src_cpu;
3098 struct rq *src_rq;
3099
3100 int dst_cpu;
3101 struct rq *dst_rq;
3102
3103 enum cpu_idle_type idle;
3104 long load_move;
3105 unsigned int flags;
3106
3107 unsigned int loop;
3108 unsigned int loop_break;
3109 unsigned int loop_max;
3110 };
3111
3112 /*
3113 * move_task - move a task from one runqueue to another runqueue.
3114 * Both runqueues must be locked.
3115 */
move_task(struct task_struct * p,struct lb_env * env)3116 static void move_task(struct task_struct *p, struct lb_env *env)
3117 {
3118 deactivate_task(env->src_rq, p, 0);
3119 set_task_cpu(p, env->dst_cpu);
3120 activate_task(env->dst_rq, p, 0);
3121 check_preempt_curr(env->dst_rq, p, 0);
3122 }
3123
3124 /*
3125 * Is this task likely cache-hot:
3126 */
3127 static int
task_hot(struct task_struct * p,u64 now,struct sched_domain * sd)3128 task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
3129 {
3130 s64 delta;
3131
3132 if (p->sched_class != &fair_sched_class)
3133 return 0;
3134
3135 if (unlikely(p->policy == SCHED_IDLE))
3136 return 0;
3137
3138 /*
3139 * Buddy candidates are cache hot:
3140 */
3141 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
3142 (&p->se == cfs_rq_of(&p->se)->next ||
3143 &p->se == cfs_rq_of(&p->se)->last))
3144 return 1;
3145
3146 if (sysctl_sched_migration_cost == -1)
3147 return 1;
3148 if (sysctl_sched_migration_cost == 0)
3149 return 0;
3150
3151 delta = now - p->se.exec_start;
3152
3153 return delta < (s64)sysctl_sched_migration_cost;
3154 }
3155
3156 /*
3157 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
3158 */
3159 static
can_migrate_task(struct task_struct * p,struct lb_env * env)3160 int can_migrate_task(struct task_struct *p, struct lb_env *env)
3161 {
3162 int tsk_cache_hot = 0;
3163 /*
3164 * We do not migrate tasks that are:
3165 * 1) running (obviously), or
3166 * 2) cannot be migrated to this CPU due to cpus_allowed, or
3167 * 3) are cache-hot on their current CPU.
3168 */
3169 if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
3170 schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
3171 return 0;
3172 }
3173 env->flags &= ~LBF_ALL_PINNED;
3174
3175 if (task_running(env->src_rq, p)) {
3176 schedstat_inc(p, se.statistics.nr_failed_migrations_running);
3177 return 0;
3178 }
3179
3180 /*
3181 * Aggressive migration if:
3182 * 1) task is cache cold, or
3183 * 2) too many balance attempts have failed.
3184 */
3185
3186 tsk_cache_hot = task_hot(p, env->src_rq->clock_task, env->sd);
3187 if (!tsk_cache_hot ||
3188 env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
3189 #ifdef CONFIG_SCHEDSTATS
3190 if (tsk_cache_hot) {
3191 schedstat_inc(env->sd, lb_hot_gained[env->idle]);
3192 schedstat_inc(p, se.statistics.nr_forced_migrations);
3193 }
3194 #endif
3195 return 1;
3196 }
3197
3198 if (tsk_cache_hot) {
3199 schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
3200 return 0;
3201 }
3202 return 1;
3203 }
3204
3205 /*
3206 * move_one_task tries to move exactly one task from busiest to this_rq, as
3207 * part of active balancing operations within "domain".
3208 * Returns 1 if successful and 0 otherwise.
3209 *
3210 * Called with both runqueues locked.
3211 */
move_one_task(struct lb_env * env)3212 static int move_one_task(struct lb_env *env)
3213 {
3214 struct task_struct *p, *n;
3215
3216 list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
3217 if (throttled_lb_pair(task_group(p), env->src_rq->cpu, env->dst_cpu))
3218 continue;
3219
3220 if (!can_migrate_task(p, env))
3221 continue;
3222
3223 move_task(p, env);
3224 /*
3225 * Right now, this is only the second place move_task()
3226 * is called, so we can safely collect move_task()
3227 * stats here rather than inside move_task().
3228 */
3229 schedstat_inc(env->sd, lb_gained[env->idle]);
3230 return 1;
3231 }
3232 return 0;
3233 }
3234
3235 static unsigned long task_h_load(struct task_struct *p);
3236
3237 static const unsigned int sched_nr_migrate_break = 32;
3238
3239 /*
3240 * move_tasks tries to move up to load_move weighted load from busiest to
3241 * this_rq, as part of a balancing operation within domain "sd".
3242 * Returns 1 if successful and 0 otherwise.
3243 *
3244 * Called with both runqueues locked.
3245 */
move_tasks(struct lb_env * env)3246 static int move_tasks(struct lb_env *env)
3247 {
3248 struct list_head *tasks = &env->src_rq->cfs_tasks;
3249 struct task_struct *p;
3250 unsigned long load;
3251 int pulled = 0;
3252
3253 if (env->load_move <= 0)
3254 return 0;
3255
3256 while (!list_empty(tasks)) {
3257 p = list_first_entry(tasks, struct task_struct, se.group_node);
3258
3259 env->loop++;
3260 /* We've more or less seen every task there is, call it quits */
3261 if (env->loop > env->loop_max)
3262 break;
3263
3264 /* take a breather every nr_migrate tasks */
3265 if (env->loop > env->loop_break) {
3266 env->loop_break += sched_nr_migrate_break;
3267 env->flags |= LBF_NEED_BREAK;
3268 break;
3269 }
3270
3271 if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
3272 goto next;
3273
3274 load = task_h_load(p);
3275
3276 if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
3277 goto next;
3278
3279 if ((load / 2) > env->load_move)
3280 goto next;
3281
3282 if (!can_migrate_task(p, env))
3283 goto next;
3284
3285 move_task(p, env);
3286 pulled++;
3287 env->load_move -= load;
3288
3289 #ifdef CONFIG_PREEMPT
3290 /*
3291 * NEWIDLE balancing is a source of latency, so preemptible
3292 * kernels will stop after the first task is pulled to minimize
3293 * the critical section.
3294 */
3295 if (env->idle == CPU_NEWLY_IDLE)
3296 break;
3297 #endif
3298
3299 /*
3300 * We only want to steal up to the prescribed amount of
3301 * weighted load.
3302 */
3303 if (env->load_move <= 0)
3304 break;
3305
3306 continue;
3307 next:
3308 list_move_tail(&p->se.group_node, tasks);
3309 }
3310
3311 /*
3312 * Right now, this is one of only two places move_task() is called,
3313 * so we can safely collect move_task() stats here rather than
3314 * inside move_task().
3315 */
3316 schedstat_add(env->sd, lb_gained[env->idle], pulled);
3317
3318 return pulled;
3319 }
3320
3321 #ifdef CONFIG_FAIR_GROUP_SCHED
3322 /*
3323 * update tg->load_weight by folding this cpu's load_avg
3324 */
update_shares_cpu(struct task_group * tg,int cpu)3325 static int update_shares_cpu(struct task_group *tg, int cpu)
3326 {
3327 struct cfs_rq *cfs_rq;
3328 unsigned long flags;
3329 struct rq *rq;
3330
3331 if (!tg->se[cpu])
3332 return 0;
3333
3334 rq = cpu_rq(cpu);
3335 cfs_rq = tg->cfs_rq[cpu];
3336
3337 raw_spin_lock_irqsave(&rq->lock, flags);
3338
3339 update_rq_clock(rq);
3340 update_cfs_load(cfs_rq, 1);
3341
3342 /*
3343 * We need to update shares after updating tg->load_weight in
3344 * order to adjust the weight of groups with long running tasks.
3345 */
3346 update_cfs_shares(cfs_rq);
3347
3348 raw_spin_unlock_irqrestore(&rq->lock, flags);
3349
3350 return 0;
3351 }
3352
update_shares(int cpu)3353 static void update_shares(int cpu)
3354 {
3355 struct cfs_rq *cfs_rq;
3356 struct rq *rq = cpu_rq(cpu);
3357
3358 rcu_read_lock();
3359 /*
3360 * Iterates the task_group tree in a bottom up fashion, see
3361 * list_add_leaf_cfs_rq() for details.
3362 */
3363 for_each_leaf_cfs_rq(rq, cfs_rq) {
3364 /* throttled entities do not contribute to load */
3365 if (throttled_hierarchy(cfs_rq))
3366 continue;
3367
3368 update_shares_cpu(cfs_rq->tg, cpu);
3369 }
3370 rcu_read_unlock();
3371 }
3372
3373 /*
3374 * Compute the cpu's hierarchical load factor for each task group.
3375 * This needs to be done in a top-down fashion because the load of a child
3376 * group is a fraction of its parents load.
3377 */
tg_load_down(struct task_group * tg,void * data)3378 static int tg_load_down(struct task_group *tg, void *data)
3379 {
3380 unsigned long load;
3381 long cpu = (long)data;
3382
3383 if (!tg->parent) {
3384 load = cpu_rq(cpu)->load.weight;
3385 } else {
3386 load = tg->parent->cfs_rq[cpu]->h_load;
3387 load *= tg->se[cpu]->load.weight;
3388 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
3389 }
3390
3391 tg->cfs_rq[cpu]->h_load = load;
3392
3393 return 0;
3394 }
3395
update_h_load(long cpu)3396 static void update_h_load(long cpu)
3397 {
3398 rcu_read_lock();
3399 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
3400 rcu_read_unlock();
3401 }
3402
task_h_load(struct task_struct * p)3403 static unsigned long task_h_load(struct task_struct *p)
3404 {
3405 struct cfs_rq *cfs_rq = task_cfs_rq(p);
3406 unsigned long load;
3407
3408 load = p->se.load.weight;
3409 load = div_u64(load * cfs_rq->h_load, cfs_rq->load.weight + 1);
3410
3411 return load;
3412 }
3413 #else
update_shares(int cpu)3414 static inline void update_shares(int cpu)
3415 {
3416 }
3417
update_h_load(long cpu)3418 static inline void update_h_load(long cpu)
3419 {
3420 }
3421
task_h_load(struct task_struct * p)3422 static unsigned long task_h_load(struct task_struct *p)
3423 {
3424 return p->se.load.weight;
3425 }
3426 #endif
3427
3428 /********** Helpers for find_busiest_group ************************/
3429 /*
3430 * sd_lb_stats - Structure to store the statistics of a sched_domain
3431 * during load balancing.
3432 */
3433 struct sd_lb_stats {
3434 struct sched_group *busiest; /* Busiest group in this sd */
3435 struct sched_group *this; /* Local group in this sd */
3436 unsigned long total_load; /* Total load of all groups in sd */
3437 unsigned long total_pwr; /* Total power of all groups in sd */
3438 unsigned long avg_load; /* Average load across all groups in sd */
3439
3440 /** Statistics of this group */
3441 unsigned long this_load;
3442 unsigned long this_load_per_task;
3443 unsigned long this_nr_running;
3444 unsigned long this_has_capacity;
3445 unsigned int this_idle_cpus;
3446
3447 /* Statistics of the busiest group */
3448 unsigned int busiest_idle_cpus;
3449 unsigned long max_load;
3450 unsigned long busiest_load_per_task;
3451 unsigned long busiest_nr_running;
3452 unsigned long busiest_group_capacity;
3453 unsigned long busiest_has_capacity;
3454 unsigned int busiest_group_weight;
3455
3456 int group_imb; /* Is there imbalance in this sd */
3457 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
3458 int power_savings_balance; /* Is powersave balance needed for this sd */
3459 struct sched_group *group_min; /* Least loaded group in sd */
3460 struct sched_group *group_leader; /* Group which relieves group_min */
3461 unsigned long min_load_per_task; /* load_per_task in group_min */
3462 unsigned long leader_nr_running; /* Nr running of group_leader */
3463 unsigned long min_nr_running; /* Nr running of group_min */
3464 #endif
3465 };
3466
3467 /*
3468 * sg_lb_stats - stats of a sched_group required for load_balancing
3469 */
3470 struct sg_lb_stats {
3471 unsigned long avg_load; /*Avg load across the CPUs of the group */
3472 unsigned long group_load; /* Total load over the CPUs of the group */
3473 unsigned long sum_nr_running; /* Nr tasks running in the group */
3474 unsigned long sum_weighted_load; /* Weighted load of group's tasks */
3475 unsigned long group_capacity;
3476 unsigned long idle_cpus;
3477 unsigned long group_weight;
3478 int group_imb; /* Is there an imbalance in the group ? */
3479 int group_has_capacity; /* Is there extra capacity in the group? */
3480 };
3481
3482 /**
3483 * get_sd_load_idx - Obtain the load index for a given sched domain.
3484 * @sd: The sched_domain whose load_idx is to be obtained.
3485 * @idle: The Idle status of the CPU for whose sd load_icx is obtained.
3486 */
get_sd_load_idx(struct sched_domain * sd,enum cpu_idle_type idle)3487 static inline int get_sd_load_idx(struct sched_domain *sd,
3488 enum cpu_idle_type idle)
3489 {
3490 int load_idx;
3491
3492 switch (idle) {
3493 case CPU_NOT_IDLE:
3494 load_idx = sd->busy_idx;
3495 break;
3496
3497 case CPU_NEWLY_IDLE:
3498 load_idx = sd->newidle_idx;
3499 break;
3500 default:
3501 load_idx = sd->idle_idx;
3502 break;
3503 }
3504
3505 return load_idx;
3506 }
3507
3508
3509 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
3510 /**
3511 * init_sd_power_savings_stats - Initialize power savings statistics for
3512 * the given sched_domain, during load balancing.
3513 *
3514 * @sd: Sched domain whose power-savings statistics are to be initialized.
3515 * @sds: Variable containing the statistics for sd.
3516 * @idle: Idle status of the CPU at which we're performing load-balancing.
3517 */
init_sd_power_savings_stats(struct sched_domain * sd,struct sd_lb_stats * sds,enum cpu_idle_type idle)3518 static inline void init_sd_power_savings_stats(struct sched_domain *sd,
3519 struct sd_lb_stats *sds, enum cpu_idle_type idle)
3520 {
3521 /*
3522 * Busy processors will not participate in power savings
3523 * balance.
3524 */
3525 if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
3526 sds->power_savings_balance = 0;
3527 else {
3528 sds->power_savings_balance = 1;
3529 sds->min_nr_running = ULONG_MAX;
3530 sds->leader_nr_running = 0;
3531 }
3532 }
3533
3534 /**
3535 * update_sd_power_savings_stats - Update the power saving stats for a
3536 * sched_domain while performing load balancing.
3537 *
3538 * @group: sched_group belonging to the sched_domain under consideration.
3539 * @sds: Variable containing the statistics of the sched_domain
3540 * @local_group: Does group contain the CPU for which we're performing
3541 * load balancing ?
3542 * @sgs: Variable containing the statistics of the group.
3543 */
update_sd_power_savings_stats(struct sched_group * group,struct sd_lb_stats * sds,int local_group,struct sg_lb_stats * sgs)3544 static inline void update_sd_power_savings_stats(struct sched_group *group,
3545 struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs)
3546 {
3547
3548 if (!sds->power_savings_balance)
3549 return;
3550
3551 /*
3552 * If the local group is idle or completely loaded
3553 * no need to do power savings balance at this domain
3554 */
3555 if (local_group && (sds->this_nr_running >= sgs->group_capacity ||
3556 !sds->this_nr_running))
3557 sds->power_savings_balance = 0;
3558
3559 /*
3560 * If a group is already running at full capacity or idle,
3561 * don't include that group in power savings calculations
3562 */
3563 if (!sds->power_savings_balance ||
3564 sgs->sum_nr_running >= sgs->group_capacity ||
3565 !sgs->sum_nr_running)
3566 return;
3567
3568 /*
3569 * Calculate the group which has the least non-idle load.
3570 * This is the group from where we need to pick up the load
3571 * for saving power
3572 */
3573 if ((sgs->sum_nr_running < sds->min_nr_running) ||
3574 (sgs->sum_nr_running == sds->min_nr_running &&
3575 group_first_cpu(group) > group_first_cpu(sds->group_min))) {
3576 sds->group_min = group;
3577 sds->min_nr_running = sgs->sum_nr_running;
3578 sds->min_load_per_task = sgs->sum_weighted_load /
3579 sgs->sum_nr_running;
3580 }
3581
3582 /*
3583 * Calculate the group which is almost near its
3584 * capacity but still has some space to pick up some load
3585 * from other group and save more power
3586 */
3587 if (sgs->sum_nr_running + 1 > sgs->group_capacity)
3588 return;
3589
3590 if (sgs->sum_nr_running > sds->leader_nr_running ||
3591 (sgs->sum_nr_running == sds->leader_nr_running &&
3592 group_first_cpu(group) < group_first_cpu(sds->group_leader))) {
3593 sds->group_leader = group;
3594 sds->leader_nr_running = sgs->sum_nr_running;
3595 }
3596 }
3597
3598 /**
3599 * check_power_save_busiest_group - see if there is potential for some power-savings balance
3600 * @sds: Variable containing the statistics of the sched_domain
3601 * under consideration.
3602 * @this_cpu: Cpu at which we're currently performing load-balancing.
3603 * @imbalance: Variable to store the imbalance.
3604 *
3605 * Description:
3606 * Check if we have potential to perform some power-savings balance.
3607 * If yes, set the busiest group to be the least loaded group in the
3608 * sched_domain, so that it's CPUs can be put to idle.
3609 *
3610 * Returns 1 if there is potential to perform power-savings balance.
3611 * Else returns 0.
3612 */
check_power_save_busiest_group(struct sd_lb_stats * sds,int this_cpu,unsigned long * imbalance)3613 static inline int check_power_save_busiest_group(struct sd_lb_stats *sds,
3614 int this_cpu, unsigned long *imbalance)
3615 {
3616 if (!sds->power_savings_balance)
3617 return 0;
3618
3619 if (sds->this != sds->group_leader ||
3620 sds->group_leader == sds->group_min)
3621 return 0;
3622
3623 *imbalance = sds->min_load_per_task;
3624 sds->busiest = sds->group_min;
3625
3626 return 1;
3627
3628 }
3629 #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
init_sd_power_savings_stats(struct sched_domain * sd,struct sd_lb_stats * sds,enum cpu_idle_type idle)3630 static inline void init_sd_power_savings_stats(struct sched_domain *sd,
3631 struct sd_lb_stats *sds, enum cpu_idle_type idle)
3632 {
3633 return;
3634 }
3635
update_sd_power_savings_stats(struct sched_group * group,struct sd_lb_stats * sds,int local_group,struct sg_lb_stats * sgs)3636 static inline void update_sd_power_savings_stats(struct sched_group *group,
3637 struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs)
3638 {
3639 return;
3640 }
3641
check_power_save_busiest_group(struct sd_lb_stats * sds,int this_cpu,unsigned long * imbalance)3642 static inline int check_power_save_busiest_group(struct sd_lb_stats *sds,
3643 int this_cpu, unsigned long *imbalance)
3644 {
3645 return 0;
3646 }
3647 #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
3648
3649
default_scale_freq_power(struct sched_domain * sd,int cpu)3650 unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu)
3651 {
3652 return SCHED_POWER_SCALE;
3653 }
3654
arch_scale_freq_power(struct sched_domain * sd,int cpu)3655 unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu)
3656 {
3657 return default_scale_freq_power(sd, cpu);
3658 }
3659
default_scale_smt_power(struct sched_domain * sd,int cpu)3660 unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu)
3661 {
3662 unsigned long weight = sd->span_weight;
3663 unsigned long smt_gain = sd->smt_gain;
3664
3665 smt_gain /= weight;
3666
3667 return smt_gain;
3668 }
3669
arch_scale_smt_power(struct sched_domain * sd,int cpu)3670 unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu)
3671 {
3672 return default_scale_smt_power(sd, cpu);
3673 }
3674
scale_rt_power(int cpu)3675 unsigned long scale_rt_power(int cpu)
3676 {
3677 struct rq *rq = cpu_rq(cpu);
3678 u64 total, available;
3679
3680 total = sched_avg_period() + (rq->clock - rq->age_stamp);
3681
3682 if (unlikely(total < rq->rt_avg)) {
3683 /* Ensures that power won't end up being negative */
3684 available = 0;
3685 } else {
3686 available = total - rq->rt_avg;
3687 }
3688
3689 if (unlikely((s64)total < SCHED_POWER_SCALE))
3690 total = SCHED_POWER_SCALE;
3691
3692 total >>= SCHED_POWER_SHIFT;
3693
3694 return div_u64(available, total);
3695 }
3696
update_cpu_power(struct sched_domain * sd,int cpu)3697 static void update_cpu_power(struct sched_domain *sd, int cpu)
3698 {
3699 unsigned long weight = sd->span_weight;
3700 unsigned long power = SCHED_POWER_SCALE;
3701 struct sched_group *sdg = sd->groups;
3702
3703 if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
3704 if (sched_feat(ARCH_POWER))
3705 power *= arch_scale_smt_power(sd, cpu);
3706 else
3707 power *= default_scale_smt_power(sd, cpu);
3708
3709 power >>= SCHED_POWER_SHIFT;
3710 }
3711
3712 sdg->sgp->power_orig = power;
3713
3714 if (sched_feat(ARCH_POWER))
3715 power *= arch_scale_freq_power(sd, cpu);
3716 else
3717 power *= default_scale_freq_power(sd, cpu);
3718
3719 power >>= SCHED_POWER_SHIFT;
3720
3721 power *= scale_rt_power(cpu);
3722 power >>= SCHED_POWER_SHIFT;
3723
3724 if (!power)
3725 power = 1;
3726
3727 cpu_rq(cpu)->cpu_power = power;
3728 sdg->sgp->power = power;
3729 }
3730
update_group_power(struct sched_domain * sd,int cpu)3731 void update_group_power(struct sched_domain *sd, int cpu)
3732 {
3733 struct sched_domain *child = sd->child;
3734 struct sched_group *group, *sdg = sd->groups;
3735 unsigned long power;
3736 unsigned long interval;
3737
3738 interval = msecs_to_jiffies(sd->balance_interval);
3739 interval = clamp(interval, 1UL, max_load_balance_interval);
3740 sdg->sgp->next_update = jiffies + interval;
3741
3742 if (!child) {
3743 update_cpu_power(sd, cpu);
3744 return;
3745 }
3746
3747 power = 0;
3748
3749 group = child->groups;
3750 do {
3751 power += group->sgp->power;
3752 group = group->next;
3753 } while (group != child->groups);
3754
3755 sdg->sgp->power = power;
3756 }
3757
3758 /*
3759 * Try and fix up capacity for tiny siblings, this is needed when
3760 * things like SD_ASYM_PACKING need f_b_g to select another sibling
3761 * which on its own isn't powerful enough.
3762 *
3763 * See update_sd_pick_busiest() and check_asym_packing().
3764 */
3765 static inline int
fix_small_capacity(struct sched_domain * sd,struct sched_group * group)3766 fix_small_capacity(struct sched_domain *sd, struct sched_group *group)
3767 {
3768 /*
3769 * Only siblings can have significantly less than SCHED_POWER_SCALE
3770 */
3771 if (!(sd->flags & SD_SHARE_CPUPOWER))
3772 return 0;
3773
3774 /*
3775 * If ~90% of the cpu_power is still there, we're good.
3776 */
3777 if (group->sgp->power * 32 > group->sgp->power_orig * 29)
3778 return 1;
3779
3780 return 0;
3781 }
3782
3783 /**
3784 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
3785 * @sd: The sched_domain whose statistics are to be updated.
3786 * @group: sched_group whose statistics are to be updated.
3787 * @this_cpu: Cpu for which load balance is currently performed.
3788 * @idle: Idle status of this_cpu
3789 * @load_idx: Load index of sched_domain of this_cpu for load calc.
3790 * @local_group: Does group contain this_cpu.
3791 * @cpus: Set of cpus considered for load balancing.
3792 * @balance: Should we balance.
3793 * @sgs: variable to hold the statistics for this group.
3794 */
update_sg_lb_stats(struct sched_domain * sd,struct sched_group * group,int this_cpu,enum cpu_idle_type idle,int load_idx,int local_group,const struct cpumask * cpus,int * balance,struct sg_lb_stats * sgs)3795 static inline void update_sg_lb_stats(struct sched_domain *sd,
3796 struct sched_group *group, int this_cpu,
3797 enum cpu_idle_type idle, int load_idx,
3798 int local_group, const struct cpumask *cpus,
3799 int *balance, struct sg_lb_stats *sgs)
3800 {
3801 unsigned long load, max_cpu_load, min_cpu_load, max_nr_running;
3802 int i;
3803 unsigned int balance_cpu = -1, first_idle_cpu = 0;
3804 unsigned long avg_load_per_task = 0;
3805
3806 if (local_group)
3807 balance_cpu = group_first_cpu(group);
3808
3809 /* Tally up the load of all CPUs in the group */
3810 max_cpu_load = 0;
3811 min_cpu_load = ~0UL;
3812 max_nr_running = 0;
3813
3814 for_each_cpu_and(i, sched_group_cpus(group), cpus) {
3815 struct rq *rq = cpu_rq(i);
3816
3817 /* Bias balancing toward cpus of our domain */
3818 if (local_group) {
3819 if (idle_cpu(i) && !first_idle_cpu) {
3820 first_idle_cpu = 1;
3821 balance_cpu = i;
3822 }
3823
3824 load = target_load(i, load_idx);
3825 } else {
3826 load = source_load(i, load_idx);
3827 if (load > max_cpu_load) {
3828 max_cpu_load = load;
3829 max_nr_running = rq->nr_running;
3830 }
3831 if (min_cpu_load > load)
3832 min_cpu_load = load;
3833 }
3834
3835 sgs->group_load += load;
3836 sgs->sum_nr_running += rq->nr_running;
3837 sgs->sum_weighted_load += weighted_cpuload(i);
3838 if (idle_cpu(i))
3839 sgs->idle_cpus++;
3840 }
3841
3842 /*
3843 * First idle cpu or the first cpu(busiest) in this sched group
3844 * is eligible for doing load balancing at this and above
3845 * domains. In the newly idle case, we will allow all the cpu's
3846 * to do the newly idle load balance.
3847 */
3848 if (local_group) {
3849 if (idle != CPU_NEWLY_IDLE) {
3850 if (balance_cpu != this_cpu) {
3851 *balance = 0;
3852 return;
3853 }
3854 update_group_power(sd, this_cpu);
3855 } else if (time_after_eq(jiffies, group->sgp->next_update))
3856 update_group_power(sd, this_cpu);
3857 }
3858
3859 /* Adjust by relative CPU power of the group */
3860 sgs->avg_load = (sgs->group_load*SCHED_POWER_SCALE) / group->sgp->power;
3861
3862 /*
3863 * Consider the group unbalanced when the imbalance is larger
3864 * than the average weight of a task.
3865 *
3866 * APZ: with cgroup the avg task weight can vary wildly and
3867 * might not be a suitable number - should we keep a
3868 * normalized nr_running number somewhere that negates
3869 * the hierarchy?
3870 */
3871 if (sgs->sum_nr_running)
3872 avg_load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
3873
3874 if ((max_cpu_load - min_cpu_load) >= avg_load_per_task && max_nr_running > 1)
3875 sgs->group_imb = 1;
3876
3877 sgs->group_capacity = DIV_ROUND_CLOSEST(group->sgp->power,
3878 SCHED_POWER_SCALE);
3879 if (!sgs->group_capacity)
3880 sgs->group_capacity = fix_small_capacity(sd, group);
3881 sgs->group_weight = group->group_weight;
3882
3883 if (sgs->group_capacity > sgs->sum_nr_running)
3884 sgs->group_has_capacity = 1;
3885 }
3886
3887 /**
3888 * update_sd_pick_busiest - return 1 on busiest group
3889 * @sd: sched_domain whose statistics are to be checked
3890 * @sds: sched_domain statistics
3891 * @sg: sched_group candidate to be checked for being the busiest
3892 * @sgs: sched_group statistics
3893 * @this_cpu: the current cpu
3894 *
3895 * Determine if @sg is a busier group than the previously selected
3896 * busiest group.
3897 */
update_sd_pick_busiest(struct sched_domain * sd,struct sd_lb_stats * sds,struct sched_group * sg,struct sg_lb_stats * sgs,int this_cpu)3898 static bool update_sd_pick_busiest(struct sched_domain *sd,
3899 struct sd_lb_stats *sds,
3900 struct sched_group *sg,
3901 struct sg_lb_stats *sgs,
3902 int this_cpu)
3903 {
3904 if (sgs->avg_load <= sds->max_load)
3905 return false;
3906
3907 if (sgs->sum_nr_running > sgs->group_capacity)
3908 return true;
3909
3910 if (sgs->group_imb)
3911 return true;
3912
3913 /*
3914 * ASYM_PACKING needs to move all the work to the lowest
3915 * numbered CPUs in the group, therefore mark all groups
3916 * higher than ourself as busy.
3917 */
3918 if ((sd->flags & SD_ASYM_PACKING) && sgs->sum_nr_running &&
3919 this_cpu < group_first_cpu(sg)) {
3920 if (!sds->busiest)
3921 return true;
3922
3923 if (group_first_cpu(sds->busiest) > group_first_cpu(sg))
3924 return true;
3925 }
3926
3927 return false;
3928 }
3929
3930 /**
3931 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
3932 * @sd: sched_domain whose statistics are to be updated.
3933 * @this_cpu: Cpu for which load balance is currently performed.
3934 * @idle: Idle status of this_cpu
3935 * @cpus: Set of cpus considered for load balancing.
3936 * @balance: Should we balance.
3937 * @sds: variable to hold the statistics for this sched_domain.
3938 */
update_sd_lb_stats(struct sched_domain * sd,int this_cpu,enum cpu_idle_type idle,const struct cpumask * cpus,int * balance,struct sd_lb_stats * sds)3939 static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu,
3940 enum cpu_idle_type idle, const struct cpumask *cpus,
3941 int *balance, struct sd_lb_stats *sds)
3942 {
3943 struct sched_domain *child = sd->child;
3944 struct sched_group *sg = sd->groups;
3945 struct sg_lb_stats sgs;
3946 int load_idx, prefer_sibling = 0;
3947
3948 if (child && child->flags & SD_PREFER_SIBLING)
3949 prefer_sibling = 1;
3950
3951 init_sd_power_savings_stats(sd, sds, idle);
3952 load_idx = get_sd_load_idx(sd, idle);
3953
3954 do {
3955 int local_group;
3956
3957 local_group = cpumask_test_cpu(this_cpu, sched_group_cpus(sg));
3958 memset(&sgs, 0, sizeof(sgs));
3959 update_sg_lb_stats(sd, sg, this_cpu, idle, load_idx,
3960 local_group, cpus, balance, &sgs);
3961
3962 if (local_group && !(*balance))
3963 return;
3964
3965 sds->total_load += sgs.group_load;
3966 sds->total_pwr += sg->sgp->power;
3967
3968 /*
3969 * In case the child domain prefers tasks go to siblings
3970 * first, lower the sg capacity to one so that we'll try
3971 * and move all the excess tasks away. We lower the capacity
3972 * of a group only if the local group has the capacity to fit
3973 * these excess tasks, i.e. nr_running < group_capacity. The
3974 * extra check prevents the case where you always pull from the
3975 * heaviest group when it is already under-utilized (possible
3976 * with a large weight task outweighs the tasks on the system).
3977 */
3978 if (prefer_sibling && !local_group && sds->this_has_capacity)
3979 sgs.group_capacity = min(sgs.group_capacity, 1UL);
3980
3981 if (local_group) {
3982 sds->this_load = sgs.avg_load;
3983 sds->this = sg;
3984 sds->this_nr_running = sgs.sum_nr_running;
3985 sds->this_load_per_task = sgs.sum_weighted_load;
3986 sds->this_has_capacity = sgs.group_has_capacity;
3987 sds->this_idle_cpus = sgs.idle_cpus;
3988 } else if (update_sd_pick_busiest(sd, sds, sg, &sgs, this_cpu)) {
3989 sds->max_load = sgs.avg_load;
3990 sds->busiest = sg;
3991 sds->busiest_nr_running = sgs.sum_nr_running;
3992 sds->busiest_idle_cpus = sgs.idle_cpus;
3993 sds->busiest_group_capacity = sgs.group_capacity;
3994 sds->busiest_load_per_task = sgs.sum_weighted_load;
3995 sds->busiest_has_capacity = sgs.group_has_capacity;
3996 sds->busiest_group_weight = sgs.group_weight;
3997 sds->group_imb = sgs.group_imb;
3998 }
3999
4000 update_sd_power_savings_stats(sg, sds, local_group, &sgs);
4001 sg = sg->next;
4002 } while (sg != sd->groups);
4003 }
4004
4005 /**
4006 * check_asym_packing - Check to see if the group is packed into the
4007 * sched doman.
4008 *
4009 * This is primarily intended to used at the sibling level. Some
4010 * cores like POWER7 prefer to use lower numbered SMT threads. In the
4011 * case of POWER7, it can move to lower SMT modes only when higher
4012 * threads are idle. When in lower SMT modes, the threads will
4013 * perform better since they share less core resources. Hence when we
4014 * have idle threads, we want them to be the higher ones.
4015 *
4016 * This packing function is run on idle threads. It checks to see if
4017 * the busiest CPU in this domain (core in the P7 case) has a higher
4018 * CPU number than the packing function is being run on. Here we are
4019 * assuming lower CPU number will be equivalent to lower a SMT thread
4020 * number.
4021 *
4022 * Returns 1 when packing is required and a task should be moved to
4023 * this CPU. The amount of the imbalance is returned in *imbalance.
4024 *
4025 * @sd: The sched_domain whose packing is to be checked.
4026 * @sds: Statistics of the sched_domain which is to be packed
4027 * @this_cpu: The cpu at whose sched_domain we're performing load-balance.
4028 * @imbalance: returns amount of imbalanced due to packing.
4029 */
check_asym_packing(struct sched_domain * sd,struct sd_lb_stats * sds,int this_cpu,unsigned long * imbalance)4030 static int check_asym_packing(struct sched_domain *sd,
4031 struct sd_lb_stats *sds,
4032 int this_cpu, unsigned long *imbalance)
4033 {
4034 int busiest_cpu;
4035
4036 if (!(sd->flags & SD_ASYM_PACKING))
4037 return 0;
4038
4039 if (!sds->busiest)
4040 return 0;
4041
4042 busiest_cpu = group_first_cpu(sds->busiest);
4043 if (this_cpu > busiest_cpu)
4044 return 0;
4045
4046 *imbalance = DIV_ROUND_CLOSEST(sds->max_load * sds->busiest->sgp->power,
4047 SCHED_POWER_SCALE);
4048 return 1;
4049 }
4050
4051 /**
4052 * fix_small_imbalance - Calculate the minor imbalance that exists
4053 * amongst the groups of a sched_domain, during
4054 * load balancing.
4055 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
4056 * @this_cpu: The cpu at whose sched_domain we're performing load-balance.
4057 * @imbalance: Variable to store the imbalance.
4058 */
fix_small_imbalance(struct sd_lb_stats * sds,int this_cpu,unsigned long * imbalance)4059 static inline void fix_small_imbalance(struct sd_lb_stats *sds,
4060 int this_cpu, unsigned long *imbalance)
4061 {
4062 unsigned long tmp, pwr_now = 0, pwr_move = 0;
4063 unsigned int imbn = 2;
4064 unsigned long scaled_busy_load_per_task;
4065
4066 if (sds->this_nr_running) {
4067 sds->this_load_per_task /= sds->this_nr_running;
4068 if (sds->busiest_load_per_task >
4069 sds->this_load_per_task)
4070 imbn = 1;
4071 } else
4072 sds->this_load_per_task =
4073 cpu_avg_load_per_task(this_cpu);
4074
4075 scaled_busy_load_per_task = sds->busiest_load_per_task
4076 * SCHED_POWER_SCALE;
4077 scaled_busy_load_per_task /= sds->busiest->sgp->power;
4078
4079 if (sds->max_load - sds->this_load + scaled_busy_load_per_task >=
4080 (scaled_busy_load_per_task * imbn)) {
4081 *imbalance = sds->busiest_load_per_task;
4082 return;
4083 }
4084
4085 /*
4086 * OK, we don't have enough imbalance to justify moving tasks,
4087 * however we may be able to increase total CPU power used by
4088 * moving them.
4089 */
4090
4091 pwr_now += sds->busiest->sgp->power *
4092 min(sds->busiest_load_per_task, sds->max_load);
4093 pwr_now += sds->this->sgp->power *
4094 min(sds->this_load_per_task, sds->this_load);
4095 pwr_now /= SCHED_POWER_SCALE;
4096
4097 /* Amount of load we'd subtract */
4098 tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE) /
4099 sds->busiest->sgp->power;
4100 if (sds->max_load > tmp)
4101 pwr_move += sds->busiest->sgp->power *
4102 min(sds->busiest_load_per_task, sds->max_load - tmp);
4103
4104 /* Amount of load we'd add */
4105 if (sds->max_load * sds->busiest->sgp->power <
4106 sds->busiest_load_per_task * SCHED_POWER_SCALE)
4107 tmp = (sds->max_load * sds->busiest->sgp->power) /
4108 sds->this->sgp->power;
4109 else
4110 tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE) /
4111 sds->this->sgp->power;
4112 pwr_move += sds->this->sgp->power *
4113 min(sds->this_load_per_task, sds->this_load + tmp);
4114 pwr_move /= SCHED_POWER_SCALE;
4115
4116 /* Move if we gain throughput */
4117 if (pwr_move > pwr_now)
4118 *imbalance = sds->busiest_load_per_task;
4119 }
4120
4121 /**
4122 * calculate_imbalance - Calculate the amount of imbalance present within the
4123 * groups of a given sched_domain during load balance.
4124 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
4125 * @this_cpu: Cpu for which currently load balance is being performed.
4126 * @imbalance: The variable to store the imbalance.
4127 */
calculate_imbalance(struct sd_lb_stats * sds,int this_cpu,unsigned long * imbalance)4128 static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu,
4129 unsigned long *imbalance)
4130 {
4131 unsigned long max_pull, load_above_capacity = ~0UL;
4132
4133 sds->busiest_load_per_task /= sds->busiest_nr_running;
4134 if (sds->group_imb) {
4135 sds->busiest_load_per_task =
4136 min(sds->busiest_load_per_task, sds->avg_load);
4137 }
4138
4139 /*
4140 * In the presence of smp nice balancing, certain scenarios can have
4141 * max load less than avg load(as we skip the groups at or below
4142 * its cpu_power, while calculating max_load..)
4143 */
4144 if (sds->max_load < sds->avg_load) {
4145 *imbalance = 0;
4146 return fix_small_imbalance(sds, this_cpu, imbalance);
4147 }
4148
4149 if (!sds->group_imb) {
4150 /*
4151 * Don't want to pull so many tasks that a group would go idle.
4152 */
4153 load_above_capacity = (sds->busiest_nr_running -
4154 sds->busiest_group_capacity);
4155
4156 load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_POWER_SCALE);
4157
4158 load_above_capacity /= sds->busiest->sgp->power;
4159 }
4160
4161 /*
4162 * We're trying to get all the cpus to the average_load, so we don't
4163 * want to push ourselves above the average load, nor do we wish to
4164 * reduce the max loaded cpu below the average load. At the same time,
4165 * we also don't want to reduce the group load below the group capacity
4166 * (so that we can implement power-savings policies etc). Thus we look
4167 * for the minimum possible imbalance.
4168 * Be careful of negative numbers as they'll appear as very large values
4169 * with unsigned longs.
4170 */
4171 max_pull = min(sds->max_load - sds->avg_load, load_above_capacity);
4172
4173 /* How much load to actually move to equalise the imbalance */
4174 *imbalance = min(max_pull * sds->busiest->sgp->power,
4175 (sds->avg_load - sds->this_load) * sds->this->sgp->power)
4176 / SCHED_POWER_SCALE;
4177
4178 /*
4179 * if *imbalance is less than the average load per runnable task
4180 * there is no guarantee that any tasks will be moved so we'll have
4181 * a think about bumping its value to force at least one task to be
4182 * moved
4183 */
4184 if (*imbalance < sds->busiest_load_per_task)
4185 return fix_small_imbalance(sds, this_cpu, imbalance);
4186
4187 }
4188
4189 /******* find_busiest_group() helpers end here *********************/
4190
4191 /**
4192 * find_busiest_group - Returns the busiest group within the sched_domain
4193 * if there is an imbalance. If there isn't an imbalance, and
4194 * the user has opted for power-savings, it returns a group whose
4195 * CPUs can be put to idle by rebalancing those tasks elsewhere, if
4196 * such a group exists.
4197 *
4198 * Also calculates the amount of weighted load which should be moved
4199 * to restore balance.
4200 *
4201 * @sd: The sched_domain whose busiest group is to be returned.
4202 * @this_cpu: The cpu for which load balancing is currently being performed.
4203 * @imbalance: Variable which stores amount of weighted load which should
4204 * be moved to restore balance/put a group to idle.
4205 * @idle: The idle status of this_cpu.
4206 * @cpus: The set of CPUs under consideration for load-balancing.
4207 * @balance: Pointer to a variable indicating if this_cpu
4208 * is the appropriate cpu to perform load balancing at this_level.
4209 *
4210 * Returns: - the busiest group if imbalance exists.
4211 * - If no imbalance and user has opted for power-savings balance,
4212 * return the least loaded group whose CPUs can be
4213 * put to idle by rebalancing its tasks onto our group.
4214 */
4215 static struct sched_group *
find_busiest_group(struct sched_domain * sd,int this_cpu,unsigned long * imbalance,enum cpu_idle_type idle,const struct cpumask * cpus,int * balance)4216 find_busiest_group(struct sched_domain *sd, int this_cpu,
4217 unsigned long *imbalance, enum cpu_idle_type idle,
4218 const struct cpumask *cpus, int *balance)
4219 {
4220 struct sd_lb_stats sds;
4221
4222 memset(&sds, 0, sizeof(sds));
4223
4224 /*
4225 * Compute the various statistics relavent for load balancing at
4226 * this level.
4227 */
4228 update_sd_lb_stats(sd, this_cpu, idle, cpus, balance, &sds);
4229
4230 /*
4231 * this_cpu is not the appropriate cpu to perform load balancing at
4232 * this level.
4233 */
4234 if (!(*balance))
4235 goto ret;
4236
4237 if ((idle == CPU_IDLE || idle == CPU_NEWLY_IDLE) &&
4238 check_asym_packing(sd, &sds, this_cpu, imbalance))
4239 return sds.busiest;
4240
4241 /* There is no busy sibling group to pull tasks from */
4242 if (!sds.busiest || sds.busiest_nr_running == 0)
4243 goto out_balanced;
4244
4245 sds.avg_load = (SCHED_POWER_SCALE * sds.total_load) / sds.total_pwr;
4246
4247 /*
4248 * If the busiest group is imbalanced the below checks don't
4249 * work because they assumes all things are equal, which typically
4250 * isn't true due to cpus_allowed constraints and the like.
4251 */
4252 if (sds.group_imb)
4253 goto force_balance;
4254
4255 /* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
4256 if (idle == CPU_NEWLY_IDLE && sds.this_has_capacity &&
4257 !sds.busiest_has_capacity)
4258 goto force_balance;
4259
4260 /*
4261 * If the local group is more busy than the selected busiest group
4262 * don't try and pull any tasks.
4263 */
4264 if (sds.this_load >= sds.max_load)
4265 goto out_balanced;
4266
4267 /*
4268 * Don't pull any tasks if this group is already above the domain
4269 * average load.
4270 */
4271 if (sds.this_load >= sds.avg_load)
4272 goto out_balanced;
4273
4274 if (idle == CPU_IDLE) {
4275 /*
4276 * This cpu is idle. If the busiest group load doesn't
4277 * have more tasks than the number of available cpu's and
4278 * there is no imbalance between this and busiest group
4279 * wrt to idle cpu's, it is balanced.
4280 */
4281 if ((sds.this_idle_cpus <= sds.busiest_idle_cpus + 1) &&
4282 sds.busiest_nr_running <= sds.busiest_group_weight)
4283 goto out_balanced;
4284 } else {
4285 /*
4286 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
4287 * imbalance_pct to be conservative.
4288 */
4289 if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load)
4290 goto out_balanced;
4291 }
4292
4293 force_balance:
4294 /* Looks like there is an imbalance. Compute it */
4295 calculate_imbalance(&sds, this_cpu, imbalance);
4296 return sds.busiest;
4297
4298 out_balanced:
4299 /*
4300 * There is no obvious imbalance. But check if we can do some balancing
4301 * to save power.
4302 */
4303 if (check_power_save_busiest_group(&sds, this_cpu, imbalance))
4304 return sds.busiest;
4305 ret:
4306 *imbalance = 0;
4307 return NULL;
4308 }
4309
4310 /*
4311 * find_busiest_queue - find the busiest runqueue among the cpus in group.
4312 */
4313 static struct rq *
find_busiest_queue(struct sched_domain * sd,struct sched_group * group,enum cpu_idle_type idle,unsigned long imbalance,const struct cpumask * cpus)4314 find_busiest_queue(struct sched_domain *sd, struct sched_group *group,
4315 enum cpu_idle_type idle, unsigned long imbalance,
4316 const struct cpumask *cpus)
4317 {
4318 struct rq *busiest = NULL, *rq;
4319 unsigned long max_load = 0;
4320 int i;
4321
4322 for_each_cpu(i, sched_group_cpus(group)) {
4323 unsigned long power = power_of(i);
4324 unsigned long capacity = DIV_ROUND_CLOSEST(power,
4325 SCHED_POWER_SCALE);
4326 unsigned long wl;
4327
4328 if (!capacity)
4329 capacity = fix_small_capacity(sd, group);
4330
4331 if (!cpumask_test_cpu(i, cpus))
4332 continue;
4333
4334 rq = cpu_rq(i);
4335 wl = weighted_cpuload(i);
4336
4337 /*
4338 * When comparing with imbalance, use weighted_cpuload()
4339 * which is not scaled with the cpu power.
4340 */
4341 if (capacity && rq->nr_running == 1 && wl > imbalance)
4342 continue;
4343
4344 /*
4345 * For the load comparisons with the other cpu's, consider
4346 * the weighted_cpuload() scaled with the cpu power, so that
4347 * the load can be moved away from the cpu that is potentially
4348 * running at a lower capacity.
4349 */
4350 wl = (wl * SCHED_POWER_SCALE) / power;
4351
4352 if (wl > max_load) {
4353 max_load = wl;
4354 busiest = rq;
4355 }
4356 }
4357
4358 return busiest;
4359 }
4360
4361 /*
4362 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
4363 * so long as it is large enough.
4364 */
4365 #define MAX_PINNED_INTERVAL 512
4366
4367 /* Working cpumask for load_balance and load_balance_newidle. */
4368 DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask);
4369
need_active_balance(struct sched_domain * sd,int idle,int busiest_cpu,int this_cpu)4370 static int need_active_balance(struct sched_domain *sd, int idle,
4371 int busiest_cpu, int this_cpu)
4372 {
4373 if (idle == CPU_NEWLY_IDLE) {
4374
4375 /*
4376 * ASYM_PACKING needs to force migrate tasks from busy but
4377 * higher numbered CPUs in order to pack all tasks in the
4378 * lowest numbered CPUs.
4379 */
4380 if ((sd->flags & SD_ASYM_PACKING) && busiest_cpu > this_cpu)
4381 return 1;
4382
4383 /*
4384 * The only task running in a non-idle cpu can be moved to this
4385 * cpu in an attempt to completely freeup the other CPU
4386 * package.
4387 *
4388 * The package power saving logic comes from
4389 * find_busiest_group(). If there are no imbalance, then
4390 * f_b_g() will return NULL. However when sched_mc={1,2} then
4391 * f_b_g() will select a group from which a running task may be
4392 * pulled to this cpu in order to make the other package idle.
4393 * If there is no opportunity to make a package idle and if
4394 * there are no imbalance, then f_b_g() will return NULL and no
4395 * action will be taken in load_balance_newidle().
4396 *
4397 * Under normal task pull operation due to imbalance, there
4398 * will be more than one task in the source run queue and
4399 * move_tasks() will succeed. ld_moved will be true and this
4400 * active balance code will not be triggered.
4401 */
4402 if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP)
4403 return 0;
4404 }
4405
4406 return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
4407 }
4408
4409 static int active_load_balance_cpu_stop(void *data);
4410
4411 /*
4412 * Check this_cpu to ensure it is balanced within domain. Attempt to move
4413 * tasks if there is an imbalance.
4414 */
load_balance(int this_cpu,struct rq * this_rq,struct sched_domain * sd,enum cpu_idle_type idle,int * balance)4415 static int load_balance(int this_cpu, struct rq *this_rq,
4416 struct sched_domain *sd, enum cpu_idle_type idle,
4417 int *balance)
4418 {
4419 int ld_moved, active_balance = 0;
4420 struct sched_group *group;
4421 unsigned long imbalance;
4422 struct rq *busiest;
4423 unsigned long flags;
4424 struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask);
4425
4426 struct lb_env env = {
4427 .sd = sd,
4428 .dst_cpu = this_cpu,
4429 .dst_rq = this_rq,
4430 .idle = idle,
4431 .loop_break = sched_nr_migrate_break,
4432 };
4433
4434 cpumask_copy(cpus, cpu_active_mask);
4435
4436 schedstat_inc(sd, lb_count[idle]);
4437
4438 redo:
4439 group = find_busiest_group(sd, this_cpu, &imbalance, idle,
4440 cpus, balance);
4441
4442 if (*balance == 0)
4443 goto out_balanced;
4444
4445 if (!group) {
4446 schedstat_inc(sd, lb_nobusyg[idle]);
4447 goto out_balanced;
4448 }
4449
4450 busiest = find_busiest_queue(sd, group, idle, imbalance, cpus);
4451 if (!busiest) {
4452 schedstat_inc(sd, lb_nobusyq[idle]);
4453 goto out_balanced;
4454 }
4455
4456 BUG_ON(busiest == this_rq);
4457
4458 schedstat_add(sd, lb_imbalance[idle], imbalance);
4459
4460 ld_moved = 0;
4461 if (busiest->nr_running > 1) {
4462 /*
4463 * Attempt to move tasks. If find_busiest_group has found
4464 * an imbalance but busiest->nr_running <= 1, the group is
4465 * still unbalanced. ld_moved simply stays zero, so it is
4466 * correctly treated as an imbalance.
4467 */
4468 env.flags |= LBF_ALL_PINNED;
4469 env.load_move = imbalance;
4470 env.src_cpu = busiest->cpu;
4471 env.src_rq = busiest;
4472 env.loop_max = min_t(unsigned long, sysctl_sched_nr_migrate, busiest->nr_running);
4473
4474 more_balance:
4475 local_irq_save(flags);
4476 double_rq_lock(this_rq, busiest);
4477 if (!env.loop)
4478 update_h_load(env.src_cpu);
4479 ld_moved += move_tasks(&env);
4480 double_rq_unlock(this_rq, busiest);
4481 local_irq_restore(flags);
4482
4483 if (env.flags & LBF_NEED_BREAK) {
4484 env.flags &= ~LBF_NEED_BREAK;
4485 goto more_balance;
4486 }
4487
4488 /*
4489 * some other cpu did the load balance for us.
4490 */
4491 if (ld_moved && this_cpu != smp_processor_id())
4492 resched_cpu(this_cpu);
4493
4494 /* All tasks on this runqueue were pinned by CPU affinity */
4495 if (unlikely(env.flags & LBF_ALL_PINNED)) {
4496 cpumask_clear_cpu(cpu_of(busiest), cpus);
4497 if (!cpumask_empty(cpus))
4498 goto redo;
4499 goto out_balanced;
4500 }
4501 }
4502
4503 if (!ld_moved) {
4504 schedstat_inc(sd, lb_failed[idle]);
4505 /*
4506 * Increment the failure counter only on periodic balance.
4507 * We do not want newidle balance, which can be very
4508 * frequent, pollute the failure counter causing
4509 * excessive cache_hot migrations and active balances.
4510 */
4511 if (idle != CPU_NEWLY_IDLE)
4512 sd->nr_balance_failed++;
4513
4514 if (need_active_balance(sd, idle, cpu_of(busiest), this_cpu)) {
4515 raw_spin_lock_irqsave(&busiest->lock, flags);
4516
4517 /* don't kick the active_load_balance_cpu_stop,
4518 * if the curr task on busiest cpu can't be
4519 * moved to this_cpu
4520 */
4521 if (!cpumask_test_cpu(this_cpu,
4522 tsk_cpus_allowed(busiest->curr))) {
4523 raw_spin_unlock_irqrestore(&busiest->lock,
4524 flags);
4525 env.flags |= LBF_ALL_PINNED;
4526 goto out_one_pinned;
4527 }
4528
4529 /*
4530 * ->active_balance synchronizes accesses to
4531 * ->active_balance_work. Once set, it's cleared
4532 * only after active load balance is finished.
4533 */
4534 if (!busiest->active_balance) {
4535 busiest->active_balance = 1;
4536 busiest->push_cpu = this_cpu;
4537 active_balance = 1;
4538 }
4539 raw_spin_unlock_irqrestore(&busiest->lock, flags);
4540
4541 if (active_balance)
4542 stop_one_cpu_nowait(cpu_of(busiest),
4543 active_load_balance_cpu_stop, busiest,
4544 &busiest->active_balance_work);
4545
4546 /*
4547 * We've kicked active balancing, reset the failure
4548 * counter.
4549 */
4550 sd->nr_balance_failed = sd->cache_nice_tries+1;
4551 }
4552 } else
4553 sd->nr_balance_failed = 0;
4554
4555 if (likely(!active_balance)) {
4556 /* We were unbalanced, so reset the balancing interval */
4557 sd->balance_interval = sd->min_interval;
4558 } else {
4559 /*
4560 * If we've begun active balancing, start to back off. This
4561 * case may not be covered by the all_pinned logic if there
4562 * is only 1 task on the busy runqueue (because we don't call
4563 * move_tasks).
4564 */
4565 if (sd->balance_interval < sd->max_interval)
4566 sd->balance_interval *= 2;
4567 }
4568
4569 goto out;
4570
4571 out_balanced:
4572 schedstat_inc(sd, lb_balanced[idle]);
4573
4574 sd->nr_balance_failed = 0;
4575
4576 out_one_pinned:
4577 /* tune up the balancing interval */
4578 if (((env.flags & LBF_ALL_PINNED) &&
4579 sd->balance_interval < MAX_PINNED_INTERVAL) ||
4580 (sd->balance_interval < sd->max_interval))
4581 sd->balance_interval *= 2;
4582
4583 ld_moved = 0;
4584 out:
4585 return ld_moved;
4586 }
4587
4588 /*
4589 * idle_balance is called by schedule() if this_cpu is about to become
4590 * idle. Attempts to pull tasks from other CPUs.
4591 */
idle_balance(int this_cpu,struct rq * this_rq)4592 void idle_balance(int this_cpu, struct rq *this_rq)
4593 {
4594 struct sched_domain *sd;
4595 int pulled_task = 0;
4596 unsigned long next_balance = jiffies + HZ;
4597
4598 this_rq->idle_stamp = this_rq->clock;
4599
4600 if (this_rq->avg_idle < sysctl_sched_migration_cost)
4601 return;
4602
4603 /*
4604 * Drop the rq->lock, but keep IRQ/preempt disabled.
4605 */
4606 raw_spin_unlock(&this_rq->lock);
4607
4608 update_shares(this_cpu);
4609 rcu_read_lock();
4610 for_each_domain(this_cpu, sd) {
4611 unsigned long interval;
4612 int balance = 1;
4613
4614 if (!(sd->flags & SD_LOAD_BALANCE))
4615 continue;
4616
4617 if (sd->flags & SD_BALANCE_NEWIDLE) {
4618 /* If we've pulled tasks over stop searching: */
4619 pulled_task = load_balance(this_cpu, this_rq,
4620 sd, CPU_NEWLY_IDLE, &balance);
4621 }
4622
4623 interval = msecs_to_jiffies(sd->balance_interval);
4624 if (time_after(next_balance, sd->last_balance + interval))
4625 next_balance = sd->last_balance + interval;
4626 if (pulled_task) {
4627 this_rq->idle_stamp = 0;
4628 break;
4629 }
4630 }
4631 rcu_read_unlock();
4632
4633 raw_spin_lock(&this_rq->lock);
4634
4635 if (pulled_task || time_after(jiffies, this_rq->next_balance)) {
4636 /*
4637 * We are going idle. next_balance may be set based on
4638 * a busy processor. So reset next_balance.
4639 */
4640 this_rq->next_balance = next_balance;
4641 }
4642 }
4643
4644 /*
4645 * active_load_balance_cpu_stop is run by cpu stopper. It pushes
4646 * running tasks off the busiest CPU onto idle CPUs. It requires at
4647 * least 1 task to be running on each physical CPU where possible, and
4648 * avoids physical / logical imbalances.
4649 */
active_load_balance_cpu_stop(void * data)4650 static int active_load_balance_cpu_stop(void *data)
4651 {
4652 struct rq *busiest_rq = data;
4653 int busiest_cpu = cpu_of(busiest_rq);
4654 int target_cpu = busiest_rq->push_cpu;
4655 struct rq *target_rq = cpu_rq(target_cpu);
4656 struct sched_domain *sd;
4657
4658 raw_spin_lock_irq(&busiest_rq->lock);
4659
4660 /* make sure the requested cpu hasn't gone down in the meantime */
4661 if (unlikely(busiest_cpu != smp_processor_id() ||
4662 !busiest_rq->active_balance))
4663 goto out_unlock;
4664
4665 /* Is there any task to move? */
4666 if (busiest_rq->nr_running <= 1)
4667 goto out_unlock;
4668
4669 /*
4670 * This condition is "impossible", if it occurs
4671 * we need to fix it. Originally reported by
4672 * Bjorn Helgaas on a 128-cpu setup.
4673 */
4674 BUG_ON(busiest_rq == target_rq);
4675
4676 /* move a task from busiest_rq to target_rq */
4677 double_lock_balance(busiest_rq, target_rq);
4678
4679 /* Search for an sd spanning us and the target CPU. */
4680 rcu_read_lock();
4681 for_each_domain(target_cpu, sd) {
4682 if ((sd->flags & SD_LOAD_BALANCE) &&
4683 cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
4684 break;
4685 }
4686
4687 if (likely(sd)) {
4688 struct lb_env env = {
4689 .sd = sd,
4690 .dst_cpu = target_cpu,
4691 .dst_rq = target_rq,
4692 .src_cpu = busiest_rq->cpu,
4693 .src_rq = busiest_rq,
4694 .idle = CPU_IDLE,
4695 };
4696
4697 schedstat_inc(sd, alb_count);
4698
4699 if (move_one_task(&env))
4700 schedstat_inc(sd, alb_pushed);
4701 else
4702 schedstat_inc(sd, alb_failed);
4703 }
4704 rcu_read_unlock();
4705 double_unlock_balance(busiest_rq, target_rq);
4706 out_unlock:
4707 busiest_rq->active_balance = 0;
4708 raw_spin_unlock_irq(&busiest_rq->lock);
4709 return 0;
4710 }
4711
4712 #ifdef CONFIG_NO_HZ
4713 /*
4714 * idle load balancing details
4715 * - When one of the busy CPUs notice that there may be an idle rebalancing
4716 * needed, they will kick the idle load balancer, which then does idle
4717 * load balancing for all the idle CPUs.
4718 */
4719 static struct {
4720 cpumask_var_t idle_cpus_mask;
4721 atomic_t nr_cpus;
4722 unsigned long next_balance; /* in jiffy units */
4723 } nohz ____cacheline_aligned;
4724
4725 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
4726 /**
4727 * lowest_flag_domain - Return lowest sched_domain containing flag.
4728 * @cpu: The cpu whose lowest level of sched domain is to
4729 * be returned.
4730 * @flag: The flag to check for the lowest sched_domain
4731 * for the given cpu.
4732 *
4733 * Returns the lowest sched_domain of a cpu which contains the given flag.
4734 */
lowest_flag_domain(int cpu,int flag)4735 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
4736 {
4737 struct sched_domain *sd;
4738
4739 for_each_domain(cpu, sd)
4740 if (sd->flags & flag)
4741 break;
4742
4743 return sd;
4744 }
4745
4746 /**
4747 * for_each_flag_domain - Iterates over sched_domains containing the flag.
4748 * @cpu: The cpu whose domains we're iterating over.
4749 * @sd: variable holding the value of the power_savings_sd
4750 * for cpu.
4751 * @flag: The flag to filter the sched_domains to be iterated.
4752 *
4753 * Iterates over all the scheduler domains for a given cpu that has the 'flag'
4754 * set, starting from the lowest sched_domain to the highest.
4755 */
4756 #define for_each_flag_domain(cpu, sd, flag) \
4757 for (sd = lowest_flag_domain(cpu, flag); \
4758 (sd && (sd->flags & flag)); sd = sd->parent)
4759
4760 /**
4761 * find_new_ilb - Finds the optimum idle load balancer for nomination.
4762 * @cpu: The cpu which is nominating a new idle_load_balancer.
4763 *
4764 * Returns: Returns the id of the idle load balancer if it exists,
4765 * Else, returns >= nr_cpu_ids.
4766 *
4767 * This algorithm picks the idle load balancer such that it belongs to a
4768 * semi-idle powersavings sched_domain. The idea is to try and avoid
4769 * completely idle packages/cores just for the purpose of idle load balancing
4770 * when there are other idle cpu's which are better suited for that job.
4771 */
find_new_ilb(int cpu)4772 static int find_new_ilb(int cpu)
4773 {
4774 int ilb = cpumask_first(nohz.idle_cpus_mask);
4775 struct sched_group *ilbg;
4776 struct sched_domain *sd;
4777
4778 /*
4779 * Have idle load balancer selection from semi-idle packages only
4780 * when power-aware load balancing is enabled
4781 */
4782 if (!(sched_smt_power_savings || sched_mc_power_savings))
4783 goto out_done;
4784
4785 /*
4786 * Optimize for the case when we have no idle CPUs or only one
4787 * idle CPU. Don't walk the sched_domain hierarchy in such cases
4788 */
4789 if (cpumask_weight(nohz.idle_cpus_mask) < 2)
4790 goto out_done;
4791
4792 rcu_read_lock();
4793 for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) {
4794 ilbg = sd->groups;
4795
4796 do {
4797 if (ilbg->group_weight !=
4798 atomic_read(&ilbg->sgp->nr_busy_cpus)) {
4799 ilb = cpumask_first_and(nohz.idle_cpus_mask,
4800 sched_group_cpus(ilbg));
4801 goto unlock;
4802 }
4803
4804 ilbg = ilbg->next;
4805
4806 } while (ilbg != sd->groups);
4807 }
4808 unlock:
4809 rcu_read_unlock();
4810
4811 out_done:
4812 if (ilb < nr_cpu_ids && idle_cpu(ilb))
4813 return ilb;
4814
4815 return nr_cpu_ids;
4816 }
4817 #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */
find_new_ilb(int call_cpu)4818 static inline int find_new_ilb(int call_cpu)
4819 {
4820 return nr_cpu_ids;
4821 }
4822 #endif
4823
4824 /*
4825 * Kick a CPU to do the nohz balancing, if it is time for it. We pick the
4826 * nohz_load_balancer CPU (if there is one) otherwise fallback to any idle
4827 * CPU (if there is one).
4828 */
nohz_balancer_kick(int cpu)4829 static void nohz_balancer_kick(int cpu)
4830 {
4831 int ilb_cpu;
4832
4833 nohz.next_balance++;
4834
4835 ilb_cpu = find_new_ilb(cpu);
4836
4837 if (ilb_cpu >= nr_cpu_ids)
4838 return;
4839
4840 if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
4841 return;
4842 /*
4843 * Use smp_send_reschedule() instead of resched_cpu().
4844 * This way we generate a sched IPI on the target cpu which
4845 * is idle. And the softirq performing nohz idle load balance
4846 * will be run before returning from the IPI.
4847 */
4848 smp_send_reschedule(ilb_cpu);
4849 return;
4850 }
4851
clear_nohz_tick_stopped(int cpu)4852 static inline void clear_nohz_tick_stopped(int cpu)
4853 {
4854 if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
4855 cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
4856 atomic_dec(&nohz.nr_cpus);
4857 clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
4858 }
4859 }
4860
set_cpu_sd_state_busy(void)4861 static inline void set_cpu_sd_state_busy(void)
4862 {
4863 struct sched_domain *sd;
4864 int cpu = smp_processor_id();
4865
4866 if (!test_bit(NOHZ_IDLE, nohz_flags(cpu)))
4867 return;
4868 clear_bit(NOHZ_IDLE, nohz_flags(cpu));
4869
4870 rcu_read_lock();
4871 for_each_domain(cpu, sd)
4872 atomic_inc(&sd->groups->sgp->nr_busy_cpus);
4873 rcu_read_unlock();
4874 }
4875
set_cpu_sd_state_idle(void)4876 void set_cpu_sd_state_idle(void)
4877 {
4878 struct sched_domain *sd;
4879 int cpu = smp_processor_id();
4880
4881 if (test_bit(NOHZ_IDLE, nohz_flags(cpu)))
4882 return;
4883 set_bit(NOHZ_IDLE, nohz_flags(cpu));
4884
4885 rcu_read_lock();
4886 for_each_domain(cpu, sd)
4887 atomic_dec(&sd->groups->sgp->nr_busy_cpus);
4888 rcu_read_unlock();
4889 }
4890
4891 /*
4892 * This routine will record that this cpu is going idle with tick stopped.
4893 * This info will be used in performing idle load balancing in the future.
4894 */
select_nohz_load_balancer(int stop_tick)4895 void select_nohz_load_balancer(int stop_tick)
4896 {
4897 int cpu = smp_processor_id();
4898
4899 /*
4900 * If this cpu is going down, then nothing needs to be done.
4901 */
4902 if (!cpu_active(cpu))
4903 return;
4904
4905 if (stop_tick) {
4906 if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
4907 return;
4908
4909 cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
4910 atomic_inc(&nohz.nr_cpus);
4911 set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
4912 }
4913 return;
4914 }
4915
sched_ilb_notifier(struct notifier_block * nfb,unsigned long action,void * hcpu)4916 static int __cpuinit sched_ilb_notifier(struct notifier_block *nfb,
4917 unsigned long action, void *hcpu)
4918 {
4919 switch (action & ~CPU_TASKS_FROZEN) {
4920 case CPU_DYING:
4921 clear_nohz_tick_stopped(smp_processor_id());
4922 return NOTIFY_OK;
4923 default:
4924 return NOTIFY_DONE;
4925 }
4926 }
4927 #endif
4928
4929 static DEFINE_SPINLOCK(balancing);
4930
4931 /*
4932 * Scale the max load_balance interval with the number of CPUs in the system.
4933 * This trades load-balance latency on larger machines for less cross talk.
4934 */
update_max_interval(void)4935 void update_max_interval(void)
4936 {
4937 max_load_balance_interval = HZ*num_online_cpus()/10;
4938 }
4939
4940 /*
4941 * It checks each scheduling domain to see if it is due to be balanced,
4942 * and initiates a balancing operation if so.
4943 *
4944 * Balancing parameters are set up in arch_init_sched_domains.
4945 */
rebalance_domains(int cpu,enum cpu_idle_type idle)4946 static void rebalance_domains(int cpu, enum cpu_idle_type idle)
4947 {
4948 int balance = 1;
4949 struct rq *rq = cpu_rq(cpu);
4950 unsigned long interval;
4951 struct sched_domain *sd;
4952 /* Earliest time when we have to do rebalance again */
4953 unsigned long next_balance = jiffies + 60*HZ;
4954 int update_next_balance = 0;
4955 int need_serialize;
4956
4957 update_shares(cpu);
4958
4959 rcu_read_lock();
4960 for_each_domain(cpu, sd) {
4961 if (!(sd->flags & SD_LOAD_BALANCE))
4962 continue;
4963
4964 interval = sd->balance_interval;
4965 if (idle != CPU_IDLE)
4966 interval *= sd->busy_factor;
4967
4968 /* scale ms to jiffies */
4969 interval = msecs_to_jiffies(interval);
4970 interval = clamp(interval, 1UL, max_load_balance_interval);
4971
4972 need_serialize = sd->flags & SD_SERIALIZE;
4973
4974 if (need_serialize) {
4975 if (!spin_trylock(&balancing))
4976 goto out;
4977 }
4978
4979 if (time_after_eq(jiffies, sd->last_balance + interval)) {
4980 if (load_balance(cpu, rq, sd, idle, &balance)) {
4981 /*
4982 * We've pulled tasks over so either we're no
4983 * longer idle.
4984 */
4985 idle = CPU_NOT_IDLE;
4986 }
4987 sd->last_balance = jiffies;
4988 }
4989 if (need_serialize)
4990 spin_unlock(&balancing);
4991 out:
4992 if (time_after(next_balance, sd->last_balance + interval)) {
4993 next_balance = sd->last_balance + interval;
4994 update_next_balance = 1;
4995 }
4996
4997 /*
4998 * Stop the load balance at this level. There is another
4999 * CPU in our sched group which is doing load balancing more
5000 * actively.
5001 */
5002 if (!balance)
5003 break;
5004 }
5005 rcu_read_unlock();
5006
5007 /*
5008 * next_balance will be updated only when there is a need.
5009 * When the cpu is attached to null domain for ex, it will not be
5010 * updated.
5011 */
5012 if (likely(update_next_balance))
5013 rq->next_balance = next_balance;
5014 }
5015
5016 #ifdef CONFIG_NO_HZ
5017 /*
5018 * In CONFIG_NO_HZ case, the idle balance kickee will do the
5019 * rebalancing for all the cpus for whom scheduler ticks are stopped.
5020 */
nohz_idle_balance(int this_cpu,enum cpu_idle_type idle)5021 static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle)
5022 {
5023 struct rq *this_rq = cpu_rq(this_cpu);
5024 struct rq *rq;
5025 int balance_cpu;
5026
5027 if (idle != CPU_IDLE ||
5028 !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
5029 goto end;
5030
5031 for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
5032 if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
5033 continue;
5034
5035 /*
5036 * If this cpu gets work to do, stop the load balancing
5037 * work being done for other cpus. Next load
5038 * balancing owner will pick it up.
5039 */
5040 if (need_resched())
5041 break;
5042
5043 raw_spin_lock_irq(&this_rq->lock);
5044 update_rq_clock(this_rq);
5045 update_idle_cpu_load(this_rq);
5046 raw_spin_unlock_irq(&this_rq->lock);
5047
5048 rebalance_domains(balance_cpu, CPU_IDLE);
5049
5050 rq = cpu_rq(balance_cpu);
5051 if (time_after(this_rq->next_balance, rq->next_balance))
5052 this_rq->next_balance = rq->next_balance;
5053 }
5054 nohz.next_balance = this_rq->next_balance;
5055 end:
5056 clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
5057 }
5058
5059 /*
5060 * Current heuristic for kicking the idle load balancer in the presence
5061 * of an idle cpu is the system.
5062 * - This rq has more than one task.
5063 * - At any scheduler domain level, this cpu's scheduler group has multiple
5064 * busy cpu's exceeding the group's power.
5065 * - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
5066 * domain span are idle.
5067 */
nohz_kick_needed(struct rq * rq,int cpu)5068 static inline int nohz_kick_needed(struct rq *rq, int cpu)
5069 {
5070 unsigned long now = jiffies;
5071 struct sched_domain *sd;
5072
5073 if (unlikely(idle_cpu(cpu)))
5074 return 0;
5075
5076 /*
5077 * We may be recently in ticked or tickless idle mode. At the first
5078 * busy tick after returning from idle, we will update the busy stats.
5079 */
5080 set_cpu_sd_state_busy();
5081 clear_nohz_tick_stopped(cpu);
5082
5083 /*
5084 * None are in tickless mode and hence no need for NOHZ idle load
5085 * balancing.
5086 */
5087 if (likely(!atomic_read(&nohz.nr_cpus)))
5088 return 0;
5089
5090 if (time_before(now, nohz.next_balance))
5091 return 0;
5092
5093 if (rq->nr_running >= 2)
5094 goto need_kick;
5095
5096 rcu_read_lock();
5097 for_each_domain(cpu, sd) {
5098 struct sched_group *sg = sd->groups;
5099 struct sched_group_power *sgp = sg->sgp;
5100 int nr_busy = atomic_read(&sgp->nr_busy_cpus);
5101
5102 if (sd->flags & SD_SHARE_PKG_RESOURCES && nr_busy > 1)
5103 goto need_kick_unlock;
5104
5105 if (sd->flags & SD_ASYM_PACKING && nr_busy != sg->group_weight
5106 && (cpumask_first_and(nohz.idle_cpus_mask,
5107 sched_domain_span(sd)) < cpu))
5108 goto need_kick_unlock;
5109
5110 if (!(sd->flags & (SD_SHARE_PKG_RESOURCES | SD_ASYM_PACKING)))
5111 break;
5112 }
5113 rcu_read_unlock();
5114 return 0;
5115
5116 need_kick_unlock:
5117 rcu_read_unlock();
5118 need_kick:
5119 return 1;
5120 }
5121 #else
nohz_idle_balance(int this_cpu,enum cpu_idle_type idle)5122 static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle) { }
5123 #endif
5124
5125 /*
5126 * run_rebalance_domains is triggered when needed from the scheduler tick.
5127 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
5128 */
run_rebalance_domains(struct softirq_action * h)5129 static void run_rebalance_domains(struct softirq_action *h)
5130 {
5131 int this_cpu = smp_processor_id();
5132 struct rq *this_rq = cpu_rq(this_cpu);
5133 enum cpu_idle_type idle = this_rq->idle_balance ?
5134 CPU_IDLE : CPU_NOT_IDLE;
5135
5136 rebalance_domains(this_cpu, idle);
5137
5138 /*
5139 * If this cpu has a pending nohz_balance_kick, then do the
5140 * balancing on behalf of the other idle cpus whose ticks are
5141 * stopped.
5142 */
5143 nohz_idle_balance(this_cpu, idle);
5144 }
5145
on_null_domain(int cpu)5146 static inline int on_null_domain(int cpu)
5147 {
5148 return !rcu_dereference_sched(cpu_rq(cpu)->sd);
5149 }
5150
5151 /*
5152 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
5153 */
trigger_load_balance(struct rq * rq,int cpu)5154 void trigger_load_balance(struct rq *rq, int cpu)
5155 {
5156 /* Don't need to rebalance while attached to NULL domain */
5157 if (time_after_eq(jiffies, rq->next_balance) &&
5158 likely(!on_null_domain(cpu)))
5159 raise_softirq(SCHED_SOFTIRQ);
5160 #ifdef CONFIG_NO_HZ
5161 if (nohz_kick_needed(rq, cpu) && likely(!on_null_domain(cpu)))
5162 nohz_balancer_kick(cpu);
5163 #endif
5164 }
5165
rq_online_fair(struct rq * rq)5166 static void rq_online_fair(struct rq *rq)
5167 {
5168 update_sysctl();
5169 }
5170
rq_offline_fair(struct rq * rq)5171 static void rq_offline_fair(struct rq *rq)
5172 {
5173 update_sysctl();
5174
5175 /* Ensure any throttled groups are reachable by pick_next_task */
5176 unthrottle_offline_cfs_rqs(rq);
5177 }
5178
5179 #endif /* CONFIG_SMP */
5180
5181 /*
5182 * scheduler tick hitting a task of our scheduling class:
5183 */
task_tick_fair(struct rq * rq,struct task_struct * curr,int queued)5184 static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
5185 {
5186 struct cfs_rq *cfs_rq;
5187 struct sched_entity *se = &curr->se;
5188
5189 for_each_sched_entity(se) {
5190 cfs_rq = cfs_rq_of(se);
5191 entity_tick(cfs_rq, se, queued);
5192 }
5193 }
5194
5195 /*
5196 * called on fork with the child task as argument from the parent's context
5197 * - child not yet on the tasklist
5198 * - preemption disabled
5199 */
task_fork_fair(struct task_struct * p)5200 static void task_fork_fair(struct task_struct *p)
5201 {
5202 struct cfs_rq *cfs_rq;
5203 struct sched_entity *se = &p->se, *curr;
5204 int this_cpu = smp_processor_id();
5205 struct rq *rq = this_rq();
5206 unsigned long flags;
5207
5208 raw_spin_lock_irqsave(&rq->lock, flags);
5209
5210 update_rq_clock(rq);
5211
5212 cfs_rq = task_cfs_rq(current);
5213 curr = cfs_rq->curr;
5214
5215 /*
5216 * Not only the cpu but also the task_group of the parent might have
5217 * been changed after parent->se.parent,cfs_rq were copied to
5218 * child->se.parent,cfs_rq. So call __set_task_cpu() to make those
5219 * of child point to valid ones.
5220 */
5221 rcu_read_lock();
5222 __set_task_cpu(p, this_cpu);
5223 rcu_read_unlock();
5224
5225 update_curr(cfs_rq);
5226
5227 if (curr)
5228 se->vruntime = curr->vruntime;
5229 place_entity(cfs_rq, se, 1);
5230
5231 if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
5232 /*
5233 * Upon rescheduling, sched_class::put_prev_task() will place
5234 * 'current' within the tree based on its new key value.
5235 */
5236 swap(curr->vruntime, se->vruntime);
5237 resched_task(rq->curr);
5238 }
5239
5240 se->vruntime -= cfs_rq->min_vruntime;
5241
5242 raw_spin_unlock_irqrestore(&rq->lock, flags);
5243 }
5244
5245 /*
5246 * Priority of the task has changed. Check to see if we preempt
5247 * the current task.
5248 */
5249 static void
prio_changed_fair(struct rq * rq,struct task_struct * p,int oldprio)5250 prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
5251 {
5252 if (!p->se.on_rq)
5253 return;
5254
5255 /*
5256 * Reschedule if we are currently running on this runqueue and
5257 * our priority decreased, or if we are not currently running on
5258 * this runqueue and our priority is higher than the current's
5259 */
5260 if (rq->curr == p) {
5261 if (p->prio > oldprio)
5262 resched_task(rq->curr);
5263 } else
5264 check_preempt_curr(rq, p, 0);
5265 }
5266
switched_from_fair(struct rq * rq,struct task_struct * p)5267 static void switched_from_fair(struct rq *rq, struct task_struct *p)
5268 {
5269 struct sched_entity *se = &p->se;
5270 struct cfs_rq *cfs_rq = cfs_rq_of(se);
5271
5272 /*
5273 * Ensure the task's vruntime is normalized, so that when it's
5274 * switched back to the fair class the enqueue_entity(.flags=0) will
5275 * do the right thing.
5276 *
5277 * If it's on_rq, then the dequeue_entity(.flags=0) will already
5278 * have normalized the vruntime, if it's !on_rq, then only when
5279 * the task is sleeping will it still have non-normalized vruntime.
5280 */
5281 if (!p->on_rq && p->state != TASK_RUNNING) {
5282 /*
5283 * Fix up our vruntime so that the current sleep doesn't
5284 * cause 'unlimited' sleep bonus.
5285 */
5286 place_entity(cfs_rq, se, 0);
5287 se->vruntime -= cfs_rq->min_vruntime;
5288 }
5289 }
5290
5291 /*
5292 * We switched to the sched_fair class.
5293 */
switched_to_fair(struct rq * rq,struct task_struct * p)5294 static void switched_to_fair(struct rq *rq, struct task_struct *p)
5295 {
5296 if (!p->se.on_rq)
5297 return;
5298
5299 /*
5300 * We were most likely switched from sched_rt, so
5301 * kick off the schedule if running, otherwise just see
5302 * if we can still preempt the current task.
5303 */
5304 if (rq->curr == p)
5305 resched_task(rq->curr);
5306 else
5307 check_preempt_curr(rq, p, 0);
5308 }
5309
5310 /* Account for a task changing its policy or group.
5311 *
5312 * This routine is mostly called to set cfs_rq->curr field when a task
5313 * migrates between groups/classes.
5314 */
set_curr_task_fair(struct rq * rq)5315 static void set_curr_task_fair(struct rq *rq)
5316 {
5317 struct sched_entity *se = &rq->curr->se;
5318
5319 for_each_sched_entity(se) {
5320 struct cfs_rq *cfs_rq = cfs_rq_of(se);
5321
5322 set_next_entity(cfs_rq, se);
5323 /* ensure bandwidth has been allocated on our new cfs_rq */
5324 account_cfs_rq_runtime(cfs_rq, 0);
5325 }
5326 }
5327
init_cfs_rq(struct cfs_rq * cfs_rq)5328 void init_cfs_rq(struct cfs_rq *cfs_rq)
5329 {
5330 cfs_rq->tasks_timeline = RB_ROOT;
5331 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
5332 #ifndef CONFIG_64BIT
5333 cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
5334 #endif
5335 }
5336
5337 #ifdef CONFIG_FAIR_GROUP_SCHED
task_move_group_fair(struct task_struct * p,int on_rq)5338 static void task_move_group_fair(struct task_struct *p, int on_rq)
5339 {
5340 /*
5341 * If the task was not on the rq at the time of this cgroup movement
5342 * it must have been asleep, sleeping tasks keep their ->vruntime
5343 * absolute on their old rq until wakeup (needed for the fair sleeper
5344 * bonus in place_entity()).
5345 *
5346 * If it was on the rq, we've just 'preempted' it, which does convert
5347 * ->vruntime to a relative base.
5348 *
5349 * Make sure both cases convert their relative position when migrating
5350 * to another cgroup's rq. This does somewhat interfere with the
5351 * fair sleeper stuff for the first placement, but who cares.
5352 */
5353 /*
5354 * When !on_rq, vruntime of the task has usually NOT been normalized.
5355 * But there are some cases where it has already been normalized:
5356 *
5357 * - Moving a forked child which is waiting for being woken up by
5358 * wake_up_new_task().
5359 * - Moving a task which has been woken up by try_to_wake_up() and
5360 * waiting for actually being woken up by sched_ttwu_pending().
5361 *
5362 * To prevent boost or penalty in the new cfs_rq caused by delta
5363 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
5364 */
5365 if (!on_rq && (!p->se.sum_exec_runtime || p->state == TASK_WAKING))
5366 on_rq = 1;
5367
5368 if (!on_rq)
5369 p->se.vruntime -= cfs_rq_of(&p->se)->min_vruntime;
5370 set_task_rq(p, task_cpu(p));
5371 if (!on_rq)
5372 p->se.vruntime += cfs_rq_of(&p->se)->min_vruntime;
5373 }
5374
free_fair_sched_group(struct task_group * tg)5375 void free_fair_sched_group(struct task_group *tg)
5376 {
5377 int i;
5378
5379 destroy_cfs_bandwidth(tg_cfs_bandwidth(tg));
5380
5381 for_each_possible_cpu(i) {
5382 if (tg->cfs_rq)
5383 kfree(tg->cfs_rq[i]);
5384 if (tg->se)
5385 kfree(tg->se[i]);
5386 }
5387
5388 kfree(tg->cfs_rq);
5389 kfree(tg->se);
5390 }
5391
alloc_fair_sched_group(struct task_group * tg,struct task_group * parent)5392 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
5393 {
5394 struct cfs_rq *cfs_rq;
5395 struct sched_entity *se;
5396 int i;
5397
5398 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
5399 if (!tg->cfs_rq)
5400 goto err;
5401 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
5402 if (!tg->se)
5403 goto err;
5404
5405 tg->shares = NICE_0_LOAD;
5406
5407 init_cfs_bandwidth(tg_cfs_bandwidth(tg));
5408
5409 for_each_possible_cpu(i) {
5410 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
5411 GFP_KERNEL, cpu_to_node(i));
5412 if (!cfs_rq)
5413 goto err;
5414
5415 se = kzalloc_node(sizeof(struct sched_entity),
5416 GFP_KERNEL, cpu_to_node(i));
5417 if (!se)
5418 goto err_free_rq;
5419
5420 init_cfs_rq(cfs_rq);
5421 init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]);
5422 }
5423
5424 return 1;
5425
5426 err_free_rq:
5427 kfree(cfs_rq);
5428 err:
5429 return 0;
5430 }
5431
unregister_fair_sched_group(struct task_group * tg,int cpu)5432 void unregister_fair_sched_group(struct task_group *tg, int cpu)
5433 {
5434 struct rq *rq = cpu_rq(cpu);
5435 unsigned long flags;
5436
5437 /*
5438 * Only empty task groups can be destroyed; so we can speculatively
5439 * check on_list without danger of it being re-added.
5440 */
5441 if (!tg->cfs_rq[cpu]->on_list)
5442 return;
5443
5444 raw_spin_lock_irqsave(&rq->lock, flags);
5445 list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
5446 raw_spin_unlock_irqrestore(&rq->lock, flags);
5447 }
5448
init_tg_cfs_entry(struct task_group * tg,struct cfs_rq * cfs_rq,struct sched_entity * se,int cpu,struct sched_entity * parent)5449 void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
5450 struct sched_entity *se, int cpu,
5451 struct sched_entity *parent)
5452 {
5453 struct rq *rq = cpu_rq(cpu);
5454
5455 cfs_rq->tg = tg;
5456 cfs_rq->rq = rq;
5457 #ifdef CONFIG_SMP
5458 /* allow initial update_cfs_load() to truncate */
5459 cfs_rq->load_stamp = 1;
5460 #endif
5461 init_cfs_rq_runtime(cfs_rq);
5462
5463 tg->cfs_rq[cpu] = cfs_rq;
5464 tg->se[cpu] = se;
5465
5466 /* se could be NULL for root_task_group */
5467 if (!se)
5468 return;
5469
5470 if (!parent)
5471 se->cfs_rq = &rq->cfs;
5472 else
5473 se->cfs_rq = parent->my_q;
5474
5475 se->my_q = cfs_rq;
5476 /* guarantee group entities always have weight */
5477 update_load_set(&se->load, NICE_0_LOAD);
5478 se->parent = parent;
5479 }
5480
5481 static DEFINE_MUTEX(shares_mutex);
5482
sched_group_set_shares(struct task_group * tg,unsigned long shares)5483 int sched_group_set_shares(struct task_group *tg, unsigned long shares)
5484 {
5485 int i;
5486 unsigned long flags;
5487
5488 /*
5489 * We can't change the weight of the root cgroup.
5490 */
5491 if (!tg->se[0])
5492 return -EINVAL;
5493
5494 shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));
5495
5496 mutex_lock(&shares_mutex);
5497 if (tg->shares == shares)
5498 goto done;
5499
5500 tg->shares = shares;
5501 for_each_possible_cpu(i) {
5502 struct rq *rq = cpu_rq(i);
5503 struct sched_entity *se;
5504
5505 se = tg->se[i];
5506 /* Propagate contribution to hierarchy */
5507 raw_spin_lock_irqsave(&rq->lock, flags);
5508 for_each_sched_entity(se)
5509 update_cfs_shares(group_cfs_rq(se));
5510 raw_spin_unlock_irqrestore(&rq->lock, flags);
5511 }
5512
5513 done:
5514 mutex_unlock(&shares_mutex);
5515 return 0;
5516 }
5517 #else /* CONFIG_FAIR_GROUP_SCHED */
5518
free_fair_sched_group(struct task_group * tg)5519 void free_fair_sched_group(struct task_group *tg) { }
5520
alloc_fair_sched_group(struct task_group * tg,struct task_group * parent)5521 int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
5522 {
5523 return 1;
5524 }
5525
unregister_fair_sched_group(struct task_group * tg,int cpu)5526 void unregister_fair_sched_group(struct task_group *tg, int cpu) { }
5527
5528 #endif /* CONFIG_FAIR_GROUP_SCHED */
5529
5530
get_rr_interval_fair(struct rq * rq,struct task_struct * task)5531 static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
5532 {
5533 struct sched_entity *se = &task->se;
5534 unsigned int rr_interval = 0;
5535
5536 /*
5537 * Time slice is 0 for SCHED_OTHER tasks that are on an otherwise
5538 * idle runqueue:
5539 */
5540 if (rq->cfs.load.weight)
5541 rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
5542
5543 return rr_interval;
5544 }
5545
5546 /*
5547 * All the scheduling class methods:
5548 */
5549 const struct sched_class fair_sched_class = {
5550 .next = &idle_sched_class,
5551 .enqueue_task = enqueue_task_fair,
5552 .dequeue_task = dequeue_task_fair,
5553 .yield_task = yield_task_fair,
5554 .yield_to_task = yield_to_task_fair,
5555
5556 .check_preempt_curr = check_preempt_wakeup,
5557
5558 .pick_next_task = pick_next_task_fair,
5559 .put_prev_task = put_prev_task_fair,
5560
5561 #ifdef CONFIG_SMP
5562 .select_task_rq = select_task_rq_fair,
5563
5564 .rq_online = rq_online_fair,
5565 .rq_offline = rq_offline_fair,
5566
5567 .task_waking = task_waking_fair,
5568 #endif
5569
5570 .set_curr_task = set_curr_task_fair,
5571 .task_tick = task_tick_fair,
5572 .task_fork = task_fork_fair,
5573
5574 .prio_changed = prio_changed_fair,
5575 .switched_from = switched_from_fair,
5576 .switched_to = switched_to_fair,
5577
5578 .get_rr_interval = get_rr_interval_fair,
5579
5580 #ifdef CONFIG_FAIR_GROUP_SCHED
5581 .task_move_group = task_move_group_fair,
5582 #endif
5583 };
5584
5585 #ifdef CONFIG_SCHED_DEBUG
print_cfs_stats(struct seq_file * m,int cpu)5586 void print_cfs_stats(struct seq_file *m, int cpu)
5587 {
5588 struct cfs_rq *cfs_rq;
5589
5590 rcu_read_lock();
5591 for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
5592 print_cfs_rq(m, cpu, cfs_rq);
5593 rcu_read_unlock();
5594 }
5595 #endif
5596
init_sched_fair_class(void)5597 __init void init_sched_fair_class(void)
5598 {
5599 #ifdef CONFIG_SMP
5600 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
5601
5602 #ifdef CONFIG_NO_HZ
5603 nohz.next_balance = jiffies;
5604 zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
5605 cpu_notifier(sched_ilb_notifier, 0);
5606 #endif
5607 #endif /* SMP */
5608
5609 }
5610