fair.c 205.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
/*
 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
 *
 *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 *
 *  Interactivity improvements by Mike Galbraith
 *  (C) 2007 Mike Galbraith <efault@gmx.de>
 *
 *  Various enhancements by Dmitry Adamushko.
 *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
 *
 *  Group scheduling enhancements by Srivatsa Vaddagiri
 *  Copyright IBM Corporation, 2007
 *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
 *
 *  Scaled math optimizations by Thomas Gleixner
 *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
18 19 20
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
21 22
 */

A
Arjan van de Ven 已提交
23
#include <linux/latencytop.h>
24
#include <linux/sched.h>
25
#include <linux/cpumask.h>
26 27 28
#include <linux/slab.h>
#include <linux/profile.h>
#include <linux/interrupt.h>
29
#include <linux/mempolicy.h>
30
#include <linux/migrate.h>
31
#include <linux/task_work.h>
32 33 34 35

#include <trace/events/sched.h>

#include "sched.h"
A
Arjan van de Ven 已提交
36

37
/*
38
 * Targeted preemption latency for CPU-bound tasks:
39
 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
40
 *
41
 * NOTE: this latency value is not the same as the concept of
I
Ingo Molnar 已提交
42 43 44
 * 'timeslice length' - timeslices in CFS are of variable length
 * and have no persistent notion like in traditional, time-slice
 * based scheduling concepts.
45
 *
I
Ingo Molnar 已提交
46 47
 * (to see the precise effective timeslice length of your workload,
 *  run vmstat and monitor the context-switches (cs) field)
48
 */
49 50
unsigned int sysctl_sched_latency = 6000000ULL;
unsigned int normalized_sysctl_sched_latency = 6000000ULL;
51

52 53 54 55 56 57 58 59 60 61 62 63
/*
 * The initial- and re-scaling of tunables is configurable
 * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
 *
 * Options are:
 * SCHED_TUNABLESCALING_NONE - unscaled, always *1
 * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
 * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
 */
enum sched_tunable_scaling sysctl_sched_tunable_scaling
	= SCHED_TUNABLESCALING_LOG;

64
/*
65
 * Minimal preemption granularity for CPU-bound tasks:
66
 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
67
 */
68 69
unsigned int sysctl_sched_min_granularity = 750000ULL;
unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
70 71

/*
72 73
 * is kept at sysctl_sched_latency / sysctl_sched_min_granularity
 */
74
static unsigned int sched_nr_latency = 8;
75 76

/*
77
 * After fork, child runs first. If set to 0 (default) then
78
 * parent will (try to) run first.
79
 */
80
unsigned int sysctl_sched_child_runs_first __read_mostly;
81 82 83

/*
 * SCHED_OTHER wake-up granularity.
84
 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
85 86 87 88 89
 *
 * This option delays the preemption effects of decoupled workloads
 * and reduces their over-scheduling. Synchronous workloads will still
 * have immediate wakeup/sleep latencies.
 */
90
unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
91
unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL;
92

93 94
const_debug unsigned int sysctl_sched_migration_cost = 500000UL;

95 96 97 98 99 100 101
/*
 * The exponential sliding  window over which load is averaged for shares
 * distribution.
 * (default: 10msec)
 */
unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;

102 103 104 105 106 107 108 109 110 111 112 113 114 115
#ifdef CONFIG_CFS_BANDWIDTH
/*
 * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
 * each time a cfs_rq requests quota.
 *
 * Note: in the case that the slice exceeds the runtime remaining (either due
 * to consumption or the quota being specified to be smaller than the slice)
 * we will always only issue the remaining available time.
 *
 * default: 5 msec, units: microseconds
  */
unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL;
#endif

116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133
static inline void update_load_add(struct load_weight *lw, unsigned long inc)
{
	lw->weight += inc;
	lw->inv_weight = 0;
}

static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
{
	lw->weight -= dec;
	lw->inv_weight = 0;
}

static inline void update_load_set(struct load_weight *lw, unsigned long w)
{
	lw->weight = w;
	lw->inv_weight = 0;
}

134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
/*
 * Increase the granularity value when there are more CPUs,
 * because with more CPUs the 'effective latency' as visible
 * to users decreases. But the relationship is not linear,
 * so pick a second-best guess by going with the log2 of the
 * number of CPUs.
 *
 * This idea comes from the SD scheduler of Con Kolivas:
 */
static int get_update_sysctl_factor(void)
{
	unsigned int cpus = min_t(int, num_online_cpus(), 8);
	unsigned int factor;

	switch (sysctl_sched_tunable_scaling) {
	case SCHED_TUNABLESCALING_NONE:
		factor = 1;
		break;
	case SCHED_TUNABLESCALING_LINEAR:
		factor = cpus;
		break;
	case SCHED_TUNABLESCALING_LOG:
	default:
		factor = 1 + ilog2(cpus);
		break;
	}

	return factor;
}

static void update_sysctl(void)
{
	unsigned int factor = get_update_sysctl_factor();

#define SET_SYSCTL(name) \
	(sysctl_##name = (factor) * normalized_sysctl_##name)
	SET_SYSCTL(sched_min_granularity);
	SET_SYSCTL(sched_latency);
	SET_SYSCTL(sched_wakeup_granularity);
#undef SET_SYSCTL
}

void sched_init_granularity(void)
{
	update_sysctl();
}

181
#define WMULT_CONST	(~0U)
182 183
#define WMULT_SHIFT	32

184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199
static void __update_inv_weight(struct load_weight *lw)
{
	unsigned long w;

	if (likely(lw->inv_weight))
		return;

	w = scale_load_down(lw->weight);

	if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
		lw->inv_weight = 1;
	else if (unlikely(!w))
		lw->inv_weight = WMULT_CONST;
	else
		lw->inv_weight = WMULT_CONST / w;
}
200 201

/*
202 203 204 205 206 207 208 209 210 211
 * delta_exec * weight / lw.weight
 *   OR
 * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT
 *
 * Either weight := NICE_0_LOAD and lw \e prio_to_wmult[], in which case
 * we're guaranteed shift stays positive because inv_weight is guaranteed to
 * fit 32 bits, and NICE_0_LOAD gives another 10 bits; therefore shift >= 22.
 *
 * Or, weight =< lw.weight (because lw.weight is the runqueue weight), thus
 * weight/lw.weight <= 1, and therefore our shift will also be positive.
212
 */
213
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
214
{
215 216
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
217

218
	__update_inv_weight(lw);
219

220 221 222 223 224
	if (unlikely(fact >> 32)) {
		while (fact >> 32) {
			fact >>= 1;
			shift--;
		}
225 226
	}

227 228
	/* hint to use a 32x32->64 mul */
	fact = (u64)(u32)fact * lw->inv_weight;
229

230 231 232 233
	while (fact >> 32) {
		fact >>= 1;
		shift--;
	}
234

235
	return mul_u64_u32_shr(delta_exec, fact, shift);
236 237 238 239
}


const struct sched_class fair_sched_class;
240

241 242 243 244
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

245
#ifdef CONFIG_FAIR_GROUP_SCHED
246

247
/* cpu runqueue to which this cfs_rq is attached */
248 249
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
250
	return cfs_rq->rq;
251 252
}

253 254
/* An entity is a task if it doesn't "own" a runqueue */
#define entity_is_task(se)	(!se->my_q)
255

256 257 258 259 260 261 262 263
static inline struct task_struct *task_of(struct sched_entity *se)
{
#ifdef CONFIG_SCHED_DEBUG
	WARN_ON_ONCE(!entity_is_task(se));
#endif
	return container_of(se, struct task_struct, se);
}

P
Peter Zijlstra 已提交
264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
/* Walk up scheduling entities hierarchy */
#define for_each_sched_entity(se) \
		for (; se; se = se->parent)

static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
{
	return p->se.cfs_rq;
}

/* runqueue on which this entity is (to be) queued */
static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
{
	return se->cfs_rq;
}

/* runqueue "owned" by this group */
static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
{
	return grp->my_q;
}

285 286
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
				       int force_update);
287

288 289 290
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
291 292 293 294 295 296 297 298 299 300 301 302
		/*
		 * Ensure we either appear before our parent (if already
		 * enqueued) or force our parent to appear after us when it is
		 * enqueued.  The fact that we always enqueue bottom-up
		 * reduces this to two cases.
		 */
		if (cfs_rq->tg->parent &&
		    cfs_rq->tg->parent->cfs_rq[cpu_of(rq_of(cfs_rq))]->on_list) {
			list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
		} else {
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
303
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
304
		}
305 306

		cfs_rq->on_list = 1;
307
		/* We should have no load, but we need to update last_decay. */
308
		update_cfs_rq_blocked_load(cfs_rq, 0);
309 310 311 312 313 314 315 316 317 318 319
	}
}

static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->on_list) {
		list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
		cfs_rq->on_list = 0;
	}
}

P
Peter Zijlstra 已提交
320 321 322 323 324
/* Iterate thr' all leaf cfs_rq's on a runqueue */
#define for_each_leaf_cfs_rq(rq, cfs_rq) \
	list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)

/* Do the two (enqueued) entities belong to the same group ? */
P
Peter Zijlstra 已提交
325
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
326 327 328
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
329
		return se->cfs_rq;
P
Peter Zijlstra 已提交
330

P
Peter Zijlstra 已提交
331
	return NULL;
P
Peter Zijlstra 已提交
332 333 334 335 336 337 338
}

static inline struct sched_entity *parent_entity(struct sched_entity *se)
{
	return se->parent;
}

339 340 341 342 343 344 345 346 347 348 349 350 351
static void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
	int se_depth, pse_depth;

	/*
	 * preemption test can be made between sibling entities who are in the
	 * same cfs_rq i.e who have a common parent. Walk up the hierarchy of
	 * both tasks until we find their ancestors who are siblings of common
	 * parent.
	 */

	/* First walk up until both entities are at same depth */
P
Peter Zijlstra 已提交
352 353
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370

	while (se_depth > pse_depth) {
		se_depth--;
		*se = parent_entity(*se);
	}

	while (pse_depth > se_depth) {
		pse_depth--;
		*pse = parent_entity(*pse);
	}

	while (!is_same_group(*se, *pse)) {
		*se = parent_entity(*se);
		*pse = parent_entity(*pse);
	}
}

371 372 373 374 375 376
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
377

378 379 380
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
381 382 383 384
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
385 386
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
387

P
Peter Zijlstra 已提交
388
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
389
{
P
Peter Zijlstra 已提交
390
	return &task_rq(p)->cfs;
391 392
}

P
Peter Zijlstra 已提交
393 394 395 396 397 398 399 400 401 402 403 404 405 406
static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
{
	struct task_struct *p = task_of(se);
	struct rq *rq = task_rq(p);

	return &rq->cfs;
}

/* runqueue "owned" by this group */
static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
{
	return NULL;
}

407 408 409 410 411 412 413 414
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
}

static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
}

P
Peter Zijlstra 已提交
415 416 417 418 419 420 421 422
#define for_each_leaf_cfs_rq(rq, cfs_rq) \
		for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)

static inline struct sched_entity *parent_entity(struct sched_entity *se)
{
	return NULL;
}

423 424 425 426 427
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
428 429
#endif	/* CONFIG_FAIR_GROUP_SCHED */

430
static __always_inline
431
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
432 433 434 435 436

/**************************************************************
 * Scheduling class tree data structure manipulation methods:
 */

437
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
438
{
439
	s64 delta = (s64)(vruntime - max_vruntime);
440
	if (delta > 0)
441
		max_vruntime = vruntime;
442

443
	return max_vruntime;
444 445
}

446
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
447 448 449 450 451 452 453 454
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

455 456 457 458 459 460
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

461 462 463 464 465 466 467 468 469 470 471 472
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
	u64 vruntime = cfs_rq->min_vruntime;

	if (cfs_rq->curr)
		vruntime = cfs_rq->curr->vruntime;

	if (cfs_rq->rb_leftmost) {
		struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost,
						   struct sched_entity,
						   run_node);

P
Peter Zijlstra 已提交
473
		if (!cfs_rq->curr)
474 475 476 477 478
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

479
	/* ensure we never gain time by being placed backwards. */
480
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
481 482 483 484
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
485 486
}

487 488 489
/*
 * Enqueue an entity into the rb-tree:
 */
490
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506
{
	struct rb_node **link = &cfs_rq->tasks_timeline.rb_node;
	struct rb_node *parent = NULL;
	struct sched_entity *entry;
	int leftmost = 1;

	/*
	 * Find the right place in the rbtree:
	 */
	while (*link) {
		parent = *link;
		entry = rb_entry(parent, struct sched_entity, run_node);
		/*
		 * We dont care about collisions. Nodes with
		 * the same key stay together.
		 */
507
		if (entity_before(se, entry)) {
508 509 510 511 512 513 514 515 516 517 518
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
			leftmost = 0;
		}
	}

	/*
	 * Maintain a cache of leftmost tree entries (it is frequently
	 * used):
	 */
519
	if (leftmost)
I
Ingo Molnar 已提交
520
		cfs_rq->rb_leftmost = &se->run_node;
521 522 523 524 525

	rb_link_node(&se->run_node, parent, link);
	rb_insert_color(&se->run_node, &cfs_rq->tasks_timeline);
}

526
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
527
{
P
Peter Zijlstra 已提交
528 529 530 531 532 533
	if (cfs_rq->rb_leftmost == &se->run_node) {
		struct rb_node *next_node;

		next_node = rb_next(&se->run_node);
		cfs_rq->rb_leftmost = next_node;
	}
I
Ingo Molnar 已提交
534

535 536 537
	rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
}

538
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
539
{
540 541 542 543 544 545
	struct rb_node *left = cfs_rq->rb_leftmost;

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
546 547
}

548 549 550 551 552 553 554 555 556 557 558
static struct sched_entity *__pick_next_entity(struct sched_entity *se)
{
	struct rb_node *next = rb_next(&se->run_node);

	if (!next)
		return NULL;

	return rb_entry(next, struct sched_entity, run_node);
}

#ifdef CONFIG_SCHED_DEBUG
559
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
560
{
561
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
562

563 564
	if (!last)
		return NULL;
565 566

	return rb_entry(last, struct sched_entity, run_node);
567 568
}

569 570 571 572
/**************************************************************
 * Scheduling class statistics methods:
 */

573
int sched_proc_update_handler(struct ctl_table *table, int write,
574
		void __user *buffer, size_t *lenp,
575 576
		loff_t *ppos)
{
577
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
578
	int factor = get_update_sysctl_factor();
579 580 581 582 583 584 585

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

586 587 588 589 590 591 592
#define WRT_SYSCTL(name) \
	(normalized_sysctl_##name = sysctl_##name / (factor))
	WRT_SYSCTL(sched_min_granularity);
	WRT_SYSCTL(sched_latency);
	WRT_SYSCTL(sched_wakeup_granularity);
#undef WRT_SYSCTL

593 594 595
	return 0;
}
#endif
596

597
/*
598
 * delta /= w
599
 */
600
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
601
{
602
	if (unlikely(se->load.weight != NICE_0_LOAD))
603
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
604 605 606 607

	return delta;
}

608 609 610
/*
 * The idea is to set a period in which each task runs once.
 *
611
 * When there are too many tasks (sched_nr_latency) we have to stretch
612 613 614 615
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
616 617 618
static u64 __sched_period(unsigned long nr_running)
{
	u64 period = sysctl_sched_latency;
619
	unsigned long nr_latency = sched_nr_latency;
620 621

	if (unlikely(nr_running > nr_latency)) {
622
		period = sysctl_sched_min_granularity;
623 624 625 626 627 628
		period *= nr_running;
	}

	return period;
}

629 630 631 632
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
633
 * s = p*P[w/rw]
634
 */
P
Peter Zijlstra 已提交
635
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
636
{
M
Mike Galbraith 已提交
637
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
638

M
Mike Galbraith 已提交
639
	for_each_sched_entity(se) {
L
Lin Ming 已提交
640
		struct load_weight *load;
641
		struct load_weight lw;
L
Lin Ming 已提交
642 643 644

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
645

M
Mike Galbraith 已提交
646
		if (unlikely(!se->on_rq)) {
647
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
648 649 650 651

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
652
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
653 654
	}
	return slice;
655 656
}

657
/*
A
Andrei Epure 已提交
658
 * We calculate the vruntime slice of a to-be-inserted task.
659
 *
660
 * vs = s/w
661
 */
662
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
663
{
664
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
665 666
}

667
#ifdef CONFIG_SMP
668
static int select_idle_sibling(struct task_struct *p, int cpu);
669 670
static unsigned long task_h_load(struct task_struct *p);

671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689
static inline void __update_task_entity_contrib(struct sched_entity *se);

/* Give new task start runnable values to heavy its load in infant time */
void init_task_runnable_average(struct task_struct *p)
{
	u32 slice;

	p->se.avg.decay_count = 0;
	slice = sched_slice(task_cfs_rq(p), &p->se) >> 10;
	p->se.avg.runnable_avg_sum = slice;
	p->se.avg.runnable_avg_period = slice;
	__update_task_entity_contrib(&p->se);
}
#else
void init_task_runnable_average(struct task_struct *p)
{
}
#endif

690
/*
691
 * Update the current task's runtime statistics.
692
 */
693
static void update_curr(struct cfs_rq *cfs_rq)
694
{
695
	struct sched_entity *curr = cfs_rq->curr;
696
	u64 now = rq_clock_task(rq_of(cfs_rq));
697
	u64 delta_exec;
698 699 700 701

	if (unlikely(!curr))
		return;

702 703
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
704
		return;
705

I
Ingo Molnar 已提交
706
	curr->exec_start = now;
707

708 709 710 711 712 713 714 715 716
	schedstat_set(curr->statistics.exec_max,
		      max(delta_exec, curr->statistics.exec_max));

	curr->sum_exec_runtime += delta_exec;
	schedstat_add(cfs_rq, exec_clock, delta_exec);

	curr->vruntime += calc_delta_fair(delta_exec, curr);
	update_min_vruntime(cfs_rq);

717 718 719
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

720
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
721
		cpuacct_charge(curtask, delta_exec);
722
		account_group_exec_runtime(curtask, delta_exec);
723
	}
724 725

	account_cfs_rq_runtime(cfs_rq, delta_exec);
726 727 728
}

static inline void
729
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
730
{
731
	schedstat_set(se->statistics.wait_start, rq_clock(rq_of(cfs_rq)));
732 733 734 735 736
}

/*
 * Task is being enqueued - update stats:
 */
737
static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
738 739 740 741 742
{
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
743
	if (se != cfs_rq->curr)
744
		update_stats_wait_start(cfs_rq, se);
745 746 747
}

static void
748
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
749
{
750
	schedstat_set(se->statistics.wait_max, max(se->statistics.wait_max,
751
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start));
752 753
	schedstat_set(se->statistics.wait_count, se->statistics.wait_count + 1);
	schedstat_set(se->statistics.wait_sum, se->statistics.wait_sum +
754
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
755 756 757
#ifdef CONFIG_SCHEDSTATS
	if (entity_is_task(se)) {
		trace_sched_stat_wait(task_of(se),
758
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
759 760
	}
#endif
761
	schedstat_set(se->statistics.wait_start, 0);
762 763 764
}

static inline void
765
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
766 767 768 769 770
{
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
771
	if (se != cfs_rq->curr)
772
		update_stats_wait_end(cfs_rq, se);
773 774 775 776 777 778
}

/*
 * We are picking a new current task - update its stats:
 */
static inline void
779
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
780 781 782 783
{
	/*
	 * We are starting a new run period:
	 */
784
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
785 786 787 788 789 790
}

/**************************************************
 * Scheduling class queueing methods:
 */

791 792
#ifdef CONFIG_NUMA_BALANCING
/*
793 794 795
 * Approximate time to scan a full NUMA task in ms. The task scan period is
 * calculated based on the tasks virtual memory size and
 * numa_balancing_scan_size.
796
 */
797 798
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
799 800 801

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
802

803 804 805
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850
static unsigned int task_nr_scan_windows(struct task_struct *p)
{
	unsigned long rss = 0;
	unsigned long nr_scan_pages;

	/*
	 * Calculations based on RSS as non-present and empty pages are skipped
	 * by the PTE scanner and NUMA hinting faults should be trapped based
	 * on resident pages
	 */
	nr_scan_pages = sysctl_numa_balancing_scan_size << (20 - PAGE_SHIFT);
	rss = get_mm_rss(p->mm);
	if (!rss)
		rss = nr_scan_pages;

	rss = round_up(rss, nr_scan_pages);
	return rss / nr_scan_pages;
}

/* For sanitys sake, never scan more PTEs than MAX_SCAN_WINDOW MB/sec. */
#define MAX_SCAN_WINDOW 2560

static unsigned int task_scan_min(struct task_struct *p)
{
	unsigned int scan, floor;
	unsigned int windows = 1;

	if (sysctl_numa_balancing_scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / sysctl_numa_balancing_scan_size;
	floor = 1000 / windows;

	scan = sysctl_numa_balancing_scan_period_min / task_nr_scan_windows(p);
	return max_t(unsigned int, floor, scan);
}

static unsigned int task_scan_max(struct task_struct *p)
{
	unsigned int smin = task_scan_min(p);
	unsigned int smax;

	/* Watch for min being lower than max due to floor calculations */
	smax = sysctl_numa_balancing_scan_period_max / task_nr_scan_windows(p);
	return max(smin, smax);
}

851 852 853 854 855 856 857 858 859 860 861 862
static void account_numa_enqueue(struct rq *rq, struct task_struct *p)
{
	rq->nr_numa_running += (p->numa_preferred_nid != -1);
	rq->nr_preferred_running += (p->numa_preferred_nid == task_node(p));
}

static void account_numa_dequeue(struct rq *rq, struct task_struct *p)
{
	rq->nr_numa_running -= (p->numa_preferred_nid != -1);
	rq->nr_preferred_running -= (p->numa_preferred_nid == task_node(p));
}

863 864 865 866 867
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
868
	pid_t gid;
869 870 871
	struct list_head task_list;

	struct rcu_head rcu;
872
	nodemask_t active_nodes;
873
	unsigned long total_faults;
874 875 876 877 878
	/*
	 * Faults_cpu is used to decide whether memory should move
	 * towards the CPU. As a consequence, these stats are weighted
	 * more by CPU use than by memory faults.
	 */
879
	unsigned long *faults_cpu;
880
	unsigned long faults[0];
881 882
};

883 884 885 886 887 888 889 890 891
/* Shared or private faults. */
#define NR_NUMA_HINT_FAULT_TYPES 2

/* Memory and CPU locality */
#define NR_NUMA_HINT_FAULT_STATS (NR_NUMA_HINT_FAULT_TYPES * 2)

/* Averaged statistics, and temporary buffers. */
#define NR_NUMA_HINT_FAULT_BUCKETS (NR_NUMA_HINT_FAULT_STATS * 2)

892 893 894 895 896
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

897 898
static inline int task_faults_idx(int nid, int priv)
{
899
	return NR_NUMA_HINT_FAULT_TYPES * nid + priv;
900 901 902 903
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
904
	if (!p->numa_faults_memory)
905 906
		return 0;

907 908
	return p->numa_faults_memory[task_faults_idx(nid, 0)] +
		p->numa_faults_memory[task_faults_idx(nid, 1)];
909 910
}

911 912 913 914 915
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

916 917
	return p->numa_group->faults[task_faults_idx(nid, 0)] +
		p->numa_group->faults[task_faults_idx(nid, 1)];
918 919
}

920 921 922 923 924 925
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
	return group->faults_cpu[task_faults_idx(nid, 0)] +
		group->faults_cpu[task_faults_idx(nid, 1)];
}

926 927 928 929 930 931 932 933 934 935
/*
 * These return the fraction of accesses done by a particular task, or
 * task group, on a particular numa node.  The group weight is given a
 * larger multiplier, in order to group tasks together that are almost
 * evenly spread out between numa nodes.
 */
static inline unsigned long task_weight(struct task_struct *p, int nid)
{
	unsigned long total_faults;

936
	if (!p->numa_faults_memory)
937 938 939 940 941 942 943 944 945 946 947 948
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

	return 1000 * task_faults(p, nid) / total_faults;
}

static inline unsigned long group_weight(struct task_struct *p, int nid)
{
949
	if (!p->numa_group || !p->numa_group->total_faults)
950 951
		return 0;

952
	return 1000 * group_faults(p, nid) / p->numa_group->total_faults;
953 954
}

955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017
bool should_numa_migrate_memory(struct task_struct *p, struct page * page,
				int src_nid, int dst_cpu)
{
	struct numa_group *ng = p->numa_group;
	int dst_nid = cpu_to_node(dst_cpu);
	int last_cpupid, this_cpupid;

	this_cpupid = cpu_pid_to_cpupid(dst_cpu, current->pid);

	/*
	 * Multi-stage node selection is used in conjunction with a periodic
	 * migration fault to build a temporal task<->page relation. By using
	 * a two-stage filter we remove short/unlikely relations.
	 *
	 * Using P(p) ~ n_p / n_t as per frequentist probability, we can equate
	 * a task's usage of a particular page (n_p) per total usage of this
	 * page (n_t) (in a given time-span) to a probability.
	 *
	 * Our periodic faults will sample this probability and getting the
	 * same result twice in a row, given these samples are fully
	 * independent, is then given by P(n)^2, provided our sample period
	 * is sufficiently short compared to the usage pattern.
	 *
	 * This quadric squishes small probabilities, making it less likely we
	 * act on an unlikely task<->page relation.
	 */
	last_cpupid = page_cpupid_xchg_last(page, this_cpupid);
	if (!cpupid_pid_unset(last_cpupid) &&
				cpupid_to_nid(last_cpupid) != dst_nid)
		return false;

	/* Always allow migrate on private faults */
	if (cpupid_match_pid(p, last_cpupid))
		return true;

	/* A shared fault, but p->numa_group has not been set up yet. */
	if (!ng)
		return true;

	/*
	 * Do not migrate if the destination is not a node that
	 * is actively used by this numa group.
	 */
	if (!node_isset(dst_nid, ng->active_nodes))
		return false;

	/*
	 * Source is a node that is not actively used by this
	 * numa group, while the destination is. Migrate.
	 */
	if (!node_isset(src_nid, ng->active_nodes))
		return true;

	/*
	 * Both source and destination are nodes in active
	 * use by this numa group. Maximize memory bandwidth
	 * by migrating from more heavily used groups, to less
	 * heavily used ones, spreading the load around.
	 * Use a 1/4 hysteresis to avoid spurious page movement.
	 */
	return group_faults(p, dst_nid) < (group_faults(p, src_nid) * 3 / 4);
}

1018
static unsigned long weighted_cpuload(const int cpu);
1019 1020
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1021
static unsigned long capacity_of(int cpu);
1022 1023
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1024
/* Cached statistics for all CPUs within a node */
1025
struct numa_stats {
1026
	unsigned long nr_running;
1027
	unsigned long load;
1028 1029

	/* Total compute capacity of CPUs on a node */
1030
	unsigned long compute_capacity;
1031 1032

	/* Approximate capacity in terms of runnable tasks on a node */
1033
	unsigned long task_capacity;
1034
	int has_free_capacity;
1035
};
1036

1037 1038 1039 1040 1041
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1042 1043
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1044 1045 1046 1047 1048 1049 1050

	memset(ns, 0, sizeof(*ns));
	for_each_cpu(cpu, cpumask_of_node(nid)) {
		struct rq *rq = cpu_rq(cpu);

		ns->nr_running += rq->nr_running;
		ns->load += weighted_cpuload(cpu);
1051
		ns->compute_capacity += capacity_of(cpu);
1052 1053

		cpus++;
1054 1055
	}

1056 1057 1058 1059 1060
	/*
	 * If we raced with hotplug and there are no CPUs left in our mask
	 * the @ns structure is NULL'ed and task_numa_compare() will
	 * not find this node attractive.
	 *
1061 1062
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1063 1064 1065 1066
	 */
	if (!cpus)
		return;

1067 1068 1069 1070 1071 1072
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, ns->compute_capacity);
	capacity = cpus / smt; /* cores */

	ns->task_capacity = min_t(unsigned, capacity,
		DIV_ROUND_CLOSEST(ns->compute_capacity, SCHED_CAPACITY_SCALE));
1073
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1074 1075
}

1076 1077
struct task_numa_env {
	struct task_struct *p;
1078

1079 1080
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1081

1082
	struct numa_stats src_stats, dst_stats;
1083

1084
	int imbalance_pct;
1085 1086 1087

	struct task_struct *best_task;
	long best_imp;
1088 1089 1090
	int best_cpu;
};

1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
static void task_numa_assign(struct task_numa_env *env,
			     struct task_struct *p, long imp)
{
	if (env->best_task)
		put_task_struct(env->best_task);
	if (p)
		get_task_struct(p);

	env->best_task = p;
	env->best_imp = imp;
	env->best_cpu = env->dst_cpu;
}

1104
static bool load_too_imbalanced(long src_load, long dst_load,
1105 1106 1107
				struct task_numa_env *env)
{
	long imb, old_imb;
1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
	long orig_src_load, orig_dst_load;
	long src_capacity, dst_capacity;

	/*
	 * The load is corrected for the CPU capacity available on each node.
	 *
	 * src_load        dst_load
	 * ------------ vs ---------
	 * src_capacity    dst_capacity
	 */
	src_capacity = env->src_stats.compute_capacity;
	dst_capacity = env->dst_stats.compute_capacity;
1120 1121 1122 1123 1124 1125

	/* We care about the slope of the imbalance, not the direction. */
	if (dst_load < src_load)
		swap(dst_load, src_load);

	/* Is the difference below the threshold? */
1126 1127
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1128 1129 1130 1131 1132 1133 1134
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
	 * Compare it with the old imbalance.
	 */
1135 1136 1137
	orig_src_load = env->src_stats.load;
	orig_dst_load = env->dst_stats.load;

1138 1139 1140
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);

1141 1142
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;
1143 1144

	/* Would this change make things worse? */
1145
	return (imb > old_imb);
1146 1147
}

1148 1149 1150 1151 1152 1153
/*
 * This checks if the overall compute and NUMA accesses of the system would
 * be improved if the source tasks was migrated to the target dst_cpu taking
 * into account that it might be best if task running on the dst_cpu should
 * be exchanged with the source task
 */
1154 1155
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1156 1157 1158 1159
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1160
	long src_load, dst_load;
1161
	long load;
1162
	long imp = env->p->numa_group ? groupimp : taskimp;
1163
	long moveimp = imp;
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181

	rcu_read_lock();
	cur = ACCESS_ONCE(dst_rq->curr);
	if (cur->pid == 0) /* idle */
		cur = NULL;

	/*
	 * "imp" is the fault differential for the source task between the
	 * source and destination node. Calculate the total differential for
	 * the source task and potential destination task. The more negative
	 * the value is, the more rmeote accesses that would be expected to
	 * be incurred if the tasks were swapped.
	 */
	if (cur) {
		/* Skip this swap candidate if cannot move to the source cpu */
		if (!cpumask_test_cpu(env->src_cpu, tsk_cpus_allowed(cur)))
			goto unlock;

1182 1183
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1184
		 * in any group then look only at task weights.
1185
		 */
1186
		if (cur->numa_group == env->p->numa_group) {
1187 1188
			imp = taskimp + task_weight(cur, env->src_nid) -
			      task_weight(cur, env->dst_nid);
1189 1190 1191 1192 1193 1194
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1195
		} else {
1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206
			/*
			 * Compare the group weights. If a task is all by
			 * itself (not part of a group), use the task weight
			 * instead.
			 */
			if (cur->numa_group)
				imp += group_weight(cur, env->src_nid) -
				       group_weight(cur, env->dst_nid);
			else
				imp += task_weight(cur, env->src_nid) -
				       task_weight(cur, env->dst_nid);
1207
		}
1208 1209
	}

1210
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1211 1212 1213 1214
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1215
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1216
		    !env->dst_stats.has_free_capacity)
1217 1218 1219 1220 1221 1222
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1223 1224
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1225 1226 1227 1228 1229 1230
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1231 1232 1233
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1234

1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
	if (moveimp > imp && moveimp > env->best_imp) {
		/*
		 * If the improvement from just moving env->p direction is
		 * better than swapping tasks around, check if a move is
		 * possible. Store a slightly smaller score than moveimp,
		 * so an actually idle CPU will win.
		 */
		if (!load_too_imbalanced(src_load, dst_load, env)) {
			imp = moveimp - 1;
			cur = NULL;
			goto assign;
		}
	}

	if (imp <= env->best_imp)
		goto unlock;

1252
	if (cur) {
1253 1254 1255
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1256 1257
	}

1258
	if (load_too_imbalanced(src_load, dst_load, env))
1259 1260
		goto unlock;

1261 1262 1263 1264 1265 1266 1267
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
	if (!cur)
		env->dst_cpu = select_idle_sibling(env->p, env->dst_cpu);

1268 1269 1270 1271 1272 1273
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1274 1275
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1276 1277 1278 1279 1280 1281 1282 1283 1284
{
	int cpu;

	for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
		/* Skip this CPU if the source task cannot migrate */
		if (!cpumask_test_cpu(cpu, tsk_cpus_allowed(env->p)))
			continue;

		env->dst_cpu = cpu;
1285
		task_numa_compare(env, taskimp, groupimp);
1286 1287 1288
	}
}

1289 1290 1291 1292
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1293

1294
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1295
		.src_nid = task_node(p),
1296 1297 1298 1299 1300 1301

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
		.best_cpu = -1
1302 1303
	};
	struct sched_domain *sd;
1304
	unsigned long taskweight, groupweight;
1305
	int nid, ret;
1306
	long taskimp, groupimp;
1307

1308
	/*
1309 1310 1311 1312 1313 1314
	 * Pick the lowest SD_NUMA domain, as that would have the smallest
	 * imbalance and would be the first to start moving tasks about.
	 *
	 * And we want to avoid any moving of tasks about, as that would create
	 * random movement of tasks -- counter the numa conditions we're trying
	 * to satisfy here.
1315 1316
	 */
	rcu_read_lock();
1317
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1318 1319
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1320 1321
	rcu_read_unlock();

1322 1323 1324 1325 1326 1327 1328
	/*
	 * Cpusets can break the scheduler domain tree into smaller
	 * balance domains, some of which do not cross NUMA boundaries.
	 * Tasks that are "trapped" in such domains cannot be migrated
	 * elsewhere, so there is no point in (re)trying.
	 */
	if (unlikely(!sd)) {
1329
		p->numa_preferred_nid = task_node(p);
1330 1331 1332
		return -EINVAL;
	}

1333 1334
	taskweight = task_weight(p, env.src_nid);
	groupweight = group_weight(p, env.src_nid);
1335
	update_numa_stats(&env.src_stats, env.src_nid);
1336
	env.dst_nid = p->numa_preferred_nid;
1337 1338
	taskimp = task_weight(p, env.dst_nid) - taskweight;
	groupimp = group_weight(p, env.dst_nid) - groupweight;
1339
	update_numa_stats(&env.dst_stats, env.dst_nid);
1340

1341 1342
	/* Try to find a spot on the preferred nid. */
	task_numa_find_cpu(&env, taskimp, groupimp);
1343 1344 1345

	/* No space available on the preferred nid. Look elsewhere. */
	if (env.best_cpu == -1) {
1346 1347 1348
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1349

1350
			/* Only consider nodes where both task and groups benefit */
1351 1352 1353
			taskimp = task_weight(p, nid) - taskweight;
			groupimp = group_weight(p, nid) - groupweight;
			if (taskimp < 0 && groupimp < 0)
1354 1355
				continue;

1356 1357
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1358
			task_numa_find_cpu(&env, taskimp, groupimp);
1359 1360 1361
		}
	}

1362 1363 1364 1365 1366 1367 1368 1369
	/*
	 * If the task is part of a workload that spans multiple NUMA nodes,
	 * and is migrating into one of the workload's active nodes, remember
	 * this node as the task's preferred numa node, so the workload can
	 * settle down.
	 * A task that migrated to a second choice node will be better off
	 * trying for a better one later. Do not set the preferred node here.
	 */
1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382
	if (p->numa_group) {
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
			nid = env.dst_nid;

		if (node_isset(nid, p->numa_group->active_nodes))
			sched_setnuma(p, env.dst_nid);
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1383

1384 1385 1386 1387 1388 1389
	/*
	 * Reset the scan period if the task is being rescheduled on an
	 * alternative node to recheck if the tasks is now properly placed.
	 */
	p->numa_scan_period = task_scan_min(p);

1390
	if (env.best_task == NULL) {
1391 1392 1393
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1394 1395 1396 1397
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1398 1399
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1400 1401
	put_task_struct(env.best_task);
	return ret;
1402 1403
}

1404 1405 1406
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1407 1408
	unsigned long interval = HZ;

1409
	/* This task has no NUMA fault statistics yet */
1410
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults_memory))
1411 1412
		return;

1413
	/* Periodically retry migrating the task to the preferred node */
1414 1415
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1416 1417

	/* Success if task is already running on preferred CPU */
1418
	if (task_node(p) == p->numa_preferred_nid)
1419 1420 1421
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1422
	task_numa_migrate(p);
1423 1424
}

1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456
/*
 * Find the nodes on which the workload is actively running. We do this by
 * tracking the nodes from which NUMA hinting faults are triggered. This can
 * be different from the set of nodes where the workload's memory is currently
 * located.
 *
 * The bitmask is used to make smarter decisions on when to do NUMA page
 * migrations, To prevent flip-flopping, and excessive page migrations, nodes
 * are added when they cause over 6/16 of the maximum number of faults, but
 * only removed when they drop below 3/16.
 */
static void update_numa_active_node_mask(struct numa_group *numa_group)
{
	unsigned long faults, max_faults = 0;
	int nid;

	for_each_online_node(nid) {
		faults = group_faults_cpu(numa_group, nid);
		if (faults > max_faults)
			max_faults = faults;
	}

	for_each_online_node(nid) {
		faults = group_faults_cpu(numa_group, nid);
		if (!node_isset(nid, numa_group->active_nodes)) {
			if (faults > max_faults * 6 / 16)
				node_set(nid, numa_group->active_nodes);
		} else if (faults < max_faults * 3 / 16)
			node_clear(nid, numa_group->active_nodes);
	}
}

1457 1458 1459
/*
 * When adapting the scan rate, the period is divided into NUMA_PERIOD_SLOTS
 * increments. The more local the fault statistics are, the higher the scan
1460 1461 1462
 * period will be for the next scan window. If local/(local+remote) ratio is
 * below NUMA_PERIOD_THRESHOLD (where range of ratio is 1..NUMA_PERIOD_SLOTS)
 * the scan period will decrease. Aim for 70% local accesses.
1463 1464
 */
#define NUMA_PERIOD_SLOTS 10
1465
#define NUMA_PERIOD_THRESHOLD 7
1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530

/*
 * Increase the scan period (slow down scanning) if the majority of
 * our memory is already on our local node, or if the majority of
 * the page accesses are shared with other processes.
 * Otherwise, decrease the scan period.
 */
static void update_task_scan_period(struct task_struct *p,
			unsigned long shared, unsigned long private)
{
	unsigned int period_slot;
	int ratio;
	int diff;

	unsigned long remote = p->numa_faults_locality[0];
	unsigned long local = p->numa_faults_locality[1];

	/*
	 * If there were no record hinting faults then either the task is
	 * completely idle or all activity is areas that are not of interest
	 * to automatic numa balancing. Scan slower
	 */
	if (local + shared == 0) {
		p->numa_scan_period = min(p->numa_scan_period_max,
			p->numa_scan_period << 1);

		p->mm->numa_next_scan = jiffies +
			msecs_to_jiffies(p->numa_scan_period);

		return;
	}

	/*
	 * Prepare to scale scan period relative to the current period.
	 *	 == NUMA_PERIOD_THRESHOLD scan period stays the same
	 *       <  NUMA_PERIOD_THRESHOLD scan period decreases (scan faster)
	 *	 >= NUMA_PERIOD_THRESHOLD scan period increases (scan slower)
	 */
	period_slot = DIV_ROUND_UP(p->numa_scan_period, NUMA_PERIOD_SLOTS);
	ratio = (local * NUMA_PERIOD_SLOTS) / (local + remote);
	if (ratio >= NUMA_PERIOD_THRESHOLD) {
		int slot = ratio - NUMA_PERIOD_THRESHOLD;
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else {
		diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot;

		/*
		 * Scale scan rate increases based on sharing. There is an
		 * inverse relationship between the degree of sharing and
		 * the adjustment made to the scanning period. Broadly
		 * speaking the intent is that there is little point
		 * scanning faster if shared accesses dominate as it may
		 * simply bounce migrations uselessly
		 */
		ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared));
		diff = (diff * ratio) / NUMA_PERIOD_SLOTS;
	}

	p->numa_scan_period = clamp(p->numa_scan_period + diff,
			task_scan_min(p), task_scan_max(p));
	memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
}

1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558
/*
 * Get the fraction of time the task has been running since the last
 * NUMA placement cycle. The scheduler keeps similar statistics, but
 * decays those on a 32ms period, which is orders of magnitude off
 * from the dozens-of-seconds NUMA balancing period. Use the scheduler
 * stats only if the task is so new there are no NUMA statistics yet.
 */
static u64 numa_get_avg_runtime(struct task_struct *p, u64 *period)
{
	u64 runtime, delta, now;
	/* Use the start of this time slice to avoid calculations. */
	now = p->se.exec_start;
	runtime = p->se.sum_exec_runtime;

	if (p->last_task_numa_placement) {
		delta = runtime - p->last_sum_exec_runtime;
		*period = now - p->last_task_numa_placement;
	} else {
		delta = p->se.avg.runnable_avg_sum;
		*period = p->se.avg.runnable_avg_period;
	}

	p->last_sum_exec_runtime = runtime;
	p->last_task_numa_placement = now;

	return delta;
}

1559 1560
static void task_numa_placement(struct task_struct *p)
{
1561 1562
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
1563
	unsigned long fault_types[2] = { 0, 0 };
1564 1565
	unsigned long total_faults;
	u64 runtime, period;
1566
	spinlock_t *group_lock = NULL;
1567

1568
	seq = ACCESS_ONCE(p->mm->numa_scan_seq);
1569 1570 1571
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
1572
	p->numa_scan_period_max = task_scan_max(p);
1573

1574 1575 1576 1577
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

1578 1579 1580
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
1581
		spin_lock_irq(group_lock);
1582 1583
	}

1584 1585
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
1586
		unsigned long faults = 0, group_faults = 0;
1587
		int priv, i;
1588

1589
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
1590
			long diff, f_diff, f_weight;
1591

1592
			i = task_faults_idx(nid, priv);
1593

1594
			/* Decay existing window, copy faults since last scan */
1595
			diff = p->numa_faults_buffer_memory[i] - p->numa_faults_memory[i] / 2;
1596 1597
			fault_types[priv] += p->numa_faults_buffer_memory[i];
			p->numa_faults_buffer_memory[i] = 0;
1598

1599 1600 1601 1602 1603 1604 1605 1606 1607 1608
			/*
			 * Normalize the faults_from, so all tasks in a group
			 * count according to CPU use, instead of by the raw
			 * number of faults. Tasks with little runtime have
			 * little over-all impact on throughput, and thus their
			 * faults are less important.
			 */
			f_weight = div64_u64(runtime << 16, period + 1);
			f_weight = (f_weight * p->numa_faults_buffer_cpu[i]) /
				   (total_faults + 1);
1609
			f_diff = f_weight - p->numa_faults_cpu[i] / 2;
1610 1611
			p->numa_faults_buffer_cpu[i] = 0;

1612 1613
			p->numa_faults_memory[i] += diff;
			p->numa_faults_cpu[i] += f_diff;
1614
			faults += p->numa_faults_memory[i];
1615
			p->total_numa_faults += diff;
1616 1617
			if (p->numa_group) {
				/* safe because we can only change our own group */
1618
				p->numa_group->faults[i] += diff;
1619
				p->numa_group->faults_cpu[i] += f_diff;
1620 1621
				p->numa_group->total_faults += diff;
				group_faults += p->numa_group->faults[i];
1622
			}
1623 1624
		}

1625 1626 1627 1628
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
1629 1630 1631 1632 1633 1634 1635

		if (group_faults > max_group_faults) {
			max_group_faults = group_faults;
			max_group_nid = nid;
		}
	}

1636 1637
	update_task_scan_period(p, fault_types[0], fault_types[1]);

1638
	if (p->numa_group) {
1639
		update_numa_active_node_mask(p->numa_group);
1640
		spin_unlock_irq(group_lock);
1641
		max_nid = max_group_nid;
1642 1643
	}

1644 1645 1646 1647 1648 1649 1650
	if (max_faults) {
		/* Set the new preferred node */
		if (max_nid != p->numa_preferred_nid)
			sched_setnuma(p, max_nid);

		if (task_node(p) != p->numa_preferred_nid)
			numa_migrate_preferred(p);
1651
	}
1652 1653
}

1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664
static inline int get_numa_group(struct numa_group *grp)
{
	return atomic_inc_not_zero(&grp->refcount);
}

static inline void put_numa_group(struct numa_group *grp)
{
	if (atomic_dec_and_test(&grp->refcount))
		kfree_rcu(grp, rcu);
}

1665 1666
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
1667 1668 1669 1670 1671 1672 1673 1674 1675
{
	struct numa_group *grp, *my_grp;
	struct task_struct *tsk;
	bool join = false;
	int cpu = cpupid_to_cpu(cpupid);
	int i;

	if (unlikely(!p->numa_group)) {
		unsigned int size = sizeof(struct numa_group) +
1676
				    4*nr_node_ids*sizeof(unsigned long);
1677 1678 1679 1680 1681 1682 1683 1684

		grp = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
		if (!grp)
			return;

		atomic_set(&grp->refcount, 1);
		spin_lock_init(&grp->lock);
		INIT_LIST_HEAD(&grp->task_list);
1685
		grp->gid = p->pid;
1686
		/* Second half of the array tracks nids where faults happen */
1687 1688
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
1689

1690 1691
		node_set(task_node(current), grp->active_nodes);

1692
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1693
			grp->faults[i] = p->numa_faults_memory[i];
1694

1695
		grp->total_faults = p->total_numa_faults;
1696

1697 1698 1699 1700 1701 1702 1703 1704 1705
		list_add(&p->numa_entry, &grp->task_list);
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
	tsk = ACCESS_ONCE(cpu_rq(cpu)->curr);

	if (!cpupid_match_pid(tsk, cpupid))
1706
		goto no_join;
1707 1708 1709

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
1710
		goto no_join;
1711 1712 1713

	my_grp = p->numa_group;
	if (grp == my_grp)
1714
		goto no_join;
1715 1716 1717 1718 1719 1720

	/*
	 * Only join the other group if its bigger; if we're the bigger group,
	 * the other task will join us.
	 */
	if (my_grp->nr_tasks > grp->nr_tasks)
1721
		goto no_join;
1722 1723 1724 1725 1726

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
1727
		goto no_join;
1728

1729 1730 1731 1732 1733 1734 1735
	/* Always join threads in the same process. */
	if (tsk->mm == current->mm)
		join = true;

	/* Simple filter to avoid false positives due to PID collisions */
	if (flags & TNF_SHARED)
		join = true;
1736

1737 1738 1739
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

1740
	if (join && !get_numa_group(grp))
1741
		goto no_join;
1742 1743 1744 1745 1746 1747

	rcu_read_unlock();

	if (!join)
		return;

1748 1749
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
1750

1751
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
1752 1753
		my_grp->faults[i] -= p->numa_faults_memory[i];
		grp->faults[i] += p->numa_faults_memory[i];
1754
	}
1755 1756
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
1757 1758 1759 1760 1761 1762

	list_move(&p->numa_entry, &grp->task_list);
	my_grp->nr_tasks--;
	grp->nr_tasks++;

	spin_unlock(&my_grp->lock);
1763
	spin_unlock_irq(&grp->lock);
1764 1765 1766 1767

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
1768 1769 1770 1771 1772
	return;

no_join:
	rcu_read_unlock();
	return;
1773 1774 1775 1776 1777
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
1778
	void *numa_faults = p->numa_faults_memory;
1779 1780
	unsigned long flags;
	int i;
1781 1782

	if (grp) {
1783
		spin_lock_irqsave(&grp->lock, flags);
1784
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1785
			grp->faults[i] -= p->numa_faults_memory[i];
1786
		grp->total_faults -= p->total_numa_faults;
1787

1788 1789
		list_del(&p->numa_entry);
		grp->nr_tasks--;
1790
		spin_unlock_irqrestore(&grp->lock, flags);
1791
		RCU_INIT_POINTER(p->numa_group, NULL);
1792 1793 1794
		put_numa_group(grp);
	}

1795 1796
	p->numa_faults_memory = NULL;
	p->numa_faults_buffer_memory = NULL;
1797 1798
	p->numa_faults_cpu= NULL;
	p->numa_faults_buffer_cpu = NULL;
1799
	kfree(numa_faults);
1800 1801
}

1802 1803 1804
/*
 * Got a PROT_NONE fault for a page on @node.
 */
1805
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
1806 1807
{
	struct task_struct *p = current;
1808
	bool migrated = flags & TNF_MIGRATED;
1809
	int cpu_node = task_node(current);
1810
	int local = !!(flags & TNF_FAULT_LOCAL);
1811
	int priv;
1812

1813
	if (!numabalancing_enabled)
1814 1815
		return;

1816 1817 1818 1819
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

1820 1821 1822 1823
	/* Do not worry about placement if exiting */
	if (p->state == TASK_DEAD)
		return;

1824
	/* Allocate buffer to track faults on a per-node basis */
1825
	if (unlikely(!p->numa_faults_memory)) {
1826 1827
		int size = sizeof(*p->numa_faults_memory) *
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
1828

1829
		p->numa_faults_memory = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
1830
		if (!p->numa_faults_memory)
1831
			return;
1832

1833
		BUG_ON(p->numa_faults_buffer_memory);
1834 1835 1836 1837 1838 1839
		/*
		 * The averaged statistics, shared & private, memory & cpu,
		 * occupy the first half of the array. The second half of the
		 * array is for current counters, which are averaged into the
		 * first set by task_numa_placement.
		 */
1840 1841 1842
		p->numa_faults_cpu = p->numa_faults_memory + (2 * nr_node_ids);
		p->numa_faults_buffer_memory = p->numa_faults_memory + (4 * nr_node_ids);
		p->numa_faults_buffer_cpu = p->numa_faults_memory + (6 * nr_node_ids);
1843
		p->total_numa_faults = 0;
1844
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
1845
	}
1846

1847 1848 1849 1850 1851 1852 1853 1854
	/*
	 * First accesses are treated as private, otherwise consider accesses
	 * to be private if the accessing pid has not changed
	 */
	if (unlikely(last_cpupid == (-1 & LAST_CPUPID_MASK))) {
		priv = 1;
	} else {
		priv = cpupid_match_pid(p, last_cpupid);
1855
		if (!priv && !(flags & TNF_NO_GROUP))
1856
			task_numa_group(p, last_cpupid, flags, &priv);
1857 1858
	}

1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869
	/*
	 * If a workload spans multiple NUMA nodes, a shared fault that
	 * occurs wholly within the set of nodes that the workload is
	 * actively using should be counted as local. This allows the
	 * scan rate to slow down when a workload has settled down.
	 */
	if (!priv && !local && p->numa_group &&
			node_isset(cpu_node, p->numa_group->active_nodes) &&
			node_isset(mem_node, p->numa_group->active_nodes))
		local = 1;

1870
	task_numa_placement(p);
1871

1872 1873 1874 1875 1876
	/*
	 * Retry task to preferred node migration periodically, in case it
	 * case it previously failed, or the scheduler moved us.
	 */
	if (time_after(jiffies, p->numa_migrate_retry))
1877 1878
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
1879 1880 1881
	if (migrated)
		p->numa_pages_migrated += pages;

1882 1883
	p->numa_faults_buffer_memory[task_faults_idx(mem_node, priv)] += pages;
	p->numa_faults_buffer_cpu[task_faults_idx(cpu_node, priv)] += pages;
1884
	p->numa_faults_locality[local] += pages;
1885 1886
}

1887 1888 1889 1890 1891 1892
static void reset_ptenuma_scan(struct task_struct *p)
{
	ACCESS_ONCE(p->mm->numa_scan_seq)++;
	p->mm->numa_scan_offset = 0;
}

1893 1894 1895 1896 1897 1898 1899 1900 1901
/*
 * The expensive part of numa migration is done from task_work context.
 * Triggered from task_tick_numa().
 */
void task_numa_work(struct callback_head *work)
{
	unsigned long migrate, next_scan, now = jiffies;
	struct task_struct *p = current;
	struct mm_struct *mm = p->mm;
1902
	struct vm_area_struct *vma;
1903
	unsigned long start, end;
1904
	unsigned long nr_pte_updates = 0;
1905
	long pages;
1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920

	WARN_ON_ONCE(p != container_of(work, struct task_struct, numa_work));

	work->next = work; /* protect against double add */
	/*
	 * Who cares about NUMA placement when they're dying.
	 *
	 * NOTE: make sure not to dereference p->mm before this check,
	 * exit_task_work() happens _after_ exit_mm() so we could be called
	 * without p->mm even though we still had it when we enqueued this
	 * work.
	 */
	if (p->flags & PF_EXITING)
		return;

1921
	if (!mm->numa_next_scan) {
1922 1923
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
1924 1925
	}

1926 1927 1928 1929 1930 1931 1932
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

1933 1934 1935 1936
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
1937

1938
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
1939 1940 1941
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

1942 1943 1944 1945 1946 1947
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

1948 1949 1950 1951 1952
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
	if (!pages)
		return;
1953

1954
	down_read(&mm->mmap_sem);
1955
	vma = find_vma(mm, start);
1956 1957
	if (!vma) {
		reset_ptenuma_scan(p);
1958
		start = 0;
1959 1960
		vma = mm->mmap;
	}
1961
	for (; vma; vma = vma->vm_next) {
1962
		if (!vma_migratable(vma) || !vma_policy_mof(p, vma))
1963 1964
			continue;

1965 1966 1967 1968 1969 1970 1971 1972 1973 1974
		/*
		 * Shared library pages mapped by multiple processes are not
		 * migrated as it is expected they are cache replicated. Avoid
		 * hinting faults in read-only file-backed mappings or the vdso
		 * as migrating the pages will be of marginal benefit.
		 */
		if (!vma->vm_mm ||
		    (vma->vm_file && (vma->vm_flags & (VM_READ|VM_WRITE)) == (VM_READ)))
			continue;

M
Mel Gorman 已提交
1975 1976 1977 1978 1979 1980
		/*
		 * Skip inaccessible VMAs to avoid any confusion between
		 * PROT_NONE and NUMA hinting ptes
		 */
		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
			continue;
1981

1982 1983 1984 1985
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
1986 1987 1988 1989 1990 1991 1992 1993 1994
			nr_pte_updates += change_prot_numa(vma, start, end);

			/*
			 * Scan sysctl_numa_balancing_scan_size but ensure that
			 * at least one PTE is updated so that unused virtual
			 * address space is quickly skipped.
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
1995

1996 1997 1998
			start = end;
			if (pages <= 0)
				goto out;
1999 2000

			cond_resched();
2001
		} while (end != vma->vm_end);
2002
	}
2003

2004
out:
2005
	/*
P
Peter Zijlstra 已提交
2006 2007 2008 2009
	 * It is possible to reach the end of the VMA list but the last few
	 * VMAs are not guaranteed to the vma_migratable. If they are not, we
	 * would find the !migratable VMA on the next scan but not reset the
	 * scanner to the start so check it now.
2010 2011
	 */
	if (vma)
2012
		mm->numa_scan_offset = start;
2013 2014 2015
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041
}

/*
 * Drive the periodic memory faults..
 */
void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
	struct callback_head *work = &curr->numa_work;
	u64 period, now;

	/*
	 * We don't care about NUMA placement if we don't have memory.
	 */
	if (!curr->mm || (curr->flags & PF_EXITING) || work->next != work)
		return;

	/*
	 * Using runtime rather than walltime has the dual advantage that
	 * we (mostly) drive the selection from busy threads and that the
	 * task needs to have done some actual work before we bother with
	 * NUMA placement.
	 */
	now = curr->se.sum_exec_runtime;
	period = (u64)curr->numa_scan_period * NSEC_PER_MSEC;

	if (now - curr->node_stamp > period) {
2042
		if (!curr->node_stamp)
2043
			curr->numa_scan_period = task_scan_min(curr);
2044
		curr->node_stamp += period;
2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055

		if (!time_before(jiffies, curr->mm->numa_next_scan)) {
			init_task_work(work, task_numa_work); /* TODO: move this into sched_fork() */
			task_work_add(curr, work, true);
		}
	}
}
#else
static void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
}
2056 2057 2058 2059 2060 2061 2062 2063

static inline void account_numa_enqueue(struct rq *rq, struct task_struct *p)
{
}

static inline void account_numa_dequeue(struct rq *rq, struct task_struct *p)
{
}
2064 2065
#endif /* CONFIG_NUMA_BALANCING */

2066 2067 2068 2069
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2070
	if (!parent_entity(se))
2071
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2072
#ifdef CONFIG_SMP
2073 2074 2075 2076 2077 2078
	if (entity_is_task(se)) {
		struct rq *rq = rq_of(cfs_rq);

		account_numa_enqueue(rq, task_of(se));
		list_add(&se->group_node, &rq->cfs_tasks);
	}
2079
#endif
2080 2081 2082 2083 2084 2085 2086
	cfs_rq->nr_running++;
}

static void
account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_sub(&cfs_rq->load, se->load.weight);
2087
	if (!parent_entity(se))
2088
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2089 2090
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2091
		list_del_init(&se->group_node);
2092
	}
2093 2094 2095
	cfs_rq->nr_running--;
}

2096 2097
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2098 2099 2100 2101 2102 2103 2104 2105 2106
static inline long calc_tg_weight(struct task_group *tg, struct cfs_rq *cfs_rq)
{
	long tg_weight;

	/*
	 * Use this CPU's actual weight instead of the last load_contribution
	 * to gain a more accurate current total weight. See
	 * update_cfs_rq_load_contribution().
	 */
2107
	tg_weight = atomic_long_read(&tg->load_avg);
2108
	tg_weight -= cfs_rq->tg_load_contrib;
2109 2110 2111 2112 2113
	tg_weight += cfs_rq->load.weight;

	return tg_weight;
}

2114
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2115
{
2116
	long tg_weight, load, shares;
2117

2118
	tg_weight = calc_tg_weight(tg, cfs_rq);
2119
	load = cfs_rq->load.weight;
2120 2121

	shares = (tg->shares * load);
2122 2123
	if (tg_weight)
		shares /= tg_weight;
2124 2125 2126 2127 2128 2129 2130 2131 2132

	if (shares < MIN_SHARES)
		shares = MIN_SHARES;
	if (shares > tg->shares)
		shares = tg->shares;

	return shares;
}
# else /* CONFIG_SMP */
2133
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2134 2135 2136 2137
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
2138 2139 2140
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2141 2142 2143 2144
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2145
		account_entity_dequeue(cfs_rq, se);
2146
	}
P
Peter Zijlstra 已提交
2147 2148 2149 2150 2151 2152 2153

	update_load_set(&se->load, weight);

	if (se->on_rq)
		account_entity_enqueue(cfs_rq, se);
}

2154 2155
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2156
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2157 2158 2159
{
	struct task_group *tg;
	struct sched_entity *se;
2160
	long shares;
P
Peter Zijlstra 已提交
2161 2162 2163

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2164
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2165
		return;
2166 2167 2168 2169
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2170
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2171 2172 2173 2174

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2175
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2176 2177 2178 2179
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2180
#ifdef CONFIG_SMP
2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208
/*
 * We choose a half-life close to 1 scheduling period.
 * Note: The tables below are dependent on this value.
 */
#define LOAD_AVG_PERIOD 32
#define LOAD_AVG_MAX 47742 /* maximum possible load avg */
#define LOAD_AVG_MAX_N 345 /* number of full periods to produce LOAD_MAX_AVG */

/* Precomputed fixed inverse multiplies for multiplication by y^n */
static const u32 runnable_avg_yN_inv[] = {
	0xffffffff, 0xfa83b2da, 0xf5257d14, 0xefe4b99a, 0xeac0c6e6, 0xe5b906e6,
	0xe0ccdeeb, 0xdbfbb796, 0xd744fcc9, 0xd2a81d91, 0xce248c14, 0xc9b9bd85,
	0xc5672a10, 0xc12c4cc9, 0xbd08a39e, 0xb8fbaf46, 0xb504f333, 0xb123f581,
	0xad583ee9, 0xa9a15ab4, 0xa5fed6a9, 0xa2704302, 0x9ef5325f, 0x9b8d39b9,
	0x9837f050, 0x94f4efa8, 0x91c3d373, 0x8ea4398a, 0x8b95c1e3, 0x88980e80,
	0x85aac367, 0x82cd8698,
};

/*
 * Precomputed \Sum y^k { 1<=k<=n }.  These are floor(true_value) to prevent
 * over-estimates when re-combining.
 */
static const u32 runnable_avg_yN_sum[] = {
	    0, 1002, 1982, 2941, 3880, 4798, 5697, 6576, 7437, 8279, 9103,
	 9909,10698,11470,12226,12966,13690,14398,15091,15769,16433,17082,
	17718,18340,18949,19545,20128,20698,21256,21802,22336,22859,23371,
};

2209 2210 2211 2212 2213 2214
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234
	unsigned int local_n;

	if (!n)
		return val;
	else if (unlikely(n > LOAD_AVG_PERIOD * 63))
		return 0;

	/* after bounds checking we can collapse to 32-bit */
	local_n = n;

	/*
	 * As y^PERIOD = 1/2, we can combine
	 *    y^n = 1/2^(n/PERIOD) * k^(n%PERIOD)
	 * With a look-up table which covers k^n (n<PERIOD)
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
2235 2236
	}

2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267
	val *= runnable_avg_yN_inv[local_n];
	/* We don't use SRR here since we always want to round down. */
	return val >> 32;
}

/*
 * For updates fully spanning n periods, the contribution to runnable
 * average will be: \Sum 1024*y^n
 *
 * We can compute this reasonably efficiently by combining:
 *   y^PERIOD = 1/2 with precomputed \Sum 1024*y^n {for  n <PERIOD}
 */
static u32 __compute_runnable_contrib(u64 n)
{
	u32 contrib = 0;

	if (likely(n <= LOAD_AVG_PERIOD))
		return runnable_avg_yN_sum[n];
	else if (unlikely(n >= LOAD_AVG_MAX_N))
		return LOAD_AVG_MAX;

	/* Compute \Sum k^n combining precomputed values for k^i, \Sum k^j */
	do {
		contrib /= 2; /* y^LOAD_AVG_PERIOD = 1/2 */
		contrib += runnable_avg_yN_sum[LOAD_AVG_PERIOD];

		n -= LOAD_AVG_PERIOD;
	} while (n > LOAD_AVG_PERIOD);

	contrib = decay_load(contrib, n);
	return contrib + runnable_avg_yN_sum[n];
2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301
}

/*
 * We can represent the historical contribution to runnable average as the
 * coefficients of a geometric series.  To do this we sub-divide our runnable
 * history into segments of approximately 1ms (1024us); label the segment that
 * occurred N-ms ago p_N, with p_0 corresponding to the current period, e.g.
 *
 * [<- 1024us ->|<- 1024us ->|<- 1024us ->| ...
 *      p0            p1           p2
 *     (now)       (~1ms ago)  (~2ms ago)
 *
 * Let u_i denote the fraction of p_i that the entity was runnable.
 *
 * We then designate the fractions u_i as our co-efficients, yielding the
 * following representation of historical load:
 *   u_0 + u_1*y + u_2*y^2 + u_3*y^3 + ...
 *
 * We choose y based on the with of a reasonably scheduling period, fixing:
 *   y^32 = 0.5
 *
 * This means that the contribution to load ~32ms ago (u_32) will be weighted
 * approximately half as much as the contribution to load within the last ms
 * (u_0).
 *
 * When a period "rolls over" and we have new u_0`, multiplying the previous
 * sum again by y is sufficient to update:
 *   load_avg = u_0` + y*(u_0 + u_1*y + u_2*y^2 + ... )
 *            = u_0 + u_1*y + u_2*y^2 + ... [re-labeling u_i --> u_{i+1}]
 */
static __always_inline int __update_entity_runnable_avg(u64 now,
							struct sched_avg *sa,
							int runnable)
{
2302 2303
	u64 delta, periods;
	u32 runnable_contrib;
2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336
	int delta_w, decayed = 0;

	delta = now - sa->last_runnable_update;
	/*
	 * This should only happen when time goes backwards, which it
	 * unfortunately does during sched clock init when we swap over to TSC.
	 */
	if ((s64)delta < 0) {
		sa->last_runnable_update = now;
		return 0;
	}

	/*
	 * Use 1024ns as the unit of measurement since it's a reasonable
	 * approximation of 1us and fast to compute.
	 */
	delta >>= 10;
	if (!delta)
		return 0;
	sa->last_runnable_update = now;

	/* delta_w is the amount already accumulated against our next period */
	delta_w = sa->runnable_avg_period % 1024;
	if (delta + delta_w >= 1024) {
		/* period roll-over */
		decayed = 1;

		/*
		 * Now that we know we're crossing a period boundary, figure
		 * out how much from delta we need to complete the current
		 * period and accrue it.
		 */
		delta_w = 1024 - delta_w;
2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356
		if (runnable)
			sa->runnable_avg_sum += delta_w;
		sa->runnable_avg_period += delta_w;

		delta -= delta_w;

		/* Figure out how many additional periods this update spans */
		periods = delta / 1024;
		delta %= 1024;

		sa->runnable_avg_sum = decay_load(sa->runnable_avg_sum,
						  periods + 1);
		sa->runnable_avg_period = decay_load(sa->runnable_avg_period,
						     periods + 1);

		/* Efficiently calculate \sum (1..n_period) 1024*y^i */
		runnable_contrib = __compute_runnable_contrib(periods);
		if (runnable)
			sa->runnable_avg_sum += runnable_contrib;
		sa->runnable_avg_period += runnable_contrib;
2357 2358 2359 2360 2361 2362 2363 2364 2365 2366
	}

	/* Remainder of delta accrued against u_0` */
	if (runnable)
		sa->runnable_avg_sum += delta;
	sa->runnable_avg_period += delta;

	return decayed;
}

2367
/* Synchronize an entity's decay with its parenting cfs_rq.*/
2368
static inline u64 __synchronize_entity_decay(struct sched_entity *se)
2369 2370 2371 2372 2373 2374
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 decays = atomic64_read(&cfs_rq->decay_counter);

	decays -= se->avg.decay_count;
	if (!decays)
2375
		return 0;
2376 2377 2378

	se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
	se->avg.decay_count = 0;
2379 2380

	return decays;
2381 2382
}

2383 2384 2385 2386 2387
#ifdef CONFIG_FAIR_GROUP_SCHED
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update)
{
	struct task_group *tg = cfs_rq->tg;
2388
	long tg_contrib;
2389 2390 2391 2392

	tg_contrib = cfs_rq->runnable_load_avg + cfs_rq->blocked_load_avg;
	tg_contrib -= cfs_rq->tg_load_contrib;

2393 2394 2395
	if (!tg_contrib)
		return;

2396 2397
	if (force_update || abs(tg_contrib) > cfs_rq->tg_load_contrib / 8) {
		atomic_long_add(tg_contrib, &tg->load_avg);
2398 2399 2400
		cfs_rq->tg_load_contrib += tg_contrib;
	}
}
2401

2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412
/*
 * Aggregate cfs_rq runnable averages into an equivalent task_group
 * representation for computing load contributions.
 */
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq)
{
	struct task_group *tg = cfs_rq->tg;
	long contrib;

	/* The fraction of a cpu used by this cfs_rq */
2413
	contrib = div_u64((u64)sa->runnable_avg_sum << NICE_0_SHIFT,
2414 2415 2416 2417 2418 2419 2420 2421 2422
			  sa->runnable_avg_period + 1);
	contrib -= cfs_rq->tg_runnable_contrib;

	if (abs(contrib) > cfs_rq->tg_runnable_contrib / 64) {
		atomic_add(contrib, &tg->runnable_avg);
		cfs_rq->tg_runnable_contrib += contrib;
	}
}

2423 2424 2425 2426
static inline void __update_group_entity_contrib(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = group_cfs_rq(se);
	struct task_group *tg = cfs_rq->tg;
2427 2428
	int runnable_avg;

2429 2430 2431
	u64 contrib;

	contrib = cfs_rq->tg_load_contrib * tg->shares;
2432 2433
	se->avg.load_avg_contrib = div_u64(contrib,
				     atomic_long_read(&tg->load_avg) + 1);
2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462

	/*
	 * For group entities we need to compute a correction term in the case
	 * that they are consuming <1 cpu so that we would contribute the same
	 * load as a task of equal weight.
	 *
	 * Explicitly co-ordinating this measurement would be expensive, but
	 * fortunately the sum of each cpus contribution forms a usable
	 * lower-bound on the true value.
	 *
	 * Consider the aggregate of 2 contributions.  Either they are disjoint
	 * (and the sum represents true value) or they are disjoint and we are
	 * understating by the aggregate of their overlap.
	 *
	 * Extending this to N cpus, for a given overlap, the maximum amount we
	 * understand is then n_i(n_i+1)/2 * w_i where n_i is the number of
	 * cpus that overlap for this interval and w_i is the interval width.
	 *
	 * On a small machine; the first term is well-bounded which bounds the
	 * total error since w_i is a subset of the period.  Whereas on a
	 * larger machine, while this first term can be larger, if w_i is the
	 * of consequential size guaranteed to see n_i*w_i quickly converge to
	 * our upper bound of 1-cpu.
	 */
	runnable_avg = atomic_read(&tg->runnable_avg);
	if (runnable_avg < NICE_0_LOAD) {
		se->avg.load_avg_contrib *= runnable_avg;
		se->avg.load_avg_contrib >>= NICE_0_SHIFT;
	}
2463
}
2464 2465 2466 2467 2468 2469

static inline void update_rq_runnable_avg(struct rq *rq, int runnable)
{
	__update_entity_runnable_avg(rq_clock_task(rq), &rq->avg, runnable);
	__update_tg_runnable_avg(&rq->avg, &rq->cfs);
}
2470
#else /* CONFIG_FAIR_GROUP_SCHED */
2471 2472
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update) {}
2473 2474
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq) {}
2475
static inline void __update_group_entity_contrib(struct sched_entity *se) {}
2476
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2477
#endif /* CONFIG_FAIR_GROUP_SCHED */
2478

2479 2480 2481 2482 2483 2484 2485 2486 2487 2488
static inline void __update_task_entity_contrib(struct sched_entity *se)
{
	u32 contrib;

	/* avoid overflowing a 32-bit type w/ SCHED_LOAD_SCALE */
	contrib = se->avg.runnable_avg_sum * scale_load_down(se->load.weight);
	contrib /= (se->avg.runnable_avg_period + 1);
	se->avg.load_avg_contrib = scale_load(contrib);
}

2489 2490 2491 2492 2493
/* Compute the current contribution to load_avg by se, return any delta */
static long __update_entity_load_avg_contrib(struct sched_entity *se)
{
	long old_contrib = se->avg.load_avg_contrib;

2494 2495 2496
	if (entity_is_task(se)) {
		__update_task_entity_contrib(se);
	} else {
2497
		__update_tg_runnable_avg(&se->avg, group_cfs_rq(se));
2498 2499
		__update_group_entity_contrib(se);
	}
2500 2501 2502 2503

	return se->avg.load_avg_contrib - old_contrib;
}

2504 2505 2506 2507 2508 2509 2510 2511 2512
static inline void subtract_blocked_load_contrib(struct cfs_rq *cfs_rq,
						 long load_contrib)
{
	if (likely(load_contrib < cfs_rq->blocked_load_avg))
		cfs_rq->blocked_load_avg -= load_contrib;
	else
		cfs_rq->blocked_load_avg = 0;
}

2513 2514
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);

2515
/* Update a sched_entity's runnable average */
2516 2517
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq)
2518
{
2519 2520
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	long contrib_delta;
2521
	u64 now;
2522

2523 2524 2525 2526 2527 2528 2529 2530 2531 2532
	/*
	 * For a group entity we need to use their owned cfs_rq_clock_task() in
	 * case they are the parent of a throttled hierarchy.
	 */
	if (entity_is_task(se))
		now = cfs_rq_clock_task(cfs_rq);
	else
		now = cfs_rq_clock_task(group_cfs_rq(se));

	if (!__update_entity_runnable_avg(now, &se->avg, se->on_rq))
2533 2534 2535
		return;

	contrib_delta = __update_entity_load_avg_contrib(se);
2536 2537 2538 2539

	if (!update_cfs_rq)
		return;

2540 2541
	if (se->on_rq)
		cfs_rq->runnable_load_avg += contrib_delta;
2542 2543 2544 2545 2546 2547 2548 2549
	else
		subtract_blocked_load_contrib(cfs_rq, -contrib_delta);
}

/*
 * Decay the load contributed by all blocked children and account this so that
 * their contribution may appropriately discounted when they wake up.
 */
2550
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq, int force_update)
2551
{
2552
	u64 now = cfs_rq_clock_task(cfs_rq) >> 20;
2553 2554 2555
	u64 decays;

	decays = now - cfs_rq->last_decay;
2556
	if (!decays && !force_update)
2557 2558
		return;

2559 2560 2561
	if (atomic_long_read(&cfs_rq->removed_load)) {
		unsigned long removed_load;
		removed_load = atomic_long_xchg(&cfs_rq->removed_load, 0);
2562 2563
		subtract_blocked_load_contrib(cfs_rq, removed_load);
	}
2564

2565 2566 2567 2568 2569 2570
	if (decays) {
		cfs_rq->blocked_load_avg = decay_load(cfs_rq->blocked_load_avg,
						      decays);
		atomic64_add(decays, &cfs_rq->decay_counter);
		cfs_rq->last_decay = now;
	}
2571 2572

	__update_cfs_rq_tg_load_contrib(cfs_rq, force_update);
2573
}
2574

2575 2576
/* Add the load generated by se into cfs_rq's child load-average */
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2577 2578
						  struct sched_entity *se,
						  int wakeup)
2579
{
2580 2581 2582 2583
	/*
	 * We track migrations using entity decay_count <= 0, on a wake-up
	 * migration we use a negative decay count to track the remote decays
	 * accumulated while sleeping.
2584 2585 2586 2587
	 *
	 * Newly forked tasks are enqueued with se->avg.decay_count == 0, they
	 * are seen by enqueue_entity_load_avg() as a migration with an already
	 * constructed load_avg_contrib.
2588 2589
	 */
	if (unlikely(se->avg.decay_count <= 0)) {
2590
		se->avg.last_runnable_update = rq_clock_task(rq_of(cfs_rq));
2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605
		if (se->avg.decay_count) {
			/*
			 * In a wake-up migration we have to approximate the
			 * time sleeping.  This is because we can't synchronize
			 * clock_task between the two cpus, and it is not
			 * guaranteed to be read-safe.  Instead, we can
			 * approximate this using our carried decays, which are
			 * explicitly atomically readable.
			 */
			se->avg.last_runnable_update -= (-se->avg.decay_count)
							<< 20;
			update_entity_load_avg(se, 0);
			/* Indicate that we're now synchronized and on-rq */
			se->avg.decay_count = 0;
		}
2606 2607
		wakeup = 0;
	} else {
2608
		__synchronize_entity_decay(se);
2609 2610
	}

2611 2612
	/* migrated tasks did not contribute to our blocked load */
	if (wakeup) {
2613
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
2614 2615
		update_entity_load_avg(se, 0);
	}
2616

2617
	cfs_rq->runnable_load_avg += se->avg.load_avg_contrib;
2618 2619
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !wakeup);
2620 2621
}

2622 2623 2624 2625 2626
/*
 * Remove se's load from this cfs_rq child load-average, if the entity is
 * transitioning to a blocked state we track its projected decay using
 * blocked_load_avg.
 */
2627
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2628 2629
						  struct sched_entity *se,
						  int sleep)
2630
{
2631
	update_entity_load_avg(se, 1);
2632 2633
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !sleep);
2634

2635
	cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib;
2636 2637 2638 2639
	if (sleep) {
		cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
		se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
	} /* migrations, e.g. sleep=0 leave decay_count == 0 */
2640
}
2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661

/*
 * Update the rq's load with the elapsed running time before entering
 * idle. if the last scheduled task is not a CFS task, idle_enter will
 * be the only way to update the runnable statistic.
 */
void idle_enter_fair(struct rq *this_rq)
{
	update_rq_runnable_avg(this_rq, 1);
}

/*
 * Update the rq's load with the elapsed idle time before a task is
 * scheduled. if the newly scheduled task is not a CFS task, idle_exit will
 * be the only way to update the runnable statistic.
 */
void idle_exit_fair(struct rq *this_rq)
{
	update_rq_runnable_avg(this_rq, 0);
}

2662 2663
static int idle_balance(struct rq *this_rq);

2664 2665
#else /* CONFIG_SMP */

2666 2667
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq) {}
2668
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2669
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2670 2671
					   struct sched_entity *se,
					   int wakeup) {}
2672
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2673 2674
					   struct sched_entity *se,
					   int sleep) {}
2675 2676
static inline void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
					      int force_update) {}
2677 2678 2679 2680 2681 2682

static inline int idle_balance(struct rq *rq)
{
	return 0;
}

2683
#endif /* CONFIG_SMP */
2684

2685
static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
2686 2687
{
#ifdef CONFIG_SCHEDSTATS
2688 2689 2690 2691 2692
	struct task_struct *tsk = NULL;

	if (entity_is_task(se))
		tsk = task_of(se);

2693
	if (se->statistics.sleep_start) {
2694
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.sleep_start;
2695 2696 2697 2698

		if ((s64)delta < 0)
			delta = 0;

2699 2700
		if (unlikely(delta > se->statistics.sleep_max))
			se->statistics.sleep_max = delta;
2701

2702
		se->statistics.sleep_start = 0;
2703
		se->statistics.sum_sleep_runtime += delta;
A
Arjan van de Ven 已提交
2704

2705
		if (tsk) {
2706
			account_scheduler_latency(tsk, delta >> 10, 1);
2707 2708
			trace_sched_stat_sleep(tsk, delta);
		}
2709
	}
2710
	if (se->statistics.block_start) {
2711
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.block_start;
2712 2713 2714 2715

		if ((s64)delta < 0)
			delta = 0;

2716 2717
		if (unlikely(delta > se->statistics.block_max))
			se->statistics.block_max = delta;
2718

2719
		se->statistics.block_start = 0;
2720
		se->statistics.sum_sleep_runtime += delta;
I
Ingo Molnar 已提交
2721

2722
		if (tsk) {
2723
			if (tsk->in_iowait) {
2724 2725
				se->statistics.iowait_sum += delta;
				se->statistics.iowait_count++;
2726
				trace_sched_stat_iowait(tsk, delta);
2727 2728
			}

2729 2730
			trace_sched_stat_blocked(tsk, delta);

2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741
			/*
			 * Blocking time is in units of nanosecs, so shift by
			 * 20 to get a milliseconds-range estimation of the
			 * amount of time that the task spent sleeping:
			 */
			if (unlikely(prof_on == SLEEP_PROFILING)) {
				profile_hits(SLEEP_PROFILING,
						(void *)get_wchan(tsk),
						delta >> 20);
			}
			account_scheduler_latency(tsk, delta >> 10, 0);
I
Ingo Molnar 已提交
2742
		}
2743 2744 2745 2746
	}
#endif
}

P
Peter Zijlstra 已提交
2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759
static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
#ifdef CONFIG_SCHED_DEBUG
	s64 d = se->vruntime - cfs_rq->min_vruntime;

	if (d < 0)
		d = -d;

	if (d > 3*sysctl_sched_latency)
		schedstat_inc(cfs_rq, nr_spread_over);
#endif
}

2760 2761 2762
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
2763
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
2764

2765 2766 2767 2768 2769 2770
	/*
	 * The 'current' period is already promised to the current tasks,
	 * however the extra weight of the new task will slow them down a
	 * little, place the new task so that it fits in the slot that
	 * stays open at the end.
	 */
P
Peter Zijlstra 已提交
2771
	if (initial && sched_feat(START_DEBIT))
2772
		vruntime += sched_vslice(cfs_rq, se);
2773

2774
	/* sleeps up to a single latency don't count. */
2775
	if (!initial) {
2776
		unsigned long thresh = sysctl_sched_latency;
2777

2778 2779 2780 2781 2782 2783
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
2784

2785
		vruntime -= thresh;
2786 2787
	}

2788
	/* ensure we never gain time by being placed backwards. */
2789
	se->vruntime = max_vruntime(se->vruntime, vruntime);
2790 2791
}

2792 2793
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

2794
static void
2795
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2796
{
2797 2798
	/*
	 * Update the normalized vruntime before updating min_vruntime
2799
	 * through calling update_curr().
2800
	 */
2801
	if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
2802 2803
		se->vruntime += cfs_rq->min_vruntime;

2804
	/*
2805
	 * Update run-time statistics of the 'current'.
2806
	 */
2807
	update_curr(cfs_rq);
2808
	enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP);
2809 2810
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
2811

2812
	if (flags & ENQUEUE_WAKEUP) {
2813
		place_entity(cfs_rq, se, 0);
2814
		enqueue_sleeper(cfs_rq, se);
I
Ingo Molnar 已提交
2815
	}
2816

2817
	update_stats_enqueue(cfs_rq, se);
P
Peter Zijlstra 已提交
2818
	check_spread(cfs_rq, se);
2819 2820
	if (se != cfs_rq->curr)
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
2821
	se->on_rq = 1;
2822

2823
	if (cfs_rq->nr_running == 1) {
2824
		list_add_leaf_cfs_rq(cfs_rq);
2825 2826
		check_enqueue_throttle(cfs_rq);
	}
2827 2828
}

2829
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
2830
{
2831 2832
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2833
		if (cfs_rq->last != se)
2834
			break;
2835 2836

		cfs_rq->last = NULL;
2837 2838
	}
}
P
Peter Zijlstra 已提交
2839

2840 2841 2842 2843
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2844
		if (cfs_rq->next != se)
2845
			break;
2846 2847

		cfs_rq->next = NULL;
2848
	}
P
Peter Zijlstra 已提交
2849 2850
}

2851 2852 2853 2854
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2855
		if (cfs_rq->skip != se)
2856
			break;
2857 2858

		cfs_rq->skip = NULL;
2859 2860 2861
	}
}

P
Peter Zijlstra 已提交
2862 2863
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2864 2865 2866 2867 2868
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
2869 2870 2871

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
2872 2873
}

2874
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
2875

2876
static void
2877
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2878
{
2879 2880 2881 2882
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
2883
	dequeue_entity_load_avg(cfs_rq, se, flags & DEQUEUE_SLEEP);
2884

2885
	update_stats_dequeue(cfs_rq, se);
2886
	if (flags & DEQUEUE_SLEEP) {
P
Peter Zijlstra 已提交
2887
#ifdef CONFIG_SCHEDSTATS
2888 2889 2890 2891
		if (entity_is_task(se)) {
			struct task_struct *tsk = task_of(se);

			if (tsk->state & TASK_INTERRUPTIBLE)
2892
				se->statistics.sleep_start = rq_clock(rq_of(cfs_rq));
2893
			if (tsk->state & TASK_UNINTERRUPTIBLE)
2894
				se->statistics.block_start = rq_clock(rq_of(cfs_rq));
2895
		}
2896
#endif
P
Peter Zijlstra 已提交
2897 2898
	}

P
Peter Zijlstra 已提交
2899
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
2900

2901
	if (se != cfs_rq->curr)
2902
		__dequeue_entity(cfs_rq, se);
2903
	se->on_rq = 0;
2904
	account_entity_dequeue(cfs_rq, se);
2905 2906 2907 2908 2909 2910

	/*
	 * Normalize the entity after updating the min_vruntime because the
	 * update can refer to the ->curr item and we need to reflect this
	 * movement in our normalized position.
	 */
2911
	if (!(flags & DEQUEUE_SLEEP))
2912
		se->vruntime -= cfs_rq->min_vruntime;
2913

2914 2915 2916
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

2917
	update_min_vruntime(cfs_rq);
2918
	update_cfs_shares(cfs_rq);
2919 2920 2921 2922 2923
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
2924
static void
I
Ingo Molnar 已提交
2925
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
2926
{
2927
	unsigned long ideal_runtime, delta_exec;
2928 2929
	struct sched_entity *se;
	s64 delta;
2930

P
Peter Zijlstra 已提交
2931
	ideal_runtime = sched_slice(cfs_rq, curr);
2932
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
2933
	if (delta_exec > ideal_runtime) {
2934
		resched_curr(rq_of(cfs_rq));
2935 2936 2937 2938 2939
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950
		return;
	}

	/*
	 * Ensure that a task that missed wakeup preemption by a
	 * narrow margin doesn't have to wait for a full slice.
	 * This also mitigates buddy induced latencies under load.
	 */
	if (delta_exec < sysctl_sched_min_granularity)
		return;

2951 2952
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
2953

2954 2955
	if (delta < 0)
		return;
2956

2957
	if (delta > ideal_runtime)
2958
		resched_curr(rq_of(cfs_rq));
2959 2960
}

2961
static void
2962
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
2963
{
2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974
	/* 'current' is not kept within the tree. */
	if (se->on_rq) {
		/*
		 * Any task has to be enqueued before it get to execute on
		 * a CPU. So account for the time it spent waiting on the
		 * runqueue.
		 */
		update_stats_wait_end(cfs_rq, se);
		__dequeue_entity(cfs_rq, se);
	}

2975
	update_stats_curr_start(cfs_rq, se);
2976
	cfs_rq->curr = se;
I
Ingo Molnar 已提交
2977 2978 2979 2980 2981 2982
#ifdef CONFIG_SCHEDSTATS
	/*
	 * Track our maximum slice length, if the CPU's load is at
	 * least twice that of our own weight (i.e. dont track it
	 * when there are only lesser-weight tasks around):
	 */
2983
	if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
2984
		se->statistics.slice_max = max(se->statistics.slice_max,
I
Ingo Molnar 已提交
2985 2986 2987
			se->sum_exec_runtime - se->prev_sum_exec_runtime);
	}
#endif
2988
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
2989 2990
}

2991 2992 2993
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

2994 2995 2996 2997 2998 2999 3000
/*
 * Pick the next process, keeping these things in mind, in this order:
 * 1) keep things fair between processes/task groups
 * 2) pick the "next" process, since someone really wants that to run
 * 3) pick the "last" process, for cache locality
 * 4) do not run the "skip" process, if something else is available
 */
3001 3002
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3003
{
3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014
	struct sched_entity *left = __pick_first_entity(cfs_rq);
	struct sched_entity *se;

	/*
	 * If curr is set we have to see if its left of the leftmost entity
	 * still in the tree, provided there was anything in the tree at all.
	 */
	if (!left || (curr && entity_before(curr, left)))
		left = curr;

	se = left; /* ideally we run the leftmost entity */
3015

3016 3017 3018 3019 3020
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3021 3022 3023 3024 3025 3026 3027 3028 3029 3030
		struct sched_entity *second;

		if (se == curr) {
			second = __pick_first_entity(cfs_rq);
		} else {
			second = __pick_next_entity(se);
			if (!second || (curr && entity_before(curr, second)))
				second = curr;
		}

3031 3032 3033
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3034

3035 3036 3037 3038 3039 3040
	/*
	 * Prefer last buddy, try to return the CPU to a preempted task.
	 */
	if (cfs_rq->last && wakeup_preempt_entity(cfs_rq->last, left) < 1)
		se = cfs_rq->last;

3041 3042 3043 3044 3045 3046
	/*
	 * Someone really wants this to run. If it's not unfair, run it.
	 */
	if (cfs_rq->next && wakeup_preempt_entity(cfs_rq->next, left) < 1)
		se = cfs_rq->next;

3047
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3048 3049

	return se;
3050 3051
}

3052
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3053

3054
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3055 3056 3057 3058 3059 3060
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3061
		update_curr(cfs_rq);
3062

3063 3064 3065
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

P
Peter Zijlstra 已提交
3066
	check_spread(cfs_rq, prev);
3067
	if (prev->on_rq) {
3068
		update_stats_wait_start(cfs_rq, prev);
3069 3070
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3071
		/* in !on_rq case, update occurred at dequeue */
3072
		update_entity_load_avg(prev, 1);
3073
	}
3074
	cfs_rq->curr = NULL;
3075 3076
}

P
Peter Zijlstra 已提交
3077 3078
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3079 3080
{
	/*
3081
	 * Update run-time statistics of the 'current'.
3082
	 */
3083
	update_curr(cfs_rq);
3084

3085 3086 3087
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3088
	update_entity_load_avg(curr, 1);
3089
	update_cfs_rq_blocked_load(cfs_rq, 1);
3090
	update_cfs_shares(cfs_rq);
3091

P
Peter Zijlstra 已提交
3092 3093 3094 3095 3096
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3097
	if (queued) {
3098
		resched_curr(rq_of(cfs_rq));
3099 3100
		return;
	}
P
Peter Zijlstra 已提交
3101 3102 3103 3104 3105 3106 3107 3108
	/*
	 * don't let the period tick interfere with the hrtick preemption
	 */
	if (!sched_feat(DOUBLE_TICK) &&
			hrtimer_active(&rq_of(cfs_rq)->hrtick_timer))
		return;
#endif

Y
Yong Zhang 已提交
3109
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3110
		check_preempt_tick(cfs_rq, curr);
3111 3112
}

3113 3114 3115 3116 3117 3118

/**************************************************
 * CFS bandwidth control machinery
 */

#ifdef CONFIG_CFS_BANDWIDTH
3119 3120

#ifdef HAVE_JUMP_LABEL
3121
static struct static_key __cfs_bandwidth_used;
3122 3123 3124

static inline bool cfs_bandwidth_used(void)
{
3125
	return static_key_false(&__cfs_bandwidth_used);
3126 3127
}

3128
void cfs_bandwidth_usage_inc(void)
3129
{
3130 3131 3132 3133 3134 3135
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3136 3137 3138 3139 3140 3141 3142
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3143 3144
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3145 3146
#endif /* HAVE_JUMP_LABEL */

3147 3148 3149 3150 3151 3152 3153 3154
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3155 3156 3157 3158 3159 3160

static inline u64 sched_cfs_bandwidth_slice(void)
{
	return (u64)sysctl_sched_cfs_bandwidth_slice * NSEC_PER_USEC;
}

P
Paul Turner 已提交
3161 3162 3163 3164 3165 3166 3167
/*
 * Replenish runtime according to assigned quota and update expiration time.
 * We use sched_clock_cpu directly instead of rq->clock to avoid adding
 * additional synchronization around rq->lock.
 *
 * requires cfs_b->lock
 */
3168
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179
{
	u64 now;

	if (cfs_b->quota == RUNTIME_INF)
		return;

	now = sched_clock_cpu(smp_processor_id());
	cfs_b->runtime = cfs_b->quota;
	cfs_b->runtime_expires = now + ktime_to_ns(cfs_b->period);
}

3180 3181 3182 3183 3184
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3185 3186 3187 3188 3189 3190
/* rq->task_clock normalized against any time this cfs_rq has spent throttled */
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
	if (unlikely(cfs_rq->throttle_count))
		return cfs_rq->throttled_clock_task;

3191
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3192 3193
}

3194 3195
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3196 3197 3198
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3199
	u64 amount = 0, min_amount, expires;
3200 3201 3202 3203 3204 3205 3206

	/* note: this is a positive sum as runtime_remaining <= 0 */
	min_amount = sched_cfs_bandwidth_slice() - cfs_rq->runtime_remaining;

	raw_spin_lock(&cfs_b->lock);
	if (cfs_b->quota == RUNTIME_INF)
		amount = min_amount;
3207
	else {
P
Paul Turner 已提交
3208 3209 3210 3211 3212 3213 3214 3215
		/*
		 * If the bandwidth pool has become inactive, then at least one
		 * period must have elapsed since the last consumption.
		 * Refresh the global state and ensure bandwidth timer becomes
		 * active.
		 */
		if (!cfs_b->timer_active) {
			__refill_cfs_bandwidth_runtime(cfs_b);
3216
			__start_cfs_bandwidth(cfs_b, false);
P
Paul Turner 已提交
3217
		}
3218 3219 3220 3221 3222 3223

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3224
	}
P
Paul Turner 已提交
3225
	expires = cfs_b->runtime_expires;
3226 3227 3228
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3229 3230 3231 3232 3233 3234 3235
	/*
	 * we may have advanced our local expiration to account for allowed
	 * spread between our sched_clock and the one on which runtime was
	 * issued.
	 */
	if ((s64)(expires - cfs_rq->runtime_expires) > 0)
		cfs_rq->runtime_expires = expires;
3236 3237

	return cfs_rq->runtime_remaining > 0;
3238 3239
}

P
Paul Turner 已提交
3240 3241 3242 3243 3244
/*
 * Note: This depends on the synchronization provided by sched_clock and the
 * fact that rq->clock snapshots this value.
 */
static void expire_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3245
{
P
Paul Turner 已提交
3246 3247 3248
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3249
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3250 3251
		return;

P
Paul Turner 已提交
3252 3253 3254 3255 3256 3257 3258 3259 3260
	if (cfs_rq->runtime_remaining < 0)
		return;

	/*
	 * If the local deadline has passed we have to consider the
	 * possibility that our sched_clock is 'fast' and the global deadline
	 * has not truly expired.
	 *
	 * Fortunately we can check determine whether this the case by checking
3261 3262 3263
	 * whether the global deadline has advanced. It is valid to compare
	 * cfs_b->runtime_expires without any locks since we only care about
	 * exact equality, so a partial write will still work.
P
Paul Turner 已提交
3264 3265
	 */

3266
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3267 3268 3269 3270 3271 3272 3273 3274
		/* extend local deadline, drift is bounded above by 2 ticks */
		cfs_rq->runtime_expires += TICK_NSEC;
	} else {
		/* global deadline is ahead, expiration has passed */
		cfs_rq->runtime_remaining = 0;
	}
}

3275
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3276 3277
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3278
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3279 3280 3281
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3282 3283
		return;

3284 3285 3286 3287 3288
	/*
	 * if we're unable to extend our runtime we resched so that the active
	 * hierarchy can be throttled
	 */
	if (!assign_cfs_rq_runtime(cfs_rq) && likely(cfs_rq->curr))
3289
		resched_curr(rq_of(cfs_rq));
3290 3291
}

3292
static __always_inline
3293
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3294
{
3295
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3296 3297 3298 3299 3300
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3301 3302
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3303
	return cfs_bandwidth_used() && cfs_rq->throttled;
3304 3305
}

3306 3307 3308
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3309
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337
}

/*
 * Ensure that neither of the group entities corresponding to src_cpu or
 * dest_cpu are members of a throttled hierarchy when performing group
 * load-balance operations.
 */
static inline int throttled_lb_pair(struct task_group *tg,
				    int src_cpu, int dest_cpu)
{
	struct cfs_rq *src_cfs_rq, *dest_cfs_rq;

	src_cfs_rq = tg->cfs_rq[src_cpu];
	dest_cfs_rq = tg->cfs_rq[dest_cpu];

	return throttled_hierarchy(src_cfs_rq) ||
	       throttled_hierarchy(dest_cfs_rq);
}

/* updated child weight may affect parent so we have to do this bottom up */
static int tg_unthrottle_up(struct task_group *tg, void *data)
{
	struct rq *rq = data;
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];

	cfs_rq->throttle_count--;
#ifdef CONFIG_SMP
	if (!cfs_rq->throttle_count) {
3338
		/* adjust cfs_rq_clock_task() */
3339
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3340
					     cfs_rq->throttled_clock_task;
3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351
	}
#endif

	return 0;
}

static int tg_throttle_down(struct task_group *tg, void *data)
{
	struct rq *rq = data;
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];

3352 3353
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3354
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3355 3356 3357 3358 3359
	cfs_rq->throttle_count++;

	return 0;
}

3360
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3361 3362 3363 3364 3365 3366 3367 3368
{
	struct rq *rq = rq_of(cfs_rq);
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	struct sched_entity *se;
	long task_delta, dequeue = 1;

	se = cfs_rq->tg->se[cpu_of(rq_of(cfs_rq))];

3369
	/* freeze hierarchy runnable averages while throttled */
3370 3371 3372
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389

	task_delta = cfs_rq->h_nr_running;
	for_each_sched_entity(se) {
		struct cfs_rq *qcfs_rq = cfs_rq_of(se);
		/* throttled entity or throttle-on-deactivate */
		if (!se->on_rq)
			break;

		if (dequeue)
			dequeue_entity(qcfs_rq, se, DEQUEUE_SLEEP);
		qcfs_rq->h_nr_running -= task_delta;

		if (qcfs_rq->load.weight)
			dequeue = 0;
	}

	if (!se)
3390
		sub_nr_running(rq, task_delta);
3391 3392

	cfs_rq->throttled = 1;
3393
	cfs_rq->throttled_clock = rq_clock(rq);
3394
	raw_spin_lock(&cfs_b->lock);
3395 3396 3397 3398 3399
	/*
	 * Add to the _head_ of the list, so that an already-started
	 * distribute_cfs_runtime will not see us
	 */
	list_add_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
3400
	if (!cfs_b->timer_active)
3401
		__start_cfs_bandwidth(cfs_b, false);
3402 3403 3404
	raw_spin_unlock(&cfs_b->lock);
}

3405
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3406 3407 3408 3409 3410 3411 3412
{
	struct rq *rq = rq_of(cfs_rq);
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	struct sched_entity *se;
	int enqueue = 1;
	long task_delta;

3413
	se = cfs_rq->tg->se[cpu_of(rq)];
3414 3415

	cfs_rq->throttled = 0;
3416 3417 3418

	update_rq_clock(rq);

3419
	raw_spin_lock(&cfs_b->lock);
3420
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
3421 3422 3423
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

3424 3425 3426
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444
	if (!cfs_rq->load.weight)
		return;

	task_delta = cfs_rq->h_nr_running;
	for_each_sched_entity(se) {
		if (se->on_rq)
			enqueue = 0;

		cfs_rq = cfs_rq_of(se);
		if (enqueue)
			enqueue_entity(cfs_rq, se, ENQUEUE_WAKEUP);
		cfs_rq->h_nr_running += task_delta;

		if (cfs_rq_throttled(cfs_rq))
			break;
	}

	if (!se)
3445
		add_nr_running(rq, task_delta);
3446 3447 3448

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
3449
		resched_curr(rq);
3450 3451 3452 3453 3454 3455
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
3456 3457
	u64 runtime;
	u64 starting_runtime = remaining;
3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487

	rcu_read_lock();
	list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
				throttled_list) {
		struct rq *rq = rq_of(cfs_rq);

		raw_spin_lock(&rq->lock);
		if (!cfs_rq_throttled(cfs_rq))
			goto next;

		runtime = -cfs_rq->runtime_remaining + 1;
		if (runtime > remaining)
			runtime = remaining;
		remaining -= runtime;

		cfs_rq->runtime_remaining += runtime;
		cfs_rq->runtime_expires = expires;

		/* we check whether we're throttled above */
		if (cfs_rq->runtime_remaining > 0)
			unthrottle_cfs_rq(cfs_rq);

next:
		raw_spin_unlock(&rq->lock);

		if (!remaining)
			break;
	}
	rcu_read_unlock();

3488
	return starting_runtime - remaining;
3489 3490
}

3491 3492 3493 3494 3495 3496 3497 3498
/*
 * Responsible for refilling a task_group's bandwidth and unthrottling its
 * cfs_rqs as appropriate. If there has been no activity within the last
 * period the timer is deactivated until scheduling resumes; cfs_b->idle is
 * used to track this state.
 */
static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun)
{
3499
	u64 runtime, runtime_expires;
3500
	int throttled;
3501 3502 3503

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
3504
		goto out_deactivate;
3505

3506
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
3507
	cfs_b->nr_periods += overrun;
3508

3509 3510 3511 3512 3513 3514
	/*
	 * idle depends on !throttled (for the case of a large deficit), and if
	 * we're going inactive then everything else can be deferred
	 */
	if (cfs_b->idle && !throttled)
		goto out_deactivate;
P
Paul Turner 已提交
3515

3516 3517 3518 3519 3520 3521 3522
	/*
	 * if we have relooped after returning idle once, we need to update our
	 * status as actually running, so that other cpus doing
	 * __start_cfs_bandwidth will stop trying to cancel us.
	 */
	cfs_b->timer_active = 1;

P
Paul Turner 已提交
3523 3524
	__refill_cfs_bandwidth_runtime(cfs_b);

3525 3526 3527
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
3528
		return 0;
3529 3530
	}

3531 3532 3533
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

3534 3535 3536
	runtime_expires = cfs_b->runtime_expires;

	/*
3537 3538 3539 3540 3541
	 * This check is repeated as we are holding onto the new bandwidth while
	 * we unthrottle. This can potentially race with an unthrottled group
	 * trying to acquire new bandwidth from the global pool. This can result
	 * in us over-using our runtime if it is all used during this loop, but
	 * only by limited amounts in that extreme case.
3542
	 */
3543 3544
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
3545 3546 3547 3548 3549 3550 3551
		raw_spin_unlock(&cfs_b->lock);
		/* we can't nest cfs_b->lock while distributing bandwidth */
		runtime = distribute_cfs_runtime(cfs_b, runtime,
						 runtime_expires);
		raw_spin_lock(&cfs_b->lock);

		throttled = !list_empty(&cfs_b->throttled_cfs_rq);
3552 3553

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3554
	}
3555

3556 3557 3558 3559 3560 3561 3562
	/*
	 * While we are ensured activity in the period following an
	 * unthrottle, this also covers the case in which the new bandwidth is
	 * insufficient to cover the existing bandwidth deficit.  (Forcing the
	 * timer to remain active while there are any throttled entities.)
	 */
	cfs_b->idle = 0;
3563

3564 3565 3566 3567 3568
	return 0;

out_deactivate:
	cfs_b->timer_active = 0;
	return 1;
3569
}
3570

3571 3572 3573 3574 3575 3576 3577
/* a cfs_rq won't donate quota below this amount */
static const u64 min_cfs_rq_runtime = 1 * NSEC_PER_MSEC;
/* minimum remaining period time to redistribute slack quota */
static const u64 min_bandwidth_expiration = 2 * NSEC_PER_MSEC;
/* how long we wait to gather additional slack before distributing */
static const u64 cfs_bandwidth_slack_period = 5 * NSEC_PER_MSEC;

3578 3579 3580 3581 3582 3583 3584
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
 * hrtimer base being cleared by __hrtimer_start_range_ns. In the case of
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640
static int runtime_refresh_within(struct cfs_bandwidth *cfs_b, u64 min_expire)
{
	struct hrtimer *refresh_timer = &cfs_b->period_timer;
	u64 remaining;

	/* if the call-back is running a quota refresh is already occurring */
	if (hrtimer_callback_running(refresh_timer))
		return 1;

	/* is a quota refresh about to occur? */
	remaining = ktime_to_ns(hrtimer_expires_remaining(refresh_timer));
	if (remaining < min_expire)
		return 1;

	return 0;
}

static void start_cfs_slack_bandwidth(struct cfs_bandwidth *cfs_b)
{
	u64 min_left = cfs_bandwidth_slack_period + min_bandwidth_expiration;

	/* if there's a quota refresh soon don't bother with slack */
	if (runtime_refresh_within(cfs_b, min_left))
		return;

	start_bandwidth_timer(&cfs_b->slack_timer,
				ns_to_ktime(cfs_bandwidth_slack_period));
}

/* we know any runtime found here is valid as update_curr() precedes return */
static void __return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	s64 slack_runtime = cfs_rq->runtime_remaining - min_cfs_rq_runtime;

	if (slack_runtime <= 0)
		return;

	raw_spin_lock(&cfs_b->lock);
	if (cfs_b->quota != RUNTIME_INF &&
	    cfs_rq->runtime_expires == cfs_b->runtime_expires) {
		cfs_b->runtime += slack_runtime;

		/* we are under rq->lock, defer unthrottling using a timer */
		if (cfs_b->runtime > sched_cfs_bandwidth_slice() &&
		    !list_empty(&cfs_b->throttled_cfs_rq))
			start_cfs_slack_bandwidth(cfs_b);
	}
	raw_spin_unlock(&cfs_b->lock);

	/* even if it's not valid for return we don't want to try again */
	cfs_rq->runtime_remaining -= slack_runtime;
}

static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
3641 3642 3643
	if (!cfs_bandwidth_used())
		return;

3644
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659
		return;

	__return_cfs_rq_runtime(cfs_rq);
}

/*
 * This is done with a timer (instead of inline with bandwidth return) since
 * it's necessary to juggle rq->locks to unthrottle their respective cfs_rqs.
 */
static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b)
{
	u64 runtime = 0, slice = sched_cfs_bandwidth_slice();
	u64 expires;

	/* confirm we're still not at a refresh boundary */
3660 3661 3662
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
3663
		return;
3664
	}
3665

3666
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
3667
		runtime = cfs_b->runtime;
3668

3669 3670 3671 3672 3673 3674 3675 3676 3677 3678
	expires = cfs_b->runtime_expires;
	raw_spin_unlock(&cfs_b->lock);

	if (!runtime)
		return;

	runtime = distribute_cfs_runtime(cfs_b, runtime, expires);

	raw_spin_lock(&cfs_b->lock);
	if (expires == cfs_b->runtime_expires)
3679
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3680 3681 3682
	raw_spin_unlock(&cfs_b->lock);
}

3683 3684 3685 3686 3687 3688 3689
/*
 * When a group wakes up we want to make sure that its quota is not already
 * expired/exceeded, otherwise it may be allowed to steal additional ticks of
 * runtime as update_curr() throttling can not not trigger until it's on-rq.
 */
static void check_enqueue_throttle(struct cfs_rq *cfs_rq)
{
3690 3691 3692
	if (!cfs_bandwidth_used())
		return;

3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707
	/* an active group must be handled by the update_curr()->put() path */
	if (!cfs_rq->runtime_enabled || cfs_rq->curr)
		return;

	/* ensure the group is not already throttled */
	if (cfs_rq_throttled(cfs_rq))
		return;

	/* update runtime allocation */
	account_cfs_rq_runtime(cfs_rq, 0);
	if (cfs_rq->runtime_remaining <= 0)
		throttle_cfs_rq(cfs_rq);
}

/* conditionally throttle active cfs_rq's from put_prev_entity() */
3708
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3709
{
3710
	if (!cfs_bandwidth_used())
3711
		return false;
3712

3713
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
3714
		return false;
3715 3716 3717 3718 3719 3720

	/*
	 * it's possible for a throttled entity to be forced into a running
	 * state (e.g. set_curr_task), in this case we're finished.
	 */
	if (cfs_rq_throttled(cfs_rq))
3721
		return true;
3722 3723

	throttle_cfs_rq(cfs_rq);
3724
	return true;
3725
}
3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743

static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, slack_timer);
	do_sched_cfs_slack_timer(cfs_b);

	return HRTIMER_NORESTART;
}

static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, period_timer);
	ktime_t now;
	int overrun;
	int idle = 0;

3744
	raw_spin_lock(&cfs_b->lock);
3745 3746 3747 3748 3749 3750 3751 3752 3753
	for (;;) {
		now = hrtimer_cb_get_time(timer);
		overrun = hrtimer_forward(timer, now, cfs_b->period);

		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
3754
	raw_spin_unlock(&cfs_b->lock);
3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779

	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
}

void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
	raw_spin_lock_init(&cfs_b->lock);
	cfs_b->runtime = 0;
	cfs_b->quota = RUNTIME_INF;
	cfs_b->period = ns_to_ktime(default_cfs_period());

	INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	cfs_b->period_timer.function = sched_cfs_period_timer;
	hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	cfs_b->slack_timer.function = sched_cfs_slack_timer;
}

static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
	cfs_rq->runtime_enabled = 0;
	INIT_LIST_HEAD(&cfs_rq->throttled_list);
}

/* requires cfs_b->lock, may release to reprogram timer */
3780
void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force)
3781 3782 3783 3784 3785 3786 3787
{
	/*
	 * The timer may be active because we're trying to set a new bandwidth
	 * period or because we're racing with the tear-down path
	 * (timer_active==0 becomes visible before the hrtimer call-back
	 * terminates).  In either case we ensure that it's re-programmed
	 */
3788 3789 3790
	while (unlikely(hrtimer_active(&cfs_b->period_timer)) &&
	       hrtimer_try_to_cancel(&cfs_b->period_timer) < 0) {
		/* bounce the lock to allow do_sched_cfs_period_timer to run */
3791
		raw_spin_unlock(&cfs_b->lock);
3792
		cpu_relax();
3793 3794
		raw_spin_lock(&cfs_b->lock);
		/* if someone else restarted the timer then we're done */
3795
		if (!force && cfs_b->timer_active)
3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808
			return;
	}

	cfs_b->timer_active = 1;
	start_bandwidth_timer(&cfs_b->period_timer, cfs_b->period);
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821
static void __maybe_unused update_runtime_enabled(struct rq *rq)
{
	struct cfs_rq *cfs_rq;

	for_each_leaf_cfs_rq(rq, cfs_rq) {
		struct cfs_bandwidth *cfs_b = &cfs_rq->tg->cfs_bandwidth;

		raw_spin_lock(&cfs_b->lock);
		cfs_rq->runtime_enabled = cfs_b->quota != RUNTIME_INF;
		raw_spin_unlock(&cfs_b->lock);
	}
}

3822
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833
{
	struct cfs_rq *cfs_rq;

	for_each_leaf_cfs_rq(rq, cfs_rq) {
		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
3834
		cfs_rq->runtime_remaining = 1;
3835 3836 3837 3838 3839 3840
		/*
		 * Offline rq is schedulable till cpu is completely disabled
		 * in take_cpu_down(), so we prevent new cfs throttling here.
		 */
		cfs_rq->runtime_enabled = 0;

3841 3842 3843 3844 3845 3846
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
3847 3848
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
3849
	return rq_clock_task(rq_of(cfs_rq));
3850 3851
}

3852
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
3853
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
3854
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
3855
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
3856 3857 3858 3859 3860

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871

static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
	return 0;
}

static inline int throttled_lb_pair(struct task_group *tg,
				    int src_cpu, int dest_cpu)
{
	return 0;
}
3872 3873 3874 3875 3876

void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}

#ifdef CONFIG_FAIR_GROUP_SCHED
static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
3877 3878
#endif

3879 3880 3881 3882 3883
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return NULL;
}
static inline void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b) {}
3884
static inline void update_runtime_enabled(struct rq *rq) {}
3885
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
3886 3887 3888

#endif /* CONFIG_CFS_BANDWIDTH */

3889 3890 3891 3892
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
3893 3894 3895 3896 3897 3898 3899 3900
#ifdef CONFIG_SCHED_HRTICK
static void hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	WARN_ON(task_rq(p) != rq);

3901
	if (cfs_rq->nr_running > 1) {
P
Peter Zijlstra 已提交
3902 3903 3904 3905 3906 3907
		u64 slice = sched_slice(cfs_rq, se);
		u64 ran = se->sum_exec_runtime - se->prev_sum_exec_runtime;
		s64 delta = slice - ran;

		if (delta < 0) {
			if (rq->curr == p)
3908
				resched_curr(rq);
P
Peter Zijlstra 已提交
3909 3910
			return;
		}
3911
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
3912 3913
	}
}
3914 3915 3916 3917 3918 3919 3920 3921 3922 3923

/*
 * called from enqueue/dequeue and updates the hrtick when the
 * current task is from our class and nr_running is low enough
 * to matter.
 */
static void hrtick_update(struct rq *rq)
{
	struct task_struct *curr = rq->curr;

3924
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
3925 3926 3927 3928 3929
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
3930
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
3931 3932 3933 3934
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
3935 3936 3937 3938

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
3939 3940
#endif

3941 3942 3943 3944 3945
/*
 * The enqueue_task method is called before nr_running is
 * increased. Here we update the fair scheduling stats and
 * then put the task into the rbtree:
 */
3946
static void
3947
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
3948 3949
{
	struct cfs_rq *cfs_rq;
3950
	struct sched_entity *se = &p->se;
3951 3952

	for_each_sched_entity(se) {
3953
		if (se->on_rq)
3954 3955
			break;
		cfs_rq = cfs_rq_of(se);
3956
		enqueue_entity(cfs_rq, se, flags);
3957 3958 3959 3960 3961 3962 3963 3964 3965

		/*
		 * end evaluation on encountering a throttled cfs_rq
		 *
		 * note: in the case of encountering a throttled cfs_rq we will
		 * post the final h_nr_running increment below.
		*/
		if (cfs_rq_throttled(cfs_rq))
			break;
3966
		cfs_rq->h_nr_running++;
3967

3968
		flags = ENQUEUE_WAKEUP;
3969
	}
P
Peter Zijlstra 已提交
3970

P
Peter Zijlstra 已提交
3971
	for_each_sched_entity(se) {
3972
		cfs_rq = cfs_rq_of(se);
3973
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
3974

3975 3976 3977
		if (cfs_rq_throttled(cfs_rq))
			break;

3978
		update_cfs_shares(cfs_rq);
3979
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
3980 3981
	}

3982 3983
	if (!se) {
		update_rq_runnable_avg(rq, rq->nr_running);
3984
		add_nr_running(rq, 1);
3985
	}
3986
	hrtick_update(rq);
3987 3988
}

3989 3990
static void set_next_buddy(struct sched_entity *se);

3991 3992 3993 3994 3995
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
3996
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
3997 3998
{
	struct cfs_rq *cfs_rq;
3999
	struct sched_entity *se = &p->se;
4000
	int task_sleep = flags & DEQUEUE_SLEEP;
4001 4002 4003

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4004
		dequeue_entity(cfs_rq, se, flags);
4005 4006 4007 4008 4009 4010 4011 4012 4013

		/*
		 * end evaluation on encountering a throttled cfs_rq
		 *
		 * note: in the case of encountering a throttled cfs_rq we will
		 * post the final h_nr_running decrement below.
		*/
		if (cfs_rq_throttled(cfs_rq))
			break;
4014
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4015

4016
		/* Don't dequeue parent if it has other entities besides us */
4017 4018 4019 4020 4021 4022 4023
		if (cfs_rq->load.weight) {
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
			if (task_sleep && parent_entity(se))
				set_next_buddy(parent_entity(se));
4024 4025 4026

			/* avoid re-evaluating load for this entity */
			se = parent_entity(se);
4027
			break;
4028
		}
4029
		flags |= DEQUEUE_SLEEP;
4030
	}
P
Peter Zijlstra 已提交
4031

P
Peter Zijlstra 已提交
4032
	for_each_sched_entity(se) {
4033
		cfs_rq = cfs_rq_of(se);
4034
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4035

4036 4037 4038
		if (cfs_rq_throttled(cfs_rq))
			break;

4039
		update_cfs_shares(cfs_rq);
4040
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
4041 4042
	}

4043
	if (!se) {
4044
		sub_nr_running(rq, 1);
4045 4046
		update_rq_runnable_avg(rq, 1);
	}
4047
	hrtick_update(rq);
4048 4049
}

4050
#ifdef CONFIG_SMP
4051 4052 4053
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
4054
	return cpu_rq(cpu)->cfs.runnable_load_avg;
4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089
}

/*
 * Return a low guess at the load of a migration-source cpu weighted
 * according to the scheduling class and "nice" value.
 *
 * We want to under-estimate the load of migration sources, to
 * balance conservatively.
 */
static unsigned long source_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long total = weighted_cpuload(cpu);

	if (type == 0 || !sched_feat(LB_BIAS))
		return total;

	return min(rq->cpu_load[type-1], total);
}

/*
 * Return a high guess at the load of a migration-target cpu weighted
 * according to the scheduling class and "nice" value.
 */
static unsigned long target_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long total = weighted_cpuload(cpu);

	if (type == 0 || !sched_feat(LB_BIAS))
		return total;

	return max(rq->cpu_load[type-1], total);
}

4090
static unsigned long capacity_of(int cpu)
4091
{
4092
	return cpu_rq(cpu)->cpu_capacity;
4093 4094 4095 4096 4097 4098
}

static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
4099
	unsigned long load_avg = rq->cfs.runnable_load_avg;
4100 4101

	if (nr_running)
4102
		return load_avg / nr_running;
4103 4104 4105 4106

	return 0;
}

4107 4108 4109 4110 4111 4112 4113
static void record_wakee(struct task_struct *p)
{
	/*
	 * Rough decay (wiping) for cost saving, don't worry
	 * about the boundary, really active task won't care
	 * about the loss.
	 */
4114
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
4115
		current->wakee_flips >>= 1;
4116 4117 4118 4119 4120 4121 4122 4123
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}
4124

4125
static void task_waking_fair(struct task_struct *p)
4126 4127 4128
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
4129 4130 4131 4132
	u64 min_vruntime;

#ifndef CONFIG_64BIT
	u64 min_vruntime_copy;
4133

4134 4135 4136 4137 4138 4139 4140 4141
	do {
		min_vruntime_copy = cfs_rq->min_vruntime_copy;
		smp_rmb();
		min_vruntime = cfs_rq->min_vruntime;
	} while (min_vruntime != min_vruntime_copy);
#else
	min_vruntime = cfs_rq->min_vruntime;
#endif
4142

4143
	se->vruntime -= min_vruntime;
4144
	record_wakee(p);
4145 4146
}

4147
#ifdef CONFIG_FAIR_GROUP_SCHED
4148 4149 4150 4151 4152 4153
/*
 * effective_load() calculates the load change as seen from the root_task_group
 *
 * Adding load to a group doesn't make a group heavier, but can cause movement
 * of group shares between cpus. Assuming the shares were perfectly aligned one
 * can calculate the shift in shares.
4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196
 *
 * Calculate the effective load difference if @wl is added (subtracted) to @tg
 * on this @cpu and results in a total addition (subtraction) of @wg to the
 * total group weight.
 *
 * Given a runqueue weight distribution (rw_i) we can compute a shares
 * distribution (s_i) using:
 *
 *   s_i = rw_i / \Sum rw_j						(1)
 *
 * Suppose we have 4 CPUs and our @tg is a direct child of the root group and
 * has 7 equal weight tasks, distributed as below (rw_i), with the resulting
 * shares distribution (s_i):
 *
 *   rw_i = {   2,   4,   1,   0 }
 *   s_i  = { 2/7, 4/7, 1/7,   0 }
 *
 * As per wake_affine() we're interested in the load of two CPUs (the CPU the
 * task used to run on and the CPU the waker is running on), we need to
 * compute the effect of waking a task on either CPU and, in case of a sync
 * wakeup, compute the effect of the current task going to sleep.
 *
 * So for a change of @wl to the local @cpu with an overall group weight change
 * of @wl we can compute the new shares distribution (s'_i) using:
 *
 *   s'_i = (rw_i + @wl) / (@wg + \Sum rw_j)				(2)
 *
 * Suppose we're interested in CPUs 0 and 1, and want to compute the load
 * differences in waking a task to CPU 0. The additional task changes the
 * weight and shares distributions like:
 *
 *   rw'_i = {   3,   4,   1,   0 }
 *   s'_i  = { 3/8, 4/8, 1/8,   0 }
 *
 * We can then compute the difference in effective weight by using:
 *
 *   dw_i = S * (s'_i - s_i)						(3)
 *
 * Where 'S' is the group weight as seen by its parent.
 *
 * Therefore the effective change in loads on CPU 0 would be 5/56 (3/8 - 2/7)
 * times the weight of the group. The effect on CPU 1 would be -4/56 (4/8 -
 * 4/7) times the weight of the group.
4197
 */
P
Peter Zijlstra 已提交
4198
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
4199
{
P
Peter Zijlstra 已提交
4200
	struct sched_entity *se = tg->se[cpu];
4201

4202
	if (!tg->parent)	/* the trivial, non-cgroup case */
4203 4204
		return wl;

P
Peter Zijlstra 已提交
4205
	for_each_sched_entity(se) {
4206
		long w, W;
P
Peter Zijlstra 已提交
4207

4208
		tg = se->my_q->tg;
4209

4210 4211 4212 4213
		/*
		 * W = @wg + \Sum rw_j
		 */
		W = wg + calc_tg_weight(tg, se->my_q);
P
Peter Zijlstra 已提交
4214

4215 4216 4217 4218
		/*
		 * w = rw_i + @wl
		 */
		w = se->my_q->load.weight + wl;
4219

4220 4221 4222 4223 4224
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
			wl = (w * tg->shares) / W;
4225 4226
		else
			wl = tg->shares;
4227

4228 4229 4230 4231 4232
		/*
		 * Per the above, wl is the new se->load.weight value; since
		 * those are clipped to [MIN_SHARES, ...) do so now. See
		 * calc_cfs_shares().
		 */
4233 4234
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
4235 4236 4237 4238

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
4239
		wl -= se->load.weight;
4240 4241 4242 4243 4244 4245 4246 4247

		/*
		 * Recursively apply this logic to all parent groups to compute
		 * the final effective load change on the root group. Since
		 * only the @tg group gets extra weight, all parent groups can
		 * only redistribute existing shares. @wl is the shift in shares
		 * resulting from this level per the above.
		 */
P
Peter Zijlstra 已提交
4248 4249
		wg = 0;
	}
4250

P
Peter Zijlstra 已提交
4251
	return wl;
4252 4253
}
#else
P
Peter Zijlstra 已提交
4254

4255
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
4256
{
4257
	return wl;
4258
}
P
Peter Zijlstra 已提交
4259

4260 4261
#endif

4262 4263
static int wake_wide(struct task_struct *p)
{
4264
	int factor = this_cpu_read(sd_llc_size);
4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283

	/*
	 * Yeah, it's the switching-frequency, could means many wakee or
	 * rapidly switch, use factor here will just help to automatically
	 * adjust the loose-degree, so bigger node will lead to more pull.
	 */
	if (p->wakee_flips > factor) {
		/*
		 * wakee is somewhat hot, it needs certain amount of cpu
		 * resource, so if waker is far more hot, prefer to leave
		 * it alone.
		 */
		if (current->wakee_flips > (factor * p->wakee_flips))
			return 1;
	}

	return 0;
}

4284
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
4285
{
4286
	s64 this_load, load;
4287
	int idx, this_cpu, prev_cpu;
4288
	unsigned long tl_per_task;
4289
	struct task_group *tg;
4290
	unsigned long weight;
4291
	int balanced;
4292

4293 4294 4295 4296 4297 4298 4299
	/*
	 * If we wake multiple tasks be careful to not bounce
	 * ourselves around too much.
	 */
	if (wake_wide(p))
		return 0;

4300 4301 4302 4303 4304
	idx	  = sd->wake_idx;
	this_cpu  = smp_processor_id();
	prev_cpu  = task_cpu(p);
	load	  = source_load(prev_cpu, idx);
	this_load = target_load(this_cpu, idx);
4305

4306 4307 4308 4309 4310
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
4311 4312 4313 4314
	if (sync) {
		tg = task_group(current);
		weight = current->se.load.weight;

4315
		this_load += effective_load(tg, this_cpu, -weight, -weight);
4316 4317
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
4318

4319 4320
	tg = task_group(p);
	weight = p->se.load.weight;
4321

4322 4323
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
4324 4325 4326
	 * due to the sync cause above having dropped this_load to 0, we'll
	 * always have an imbalance, but there's really nothing you can do
	 * about that, so that's good too.
4327 4328 4329 4330
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
4331 4332
	if (this_load > 0) {
		s64 this_eff_load, prev_eff_load;
4333 4334

		this_eff_load = 100;
4335
		this_eff_load *= capacity_of(prev_cpu);
4336 4337 4338 4339
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
4340
		prev_eff_load *= capacity_of(this_cpu);
4341 4342 4343 4344 4345
		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);

		balanced = this_eff_load <= prev_eff_load;
	} else
		balanced = true;
4346

4347
	/*
I
Ingo Molnar 已提交
4348 4349 4350
	 * If the currently running task will sleep within
	 * a reasonable amount of time then attract this newly
	 * woken task:
4351
	 */
4352 4353
	if (sync && balanced)
		return 1;
4354

4355
	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
4356 4357
	tl_per_task = cpu_avg_load_per_task(this_cpu);

4358 4359 4360
	if (balanced ||
	    (this_load <= load &&
	     this_load + target_load(prev_cpu, idx) <= tl_per_task)) {
4361 4362 4363 4364 4365
		/*
		 * This domain has SD_WAKE_AFFINE and
		 * p is cache cold in this domain, and
		 * there is no bad imbalance.
		 */
4366
		schedstat_inc(sd, ttwu_move_affine);
4367
		schedstat_inc(p, se.statistics.nr_wakeups_affine);
4368 4369 4370 4371 4372 4373

		return 1;
	}
	return 0;
}

4374 4375 4376 4377 4378
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
4379
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
4380
		  int this_cpu, int sd_flag)
4381
{
4382
	struct sched_group *idlest = NULL, *group = sd->groups;
4383
	unsigned long min_load = ULONG_MAX, this_load = 0;
4384
	int load_idx = sd->forkexec_idx;
4385
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
4386

4387 4388 4389
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

4390 4391 4392 4393
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
4394

4395 4396
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
4397
					tsk_cpus_allowed(p)))
4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415
			continue;

		local_group = cpumask_test_cpu(this_cpu,
					       sched_group_cpus(group));

		/* Tally up the load of all CPUs in the group */
		avg_load = 0;

		for_each_cpu(i, sched_group_cpus(group)) {
			/* Bias balancing toward cpus of our domain */
			if (local_group)
				load = source_load(i, load_idx);
			else
				load = target_load(i, load_idx);

			avg_load += load;
		}

4416
		/* Adjust by relative CPU capacity of the group */
4417
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442

		if (local_group) {
			this_load = avg_load;
		} else if (avg_load < min_load) {
			min_load = avg_load;
			idlest = group;
		}
	} while (group = group->next, group != sd->groups);

	if (!idlest || 100*this_load < imbalance*min_load)
		return NULL;
	return idlest;
}

/*
 * find_idlest_cpu - find the idlest cpu among the cpus in group.
 */
static int
find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
{
	unsigned long load, min_load = ULONG_MAX;
	int idlest = -1;
	int i;

	/* Traverse only the allowed CPUs */
4443
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
4444 4445 4446 4447 4448
		load = weighted_cpuload(i);

		if (load < min_load || (load == min_load && i == this_cpu)) {
			min_load = load;
			idlest = i;
4449 4450 4451
		}
	}

4452 4453
	return idlest;
}
4454

4455 4456 4457
/*
 * Try and locate an idle CPU in the sched_domain.
 */
4458
static int select_idle_sibling(struct task_struct *p, int target)
4459
{
4460
	struct sched_domain *sd;
4461
	struct sched_group *sg;
4462
	int i = task_cpu(p);
4463

4464 4465
	if (idle_cpu(target))
		return target;
4466 4467

	/*
4468
	 * If the prevous cpu is cache affine and idle, don't be stupid.
4469
	 */
4470 4471
	if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
		return i;
4472 4473

	/*
4474
	 * Otherwise, iterate the domains and find an elegible idle cpu.
4475
	 */
4476
	sd = rcu_dereference(per_cpu(sd_llc, target));
4477
	for_each_lower_domain(sd) {
4478 4479 4480 4481 4482 4483 4484
		sg = sd->groups;
		do {
			if (!cpumask_intersects(sched_group_cpus(sg),
						tsk_cpus_allowed(p)))
				goto next;

			for_each_cpu(i, sched_group_cpus(sg)) {
4485
				if (i == target || !idle_cpu(i))
4486 4487
					goto next;
			}
4488

4489 4490 4491 4492 4493 4494 4495 4496
			target = cpumask_first_and(sched_group_cpus(sg),
					tsk_cpus_allowed(p));
			goto done;
next:
			sg = sg->next;
		} while (sg != sd->groups);
	}
done:
4497 4498 4499
	return target;
}

4500
/*
4501 4502 4503
 * select_task_rq_fair: Select target runqueue for the waking task in domains
 * that have the 'sd_flag' flag set. In practice, this is SD_BALANCE_WAKE,
 * SD_BALANCE_FORK, or SD_BALANCE_EXEC.
4504
 *
4505 4506
 * Balances load by selecting the idlest cpu in the idlest group, or under
 * certain conditions an idle sibling cpu if the domain has SD_WAKE_AFFINE set.
4507
 *
4508
 * Returns the target cpu number.
4509 4510 4511
 *
 * preempt must be disabled.
 */
4512
static int
4513
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
4514
{
4515
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
4516 4517
	int cpu = smp_processor_id();
	int new_cpu = cpu;
4518
	int want_affine = 0;
4519
	int sync = wake_flags & WF_SYNC;
4520

4521
	if (p->nr_cpus_allowed == 1)
4522 4523
		return prev_cpu;

4524 4525
	if (sd_flag & SD_BALANCE_WAKE)
		want_affine = cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
4526

4527
	rcu_read_lock();
4528
	for_each_domain(cpu, tmp) {
4529 4530 4531
		if (!(tmp->flags & SD_LOAD_BALANCE))
			continue;

4532
		/*
4533 4534
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
4535
		 */
4536 4537 4538
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
4539
			break;
4540
		}
4541

4542
		if (tmp->flags & sd_flag)
4543 4544 4545
			sd = tmp;
	}

4546 4547
	if (affine_sd && cpu != prev_cpu && wake_affine(affine_sd, p, sync))
		prev_cpu = cpu;
4548

4549
	if (sd_flag & SD_BALANCE_WAKE) {
4550 4551
		new_cpu = select_idle_sibling(p, prev_cpu);
		goto unlock;
4552
	}
4553

4554 4555
	while (sd) {
		struct sched_group *group;
4556
		int weight;
4557

4558
		if (!(sd->flags & sd_flag)) {
4559 4560 4561
			sd = sd->child;
			continue;
		}
4562

4563
		group = find_idlest_group(sd, p, cpu, sd_flag);
4564 4565 4566 4567
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
4568

4569
		new_cpu = find_idlest_cpu(group, p, cpu);
4570 4571 4572 4573
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
4574
		}
4575 4576 4577

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
4578
		weight = sd->span_weight;
4579 4580
		sd = NULL;
		for_each_domain(cpu, tmp) {
4581
			if (weight <= tmp->span_weight)
4582
				break;
4583
			if (tmp->flags & sd_flag)
4584 4585 4586
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
4587
	}
4588 4589
unlock:
	rcu_read_unlock();
4590

4591
	return new_cpu;
4592
}
4593 4594 4595 4596 4597 4598 4599 4600 4601 4602

/*
 * Called immediately before a task is migrated to a new cpu; task_cpu(p) and
 * cfs_rq_of(p) references at time of call are still valid and identify the
 * previous cpu.  However, the caller only guarantees p->pi_lock is held; no
 * other assumptions, including the state of rq->lock, should be made.
 */
static void
migrate_task_rq_fair(struct task_struct *p, int next_cpu)
{
4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	/*
	 * Load tracking: accumulate removed load so that it can be processed
	 * when we next update owning cfs_rq under rq->lock.  Tasks contribute
	 * to blocked load iff they have a positive decay-count.  It can never
	 * be negative here since on-rq tasks have decay-count == 0.
	 */
	if (se->avg.decay_count) {
		se->avg.decay_count = -__synchronize_entity_decay(se);
4614 4615
		atomic_long_add(se->avg.load_avg_contrib,
						&cfs_rq->removed_load);
4616
	}
4617 4618 4619

	/* We have migrated, no longer consider this task hot */
	se->exec_start = 0;
4620
}
4621 4622
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
4623 4624
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
4625 4626 4627 4628
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
4629 4630
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
4631 4632 4633 4634 4635 4636 4637 4638 4639
	 *
	 * By using 'se' instead of 'curr' we penalize light tasks, so
	 * they get preempted easier. That is, if 'se' < 'curr' then
	 * the resulting gran will be larger, therefore penalizing the
	 * lighter, if otoh 'se' > 'curr' then the resulting gran will
	 * be smaller, again penalizing the lighter task.
	 *
	 * This is especially important for buddies when the leftmost
	 * task is higher priority than the buddy.
4640
	 */
4641
	return calc_delta_fair(gran, se);
4642 4643
}

4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665
/*
 * Should 'se' preempt 'curr'.
 *
 *             |s1
 *        |s2
 *   |s3
 *         g
 *      |<--->|c
 *
 *  w(c, s1) = -1
 *  w(c, s2) =  0
 *  w(c, s3) =  1
 *
 */
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
{
	s64 gran, vdiff = curr->vruntime - se->vruntime;

	if (vdiff <= 0)
		return -1;

P
Peter Zijlstra 已提交
4666
	gran = wakeup_gran(curr, se);
4667 4668 4669 4670 4671 4672
	if (vdiff > gran)
		return 1;

	return 0;
}

4673 4674
static void set_last_buddy(struct sched_entity *se)
{
4675 4676 4677 4678 4679
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
4680 4681 4682 4683
}

static void set_next_buddy(struct sched_entity *se)
{
4684 4685 4686 4687 4688
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
4689 4690
}

4691 4692
static void set_skip_buddy(struct sched_entity *se)
{
4693 4694
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
4695 4696
}

4697 4698 4699
/*
 * Preempt the current task with a newly woken task if needed:
 */
4700
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
4701 4702
{
	struct task_struct *curr = rq->curr;
4703
	struct sched_entity *se = &curr->se, *pse = &p->se;
4704
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
4705
	int scale = cfs_rq->nr_running >= sched_nr_latency;
4706
	int next_buddy_marked = 0;
4707

I
Ingo Molnar 已提交
4708 4709 4710
	if (unlikely(se == pse))
		return;

4711
	/*
4712
	 * This is possible from callers such as attach_tasks(), in which we
4713 4714 4715 4716 4717 4718 4719
	 * unconditionally check_prempt_curr() after an enqueue (which may have
	 * lead to a throttle).  This both saves work and prevents false
	 * next-buddy nomination below.
	 */
	if (unlikely(throttled_hierarchy(cfs_rq_of(pse))))
		return;

4720
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
4721
		set_next_buddy(pse);
4722 4723
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
4724

4725 4726 4727
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
4728 4729 4730 4731 4732 4733
	 *
	 * Note: this also catches the edge-case of curr being in a throttled
	 * group (e.g. via set_curr_task), since update_curr() (in the
	 * enqueue of curr) will have resulted in resched being set.  This
	 * prevents us from potentially nominating it as a false LAST_BUDDY
	 * below.
4734 4735 4736 4737
	 */
	if (test_tsk_need_resched(curr))
		return;

4738 4739 4740 4741 4742
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

4743
	/*
4744 4745
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
4746
	 */
4747
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
4748
		return;
4749

4750
	find_matching_se(&se, &pse);
4751
	update_curr(cfs_rq_of(se));
4752
	BUG_ON(!pse);
4753 4754 4755 4756 4757 4758 4759
	if (wakeup_preempt_entity(se, pse) == 1) {
		/*
		 * Bias pick_next to pick the sched entity that is
		 * triggering this preemption.
		 */
		if (!next_buddy_marked)
			set_next_buddy(pse);
4760
		goto preempt;
4761
	}
4762

4763
	return;
4764

4765
preempt:
4766
	resched_curr(rq);
4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780
	/*
	 * Only set the backward buddy when the current task is still
	 * on the rq. This can happen when a wakeup gets interleaved
	 * with schedule on the ->pre_schedule() or idle_balance()
	 * point, either of which can * drop the rq lock.
	 *
	 * Also, during early boot the idle thread is in the fair class,
	 * for obvious reasons its a bad idea to schedule back to it.
	 */
	if (unlikely(!se->on_rq || curr == rq->idle))
		return;

	if (sched_feat(LAST_BUDDY) && scale && entity_is_task(se))
		set_last_buddy(se);
4781 4782
}

4783 4784
static struct task_struct *
pick_next_task_fair(struct rq *rq, struct task_struct *prev)
4785 4786 4787
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
4788
	struct task_struct *p;
4789
	int new_tasks;
4790

4791
again:
4792 4793
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
4794
		goto idle;
4795

4796
	if (prev->sched_class != &fair_sched_class)
4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867
		goto simple;

	/*
	 * Because of the set_next_buddy() in dequeue_task_fair() it is rather
	 * likely that a next task is from the same cgroup as the current.
	 *
	 * Therefore attempt to avoid putting and setting the entire cgroup
	 * hierarchy, only change the part that actually changes.
	 */

	do {
		struct sched_entity *curr = cfs_rq->curr;

		/*
		 * Since we got here without doing put_prev_entity() we also
		 * have to consider cfs_rq->curr. If it is still a runnable
		 * entity, update_curr() will update its vruntime, otherwise
		 * forget we've ever seen it.
		 */
		if (curr && curr->on_rq)
			update_curr(cfs_rq);
		else
			curr = NULL;

		/*
		 * This call to check_cfs_rq_runtime() will do the throttle and
		 * dequeue its entity in the parent(s). Therefore the 'simple'
		 * nr_running test will indeed be correct.
		 */
		if (unlikely(check_cfs_rq_runtime(cfs_rq)))
			goto simple;

		se = pick_next_entity(cfs_rq, curr);
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

	p = task_of(se);

	/*
	 * Since we haven't yet done put_prev_entity and if the selected task
	 * is a different task than we started out with, try and touch the
	 * least amount of cfs_rqs.
	 */
	if (prev != p) {
		struct sched_entity *pse = &prev->se;

		while (!(cfs_rq = is_same_group(se, pse))) {
			int se_depth = se->depth;
			int pse_depth = pse->depth;

			if (se_depth <= pse_depth) {
				put_prev_entity(cfs_rq_of(pse), pse);
				pse = parent_entity(pse);
			}
			if (se_depth >= pse_depth) {
				set_next_entity(cfs_rq_of(se), se);
				se = parent_entity(se);
			}
		}

		put_prev_entity(cfs_rq, pse);
		set_next_entity(cfs_rq, se);
	}

	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);

	return p;
simple:
	cfs_rq = &rq->cfs;
#endif
4868

4869
	if (!cfs_rq->nr_running)
4870
		goto idle;
4871

4872
	put_prev_task(rq, prev);
4873

4874
	do {
4875
		se = pick_next_entity(cfs_rq, NULL);
4876
		set_next_entity(cfs_rq, se);
4877 4878 4879
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
4880
	p = task_of(se);
4881

4882 4883
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
4884 4885

	return p;
4886 4887

idle:
4888
	new_tasks = idle_balance(rq);
4889 4890 4891 4892 4893
	/*
	 * Because idle_balance() releases (and re-acquires) rq->lock, it is
	 * possible for any higher priority task to appear. In that case we
	 * must re-start the pick_next_entity() loop.
	 */
4894
	if (new_tasks < 0)
4895 4896
		return RETRY_TASK;

4897
	if (new_tasks > 0)
4898 4899 4900
		goto again;

	return NULL;
4901 4902 4903 4904 4905
}

/*
 * Account for a descheduled task:
 */
4906
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
4907 4908 4909 4910 4911 4912
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4913
		put_prev_entity(cfs_rq, se);
4914 4915 4916
	}
}

4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941
/*
 * sched_yield() is very simple
 *
 * The magic of dealing with the ->skip buddy is in pick_next_entity.
 */
static void yield_task_fair(struct rq *rq)
{
	struct task_struct *curr = rq->curr;
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
	struct sched_entity *se = &curr->se;

	/*
	 * Are we the only task in the tree?
	 */
	if (unlikely(rq->nr_running == 1))
		return;

	clear_buddies(cfs_rq, se);

	if (curr->policy != SCHED_BATCH) {
		update_rq_clock(rq);
		/*
		 * Update run-time statistics of the 'current'.
		 */
		update_curr(cfs_rq);
4942 4943 4944 4945 4946 4947
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
		 rq->skip_clock_update = 1;
4948 4949 4950 4951 4952
	}

	set_skip_buddy(se);
}

4953 4954 4955 4956
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

4957 4958
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
4959 4960 4961 4962 4963 4964 4965 4966 4967 4968
		return false;

	/* Tell the scheduler that we'd really like pse to run next. */
	set_next_buddy(se);

	yield_task_fair(rq);

	return true;
}

4969
#ifdef CONFIG_SMP
4970
/**************************************************
P
Peter Zijlstra 已提交
4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993
 * Fair scheduling class load-balancing methods.
 *
 * BASICS
 *
 * The purpose of load-balancing is to achieve the same basic fairness the
 * per-cpu scheduler provides, namely provide a proportional amount of compute
 * time to each task. This is expressed in the following equation:
 *
 *   W_i,n/P_i == W_j,n/P_j for all i,j                               (1)
 *
 * Where W_i,n is the n-th weight average for cpu i. The instantaneous weight
 * W_i,0 is defined as:
 *
 *   W_i,0 = \Sum_j w_i,j                                             (2)
 *
 * Where w_i,j is the weight of the j-th runnable task on cpu i. This weight
 * is derived from the nice value as per prio_to_weight[].
 *
 * The weight average is an exponential decay average of the instantaneous
 * weight:
 *
 *   W'_i,n = (2^n - 1) / 2^n * W_i,n + 1 / 2^n * W_i,0               (3)
 *
4994
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
4995 4996 4997 4998 4999 5000
 * fraction of 'recent' time available for SCHED_OTHER task execution. But it
 * can also include other factors [XXX].
 *
 * To achieve this balance we define a measure of imbalance which follows
 * directly from (1):
 *
5001
 *   imb_i,j = max{ avg(W/C), W_i/C_i } - min{ avg(W/C), W_j/C_j }    (4)
P
Peter Zijlstra 已提交
5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086
 *
 * We them move tasks around to minimize the imbalance. In the continuous
 * function space it is obvious this converges, in the discrete case we get
 * a few fun cases generally called infeasible weight scenarios.
 *
 * [XXX expand on:
 *     - infeasible weights;
 *     - local vs global optima in the discrete case. ]
 *
 *
 * SCHED DOMAINS
 *
 * In order to solve the imbalance equation (4), and avoid the obvious O(n^2)
 * for all i,j solution, we create a tree of cpus that follows the hardware
 * topology where each level pairs two lower groups (or better). This results
 * in O(log n) layers. Furthermore we reduce the number of cpus going up the
 * tree to only the first of the previous level and we decrease the frequency
 * of load-balance at each level inv. proportional to the number of cpus in
 * the groups.
 *
 * This yields:
 *
 *     log_2 n     1     n
 *   \Sum       { --- * --- * 2^i } = O(n)                            (5)
 *     i = 0      2^i   2^i
 *                               `- size of each group
 *         |         |     `- number of cpus doing load-balance
 *         |         `- freq
 *         `- sum over all levels
 *
 * Coupled with a limit on how many tasks we can migrate every balance pass,
 * this makes (5) the runtime complexity of the balancer.
 *
 * An important property here is that each CPU is still (indirectly) connected
 * to every other cpu in at most O(log n) steps:
 *
 * The adjacency matrix of the resulting graph is given by:
 *
 *             log_2 n     
 *   A_i,j = \Union     (i % 2^k == 0) && i / 2^(k+1) == j / 2^(k+1)  (6)
 *             k = 0
 *
 * And you'll find that:
 *
 *   A^(log_2 n)_i,j != 0  for all i,j                                (7)
 *
 * Showing there's indeed a path between every cpu in at most O(log n) steps.
 * The task movement gives a factor of O(m), giving a convergence complexity
 * of:
 *
 *   O(nm log n),  n := nr_cpus, m := nr_tasks                        (8)
 *
 *
 * WORK CONSERVING
 *
 * In order to avoid CPUs going idle while there's still work to do, new idle
 * balancing is more aggressive and has the newly idle cpu iterate up the domain
 * tree itself instead of relying on other CPUs to bring it work.
 *
 * This adds some complexity to both (5) and (8) but it reduces the total idle
 * time.
 *
 * [XXX more?]
 *
 *
 * CGROUPS
 *
 * Cgroups make a horror show out of (2), instead of a simple sum we get:
 *
 *                                s_k,i
 *   W_i,0 = \Sum_j \Prod_k w_k * -----                               (9)
 *                                 S_k
 *
 * Where
 *
 *   s_k,i = \Sum_j w_i,j,k  and  S_k = \Sum_i s_k,i                 (10)
 *
 * w_i,j,k is the weight of the j-th runnable task in the k-th cgroup on cpu i.
 *
 * The big problem is S_k, its a global sum needed to compute a local (W_i)
 * property.
 *
 * [XXX write more on how we solve this.. _after_ merging pjt's patches that
 *      rewrite all of this once again.]
 */ 
5087

5088 5089
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

5090 5091
enum fbq_type { regular, remote, all };

5092
#define LBF_ALL_PINNED	0x01
5093
#define LBF_NEED_BREAK	0x02
5094 5095
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
5096 5097 5098 5099 5100

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
5101
	int			src_cpu;
5102 5103 5104 5105

	int			dst_cpu;
	struct rq		*dst_rq;

5106 5107
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
5108
	enum cpu_idle_type	idle;
5109
	long			imbalance;
5110 5111 5112
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

5113
	unsigned int		flags;
5114 5115 5116 5117

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
5118 5119

	enum fbq_type		fbq_type;
5120
	struct list_head	tasks;
5121 5122
};

5123 5124 5125
/*
 * Is this task likely cache-hot:
 */
5126
static int task_hot(struct task_struct *p, struct lb_env *env)
5127 5128 5129
{
	s64 delta;

5130 5131
	lockdep_assert_held(&env->src_rq->lock);

5132 5133 5134 5135 5136 5137 5138 5139 5140
	if (p->sched_class != &fair_sched_class)
		return 0;

	if (unlikely(p->policy == SCHED_IDLE))
		return 0;

	/*
	 * Buddy candidates are cache hot:
	 */
5141
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
5142 5143 5144 5145 5146 5147 5148 5149 5150
			(&p->se == cfs_rq_of(&p->se)->next ||
			 &p->se == cfs_rq_of(&p->se)->last))
		return 1;

	if (sysctl_sched_migration_cost == -1)
		return 1;
	if (sysctl_sched_migration_cost == 0)
		return 0;

5151
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
5152 5153 5154 5155

	return delta < (s64)sysctl_sched_migration_cost;
}

5156 5157 5158 5159
#ifdef CONFIG_NUMA_BALANCING
/* Returns true if the destination node has incurred more faults */
static bool migrate_improves_locality(struct task_struct *p, struct lb_env *env)
{
5160
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5161 5162
	int src_nid, dst_nid;

5163
	if (!sched_feat(NUMA_FAVOUR_HIGHER) || !p->numa_faults_memory ||
5164 5165 5166 5167 5168 5169 5170
	    !(env->sd->flags & SD_NUMA)) {
		return false;
	}

	src_nid = cpu_to_node(env->src_cpu);
	dst_nid = cpu_to_node(env->dst_cpu);

5171
	if (src_nid == dst_nid)
5172 5173
		return false;

5174 5175 5176 5177
	if (numa_group) {
		/* Task is already in the group's interleave set. */
		if (node_isset(src_nid, numa_group->active_nodes))
			return false;
5178

5179 5180 5181
		/* Task is moving into the group's interleave set. */
		if (node_isset(dst_nid, numa_group->active_nodes))
			return true;
5182

5183 5184 5185 5186 5187
		return group_faults(p, dst_nid) > group_faults(p, src_nid);
	}

	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
5188 5189
		return true;

5190
	return task_faults(p, dst_nid) > task_faults(p, src_nid);
5191
}
5192 5193 5194 5195


static bool migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
{
5196
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5197 5198 5199 5200 5201
	int src_nid, dst_nid;

	if (!sched_feat(NUMA) || !sched_feat(NUMA_RESIST_LOWER))
		return false;

5202
	if (!p->numa_faults_memory || !(env->sd->flags & SD_NUMA))
5203 5204 5205 5206 5207
		return false;

	src_nid = cpu_to_node(env->src_cpu);
	dst_nid = cpu_to_node(env->dst_cpu);

5208
	if (src_nid == dst_nid)
5209 5210
		return false;

5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222
	if (numa_group) {
		/* Task is moving within/into the group's interleave set. */
		if (node_isset(dst_nid, numa_group->active_nodes))
			return false;

		/* Task is moving out of the group's interleave set. */
		if (node_isset(src_nid, numa_group->active_nodes))
			return true;

		return group_faults(p, dst_nid) < group_faults(p, src_nid);
	}

5223 5224 5225 5226
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid)
		return true;

5227
	return task_faults(p, dst_nid) < task_faults(p, src_nid);
5228 5229
}

5230 5231 5232 5233 5234 5235
#else
static inline bool migrate_improves_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5236 5237 5238 5239 5240 5241

static inline bool migrate_degrades_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5242 5243
#endif

5244 5245 5246 5247
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
5248
int can_migrate_task(struct task_struct *p, struct lb_env *env)
5249 5250
{
	int tsk_cache_hot = 0;
5251 5252 5253

	lockdep_assert_held(&env->src_rq->lock);

5254 5255
	/*
	 * We do not migrate tasks that are:
5256
	 * 1) throttled_lb_pair, or
5257
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
5258 5259
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
5260
	 */
5261 5262 5263
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

5264
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
5265
		int cpu;
5266

5267
		schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
5268

5269 5270
		env->flags |= LBF_SOME_PINNED;

5271 5272 5273 5274 5275 5276 5277 5278
		/*
		 * Remember if this task can be migrated to any other cpu in
		 * our sched_group. We may want to revisit it if we couldn't
		 * meet load balance goals by pulling other tasks on src_cpu.
		 *
		 * Also avoid computing new_dst_cpu if we have already computed
		 * one in current iteration.
		 */
5279
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
5280 5281
			return 0;

5282 5283 5284
		/* Prevent to re-select dst_cpu via env's cpus */
		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
			if (cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) {
5285
				env->flags |= LBF_DST_PINNED;
5286 5287 5288
				env->new_dst_cpu = cpu;
				break;
			}
5289
		}
5290

5291 5292
		return 0;
	}
5293 5294

	/* Record that we found atleast one task that could run on dst_cpu */
5295
	env->flags &= ~LBF_ALL_PINNED;
5296

5297
	if (task_running(env->src_rq, p)) {
5298
		schedstat_inc(p, se.statistics.nr_failed_migrations_running);
5299 5300 5301 5302 5303
		return 0;
	}

	/*
	 * Aggressive migration if:
5304 5305 5306
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
5307
	 */
5308
	tsk_cache_hot = task_hot(p, env);
5309 5310
	if (!tsk_cache_hot)
		tsk_cache_hot = migrate_degrades_locality(p, env);
5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321

	if (migrate_improves_locality(p, env)) {
#ifdef CONFIG_SCHEDSTATS
		if (tsk_cache_hot) {
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
			schedstat_inc(p, se.statistics.nr_forced_migrations);
		}
#endif
		return 1;
	}

5322
	if (!tsk_cache_hot ||
5323
		env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
Z
Zhang Hang 已提交
5324

5325
		if (tsk_cache_hot) {
5326
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
5327
			schedstat_inc(p, se.statistics.nr_forced_migrations);
5328
		}
Z
Zhang Hang 已提交
5329

5330 5331 5332
		return 1;
	}

Z
Zhang Hang 已提交
5333 5334
	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
	return 0;
5335 5336
}

5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348
/*
 * detach_task() -- detach the task for the migration specified in env
 */
static void detach_task(struct task_struct *p, struct lb_env *env)
{
	lockdep_assert_held(&env->src_rq->lock);

	deactivate_task(env->src_rq, p, 0);
	p->on_rq = TASK_ON_RQ_MIGRATING;
	set_task_cpu(p, env->dst_cpu);
}

5349
/*
5350
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
5351 5352
 * part of active balancing operations within "domain".
 *
5353
 * Returns a task if successful and NULL otherwise.
5354
 */
5355
static struct task_struct *detach_one_task(struct lb_env *env)
5356 5357 5358
{
	struct task_struct *p, *n;

5359 5360
	lockdep_assert_held(&env->src_rq->lock);

5361 5362 5363
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
5364

5365
		detach_task(p, env);
5366

5367
		/*
5368
		 * Right now, this is only the second place where
5369
		 * lb_gained[env->idle] is updated (other is detach_tasks)
5370
		 * so we can safely collect stats here rather than
5371
		 * inside detach_tasks().
5372 5373
		 */
		schedstat_inc(env->sd, lb_gained[env->idle]);
5374
		return p;
5375
	}
5376 5377 5378
	return NULL;
}

5379 5380
static const unsigned int sched_nr_migrate_break = 32;

5381
/*
5382 5383
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
5384
 *
5385
 * Returns number of detached tasks if successful and 0 otherwise.
5386
 */
5387
static int detach_tasks(struct lb_env *env)
5388
{
5389 5390
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
5391
	unsigned long load;
5392 5393 5394
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
5395

5396
	if (env->imbalance <= 0)
5397
		return 0;
5398

5399 5400
	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
5401

5402 5403
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
5404
		if (env->loop > env->loop_max)
5405
			break;
5406 5407

		/* take a breather every nr_migrate tasks */
5408
		if (env->loop > env->loop_break) {
5409
			env->loop_break += sched_nr_migrate_break;
5410
			env->flags |= LBF_NEED_BREAK;
5411
			break;
5412
		}
5413

5414
		if (!can_migrate_task(p, env))
5415 5416 5417
			goto next;

		load = task_h_load(p);
5418

5419
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
5420 5421
			goto next;

5422
		if ((load / 2) > env->imbalance)
5423
			goto next;
5424

5425 5426 5427 5428
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
5429
		env->imbalance -= load;
5430 5431

#ifdef CONFIG_PREEMPT
5432 5433
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
5434
		 * kernels will stop after the first task is detached to minimize
5435 5436
		 * the critical section.
		 */
5437
		if (env->idle == CPU_NEWLY_IDLE)
5438
			break;
5439 5440
#endif

5441 5442 5443 5444
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
5445
		if (env->imbalance <= 0)
5446
			break;
5447 5448 5449

		continue;
next:
5450
		list_move_tail(&p->se.group_node, tasks);
5451
	}
5452

5453
	/*
5454 5455 5456
	 * Right now, this is one of only two places we collect this stat
	 * so we can safely collect detach_one_task() stats here rather
	 * than inside detach_one_task().
5457
	 */
5458
	schedstat_add(env->sd, lb_gained[env->idle], detached);
5459

5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505
	return detached;
}

/*
 * attach_task() -- attach the task detached by detach_task() to its new rq.
 */
static void attach_task(struct rq *rq, struct task_struct *p)
{
	lockdep_assert_held(&rq->lock);

	BUG_ON(task_rq(p) != rq);
	p->on_rq = TASK_ON_RQ_QUEUED;
	activate_task(rq, p, 0);
	check_preempt_curr(rq, p, 0);
}

/*
 * attach_one_task() -- attaches the task returned from detach_one_task() to
 * its new rq.
 */
static void attach_one_task(struct rq *rq, struct task_struct *p)
{
	raw_spin_lock(&rq->lock);
	attach_task(rq, p);
	raw_spin_unlock(&rq->lock);
}

/*
 * attach_tasks() -- attaches all tasks detached by detach_tasks() to their
 * new rq.
 */
static void attach_tasks(struct lb_env *env)
{
	struct list_head *tasks = &env->tasks;
	struct task_struct *p;

	raw_spin_lock(&env->dst_rq->lock);

	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
		list_del_init(&p->se.group_node);

		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
5506 5507
}

P
Peter Zijlstra 已提交
5508
#ifdef CONFIG_FAIR_GROUP_SCHED
5509 5510 5511
/*
 * update tg->load_weight by folding this cpu's load_avg
 */
5512
static void __update_blocked_averages_cpu(struct task_group *tg, int cpu)
5513
{
5514 5515
	struct sched_entity *se = tg->se[cpu];
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
5516

5517 5518 5519
	/* throttled entities do not contribute to load */
	if (throttled_hierarchy(cfs_rq))
		return;
5520

5521
	update_cfs_rq_blocked_load(cfs_rq, 1);
5522

5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536
	if (se) {
		update_entity_load_avg(se, 1);
		/*
		 * We pivot on our runnable average having decayed to zero for
		 * list removal.  This generally implies that all our children
		 * have also been removed (modulo rounding error or bandwidth
		 * control); however, such cases are rare and we can fix these
		 * at enqueue.
		 *
		 * TODO: fix up out-of-order children on enqueue.
		 */
		if (!se->avg.runnable_avg_sum && !cfs_rq->nr_running)
			list_del_leaf_cfs_rq(cfs_rq);
	} else {
5537
		struct rq *rq = rq_of(cfs_rq);
5538 5539
		update_rq_runnable_avg(rq, rq->nr_running);
	}
5540 5541
}

5542
static void update_blocked_averages(int cpu)
5543 5544
{
	struct rq *rq = cpu_rq(cpu);
5545 5546
	struct cfs_rq *cfs_rq;
	unsigned long flags;
5547

5548 5549
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
5550 5551 5552 5553
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
5554
	for_each_leaf_cfs_rq(rq, cfs_rq) {
5555 5556 5557 5558 5559 5560
		/*
		 * Note: We may want to consider periodically releasing
		 * rq->lock about these updates so that creating many task
		 * groups does not result in continually extending hold time.
		 */
		__update_blocked_averages_cpu(cfs_rq->tg, rq->cpu);
5561
	}
5562 5563

	raw_spin_unlock_irqrestore(&rq->lock, flags);
5564 5565
}

5566
/*
5567
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
5568 5569 5570
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
5571
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
5572
{
5573 5574
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
5575
	unsigned long now = jiffies;
5576
	unsigned long load;
5577

5578
	if (cfs_rq->last_h_load_update == now)
5579 5580
		return;

5581 5582 5583 5584 5585 5586 5587
	cfs_rq->h_load_next = NULL;
	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
		cfs_rq->h_load_next = se;
		if (cfs_rq->last_h_load_update == now)
			break;
	}
5588

5589
	if (!se) {
5590
		cfs_rq->h_load = cfs_rq->runnable_load_avg;
5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
		load = div64_ul(load * se->avg.load_avg_contrib,
				cfs_rq->runnable_load_avg + 1);
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
5602 5603
}

5604
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
5605
{
5606
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
5607

5608
	update_cfs_rq_h_load(cfs_rq);
5609 5610
	return div64_ul(p->se.avg.load_avg_contrib * cfs_rq->h_load,
			cfs_rq->runnable_load_avg + 1);
P
Peter Zijlstra 已提交
5611 5612
}
#else
5613
static inline void update_blocked_averages(int cpu)
5614 5615 5616
{
}

5617
static unsigned long task_h_load(struct task_struct *p)
5618
{
5619
	return p->se.avg.load_avg_contrib;
5620
}
P
Peter Zijlstra 已提交
5621
#endif
5622 5623

/********** Helpers for find_busiest_group ************************/
5624 5625 5626 5627 5628 5629 5630

enum group_type {
	group_other = 0,
	group_imbalanced,
	group_overloaded,
};

5631 5632 5633 5634 5635 5636 5637
/*
 * sg_lb_stats - stats of a sched_group required for load_balancing
 */
struct sg_lb_stats {
	unsigned long avg_load; /*Avg load across the CPUs of the group */
	unsigned long group_load; /* Total load over the CPUs of the group */
	unsigned long sum_weighted_load; /* Weighted load of group's tasks */
J
Joonsoo Kim 已提交
5638
	unsigned long load_per_task;
5639
	unsigned long group_capacity;
5640
	unsigned int sum_nr_running; /* Nr tasks running in the group */
5641
	unsigned int group_capacity_factor;
5642 5643
	unsigned int idle_cpus;
	unsigned int group_weight;
5644
	enum group_type group_type;
5645
	int group_has_free_capacity;
5646 5647 5648 5649
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
5650 5651
};

J
Joonsoo Kim 已提交
5652 5653 5654 5655 5656 5657 5658 5659
/*
 * sd_lb_stats - Structure to store the statistics of a sched_domain
 *		 during load balancing.
 */
struct sd_lb_stats {
	struct sched_group *busiest;	/* Busiest group in this sd */
	struct sched_group *local;	/* Local group in this sd */
	unsigned long total_load;	/* Total load of all groups in sd */
5660
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
5661 5662 5663
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
5664
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
5665 5666
};

5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678
static inline void init_sd_lb_stats(struct sd_lb_stats *sds)
{
	/*
	 * Skimp on the clearing to avoid duplicate work. We can avoid clearing
	 * local_stat because update_sg_lb_stats() does a full clear/assignment.
	 * We must however clear busiest_stat::avg_load because
	 * update_sd_pick_busiest() reads this before assignment.
	 */
	*sds = (struct sd_lb_stats){
		.busiest = NULL,
		.local = NULL,
		.total_load = 0UL,
5679
		.total_capacity = 0UL,
5680 5681
		.busiest_stat = {
			.avg_load = 0UL,
5682 5683
			.sum_nr_running = 0,
			.group_type = group_other,
5684 5685 5686 5687
		},
	};
}

5688 5689 5690
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
5691
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
5692 5693
 *
 * Return: The load index.
5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715
 */
static inline int get_sd_load_idx(struct sched_domain *sd,
					enum cpu_idle_type idle)
{
	int load_idx;

	switch (idle) {
	case CPU_NOT_IDLE:
		load_idx = sd->busy_idx;
		break;

	case CPU_NEWLY_IDLE:
		load_idx = sd->newidle_idx;
		break;
	default:
		load_idx = sd->idle_idx;
		break;
	}

	return load_idx;
}

5716
static unsigned long default_scale_capacity(struct sched_domain *sd, int cpu)
5717
{
5718
	return SCHED_CAPACITY_SCALE;
5719 5720
}

5721
unsigned long __weak arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
5722
{
5723
	return default_scale_capacity(sd, cpu);
5724 5725
}

5726
static unsigned long default_scale_smt_capacity(struct sched_domain *sd, int cpu)
5727
{
5728
	unsigned long weight = sd->span_weight;
5729 5730 5731 5732 5733 5734 5735
	unsigned long smt_gain = sd->smt_gain;

	smt_gain /= weight;

	return smt_gain;
}

5736
unsigned long __weak arch_scale_smt_capacity(struct sched_domain *sd, int cpu)
5737
{
5738
	return default_scale_smt_capacity(sd, cpu);
5739 5740
}

5741
static unsigned long scale_rt_capacity(int cpu)
5742 5743
{
	struct rq *rq = cpu_rq(cpu);
5744
	u64 total, available, age_stamp, avg;
5745
	s64 delta;
5746

5747 5748 5749 5750 5751 5752 5753
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
	age_stamp = ACCESS_ONCE(rq->age_stamp);
	avg = ACCESS_ONCE(rq->rt_avg);

5754 5755 5756 5757 5758
	delta = rq_clock(rq) - age_stamp;
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
5759

5760
	if (unlikely(total < avg)) {
5761
		/* Ensures that capacity won't end up being negative */
5762 5763
		available = 0;
	} else {
5764
		available = total - avg;
5765
	}
5766

5767 5768
	if (unlikely((s64)total < SCHED_CAPACITY_SCALE))
		total = SCHED_CAPACITY_SCALE;
5769

5770
	total >>= SCHED_CAPACITY_SHIFT;
5771 5772 5773 5774

	return div_u64(available, total);
}

5775
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
5776
{
5777
	unsigned long weight = sd->span_weight;
5778
	unsigned long capacity = SCHED_CAPACITY_SCALE;
5779 5780
	struct sched_group *sdg = sd->groups;

5781 5782
	if ((sd->flags & SD_SHARE_CPUCAPACITY) && weight > 1) {
		if (sched_feat(ARCH_CAPACITY))
5783
			capacity *= arch_scale_smt_capacity(sd, cpu);
5784
		else
5785
			capacity *= default_scale_smt_capacity(sd, cpu);
5786

5787
		capacity >>= SCHED_CAPACITY_SHIFT;
5788 5789
	}

5790
	sdg->sgc->capacity_orig = capacity;
5791

5792
	if (sched_feat(ARCH_CAPACITY))
5793
		capacity *= arch_scale_freq_capacity(sd, cpu);
5794
	else
5795
		capacity *= default_scale_capacity(sd, cpu);
5796

5797
	capacity >>= SCHED_CAPACITY_SHIFT;
5798

5799
	capacity *= scale_rt_capacity(cpu);
5800
	capacity >>= SCHED_CAPACITY_SHIFT;
5801

5802 5803
	if (!capacity)
		capacity = 1;
5804

5805 5806
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
5807 5808
}

5809
void update_group_capacity(struct sched_domain *sd, int cpu)
5810 5811 5812
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
5813
	unsigned long capacity, capacity_orig;
5814 5815 5816 5817
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
5818
	sdg->sgc->next_update = jiffies + interval;
5819 5820

	if (!child) {
5821
		update_cpu_capacity(sd, cpu);
5822 5823 5824
		return;
	}

5825
	capacity_orig = capacity = 0;
5826

P
Peter Zijlstra 已提交
5827 5828 5829 5830 5831 5832
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

5833
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
5834
			struct sched_group_capacity *sgc;
5835
			struct rq *rq = cpu_rq(cpu);
5836

5837
			/*
5838
			 * build_sched_domains() -> init_sched_groups_capacity()
5839 5840 5841
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
5842 5843
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
5844
			 *
5845
			 * This avoids capacity/capacity_orig from being 0 and
5846 5847
			 * causing divide-by-zero issues on boot.
			 *
5848
			 * Runtime updates will correct capacity_orig.
5849 5850
			 */
			if (unlikely(!rq->sd)) {
5851 5852
				capacity_orig += capacity_of(cpu);
				capacity += capacity_of(cpu);
5853 5854
				continue;
			}
5855

5856 5857 5858
			sgc = rq->sd->groups->sgc;
			capacity_orig += sgc->capacity_orig;
			capacity += sgc->capacity;
5859
		}
P
Peter Zijlstra 已提交
5860 5861 5862 5863 5864 5865 5866 5867
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
		 */ 

		group = child->groups;
		do {
5868 5869
			capacity_orig += group->sgc->capacity_orig;
			capacity += group->sgc->capacity;
P
Peter Zijlstra 已提交
5870 5871 5872
			group = group->next;
		} while (group != child->groups);
	}
5873

5874 5875
	sdg->sgc->capacity_orig = capacity_orig;
	sdg->sgc->capacity = capacity;
5876 5877
}

5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888
/*
 * Try and fix up capacity for tiny siblings, this is needed when
 * things like SD_ASYM_PACKING need f_b_g to select another sibling
 * which on its own isn't powerful enough.
 *
 * See update_sd_pick_busiest() and check_asym_packing().
 */
static inline int
fix_small_capacity(struct sched_domain *sd, struct sched_group *group)
{
	/*
5889
	 * Only siblings can have significantly less than SCHED_CAPACITY_SCALE
5890
	 */
5891
	if (!(sd->flags & SD_SHARE_CPUCAPACITY))
5892 5893 5894
		return 0;

	/*
5895
	 * If ~90% of the cpu_capacity is still there, we're good.
5896
	 */
5897
	if (group->sgc->capacity * 32 > group->sgc->capacity_orig * 29)
5898 5899 5900 5901 5902
		return 1;

	return 0;
}

5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918
/*
 * Group imbalance indicates (and tries to solve) the problem where balancing
 * groups is inadequate due to tsk_cpus_allowed() constraints.
 *
 * Imagine a situation of two groups of 4 cpus each and 4 tasks each with a
 * cpumask covering 1 cpu of the first group and 3 cpus of the second group.
 * Something like:
 *
 * 	{ 0 1 2 3 } { 4 5 6 7 }
 * 	        *     * * *
 *
 * If we were to balance group-wise we'd place two tasks in the first group and
 * two tasks in the second group. Clearly this is undesired as it will overload
 * cpu 3 and leave one of the cpus in the second group unused.
 *
 * The current solution to this issue is detecting the skew in the first group
5919 5920
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
5921 5922
 *
 * When this is so detected; this group becomes a candidate for busiest; see
5923
 * update_sd_pick_busiest(). And calculate_imbalance() and
5924
 * find_busiest_group() avoid some of the usual balance conditions to allow it
5925 5926 5927 5928 5929 5930 5931
 * to create an effective group imbalance.
 *
 * This is a somewhat tricky proposition since the next run might not find the
 * group imbalance and decide the groups need to be balanced again. A most
 * subtle and fragile situation.
 */

5932
static inline int sg_imbalanced(struct sched_group *group)
5933
{
5934
	return group->sgc->imbalance;
5935 5936
}

5937
/*
5938
 * Compute the group capacity factor.
5939
 *
5940
 * Avoid the issue where N*frac(smt_capacity) >= 1 creates 'phantom' cores by
5941
 * first dividing out the smt factor and computing the actual number of cores
5942
 * and limit unit capacity with that.
5943
 */
5944
static inline int sg_capacity_factor(struct lb_env *env, struct sched_group *group)
5945
{
5946
	unsigned int capacity_factor, smt, cpus;
5947
	unsigned int capacity, capacity_orig;
5948

5949 5950
	capacity = group->sgc->capacity;
	capacity_orig = group->sgc->capacity_orig;
5951
	cpus = group->group_weight;
5952

5953
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_capacity < 2 */
5954
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, capacity_orig);
5955
	capacity_factor = cpus / smt; /* cores */
5956

5957
	capacity_factor = min_t(unsigned,
5958
		capacity_factor, DIV_ROUND_CLOSEST(capacity, SCHED_CAPACITY_SCALE));
5959 5960
	if (!capacity_factor)
		capacity_factor = fix_small_capacity(env->sd, group);
5961

5962
	return capacity_factor;
5963 5964
}

5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976
static enum group_type
group_classify(struct sched_group *group, struct sg_lb_stats *sgs)
{
	if (sgs->sum_nr_running > sgs->group_capacity_factor)
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

5977 5978
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
5979
 * @env: The load balancing environment.
5980 5981 5982 5983
 * @group: sched_group whose statistics are to be updated.
 * @load_idx: Load index of sched_domain of this_cpu for load calc.
 * @local_group: Does group contain this_cpu.
 * @sgs: variable to hold the statistics for this group.
5984
 * @overload: Indicate more than one runnable task for any CPU.
5985
 */
5986 5987
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
5988 5989
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
5990
{
5991
	unsigned long load;
5992
	int i;
5993

5994 5995
	memset(sgs, 0, sizeof(*sgs));

5996
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
5997 5998 5999
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
6000
		if (local_group)
6001
			load = target_load(i, load_idx);
6002
		else
6003 6004 6005
			load = source_load(i, load_idx);

		sgs->group_load += load;
6006
		sgs->sum_nr_running += rq->nr_running;
6007 6008 6009 6010

		if (rq->nr_running > 1)
			*overload = true;

6011 6012 6013 6014
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
6015
		sgs->sum_weighted_load += weighted_cpuload(i);
6016 6017
		if (idle_cpu(i))
			sgs->idle_cpus++;
6018 6019
	}

6020 6021
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
6022
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
6023

6024
	if (sgs->sum_nr_running)
6025
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
6026

6027
	sgs->group_weight = group->group_weight;
6028
	sgs->group_capacity_factor = sg_capacity_factor(env, group);
6029
	sgs->group_type = group_classify(group, sgs);
6030

6031
	if (sgs->group_capacity_factor > sgs->sum_nr_running)
6032
		sgs->group_has_free_capacity = 1;
6033 6034
}

6035 6036
/**
 * update_sd_pick_busiest - return 1 on busiest group
6037
 * @env: The load balancing environment.
6038 6039
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
6040
 * @sgs: sched_group statistics
6041 6042 6043
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
6044 6045 6046
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
6047
 */
6048
static bool update_sd_pick_busiest(struct lb_env *env,
6049 6050
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
6051
				   struct sg_lb_stats *sgs)
6052
{
6053
	struct sg_lb_stats *busiest = &sds->busiest_stat;
6054

6055
	if (sgs->group_type > busiest->group_type)
6056 6057
		return true;

6058 6059 6060 6061 6062 6063 6064 6065
	if (sgs->group_type < busiest->group_type)
		return false;

	if (sgs->avg_load <= busiest->avg_load)
		return false;

	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
6066 6067 6068 6069 6070 6071 6072
		return true;

	/*
	 * ASYM_PACKING needs to move all the work to the lowest
	 * numbered CPUs in the group, therefore mark all groups
	 * higher than ourself as busy.
	 */
6073
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
6074 6075 6076 6077 6078 6079 6080 6081 6082 6083
		if (!sds->busiest)
			return true;

		if (group_first_cpu(sds->busiest) > group_first_cpu(sg))
			return true;
	}

	return false;
}

6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113
#ifdef CONFIG_NUMA_BALANCING
static inline enum fbq_type fbq_classify_group(struct sg_lb_stats *sgs)
{
	if (sgs->sum_nr_running > sgs->nr_numa_running)
		return regular;
	if (sgs->sum_nr_running > sgs->nr_preferred_running)
		return remote;
	return all;
}

static inline enum fbq_type fbq_classify_rq(struct rq *rq)
{
	if (rq->nr_running > rq->nr_numa_running)
		return regular;
	if (rq->nr_running > rq->nr_preferred_running)
		return remote;
	return all;
}
#else
static inline enum fbq_type fbq_classify_group(struct sg_lb_stats *sgs)
{
	return all;
}

static inline enum fbq_type fbq_classify_rq(struct rq *rq)
{
	return regular;
}
#endif /* CONFIG_NUMA_BALANCING */

6114
/**
6115
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
6116
 * @env: The load balancing environment.
6117 6118
 * @sds: variable to hold the statistics for this sched_domain.
 */
6119
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
6120
{
6121 6122
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
6123
	struct sg_lb_stats tmp_sgs;
6124
	int load_idx, prefer_sibling = 0;
6125
	bool overload = false;
6126 6127 6128 6129

	if (child && child->flags & SD_PREFER_SIBLING)
		prefer_sibling = 1;

6130
	load_idx = get_sd_load_idx(env->sd, env->idle);
6131 6132

	do {
J
Joonsoo Kim 已提交
6133
		struct sg_lb_stats *sgs = &tmp_sgs;
6134 6135
		int local_group;

6136
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
6137 6138 6139
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
6140 6141

			if (env->idle != CPU_NEWLY_IDLE ||
6142 6143
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
6144
		}
6145

6146 6147
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
6148

6149 6150 6151
		if (local_group)
			goto next_group;

6152 6153
		/*
		 * In case the child domain prefers tasks go to siblings
6154
		 * first, lower the sg capacity factor to one so that we'll try
6155 6156
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
6157
		 * these excess tasks, i.e. nr_running < group_capacity_factor. The
6158 6159 6160
		 * extra check prevents the case where you always pull from the
		 * heaviest group when it is already under-utilized (possible
		 * with a large weight task outweighs the tasks on the system).
6161
		 */
6162
		if (prefer_sibling && sds->local &&
6163
		    sds->local_stat.group_has_free_capacity)
6164
			sgs->group_capacity_factor = min(sgs->group_capacity_factor, 1U);
6165

6166
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
6167
			sds->busiest = sg;
J
Joonsoo Kim 已提交
6168
			sds->busiest_stat = *sgs;
6169 6170
		}

6171 6172 6173
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
6174
		sds->total_capacity += sgs->group_capacity;
6175

6176
		sg = sg->next;
6177
	} while (sg != env->sd->groups);
6178 6179 6180

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
6181 6182 6183 6184 6185 6186 6187

	if (!env->sd->parent) {
		/* update overload indicator if we are at root domain */
		if (env->dst_rq->rd->overload != overload)
			env->dst_rq->rd->overload = overload;
	}

6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206
}

/**
 * check_asym_packing - Check to see if the group is packed into the
 *			sched doman.
 *
 * This is primarily intended to used at the sibling level.  Some
 * cores like POWER7 prefer to use lower numbered SMT threads.  In the
 * case of POWER7, it can move to lower SMT modes only when higher
 * threads are idle.  When in lower SMT modes, the threads will
 * perform better since they share less core resources.  Hence when we
 * have idle threads, we want them to be the higher ones.
 *
 * This packing function is run on idle threads.  It checks to see if
 * the busiest CPU in this domain (core in the P7 case) has a higher
 * CPU number than the packing function is being run on.  Here we are
 * assuming lower CPU number will be equivalent to lower a SMT thread
 * number.
 *
6207
 * Return: 1 when packing is required and a task should be moved to
6208 6209
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
6210
 * @env: The load balancing environment.
6211 6212
 * @sds: Statistics of the sched_domain which is to be packed
 */
6213
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
6214 6215 6216
{
	int busiest_cpu;

6217
	if (!(env->sd->flags & SD_ASYM_PACKING))
6218 6219 6220 6221 6222 6223
		return 0;

	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
6224
	if (env->dst_cpu > busiest_cpu)
6225 6226
		return 0;

6227
	env->imbalance = DIV_ROUND_CLOSEST(
6228
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
6229
		SCHED_CAPACITY_SCALE);
6230

6231
	return 1;
6232 6233 6234 6235 6236 6237
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
6238
 * @env: The load balancing environment.
6239 6240
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
6241 6242
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6243
{
6244
	unsigned long tmp, capa_now = 0, capa_move = 0;
6245
	unsigned int imbn = 2;
6246
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
6247
	struct sg_lb_stats *local, *busiest;
6248

J
Joonsoo Kim 已提交
6249 6250
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6251

J
Joonsoo Kim 已提交
6252 6253 6254 6255
	if (!local->sum_nr_running)
		local->load_per_task = cpu_avg_load_per_task(env->dst_cpu);
	else if (busiest->load_per_task > local->load_per_task)
		imbn = 1;
6256

J
Joonsoo Kim 已提交
6257
	scaled_busy_load_per_task =
6258
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6259
		busiest->group_capacity;
J
Joonsoo Kim 已提交
6260

6261 6262
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
6263
		env->imbalance = busiest->load_per_task;
6264 6265 6266 6267 6268
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
6269
	 * however we may be able to increase total CPU capacity used by
6270 6271 6272
	 * moving them.
	 */

6273
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
6274
			min(busiest->load_per_task, busiest->avg_load);
6275
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
6276
			min(local->load_per_task, local->avg_load);
6277
	capa_now /= SCHED_CAPACITY_SCALE;
6278 6279

	/* Amount of load we'd subtract */
6280
	if (busiest->avg_load > scaled_busy_load_per_task) {
6281
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
6282
			    min(busiest->load_per_task,
6283
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
6284
	}
6285 6286

	/* Amount of load we'd add */
6287
	if (busiest->avg_load * busiest->group_capacity <
6288
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
6289 6290
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
6291
	} else {
6292
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6293
		      local->group_capacity;
J
Joonsoo Kim 已提交
6294
	}
6295
	capa_move += local->group_capacity *
6296
		    min(local->load_per_task, local->avg_load + tmp);
6297
	capa_move /= SCHED_CAPACITY_SCALE;
6298 6299

	/* Move if we gain throughput */
6300
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
6301
		env->imbalance = busiest->load_per_task;
6302 6303 6304 6305 6306
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
6307
 * @env: load balance environment
6308 6309
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
6310
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6311
{
6312
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
6313 6314 6315 6316
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6317

6318
	if (busiest->group_type == group_imbalanced) {
6319 6320 6321 6322
		/*
		 * In the group_imb case we cannot rely on group-wide averages
		 * to ensure cpu-load equilibrium, look at wider averages. XXX
		 */
J
Joonsoo Kim 已提交
6323 6324
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
6325 6326
	}

6327 6328 6329
	/*
	 * In the presence of smp nice balancing, certain scenarios can have
	 * max load less than avg load(as we skip the groups at or below
6330
	 * its cpu_capacity, while calculating max_load..)
6331
	 */
6332 6333
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
6334 6335
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
6336 6337
	}

6338 6339 6340 6341 6342
	/*
	 * If there aren't any idle cpus, avoid creating some.
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
J
Joonsoo Kim 已提交
6343
		load_above_capacity =
6344
			(busiest->sum_nr_running - busiest->group_capacity_factor);
6345

6346
		load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_CAPACITY_SCALE);
6347
		load_above_capacity /= busiest->group_capacity;
6348 6349 6350 6351 6352 6353 6354 6355 6356 6357
	}

	/*
	 * We're trying to get all the cpus to the average_load, so we don't
	 * want to push ourselves above the average load, nor do we wish to
	 * reduce the max loaded cpu below the average load. At the same time,
	 * we also don't want to reduce the group load below the group capacity
	 * (so that we can implement power-savings policies etc). Thus we look
	 * for the minimum possible imbalance.
	 */
6358
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
6359 6360

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
6361
	env->imbalance = min(
6362 6363
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
6364
	) / SCHED_CAPACITY_SCALE;
6365 6366 6367

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
6368
	 * there is no guarantee that any tasks will be moved so we'll have
6369 6370 6371
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
6372
	if (env->imbalance < busiest->load_per_task)
6373
		return fix_small_imbalance(env, sds);
6374
}
6375

6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
 * if there is an imbalance. If there isn't an imbalance, and
 * the user has opted for power-savings, it returns a group whose
 * CPUs can be put to idle by rebalancing those tasks elsewhere, if
 * such a group exists.
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
6388
 * @env: The load balancing environment.
6389
 *
6390
 * Return:	- The busiest group if imbalance exists.
6391 6392 6393 6394
 *		- If no imbalance and user has opted for power-savings balance,
 *		   return the least loaded group whose CPUs can be
 *		   put to idle by rebalancing its tasks onto our group.
 */
J
Joonsoo Kim 已提交
6395
static struct sched_group *find_busiest_group(struct lb_env *env)
6396
{
J
Joonsoo Kim 已提交
6397
	struct sg_lb_stats *local, *busiest;
6398 6399
	struct sd_lb_stats sds;

6400
	init_sd_lb_stats(&sds);
6401 6402 6403 6404 6405

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
6406
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
6407 6408
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
6409

6410 6411
	if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
	    check_asym_packing(env, &sds))
6412 6413
		return sds.busiest;

6414
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
6415
	if (!sds.busiest || busiest->sum_nr_running == 0)
6416 6417
		goto out_balanced;

6418 6419
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
6420

P
Peter Zijlstra 已提交
6421 6422
	/*
	 * If the busiest group is imbalanced the below checks don't
6423
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
6424 6425
	 * isn't true due to cpus_allowed constraints and the like.
	 */
6426
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
6427 6428
		goto force_balance;

6429
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
6430 6431
	if (env->idle == CPU_NEWLY_IDLE && local->group_has_free_capacity &&
	    !busiest->group_has_free_capacity)
6432 6433
		goto force_balance;

6434 6435 6436 6437
	/*
	 * If the local group is more busy than the selected busiest group
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
6438
	if (local->avg_load >= busiest->avg_load)
6439 6440
		goto out_balanced;

6441 6442 6443 6444
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
6445
	if (local->avg_load >= sds.avg_load)
6446 6447
		goto out_balanced;

6448
	if (env->idle == CPU_IDLE) {
6449 6450 6451 6452 6453 6454
		/*
		 * This cpu is idle. If the busiest group load doesn't
		 * have more tasks than the number of available cpu's and
		 * there is no imbalance between this and busiest group
		 * wrt to idle cpu's, it is balanced.
		 */
J
Joonsoo Kim 已提交
6455 6456
		if ((local->idle_cpus < busiest->idle_cpus) &&
		    busiest->sum_nr_running <= busiest->group_weight)
6457
			goto out_balanced;
6458 6459 6460 6461 6462
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
6463 6464
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
6465
			goto out_balanced;
6466
	}
6467

6468
force_balance:
6469
	/* Looks like there is an imbalance. Compute it */
6470
	calculate_imbalance(env, &sds);
6471 6472 6473
	return sds.busiest;

out_balanced:
6474
	env->imbalance = 0;
6475 6476 6477 6478 6479 6480
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
6481
static struct rq *find_busiest_queue(struct lb_env *env,
6482
				     struct sched_group *group)
6483 6484
{
	struct rq *busiest = NULL, *rq;
6485
	unsigned long busiest_load = 0, busiest_capacity = 1;
6486 6487
	int i;

6488
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6489
		unsigned long capacity, capacity_factor, wl;
6490 6491 6492 6493
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
6494

6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516
		/*
		 * We classify groups/runqueues into three groups:
		 *  - regular: there are !numa tasks
		 *  - remote:  there are numa tasks that run on the 'wrong' node
		 *  - all:     there is no distinction
		 *
		 * In order to avoid migrating ideally placed numa tasks,
		 * ignore those when there's better options.
		 *
		 * If we ignore the actual busiest queue to migrate another
		 * task, the next balance pass can still reduce the busiest
		 * queue by moving tasks around inside the node.
		 *
		 * If we cannot move enough load due to this classification
		 * the next pass will adjust the group classification and
		 * allow migration of more tasks.
		 *
		 * Both cases only affect the total convergence complexity.
		 */
		if (rt > env->fbq_type)
			continue;

6517
		capacity = capacity_of(i);
6518
		capacity_factor = DIV_ROUND_CLOSEST(capacity, SCHED_CAPACITY_SCALE);
6519 6520
		if (!capacity_factor)
			capacity_factor = fix_small_capacity(env->sd, group);
6521

6522
		wl = weighted_cpuload(i);
6523

6524 6525
		/*
		 * When comparing with imbalance, use weighted_cpuload()
6526
		 * which is not scaled with the cpu capacity.
6527
		 */
6528
		if (capacity_factor && rq->nr_running == 1 && wl > env->imbalance)
6529 6530
			continue;

6531 6532
		/*
		 * For the load comparisons with the other cpu's, consider
6533 6534 6535
		 * the weighted_cpuload() scaled with the cpu capacity, so
		 * that the load can be moved away from the cpu that is
		 * potentially running at a lower capacity.
6536
		 *
6537
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
6538
		 * multiplication to rid ourselves of the division works out
6539 6540
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
6541
		 */
6542
		if (wl * busiest_capacity > busiest_load * capacity) {
6543
			busiest_load = wl;
6544
			busiest_capacity = capacity;
6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558
			busiest = rq;
		}
	}

	return busiest;
}

/*
 * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
 * so long as it is large enough.
 */
#define MAX_PINNED_INTERVAL	512

/* Working cpumask for load_balance and load_balance_newidle. */
6559
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
6560

6561
static int need_active_balance(struct lb_env *env)
6562
{
6563 6564 6565
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
6566 6567 6568 6569 6570 6571

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
6572
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
6573
			return 1;
6574 6575 6576 6577 6578
	}

	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

6579 6580
static int active_load_balance_cpu_stop(void *data);

6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611
static int should_we_balance(struct lb_env *env)
{
	struct sched_group *sg = env->sd->groups;
	struct cpumask *sg_cpus, *sg_mask;
	int cpu, balance_cpu = -1;

	/*
	 * In the newly idle case, we will allow all the cpu's
	 * to do the newly idle load balance.
	 */
	if (env->idle == CPU_NEWLY_IDLE)
		return 1;

	sg_cpus = sched_group_cpus(sg);
	sg_mask = sched_group_mask(sg);
	/* Try to find first idle cpu */
	for_each_cpu_and(cpu, sg_cpus, env->cpus) {
		if (!cpumask_test_cpu(cpu, sg_mask) || !idle_cpu(cpu))
			continue;

		balance_cpu = cpu;
		break;
	}

	if (balance_cpu == -1)
		balance_cpu = group_balance_cpu(sg);

	/*
	 * First idle cpu or the first cpu(busiest) in this sched group
	 * is eligible for doing load balancing at this and above domains.
	 */
6612
	return balance_cpu == env->dst_cpu;
6613 6614
}

6615 6616 6617 6618 6619 6620
/*
 * Check this_cpu to ensure it is balanced within domain. Attempt to move
 * tasks if there is an imbalance.
 */
static int load_balance(int this_cpu, struct rq *this_rq,
			struct sched_domain *sd, enum cpu_idle_type idle,
6621
			int *continue_balancing)
6622
{
6623
	int ld_moved, cur_ld_moved, active_balance = 0;
6624
	struct sched_domain *sd_parent = sd->parent;
6625 6626 6627
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
6628
	struct cpumask *cpus = __get_cpu_var(load_balance_mask);
6629

6630 6631
	struct lb_env env = {
		.sd		= sd,
6632 6633
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
6634
		.dst_grpmask    = sched_group_cpus(sd->groups),
6635
		.idle		= idle,
6636
		.loop_break	= sched_nr_migrate_break,
6637
		.cpus		= cpus,
6638
		.fbq_type	= all,
6639
		.tasks		= LIST_HEAD_INIT(env.tasks),
6640 6641
	};

6642 6643 6644 6645
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
6646
	if (idle == CPU_NEWLY_IDLE)
6647 6648
		env.dst_grpmask = NULL;

6649 6650 6651 6652 6653
	cpumask_copy(cpus, cpu_active_mask);

	schedstat_inc(sd, lb_count[idle]);

redo:
6654 6655
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
6656
		goto out_balanced;
6657
	}
6658

6659
	group = find_busiest_group(&env);
6660 6661 6662 6663 6664
	if (!group) {
		schedstat_inc(sd, lb_nobusyg[idle]);
		goto out_balanced;
	}

6665
	busiest = find_busiest_queue(&env, group);
6666 6667 6668 6669 6670
	if (!busiest) {
		schedstat_inc(sd, lb_nobusyq[idle]);
		goto out_balanced;
	}

6671
	BUG_ON(busiest == env.dst_rq);
6672

6673
	schedstat_add(sd, lb_imbalance[idle], env.imbalance);
6674 6675 6676 6677 6678 6679 6680 6681 6682

	ld_moved = 0;
	if (busiest->nr_running > 1) {
		/*
		 * Attempt to move tasks. If find_busiest_group has found
		 * an imbalance but busiest->nr_running <= 1, the group is
		 * still unbalanced. ld_moved simply stays zero, so it is
		 * correctly treated as an imbalance.
		 */
6683
		env.flags |= LBF_ALL_PINNED;
6684 6685 6686
		env.src_cpu   = busiest->cpu;
		env.src_rq    = busiest;
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
6687

6688
more_balance:
6689
		raw_spin_lock_irqsave(&busiest->lock, flags);
6690 6691 6692 6693 6694

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711
		cur_ld_moved = detach_tasks(&env);

		/*
		 * We've detached some tasks from busiest_rq. Every
		 * task is masked "TASK_ON_RQ_MIGRATING", so we can safely
		 * unlock busiest->lock, and we are able to be sure
		 * that nobody can manipulate the tasks in parallel.
		 * See task_rq_lock() family for the details.
		 */

		raw_spin_unlock(&busiest->lock);

		if (cur_ld_moved) {
			attach_tasks(&env);
			ld_moved += cur_ld_moved;
		}

6712 6713 6714 6715 6716
		local_irq_restore(flags);

		/*
		 * some other cpu did the load balance for us.
		 */
6717 6718 6719
		if (cur_ld_moved && env.dst_cpu != smp_processor_id())
			resched_cpu(env.dst_cpu);

6720 6721 6722 6723 6724
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743
		/*
		 * Revisit (affine) tasks on src_cpu that couldn't be moved to
		 * us and move them to an alternate dst_cpu in our sched_group
		 * where they can run. The upper limit on how many times we
		 * iterate on same src_cpu is dependent on number of cpus in our
		 * sched_group.
		 *
		 * This changes load balance semantics a bit on who can move
		 * load to a given_cpu. In addition to the given_cpu itself
		 * (or a ilb_cpu acting on its behalf where given_cpu is
		 * nohz-idle), we now have balance_cpu in a position to move
		 * load to given_cpu. In rare situations, this may cause
		 * conflicts (balance_cpu and given_cpu/ilb_cpu deciding
		 * _independently_ and at _same_ time to move some load to
		 * given_cpu) causing exceess load to be moved to given_cpu.
		 * This however should not happen so much in practice and
		 * moreover subsequent load balance cycles should correct the
		 * excess load moved.
		 */
6744
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
6745

6746 6747 6748
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

6749
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
6750
			env.dst_cpu	 = env.new_dst_cpu;
6751
			env.flags	&= ~LBF_DST_PINNED;
6752 6753
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
6754

6755 6756 6757 6758 6759 6760
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
6761

6762 6763 6764 6765
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
6766
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
6767

6768
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
6769 6770 6771
				*group_imbalance = 1;
		}

6772
		/* All tasks on this runqueue were pinned by CPU affinity */
6773
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
6774
			cpumask_clear_cpu(cpu_of(busiest), cpus);
6775 6776 6777
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
6778
				goto redo;
6779
			}
6780
			goto out_all_pinned;
6781 6782 6783 6784 6785
		}
	}

	if (!ld_moved) {
		schedstat_inc(sd, lb_failed[idle]);
6786 6787 6788 6789 6790 6791 6792 6793
		/*
		 * Increment the failure counter only on periodic balance.
		 * We do not want newidle balance, which can be very
		 * frequent, pollute the failure counter causing
		 * excessive cache_hot migrations and active balances.
		 */
		if (idle != CPU_NEWLY_IDLE)
			sd->nr_balance_failed++;
6794

6795
		if (need_active_balance(&env)) {
6796 6797
			raw_spin_lock_irqsave(&busiest->lock, flags);

6798 6799 6800
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
6801 6802
			 */
			if (!cpumask_test_cpu(this_cpu,
6803
					tsk_cpus_allowed(busiest->curr))) {
6804 6805
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
6806
				env.flags |= LBF_ALL_PINNED;
6807 6808 6809
				goto out_one_pinned;
			}

6810 6811 6812 6813 6814
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
6815 6816 6817 6818 6819 6820
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
6821

6822
			if (active_balance) {
6823 6824 6825
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
6826
			}
6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844

			/*
			 * We've kicked active balancing, reset the failure
			 * counter.
			 */
			sd->nr_balance_failed = sd->cache_nice_tries+1;
		}
	} else
		sd->nr_balance_failed = 0;

	if (likely(!active_balance)) {
		/* We were unbalanced, so reset the balancing interval */
		sd->balance_interval = sd->min_interval;
	} else {
		/*
		 * If we've begun active balancing, start to back off. This
		 * case may not be covered by the all_pinned logic if there
		 * is only 1 task on the busy runqueue (because we don't call
6845
		 * detach_tasks).
6846 6847 6848 6849 6850 6851 6852 6853
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870
	/*
	 * We reach balance although we may have faced some affinity
	 * constraints. Clear the imbalance flag if it was set.
	 */
	if (sd_parent) {
		int *group_imbalance = &sd_parent->groups->sgc->imbalance;

		if (*group_imbalance)
			*group_imbalance = 0;
	}

out_all_pinned:
	/*
	 * We reach balance because all tasks are pinned at this level so
	 * we can't migrate them. Let the imbalance flag set so parent level
	 * can try to migrate them.
	 */
6871 6872 6873 6874 6875 6876
	schedstat_inc(sd, lb_balanced[idle]);

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
6877
	if (((env.flags & LBF_ALL_PINNED) &&
6878
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
6879 6880 6881
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

6882
	ld_moved = 0;
6883 6884 6885 6886
out:
	return ld_moved;
}

6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909 6910 6911 6912 6913
static inline unsigned long
get_sd_balance_interval(struct sched_domain *sd, int cpu_busy)
{
	unsigned long interval = sd->balance_interval;

	if (cpu_busy)
		interval *= sd->busy_factor;

	/* scale ms to jiffies */
	interval = msecs_to_jiffies(interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);

	return interval;
}

static inline void
update_next_balance(struct sched_domain *sd, int cpu_busy, unsigned long *next_balance)
{
	unsigned long interval, next;

	interval = get_sd_balance_interval(sd, cpu_busy);
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

6914 6915 6916 6917
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
6918
static int idle_balance(struct rq *this_rq)
6919
{
6920 6921
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
6922 6923
	struct sched_domain *sd;
	int pulled_task = 0;
6924
	u64 curr_cost = 0;
6925

6926
	idle_enter_fair(this_rq);
6927

6928 6929 6930 6931 6932 6933
	/*
	 * We must set idle_stamp _before_ calling idle_balance(), such that we
	 * measure the duration of idle_balance() as idle time.
	 */
	this_rq->idle_stamp = rq_clock(this_rq);

6934 6935
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
6936 6937 6938 6939 6940 6941
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, 0, &next_balance);
		rcu_read_unlock();

6942
		goto out;
6943
	}
6944

6945 6946 6947 6948 6949
	/*
	 * Drop the rq->lock, but keep IRQ/preempt disabled.
	 */
	raw_spin_unlock(&this_rq->lock);

6950
	update_blocked_averages(this_cpu);
6951
	rcu_read_lock();
6952
	for_each_domain(this_cpu, sd) {
6953
		int continue_balancing = 1;
6954
		u64 t0, domain_cost;
6955 6956 6957 6958

		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

6959 6960
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, 0, &next_balance);
6961
			break;
6962
		}
6963

6964
		if (sd->flags & SD_BALANCE_NEWIDLE) {
6965 6966
			t0 = sched_clock_cpu(this_cpu);

6967
			pulled_task = load_balance(this_cpu, this_rq,
6968 6969
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
6970 6971 6972 6973 6974 6975

			domain_cost = sched_clock_cpu(this_cpu) - t0;
			if (domain_cost > sd->max_newidle_lb_cost)
				sd->max_newidle_lb_cost = domain_cost;

			curr_cost += domain_cost;
6976
		}
6977

6978
		update_next_balance(sd, 0, &next_balance);
6979 6980 6981 6982 6983 6984

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
6985 6986
			break;
	}
6987
	rcu_read_unlock();
6988 6989 6990

	raw_spin_lock(&this_rq->lock);

6991 6992 6993
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

6994
	/*
6995 6996 6997
	 * While browsing the domains, we released the rq lock, a task could
	 * have been enqueued in the meantime. Since we're not going idle,
	 * pretend we pulled a task.
6998
	 */
6999
	if (this_rq->cfs.h_nr_running && !pulled_task)
7000
		pulled_task = 1;
7001

7002 7003 7004
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
7005
		this_rq->next_balance = next_balance;
7006

7007
	/* Is there a task of a high priority class? */
7008
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
7009 7010 7011 7012
		pulled_task = -1;

	if (pulled_task) {
		idle_exit_fair(this_rq);
7013
		this_rq->idle_stamp = 0;
7014
	}
7015

7016
	return pulled_task;
7017 7018 7019
}

/*
7020 7021 7022 7023
 * active_load_balance_cpu_stop is run by cpu stopper. It pushes
 * running tasks off the busiest CPU onto idle CPUs. It requires at
 * least 1 task to be running on each physical CPU where possible, and
 * avoids physical / logical imbalances.
7024
 */
7025
static int active_load_balance_cpu_stop(void *data)
7026
{
7027 7028
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
7029
	int target_cpu = busiest_rq->push_cpu;
7030
	struct rq *target_rq = cpu_rq(target_cpu);
7031
	struct sched_domain *sd;
7032
	struct task_struct *p = NULL;
7033 7034 7035 7036 7037 7038 7039

	raw_spin_lock_irq(&busiest_rq->lock);

	/* make sure the requested cpu hasn't gone down in the meantime */
	if (unlikely(busiest_cpu != smp_processor_id() ||
		     !busiest_rq->active_balance))
		goto out_unlock;
7040 7041 7042

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
7043
		goto out_unlock;
7044 7045 7046 7047 7048 7049 7050 7051 7052

	/*
	 * This condition is "impossible", if it occurs
	 * we need to fix it. Originally reported by
	 * Bjorn Helgaas on a 128-cpu setup.
	 */
	BUG_ON(busiest_rq == target_rq);

	/* Search for an sd spanning us and the target CPU. */
7053
	rcu_read_lock();
7054 7055 7056 7057 7058 7059 7060
	for_each_domain(target_cpu, sd) {
		if ((sd->flags & SD_LOAD_BALANCE) &&
		    cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
				break;
	}

	if (likely(sd)) {
7061 7062
		struct lb_env env = {
			.sd		= sd,
7063 7064 7065 7066
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
7067 7068 7069
			.idle		= CPU_IDLE,
		};

7070 7071
		schedstat_inc(sd, alb_count);

7072 7073
		p = detach_one_task(&env);
		if (p)
7074 7075 7076 7077
			schedstat_inc(sd, alb_pushed);
		else
			schedstat_inc(sd, alb_failed);
	}
7078
	rcu_read_unlock();
7079 7080
out_unlock:
	busiest_rq->active_balance = 0;
7081 7082 7083 7084 7085 7086 7087
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

7088
	return 0;
7089 7090
}

7091 7092 7093 7094 7095
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

7096
#ifdef CONFIG_NO_HZ_COMMON
7097 7098 7099 7100 7101 7102
/*
 * idle load balancing details
 * - When one of the busy CPUs notice that there may be an idle rebalancing
 *   needed, they will kick the idle load balancer, which then does idle
 *   load balancing for all the idle CPUs.
 */
7103
static struct {
7104
	cpumask_var_t idle_cpus_mask;
7105
	atomic_t nr_cpus;
7106 7107
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
7108

7109
static inline int find_new_ilb(void)
7110
{
7111
	int ilb = cpumask_first(nohz.idle_cpus_mask);
7112

7113 7114 7115 7116
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
7117 7118
}

7119 7120 7121 7122 7123
/*
 * Kick a CPU to do the nohz balancing, if it is time for it. We pick the
 * nohz_load_balancer CPU (if there is one) otherwise fallback to any idle
 * CPU (if there is one).
 */
7124
static void nohz_balancer_kick(void)
7125 7126 7127 7128 7129
{
	int ilb_cpu;

	nohz.next_balance++;

7130
	ilb_cpu = find_new_ilb();
7131

7132 7133
	if (ilb_cpu >= nr_cpu_ids)
		return;
7134

7135
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
7136 7137 7138 7139 7140 7141 7142 7143
		return;
	/*
	 * Use smp_send_reschedule() instead of resched_cpu().
	 * This way we generate a sched IPI on the target cpu which
	 * is idle. And the softirq performing nohz idle load balance
	 * will be run before returning from the IPI.
	 */
	smp_send_reschedule(ilb_cpu);
7144 7145 7146
	return;
}

7147
static inline void nohz_balance_exit_idle(int cpu)
7148 7149
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
7150 7151 7152 7153 7154 7155 7156
		/*
		 * Completely isolated CPUs don't ever set, so we must test.
		 */
		if (likely(cpumask_test_cpu(cpu, nohz.idle_cpus_mask))) {
			cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
			atomic_dec(&nohz.nr_cpus);
		}
7157 7158 7159 7160
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

7161 7162 7163
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
7164
	int cpu = smp_processor_id();
7165 7166

	rcu_read_lock();
7167
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7168 7169 7170 7171 7172

	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

7173
	atomic_inc(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7174
unlock:
7175 7176 7177 7178 7179 7180
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
7181
	int cpu = smp_processor_id();
7182 7183

	rcu_read_lock();
7184
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7185 7186 7187 7188 7189

	if (!sd || sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 1;

7190
	atomic_dec(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7191
unlock:
7192 7193 7194
	rcu_read_unlock();
}

7195
/*
7196
 * This routine will record that the cpu is going idle with tick stopped.
7197
 * This info will be used in performing idle load balancing in the future.
7198
 */
7199
void nohz_balance_enter_idle(int cpu)
7200
{
7201 7202 7203 7204 7205 7206
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

7207 7208
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
7209

7210 7211 7212 7213 7214 7215
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

7216 7217 7218
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
7219
}
7220

7221
static int sched_ilb_notifier(struct notifier_block *nfb,
7222 7223 7224 7225
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DYING:
7226
		nohz_balance_exit_idle(smp_processor_id());
7227 7228 7229 7230 7231
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}
7232 7233 7234 7235
#endif

static DEFINE_SPINLOCK(balancing);

7236 7237 7238 7239
/*
 * Scale the max load_balance interval with the number of CPUs in the system.
 * This trades load-balance latency on larger machines for less cross talk.
 */
7240
void update_max_interval(void)
7241 7242 7243 7244
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

7245 7246 7247 7248
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
7249
 * Balancing parameters are set up in init_sched_domains.
7250
 */
7251
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
7252
{
7253
	int continue_balancing = 1;
7254
	int cpu = rq->cpu;
7255
	unsigned long interval;
7256
	struct sched_domain *sd;
7257 7258 7259
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
7260 7261
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
7262

7263
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
7264

7265
	rcu_read_lock();
7266
	for_each_domain(cpu, sd) {
7267 7268 7269 7270 7271 7272 7273 7274 7275 7276 7277 7278
		/*
		 * Decay the newidle max times here because this is a regular
		 * visit to all the domains. Decay ~1% per second.
		 */
		if (time_after(jiffies, sd->next_decay_max_lb_cost)) {
			sd->max_newidle_lb_cost =
				(sd->max_newidle_lb_cost * 253) / 256;
			sd->next_decay_max_lb_cost = jiffies + HZ;
			need_decay = 1;
		}
		max_cost += sd->max_newidle_lb_cost;

7279 7280 7281
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

7282 7283 7284 7285 7286 7287 7288 7289 7290 7291 7292
		/*
		 * Stop the load balance at this level. There is another
		 * CPU in our sched group which is doing load balancing more
		 * actively.
		 */
		if (!continue_balancing) {
			if (need_decay)
				continue;
			break;
		}

7293
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7294 7295 7296 7297 7298 7299 7300 7301

		need_serialize = sd->flags & SD_SERIALIZE;
		if (need_serialize) {
			if (!spin_trylock(&balancing))
				goto out;
		}

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
7302
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
7303
				/*
7304
				 * The LBF_DST_PINNED logic could have changed
7305 7306
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
7307
				 */
7308
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
7309 7310
			}
			sd->last_balance = jiffies;
7311
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7312 7313 7314 7315 7316 7317 7318 7319
		}
		if (need_serialize)
			spin_unlock(&balancing);
out:
		if (time_after(next_balance, sd->last_balance + interval)) {
			next_balance = sd->last_balance + interval;
			update_next_balance = 1;
		}
7320 7321
	}
	if (need_decay) {
7322
		/*
7323 7324
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
7325
		 */
7326 7327
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
7328
	}
7329
	rcu_read_unlock();
7330 7331 7332 7333 7334 7335 7336 7337 7338 7339

	/*
	 * next_balance will be updated only when there is a need.
	 * When the cpu is attached to null domain for ex, it will not be
	 * updated.
	 */
	if (likely(update_next_balance))
		rq->next_balance = next_balance;
}

7340
#ifdef CONFIG_NO_HZ_COMMON
7341
/*
7342
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
7343 7344
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
7345
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
7346
{
7347
	int this_cpu = this_rq->cpu;
7348 7349 7350
	struct rq *rq;
	int balance_cpu;

7351 7352 7353
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
7354 7355

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
7356
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
7357 7358 7359 7360 7361 7362 7363
			continue;

		/*
		 * If this cpu gets work to do, stop the load balancing
		 * work being done for other cpus. Next load
		 * balancing owner will pick it up.
		 */
7364
		if (need_resched())
7365 7366
			break;

V
Vincent Guittot 已提交
7367 7368
		rq = cpu_rq(balance_cpu);

7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379
		/*
		 * If time for next balance is due,
		 * do the balance.
		 */
		if (time_after_eq(jiffies, rq->next_balance)) {
			raw_spin_lock_irq(&rq->lock);
			update_rq_clock(rq);
			update_idle_cpu_load(rq);
			raw_spin_unlock_irq(&rq->lock);
			rebalance_domains(rq, CPU_IDLE);
		}
7380 7381 7382 7383 7384

		if (time_after(this_rq->next_balance, rq->next_balance))
			this_rq->next_balance = rq->next_balance;
	}
	nohz.next_balance = this_rq->next_balance;
7385 7386
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
7387 7388 7389
}

/*
7390 7391 7392 7393
 * Current heuristic for kicking the idle load balancer in the presence
 * of an idle cpu is the system.
 *   - This rq has more than one task.
 *   - At any scheduler domain level, this cpu's scheduler group has multiple
7394
 *     busy cpu's exceeding the group's capacity.
7395 7396
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
7397
 */
7398
static inline int nohz_kick_needed(struct rq *rq)
7399 7400
{
	unsigned long now = jiffies;
7401
	struct sched_domain *sd;
7402
	struct sched_group_capacity *sgc;
7403
	int nr_busy, cpu = rq->cpu;
7404

7405
	if (unlikely(rq->idle_balance))
7406 7407
		return 0;

7408 7409 7410 7411
       /*
	* We may be recently in ticked or tickless idle mode. At the first
	* busy tick after returning from idle, we will update the busy stats.
	*/
7412
	set_cpu_sd_state_busy();
7413
	nohz_balance_exit_idle(cpu);
7414 7415 7416 7417 7418 7419 7420

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
		return 0;
7421 7422

	if (time_before(now, nohz.next_balance))
7423 7424
		return 0;

7425 7426
	if (rq->nr_running >= 2)
		goto need_kick;
7427

7428
	rcu_read_lock();
7429
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
7430

7431
	if (sd) {
7432 7433
		sgc = sd->groups->sgc;
		nr_busy = atomic_read(&sgc->nr_busy_cpus);
7434

7435
		if (nr_busy > 1)
7436
			goto need_kick_unlock;
7437
	}
7438 7439 7440 7441 7442 7443 7444

	sd = rcu_dereference(per_cpu(sd_asym, cpu));

	if (sd && (cpumask_first_and(nohz.idle_cpus_mask,
				  sched_domain_span(sd)) < cpu))
		goto need_kick_unlock;

7445
	rcu_read_unlock();
7446
	return 0;
7447 7448 7449

need_kick_unlock:
	rcu_read_unlock();
7450 7451
need_kick:
	return 1;
7452 7453
}
#else
7454
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
7455 7456 7457 7458 7459 7460
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
7461 7462
static void run_rebalance_domains(struct softirq_action *h)
{
7463
	struct rq *this_rq = this_rq();
7464
	enum cpu_idle_type idle = this_rq->idle_balance ?
7465 7466
						CPU_IDLE : CPU_NOT_IDLE;

7467
	rebalance_domains(this_rq, idle);
7468 7469

	/*
7470
	 * If this cpu has a pending nohz_balance_kick, then do the
7471 7472 7473
	 * balancing on behalf of the other idle cpus whose ticks are
	 * stopped.
	 */
7474
	nohz_idle_balance(this_rq, idle);
7475 7476 7477 7478 7479
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
7480
void trigger_load_balance(struct rq *rq)
7481 7482
{
	/* Don't need to rebalance while attached to NULL domain */
7483 7484 7485 7486
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
7487
		raise_softirq(SCHED_SOFTIRQ);
7488
#ifdef CONFIG_NO_HZ_COMMON
7489
	if (nohz_kick_needed(rq))
7490
		nohz_balancer_kick();
7491
#endif
7492 7493
}

7494 7495 7496
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
7497 7498

	update_runtime_enabled(rq);
7499 7500 7501 7502 7503
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
7504 7505 7506

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
7507 7508
}

7509
#endif /* CONFIG_SMP */
7510

7511 7512 7513
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
7514
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
7515 7516 7517 7518 7519 7520
{
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &curr->se;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
P
Peter Zijlstra 已提交
7521
		entity_tick(cfs_rq, se, queued);
7522
	}
7523

7524
	if (numabalancing_enabled)
7525
		task_tick_numa(rq, curr);
7526

7527
	update_rq_runnable_avg(rq, 1);
7528 7529 7530
}

/*
P
Peter Zijlstra 已提交
7531 7532 7533
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
7534
 */
P
Peter Zijlstra 已提交
7535
static void task_fork_fair(struct task_struct *p)
7536
{
7537 7538
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
7539
	int this_cpu = smp_processor_id();
P
Peter Zijlstra 已提交
7540 7541 7542
	struct rq *rq = this_rq();
	unsigned long flags;

7543
	raw_spin_lock_irqsave(&rq->lock, flags);
7544

7545 7546
	update_rq_clock(rq);

7547 7548 7549
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;

7550 7551 7552 7553 7554 7555 7556 7557 7558
	/*
	 * Not only the cpu but also the task_group of the parent might have
	 * been changed after parent->se.parent,cfs_rq were copied to
	 * child->se.parent,cfs_rq. So call __set_task_cpu() to make those
	 * of child point to valid ones.
	 */
	rcu_read_lock();
	__set_task_cpu(p, this_cpu);
	rcu_read_unlock();
7559

7560
	update_curr(cfs_rq);
P
Peter Zijlstra 已提交
7561

7562 7563
	if (curr)
		se->vruntime = curr->vruntime;
7564
	place_entity(cfs_rq, se, 1);
7565

P
Peter Zijlstra 已提交
7566
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
7567
		/*
7568 7569 7570
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
7571
		swap(curr->vruntime, se->vruntime);
7572
		resched_curr(rq);
7573
	}
7574

7575 7576
	se->vruntime -= cfs_rq->min_vruntime;

7577
	raw_spin_unlock_irqrestore(&rq->lock, flags);
7578 7579
}

7580 7581 7582 7583
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
7584 7585
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
7586
{
7587
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7588 7589
		return;

7590 7591 7592 7593 7594
	/*
	 * Reschedule if we are currently running on this runqueue and
	 * our priority decreased, or if we are not currently running on
	 * this runqueue and our priority is higher than the current's
	 */
P
Peter Zijlstra 已提交
7595
	if (rq->curr == p) {
7596
		if (p->prio > oldprio)
7597
			resched_curr(rq);
7598
	} else
7599
		check_preempt_curr(rq, p, 0);
7600 7601
}

P
Peter Zijlstra 已提交
7602 7603 7604 7605 7606 7607
static void switched_from_fair(struct rq *rq, struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	/*
7608
	 * Ensure the task's vruntime is normalized, so that when it's
P
Peter Zijlstra 已提交
7609 7610 7611
	 * switched back to the fair class the enqueue_entity(.flags=0) will
	 * do the right thing.
	 *
7612 7613
	 * If it's queued, then the dequeue_entity(.flags=0) will already
	 * have normalized the vruntime, if it's !queued, then only when
P
Peter Zijlstra 已提交
7614 7615
	 * the task is sleeping will it still have non-normalized vruntime.
	 */
7616
	if (!task_on_rq_queued(p) && p->state != TASK_RUNNING) {
P
Peter Zijlstra 已提交
7617 7618 7619 7620 7621 7622 7623
		/*
		 * Fix up our vruntime so that the current sleep doesn't
		 * cause 'unlimited' sleep bonus.
		 */
		place_entity(cfs_rq, se, 0);
		se->vruntime -= cfs_rq->min_vruntime;
	}
7624

7625
#ifdef CONFIG_SMP
7626 7627 7628 7629 7630
	/*
	* Remove our load from contribution when we leave sched_fair
	* and ensure we don't carry in an old decay_count if we
	* switch back.
	*/
7631 7632 7633
	if (se->avg.decay_count) {
		__synchronize_entity_decay(se);
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
7634 7635
	}
#endif
P
Peter Zijlstra 已提交
7636 7637
}

7638 7639 7640
/*
 * We switched to the sched_fair class.
 */
P
Peter Zijlstra 已提交
7641
static void switched_to_fair(struct rq *rq, struct task_struct *p)
7642
{
7643
#ifdef CONFIG_FAIR_GROUP_SCHED
7644
	struct sched_entity *se = &p->se;
7645 7646 7647 7648 7649 7650
	/*
	 * Since the real-depth could have been changed (only FAIR
	 * class maintain depth value), reset depth properly.
	 */
	se->depth = se->parent ? se->parent->depth + 1 : 0;
#endif
7651
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7652 7653
		return;

7654 7655 7656 7657 7658
	/*
	 * We were most likely switched from sched_rt, so
	 * kick off the schedule if running, otherwise just see
	 * if we can still preempt the current task.
	 */
P
Peter Zijlstra 已提交
7659
	if (rq->curr == p)
7660
		resched_curr(rq);
7661
	else
7662
		check_preempt_curr(rq, p, 0);
7663 7664
}

7665 7666 7667 7668 7669 7670 7671 7672 7673
/* Account for a task changing its policy or group.
 *
 * This routine is mostly called to set cfs_rq->curr field when a task
 * migrates between groups/classes.
 */
static void set_curr_task_fair(struct rq *rq)
{
	struct sched_entity *se = &rq->curr->se;

7674 7675 7676 7677 7678 7679 7680
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);

		set_next_entity(cfs_rq, se);
		/* ensure bandwidth has been allocated on our new cfs_rq */
		account_cfs_rq_runtime(cfs_rq, 0);
	}
7681 7682
}

7683 7684 7685 7686 7687 7688 7689
void init_cfs_rq(struct cfs_rq *cfs_rq)
{
	cfs_rq->tasks_timeline = RB_ROOT;
	cfs_rq->min_vruntime = (u64)(-(1LL << 20));
#ifndef CONFIG_64BIT
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
7690
#ifdef CONFIG_SMP
7691
	atomic64_set(&cfs_rq->decay_counter, 1);
7692
	atomic_long_set(&cfs_rq->removed_load, 0);
7693
#endif
7694 7695
}

P
Peter Zijlstra 已提交
7696
#ifdef CONFIG_FAIR_GROUP_SCHED
7697
static void task_move_group_fair(struct task_struct *p, int queued)
P
Peter Zijlstra 已提交
7698
{
P
Peter Zijlstra 已提交
7699
	struct sched_entity *se = &p->se;
7700
	struct cfs_rq *cfs_rq;
P
Peter Zijlstra 已提交
7701

7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714
	/*
	 * If the task was not on the rq at the time of this cgroup movement
	 * it must have been asleep, sleeping tasks keep their ->vruntime
	 * absolute on their old rq until wakeup (needed for the fair sleeper
	 * bonus in place_entity()).
	 *
	 * If it was on the rq, we've just 'preempted' it, which does convert
	 * ->vruntime to a relative base.
	 *
	 * Make sure both cases convert their relative position when migrating
	 * to another cgroup's rq. This does somewhat interfere with the
	 * fair sleeper stuff for the first placement, but who cares.
	 */
7715
	/*
7716
	 * When !queued, vruntime of the task has usually NOT been normalized.
7717 7718 7719 7720
	 * But there are some cases where it has already been normalized:
	 *
	 * - Moving a forked child which is waiting for being woken up by
	 *   wake_up_new_task().
7721 7722
	 * - Moving a task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
7723 7724 7725 7726
	 *
	 * To prevent boost or penalty in the new cfs_rq caused by delta
	 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
	 */
7727 7728
	if (!queued && (!se->sum_exec_runtime || p->state == TASK_WAKING))
		queued = 1;
7729

7730
	if (!queued)
P
Peter Zijlstra 已提交
7731
		se->vruntime -= cfs_rq_of(se)->min_vruntime;
7732
	set_task_rq(p, task_cpu(p));
P
Peter Zijlstra 已提交
7733
	se->depth = se->parent ? se->parent->depth + 1 : 0;
7734
	if (!queued) {
P
Peter Zijlstra 已提交
7735 7736
		cfs_rq = cfs_rq_of(se);
		se->vruntime += cfs_rq->min_vruntime;
7737 7738 7739 7740 7741 7742
#ifdef CONFIG_SMP
		/*
		 * migrate_task_rq_fair() will have removed our previous
		 * contribution, but we must synchronize for ongoing future
		 * decay.
		 */
P
Peter Zijlstra 已提交
7743 7744
		se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
		cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
7745 7746
#endif
	}
P
Peter Zijlstra 已提交
7747
}
7748 7749 7750 7751 7752 7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767 7768 7769 7770 7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800 7801 7802 7803 7804 7805 7806 7807 7808 7809 7810 7811 7812 7813 7814 7815 7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839

void free_fair_sched_group(struct task_group *tg)
{
	int i;

	destroy_cfs_bandwidth(tg_cfs_bandwidth(tg));

	for_each_possible_cpu(i) {
		if (tg->cfs_rq)
			kfree(tg->cfs_rq[i]);
		if (tg->se)
			kfree(tg->se[i]);
	}

	kfree(tg->cfs_rq);
	kfree(tg->se);
}

int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
{
	struct cfs_rq *cfs_rq;
	struct sched_entity *se;
	int i;

	tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->cfs_rq)
		goto err;
	tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->se)
		goto err;

	tg->shares = NICE_0_LOAD;

	init_cfs_bandwidth(tg_cfs_bandwidth(tg));

	for_each_possible_cpu(i) {
		cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
				      GFP_KERNEL, cpu_to_node(i));
		if (!cfs_rq)
			goto err;

		se = kzalloc_node(sizeof(struct sched_entity),
				  GFP_KERNEL, cpu_to_node(i));
		if (!se)
			goto err_free_rq;

		init_cfs_rq(cfs_rq);
		init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]);
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

void unregister_fair_sched_group(struct task_group *tg, int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

	/*
	* Only empty task groups can be destroyed; so we can speculatively
	* check on_list without danger of it being re-added.
	*/
	if (!tg->cfs_rq[cpu]->on_list)
		return;

	raw_spin_lock_irqsave(&rq->lock, flags);
	list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
			struct sched_entity *se, int cpu,
			struct sched_entity *parent)
{
	struct rq *rq = cpu_rq(cpu);

	cfs_rq->tg = tg;
	cfs_rq->rq = rq;
	init_cfs_rq_runtime(cfs_rq);

	tg->cfs_rq[cpu] = cfs_rq;
	tg->se[cpu] = se;

	/* se could be NULL for root_task_group */
	if (!se)
		return;

P
Peter Zijlstra 已提交
7840
	if (!parent) {
7841
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
7842 7843
		se->depth = 0;
	} else {
7844
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
7845 7846
		se->depth = parent->depth + 1;
	}
7847 7848

	se->my_q = cfs_rq;
7849 7850
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880
	se->parent = parent;
}

static DEFINE_MUTEX(shares_mutex);

int sched_group_set_shares(struct task_group *tg, unsigned long shares)
{
	int i;
	unsigned long flags;

	/*
	 * We can't change the weight of the root cgroup.
	 */
	if (!tg->se[0])
		return -EINVAL;

	shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));

	mutex_lock(&shares_mutex);
	if (tg->shares == shares)
		goto done;

	tg->shares = shares;
	for_each_possible_cpu(i) {
		struct rq *rq = cpu_rq(i);
		struct sched_entity *se;

		se = tg->se[i];
		/* Propagate contribution to hierarchy */
		raw_spin_lock_irqsave(&rq->lock, flags);
7881 7882 7883

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
7884
		for_each_sched_entity(se)
7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897 7898 7899 7900 7901 7902 7903 7904 7905
			update_cfs_shares(group_cfs_rq(se));
		raw_spin_unlock_irqrestore(&rq->lock, flags);
	}

done:
	mutex_unlock(&shares_mutex);
	return 0;
}
#else /* CONFIG_FAIR_GROUP_SCHED */

void free_fair_sched_group(struct task_group *tg) { }

int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
{
	return 1;
}

void unregister_fair_sched_group(struct task_group *tg, int cpu) { }

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
7906

7907
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
7908 7909 7910 7911 7912 7913 7914 7915 7916
{
	struct sched_entity *se = &task->se;
	unsigned int rr_interval = 0;

	/*
	 * Time slice is 0 for SCHED_OTHER tasks that are on an otherwise
	 * idle runqueue:
	 */
	if (rq->cfs.load.weight)
7917
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
7918 7919 7920 7921

	return rr_interval;
}

7922 7923 7924
/*
 * All the scheduling class methods:
 */
7925
const struct sched_class fair_sched_class = {
7926
	.next			= &idle_sched_class,
7927 7928 7929
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
7930
	.yield_to_task		= yield_to_task_fair,
7931

I
Ingo Molnar 已提交
7932
	.check_preempt_curr	= check_preempt_wakeup,
7933 7934 7935 7936

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

7937
#ifdef CONFIG_SMP
L
Li Zefan 已提交
7938
	.select_task_rq		= select_task_rq_fair,
7939
	.migrate_task_rq	= migrate_task_rq_fair,
7940

7941 7942
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
7943 7944

	.task_waking		= task_waking_fair,
7945
#endif
7946

7947
	.set_curr_task          = set_curr_task_fair,
7948
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
7949
	.task_fork		= task_fork_fair,
7950 7951

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
7952
	.switched_from		= switched_from_fair,
7953
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
7954

7955 7956
	.get_rr_interval	= get_rr_interval_fair,

P
Peter Zijlstra 已提交
7957
#ifdef CONFIG_FAIR_GROUP_SCHED
7958
	.task_move_group	= task_move_group_fair,
P
Peter Zijlstra 已提交
7959
#endif
7960 7961 7962
};

#ifdef CONFIG_SCHED_DEBUG
7963
void print_cfs_stats(struct seq_file *m, int cpu)
7964 7965 7966
{
	struct cfs_rq *cfs_rq;

7967
	rcu_read_lock();
7968
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
7969
		print_cfs_rq(m, cpu, cfs_rq);
7970
	rcu_read_unlock();
7971 7972
}
#endif
7973 7974 7975 7976 7977 7978

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

7979
#ifdef CONFIG_NO_HZ_COMMON
7980
	nohz.next_balance = jiffies;
7981
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
7982
	cpu_notifier(sched_ilb_notifier, 0);
7983 7984 7985 7986
#endif
#endif /* SMP */

}