fair.c 206.3 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
#include <linux/cpuidle.h>
27 28 29
#include <linux/slab.h>
#include <linux/profile.h>
#include <linux/interrupt.h>
30
#include <linux/mempolicy.h>
31
#include <linux/migrate.h>
32
#include <linux/task_work.h>
33 34 35 36

#include <trace/events/sched.h>

#include "sched.h"
A
Arjan van de Ven 已提交
37

38
/*
39
 * Targeted preemption latency for CPU-bound tasks:
40
 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
41
 *
42
 * NOTE: this latency value is not the same as the concept of
I
Ingo Molnar 已提交
43 44 45
 * 'timeslice length' - timeslices in CFS are of variable length
 * and have no persistent notion like in traditional, time-slice
 * based scheduling concepts.
46
 *
I
Ingo Molnar 已提交
47 48
 * (to see the precise effective timeslice length of your workload,
 *  run vmstat and monitor the context-switches (cs) field)
49
 */
50 51
unsigned int sysctl_sched_latency = 6000000ULL;
unsigned int normalized_sysctl_sched_latency = 6000000ULL;
52

53 54 55 56 57 58 59 60 61 62 63 64
/*
 * 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;

65
/*
66
 * Minimal preemption granularity for CPU-bound tasks:
67
 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
68
 */
69 70
unsigned int sysctl_sched_min_granularity = 750000ULL;
unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
71 72

/*
73 74
 * is kept at sysctl_sched_latency / sysctl_sched_min_granularity
 */
75
static unsigned int sched_nr_latency = 8;
76 77

/*
78
 * After fork, child runs first. If set to 0 (default) then
79
 * parent will (try to) run first.
80
 */
81
unsigned int sysctl_sched_child_runs_first __read_mostly;
82 83 84

/*
 * SCHED_OTHER wake-up granularity.
85
 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
86 87 88 89 90
 *
 * This option delays the preemption effects of decoupled workloads
 * and reduces their over-scheduling. Synchronous workloads will still
 * have immediate wakeup/sleep latencies.
 */
91
unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
92
unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL;
93

94 95
const_debug unsigned int sysctl_sched_migration_cost = 500000UL;

96 97 98 99 100 101 102
/*
 * The exponential sliding  window over which load is averaged for shares
 * distribution.
 * (default: 10msec)
 */
unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;

103 104 105 106 107 108 109 110 111 112 113 114 115 116
#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

117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134
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;
}

135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
/*
 * 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();
}

182
#define WMULT_CONST	(~0U)
183 184
#define WMULT_SHIFT	32

185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
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;
}
201 202

/*
203 204 205 206 207 208 209 210 211 212
 * 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.
213
 */
214
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
215
{
216 217
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
218

219
	__update_inv_weight(lw);
220

221 222 223 224 225
	if (unlikely(fact >> 32)) {
		while (fact >> 32) {
			fact >>= 1;
			shift--;
		}
226 227
	}

228 229
	/* hint to use a 32x32->64 mul */
	fact = (u64)(u32)fact * lw->inv_weight;
230

231 232 233 234
	while (fact >> 32) {
		fact >>= 1;
		shift--;
	}
235

236
	return mul_u64_u32_shr(delta_exec, fact, shift);
237 238 239 240
}


const struct sched_class fair_sched_class;
241

242 243 244 245
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

246
#ifdef CONFIG_FAIR_GROUP_SCHED
247

248
/* cpu runqueue to which this cfs_rq is attached */
249 250
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
251
	return cfs_rq->rq;
252 253
}

254 255
/* An entity is a task if it doesn't "own" a runqueue */
#define entity_is_task(se)	(!se->my_q)
256

257 258 259 260 261 262 263 264
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
265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
/* 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;
}

286 287
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
				       int force_update);
288

289 290 291
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
292 293 294 295 296 297 298 299 300 301 302 303
		/*
		 * 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,
304
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
305
		}
306 307

		cfs_rq->on_list = 1;
308
		/* We should have no load, but we need to update last_decay. */
309
		update_cfs_rq_blocked_load(cfs_rq, 0);
310 311 312 313 314 315 316 317 318 319 320
	}
}

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
321 322 323 324 325
/* 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
326
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
327 328 329
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
330
		return se->cfs_rq;
P
Peter Zijlstra 已提交
331

P
Peter Zijlstra 已提交
332
	return NULL;
P
Peter Zijlstra 已提交
333 334 335 336 337 338 339
}

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

340 341 342 343 344 345 346 347 348 349 350 351 352
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
353 354
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371

	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);
	}
}

372 373 374 375 376 377
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
378

379 380 381
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
382 383 384 385
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
386 387
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
388

P
Peter Zijlstra 已提交
389
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
390
{
P
Peter Zijlstra 已提交
391
	return &task_rq(p)->cfs;
392 393
}

P
Peter Zijlstra 已提交
394 395 396 397 398 399 400 401 402 403 404 405 406 407
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;
}

408 409 410 411 412 413 414 415
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
416 417 418 419 420 421 422 423
#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;
}

424 425 426 427 428
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
429 430
#endif	/* CONFIG_FAIR_GROUP_SCHED */

431
static __always_inline
432
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
433 434 435 436 437

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

438
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
439
{
440
	s64 delta = (s64)(vruntime - max_vruntime);
441
	if (delta > 0)
442
		max_vruntime = vruntime;
443

444
	return max_vruntime;
445 446
}

447
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
448 449 450 451 452 453 454 455
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

456 457 458 459 460 461
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

462 463 464 465 466 467 468 469 470 471 472 473
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
474
		if (!cfs_rq->curr)
475 476 477 478 479
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

480
	/* ensure we never gain time by being placed backwards. */
481
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
482 483 484 485
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
486 487
}

488 489 490
/*
 * Enqueue an entity into the rb-tree:
 */
491
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507
{
	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.
		 */
508
		if (entity_before(se, entry)) {
509 510 511 512 513 514 515 516 517 518 519
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
			leftmost = 0;
		}
	}

	/*
	 * Maintain a cache of leftmost tree entries (it is frequently
	 * used):
	 */
520
	if (leftmost)
I
Ingo Molnar 已提交
521
		cfs_rq->rb_leftmost = &se->run_node;
522 523 524 525 526

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

527
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
528
{
P
Peter Zijlstra 已提交
529 530 531 532 533 534
	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
535

536 537 538
	rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
}

539
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
540
{
541 542 543 544 545 546
	struct rb_node *left = cfs_rq->rb_leftmost;

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
547 548
}

549 550 551 552 553 554 555 556 557 558 559
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
560
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
561
{
562
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
563

564 565
	if (!last)
		return NULL;
566 567

	return rb_entry(last, struct sched_entity, run_node);
568 569
}

570 571 572 573
/**************************************************************
 * Scheduling class statistics methods:
 */

574
int sched_proc_update_handler(struct ctl_table *table, int write,
575
		void __user *buffer, size_t *lenp,
576 577
		loff_t *ppos)
{
578
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
579
	int factor = get_update_sysctl_factor();
580 581 582 583 584 585 586

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

587 588 589 590 591 592 593
#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

594 595 596
	return 0;
}
#endif
597

598
/*
599
 * delta /= w
600
 */
601
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
602
{
603
	if (unlikely(se->load.weight != NICE_0_LOAD))
604
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
605 606 607 608

	return delta;
}

609 610 611
/*
 * The idea is to set a period in which each task runs once.
 *
612
 * When there are too many tasks (sched_nr_latency) we have to stretch
613 614 615 616
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
617 618 619
static u64 __sched_period(unsigned long nr_running)
{
	u64 period = sysctl_sched_latency;
620
	unsigned long nr_latency = sched_nr_latency;
621 622

	if (unlikely(nr_running > nr_latency)) {
623
		period = sysctl_sched_min_granularity;
624 625 626 627 628 629
		period *= nr_running;
	}

	return period;
}

630 631 632 633
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
634
 * s = p*P[w/rw]
635
 */
P
Peter Zijlstra 已提交
636
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
637
{
M
Mike Galbraith 已提交
638
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
639

M
Mike Galbraith 已提交
640
	for_each_sched_entity(se) {
L
Lin Ming 已提交
641
		struct load_weight *load;
642
		struct load_weight lw;
L
Lin Ming 已提交
643 644 645

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
646

M
Mike Galbraith 已提交
647
		if (unlikely(!se->on_rq)) {
648
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
649 650 651 652

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
653
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
654 655
	}
	return slice;
656 657
}

658
/*
A
Andrei Epure 已提交
659
 * We calculate the vruntime slice of a to-be-inserted task.
660
 *
661
 * vs = s/w
662
 */
663
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
664
{
665
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
666 667
}

668
#ifdef CONFIG_SMP
669
static int select_idle_sibling(struct task_struct *p, int cpu);
670 671
static unsigned long task_h_load(struct task_struct *p);

672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690
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

691
/*
692
 * Update the current task's runtime statistics.
693
 */
694
static void update_curr(struct cfs_rq *cfs_rq)
695
{
696
	struct sched_entity *curr = cfs_rq->curr;
697
	u64 now = rq_clock_task(rq_of(cfs_rq));
698
	u64 delta_exec;
699 700 701 702

	if (unlikely(!curr))
		return;

703 704
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
705
		return;
706

I
Ingo Molnar 已提交
707
	curr->exec_start = now;
708

709 710 711 712 713 714 715 716 717
	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);

718 719 720
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

721
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
722
		cpuacct_charge(curtask, delta_exec);
723
		account_group_exec_runtime(curtask, delta_exec);
724
	}
725 726

	account_cfs_rq_runtime(cfs_rq, delta_exec);
727 728 729
}

static inline void
730
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
731
{
732
	schedstat_set(se->statistics.wait_start, rq_clock(rq_of(cfs_rq)));
733 734 735 736 737
}

/*
 * Task is being enqueued - update stats:
 */
738
static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
739 740 741 742 743
{
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
744
	if (se != cfs_rq->curr)
745
		update_stats_wait_start(cfs_rq, se);
746 747 748
}

static void
749
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
750
{
751
	schedstat_set(se->statistics.wait_max, max(se->statistics.wait_max,
752
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start));
753 754
	schedstat_set(se->statistics.wait_count, se->statistics.wait_count + 1);
	schedstat_set(se->statistics.wait_sum, se->statistics.wait_sum +
755
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
756 757 758
#ifdef CONFIG_SCHEDSTATS
	if (entity_is_task(se)) {
		trace_sched_stat_wait(task_of(se),
759
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
760 761
	}
#endif
762
	schedstat_set(se->statistics.wait_start, 0);
763 764 765
}

static inline void
766
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
767 768 769 770 771
{
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
772
	if (se != cfs_rq->curr)
773
		update_stats_wait_end(cfs_rq, se);
774 775 776 777 778 779
}

/*
 * We are picking a new current task - update its stats:
 */
static inline void
780
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
781 782 783 784
{
	/*
	 * We are starting a new run period:
	 */
785
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
786 787 788 789 790 791
}

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

792 793
#ifdef CONFIG_NUMA_BALANCING
/*
794 795 796
 * 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.
797
 */
798 799
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
800 801 802

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
803

804 805 806
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851
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);
}

852 853 854 855 856 857 858 859 860 861 862 863
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));
}

864 865 866 867 868
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
869
	pid_t gid;
870 871 872
	struct list_head task_list;

	struct rcu_head rcu;
873
	nodemask_t active_nodes;
874
	unsigned long total_faults;
875 876 877 878 879
	/*
	 * 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.
	 */
880
	unsigned long *faults_cpu;
881
	unsigned long faults[0];
882 883
};

884 885 886 887 888 889 890 891 892
/* 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)

893 894 895 896 897
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

898 899
static inline int task_faults_idx(int nid, int priv)
{
900
	return NR_NUMA_HINT_FAULT_TYPES * nid + priv;
901 902 903 904
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
905
	if (!p->numa_faults_memory)
906 907
		return 0;

908 909
	return p->numa_faults_memory[task_faults_idx(nid, 0)] +
		p->numa_faults_memory[task_faults_idx(nid, 1)];
910 911
}

912 913 914 915 916
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

917 918
	return p->numa_group->faults[task_faults_idx(nid, 0)] +
		p->numa_group->faults[task_faults_idx(nid, 1)];
919 920
}

921 922 923 924 925 926
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)];
}

927 928 929 930 931 932 933 934 935 936
/*
 * 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;

937
	if (!p->numa_faults_memory)
938 939 940 941 942 943 944 945 946 947 948 949
		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)
{
950
	if (!p->numa_group || !p->numa_group->total_faults)
951 952
		return 0;

953
	return 1000 * group_faults(p, nid) / p->numa_group->total_faults;
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 1018
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);
}

1019
static unsigned long weighted_cpuload(const int cpu);
1020 1021
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1022
static unsigned long capacity_of(int cpu);
1023 1024
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1025
/* Cached statistics for all CPUs within a node */
1026
struct numa_stats {
1027
	unsigned long nr_running;
1028
	unsigned long load;
1029 1030

	/* Total compute capacity of CPUs on a node */
1031
	unsigned long compute_capacity;
1032 1033

	/* Approximate capacity in terms of runnable tasks on a node */
1034
	unsigned long task_capacity;
1035
	int has_free_capacity;
1036
};
1037

1038 1039 1040 1041 1042
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1043 1044
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1045 1046 1047 1048 1049 1050 1051

	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);
1052
		ns->compute_capacity += capacity_of(cpu);
1053 1054

		cpus++;
1055 1056
	}

1057 1058 1059 1060 1061
	/*
	 * 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.
	 *
1062 1063
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1064 1065 1066 1067
	 */
	if (!cpus)
		return;

1068 1069 1070 1071 1072 1073
	/* 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));
1074
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1075 1076
}

1077 1078
struct task_numa_env {
	struct task_struct *p;
1079

1080 1081
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1082

1083
	struct numa_stats src_stats, dst_stats;
1084

1085
	int imbalance_pct;
1086 1087 1088

	struct task_struct *best_task;
	long best_imp;
1089 1090 1091
	int best_cpu;
};

1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104
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;
}

1105
static bool load_too_imbalanced(long src_load, long dst_load,
1106 1107 1108
				struct task_numa_env *env)
{
	long imb, old_imb;
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120
	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;
1121 1122 1123 1124 1125 1126

	/* 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? */
1127 1128
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1129 1130 1131 1132 1133 1134 1135
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
	 * Compare it with the old imbalance.
	 */
1136 1137 1138
	orig_src_load = env->src_stats.load;
	orig_dst_load = env->dst_stats.load;

1139 1140 1141
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);

1142 1143
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;
1144 1145

	/* Would this change make things worse? */
1146
	return (imb > old_imb);
1147 1148
}

1149 1150 1151 1152 1153 1154
/*
 * 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
 */
1155 1156
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1157 1158 1159 1160
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1161
	long src_load, dst_load;
1162
	long load;
1163
	long imp = env->p->numa_group ? groupimp : taskimp;
1164
	long moveimp = imp;
1165 1166

	rcu_read_lock();
1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177

	raw_spin_lock_irq(&dst_rq->lock);
	cur = dst_rq->curr;
	/*
	 * No need to move the exiting task, and this ensures that ->curr
	 * wasn't reaped and thus get_task_struct() in task_numa_assign()
	 * is safe under RCU read lock.
	 * Note that rcu_read_lock() itself can't protect from the final
	 * put_task_struct() after the last schedule().
	 */
	if ((cur->flags & PF_EXITING) || is_idle_task(cur))
1178
		cur = NULL;
1179
	raw_spin_unlock_irq(&dst_rq->lock);
1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192

	/*
	 * "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;

1193 1194
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1195
		 * in any group then look only at task weights.
1196
		 */
1197
		if (cur->numa_group == env->p->numa_group) {
1198 1199
			imp = taskimp + task_weight(cur, env->src_nid) -
			      task_weight(cur, env->dst_nid);
1200 1201 1202 1203 1204 1205
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1206
		} else {
1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217
			/*
			 * 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);
1218
		}
1219 1220
	}

1221
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1222 1223 1224 1225
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1226
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1227
		    !env->dst_stats.has_free_capacity)
1228 1229 1230 1231 1232 1233
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1234 1235
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1236 1237 1238 1239 1240 1241
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1242 1243 1244
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1245

1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
	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;

1263
	if (cur) {
1264 1265 1266
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1267 1268
	}

1269
	if (load_too_imbalanced(src_load, dst_load, env))
1270 1271
		goto unlock;

1272 1273 1274 1275 1276 1277 1278
	/*
	 * 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);

1279 1280 1281 1282 1283 1284
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1285 1286
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1287 1288 1289 1290 1291 1292 1293 1294 1295
{
	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;
1296
		task_numa_compare(env, taskimp, groupimp);
1297 1298 1299
	}
}

1300 1301 1302 1303
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1304

1305
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1306
		.src_nid = task_node(p),
1307 1308 1309 1310 1311 1312

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
		.best_cpu = -1
1313 1314
	};
	struct sched_domain *sd;
1315
	unsigned long taskweight, groupweight;
1316
	int nid, ret;
1317
	long taskimp, groupimp;
1318

1319
	/*
1320 1321 1322 1323 1324 1325
	 * 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.
1326 1327
	 */
	rcu_read_lock();
1328
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1329 1330
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1331 1332
	rcu_read_unlock();

1333 1334 1335 1336 1337 1338 1339
	/*
	 * 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)) {
1340
		p->numa_preferred_nid = task_node(p);
1341 1342 1343
		return -EINVAL;
	}

1344 1345
	taskweight = task_weight(p, env.src_nid);
	groupweight = group_weight(p, env.src_nid);
1346
	update_numa_stats(&env.src_stats, env.src_nid);
1347
	env.dst_nid = p->numa_preferred_nid;
1348 1349
	taskimp = task_weight(p, env.dst_nid) - taskweight;
	groupimp = group_weight(p, env.dst_nid) - groupweight;
1350
	update_numa_stats(&env.dst_stats, env.dst_nid);
1351

1352 1353
	/* Try to find a spot on the preferred nid. */
	task_numa_find_cpu(&env, taskimp, groupimp);
1354 1355 1356

	/* No space available on the preferred nid. Look elsewhere. */
	if (env.best_cpu == -1) {
1357 1358 1359
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1360

1361
			/* Only consider nodes where both task and groups benefit */
1362 1363 1364
			taskimp = task_weight(p, nid) - taskweight;
			groupimp = group_weight(p, nid) - groupweight;
			if (taskimp < 0 && groupimp < 0)
1365 1366
				continue;

1367 1368
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1369
			task_numa_find_cpu(&env, taskimp, groupimp);
1370 1371 1372
		}
	}

1373 1374 1375 1376 1377 1378 1379 1380
	/*
	 * 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.
	 */
1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393
	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;
1394

1395 1396 1397 1398 1399 1400
	/*
	 * 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);

1401
	if (env.best_task == NULL) {
1402 1403 1404
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1405 1406 1407 1408
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1409 1410
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1411 1412
	put_task_struct(env.best_task);
	return ret;
1413 1414
}

1415 1416 1417
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1418 1419
	unsigned long interval = HZ;

1420
	/* This task has no NUMA fault statistics yet */
1421
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults_memory))
1422 1423
		return;

1424
	/* Periodically retry migrating the task to the preferred node */
1425 1426
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1427 1428

	/* Success if task is already running on preferred CPU */
1429
	if (task_node(p) == p->numa_preferred_nid)
1430 1431 1432
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1433
	task_numa_migrate(p);
1434 1435
}

1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467
/*
 * 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);
	}
}

1468 1469 1470
/*
 * 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
1471 1472 1473
 * 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.
1474 1475
 */
#define NUMA_PERIOD_SLOTS 10
1476
#define NUMA_PERIOD_THRESHOLD 7
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 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541

/*
 * 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));
}

1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569
/*
 * 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;
}

1570 1571
static void task_numa_placement(struct task_struct *p)
{
1572 1573
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
1574
	unsigned long fault_types[2] = { 0, 0 };
1575 1576
	unsigned long total_faults;
	u64 runtime, period;
1577
	spinlock_t *group_lock = NULL;
1578

1579
	seq = ACCESS_ONCE(p->mm->numa_scan_seq);
1580 1581 1582
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
1583
	p->numa_scan_period_max = task_scan_max(p);
1584

1585 1586 1587 1588
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

1589 1590 1591
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
1592
		spin_lock_irq(group_lock);
1593 1594
	}

1595 1596
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
1597
		unsigned long faults = 0, group_faults = 0;
1598
		int priv, i;
1599

1600
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
1601
			long diff, f_diff, f_weight;
1602

1603
			i = task_faults_idx(nid, priv);
1604

1605
			/* Decay existing window, copy faults since last scan */
1606
			diff = p->numa_faults_buffer_memory[i] - p->numa_faults_memory[i] / 2;
1607 1608
			fault_types[priv] += p->numa_faults_buffer_memory[i];
			p->numa_faults_buffer_memory[i] = 0;
1609

1610 1611 1612 1613 1614 1615 1616 1617 1618 1619
			/*
			 * 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);
1620
			f_diff = f_weight - p->numa_faults_cpu[i] / 2;
1621 1622
			p->numa_faults_buffer_cpu[i] = 0;

1623 1624
			p->numa_faults_memory[i] += diff;
			p->numa_faults_cpu[i] += f_diff;
1625
			faults += p->numa_faults_memory[i];
1626
			p->total_numa_faults += diff;
1627 1628
			if (p->numa_group) {
				/* safe because we can only change our own group */
1629
				p->numa_group->faults[i] += diff;
1630
				p->numa_group->faults_cpu[i] += f_diff;
1631 1632
				p->numa_group->total_faults += diff;
				group_faults += p->numa_group->faults[i];
1633
			}
1634 1635
		}

1636 1637 1638 1639
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
1640 1641 1642 1643 1644 1645 1646

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

1647 1648
	update_task_scan_period(p, fault_types[0], fault_types[1]);

1649
	if (p->numa_group) {
1650
		update_numa_active_node_mask(p->numa_group);
1651
		spin_unlock_irq(group_lock);
1652
		max_nid = max_group_nid;
1653 1654
	}

1655 1656 1657 1658 1659 1660 1661
	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);
1662
	}
1663 1664
}

1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675
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);
}

1676 1677
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
1678 1679 1680 1681 1682 1683 1684 1685 1686
{
	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) +
1687
				    4*nr_node_ids*sizeof(unsigned long);
1688 1689 1690 1691 1692 1693 1694 1695

		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);
1696
		grp->gid = p->pid;
1697
		/* Second half of the array tracks nids where faults happen */
1698 1699
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
1700

1701 1702
		node_set(task_node(current), grp->active_nodes);

1703
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1704
			grp->faults[i] = p->numa_faults_memory[i];
1705

1706
		grp->total_faults = p->total_numa_faults;
1707

1708 1709 1710 1711 1712 1713 1714 1715 1716
		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))
1717
		goto no_join;
1718 1719 1720

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
1721
		goto no_join;
1722 1723 1724

	my_grp = p->numa_group;
	if (grp == my_grp)
1725
		goto no_join;
1726 1727 1728 1729 1730 1731

	/*
	 * 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)
1732
		goto no_join;
1733 1734 1735 1736 1737

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
1738
		goto no_join;
1739

1740 1741 1742 1743 1744 1745 1746
	/* 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;
1747

1748 1749 1750
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

1751
	if (join && !get_numa_group(grp))
1752
		goto no_join;
1753 1754 1755 1756 1757 1758

	rcu_read_unlock();

	if (!join)
		return;

1759 1760
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
1761

1762
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
1763 1764
		my_grp->faults[i] -= p->numa_faults_memory[i];
		grp->faults[i] += p->numa_faults_memory[i];
1765
	}
1766 1767
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
1768 1769 1770 1771 1772 1773

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

	spin_unlock(&my_grp->lock);
1774
	spin_unlock_irq(&grp->lock);
1775 1776 1777 1778

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
1779 1780 1781 1782 1783
	return;

no_join:
	rcu_read_unlock();
	return;
1784 1785 1786 1787 1788
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
1789
	void *numa_faults = p->numa_faults_memory;
1790 1791
	unsigned long flags;
	int i;
1792 1793

	if (grp) {
1794
		spin_lock_irqsave(&grp->lock, flags);
1795
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1796
			grp->faults[i] -= p->numa_faults_memory[i];
1797
		grp->total_faults -= p->total_numa_faults;
1798

1799 1800
		list_del(&p->numa_entry);
		grp->nr_tasks--;
1801
		spin_unlock_irqrestore(&grp->lock, flags);
1802
		RCU_INIT_POINTER(p->numa_group, NULL);
1803 1804 1805
		put_numa_group(grp);
	}

1806 1807
	p->numa_faults_memory = NULL;
	p->numa_faults_buffer_memory = NULL;
1808 1809
	p->numa_faults_cpu= NULL;
	p->numa_faults_buffer_cpu = NULL;
1810
	kfree(numa_faults);
1811 1812
}

1813 1814 1815
/*
 * Got a PROT_NONE fault for a page on @node.
 */
1816
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
1817 1818
{
	struct task_struct *p = current;
1819
	bool migrated = flags & TNF_MIGRATED;
1820
	int cpu_node = task_node(current);
1821
	int local = !!(flags & TNF_FAULT_LOCAL);
1822
	int priv;
1823

1824
	if (!numabalancing_enabled)
1825 1826
		return;

1827 1828 1829 1830
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

1831
	/* Allocate buffer to track faults on a per-node basis */
1832
	if (unlikely(!p->numa_faults_memory)) {
1833 1834
		int size = sizeof(*p->numa_faults_memory) *
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
1835

1836
		p->numa_faults_memory = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
1837
		if (!p->numa_faults_memory)
1838
			return;
1839

1840
		BUG_ON(p->numa_faults_buffer_memory);
1841 1842 1843 1844 1845 1846
		/*
		 * 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.
		 */
1847 1848 1849
		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);
1850
		p->total_numa_faults = 0;
1851
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
1852
	}
1853

1854 1855 1856 1857 1858 1859 1860 1861
	/*
	 * 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);
1862
		if (!priv && !(flags & TNF_NO_GROUP))
1863
			task_numa_group(p, last_cpupid, flags, &priv);
1864 1865
	}

1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876
	/*
	 * 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;

1877
	task_numa_placement(p);
1878

1879 1880 1881 1882 1883
	/*
	 * 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))
1884 1885
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
1886 1887 1888
	if (migrated)
		p->numa_pages_migrated += pages;

1889 1890
	p->numa_faults_buffer_memory[task_faults_idx(mem_node, priv)] += pages;
	p->numa_faults_buffer_cpu[task_faults_idx(cpu_node, priv)] += pages;
1891
	p->numa_faults_locality[local] += pages;
1892 1893
}

1894 1895 1896 1897 1898 1899
static void reset_ptenuma_scan(struct task_struct *p)
{
	ACCESS_ONCE(p->mm->numa_scan_seq)++;
	p->mm->numa_scan_offset = 0;
}

1900 1901 1902 1903 1904 1905 1906 1907 1908
/*
 * 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;
1909
	struct vm_area_struct *vma;
1910
	unsigned long start, end;
1911
	unsigned long nr_pte_updates = 0;
1912
	long pages;
1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927

	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;

1928
	if (!mm->numa_next_scan) {
1929 1930
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
1931 1932
	}

1933 1934 1935 1936 1937 1938 1939
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

1940 1941 1942 1943
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
1944

1945
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
1946 1947 1948
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

1949 1950 1951 1952 1953 1954
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

1955 1956 1957 1958 1959
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
	if (!pages)
		return;
1960

1961
	down_read(&mm->mmap_sem);
1962
	vma = find_vma(mm, start);
1963 1964
	if (!vma) {
		reset_ptenuma_scan(p);
1965
		start = 0;
1966 1967
		vma = mm->mmap;
	}
1968
	for (; vma; vma = vma->vm_next) {
1969
		if (!vma_migratable(vma) || !vma_policy_mof(vma))
1970 1971
			continue;

1972 1973 1974 1975 1976 1977 1978 1979 1980 1981
		/*
		 * 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
1982 1983 1984 1985 1986 1987
		/*
		 * 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;
1988

1989 1990 1991 1992
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
1993 1994 1995 1996 1997 1998 1999 2000 2001
			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;
2002

2003 2004 2005
			start = end;
			if (pages <= 0)
				goto out;
2006 2007

			cond_resched();
2008
		} while (end != vma->vm_end);
2009
	}
2010

2011
out:
2012
	/*
P
Peter Zijlstra 已提交
2013 2014 2015 2016
	 * 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.
2017 2018
	 */
	if (vma)
2019
		mm->numa_scan_offset = start;
2020 2021 2022
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048
}

/*
 * 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) {
2049
		if (!curr->node_stamp)
2050
			curr->numa_scan_period = task_scan_min(curr);
2051
		curr->node_stamp += period;
2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062

		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)
{
}
2063 2064 2065 2066 2067 2068 2069 2070

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)
{
}
2071 2072
#endif /* CONFIG_NUMA_BALANCING */

2073 2074 2075 2076
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2077
	if (!parent_entity(se))
2078
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2079
#ifdef CONFIG_SMP
2080 2081 2082 2083 2084 2085
	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);
	}
2086
#endif
2087 2088 2089 2090 2091 2092 2093
	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);
2094
	if (!parent_entity(se))
2095
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2096 2097
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2098
		list_del_init(&se->group_node);
2099
	}
2100 2101 2102
	cfs_rq->nr_running--;
}

2103 2104
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2105 2106 2107 2108 2109 2110 2111 2112 2113
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().
	 */
2114
	tg_weight = atomic_long_read(&tg->load_avg);
2115
	tg_weight -= cfs_rq->tg_load_contrib;
2116 2117 2118 2119 2120
	tg_weight += cfs_rq->load.weight;

	return tg_weight;
}

2121
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2122
{
2123
	long tg_weight, load, shares;
2124

2125
	tg_weight = calc_tg_weight(tg, cfs_rq);
2126
	load = cfs_rq->load.weight;
2127 2128

	shares = (tg->shares * load);
2129 2130
	if (tg_weight)
		shares /= tg_weight;
2131 2132 2133 2134 2135 2136 2137 2138 2139

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

	return shares;
}
# else /* CONFIG_SMP */
2140
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2141 2142 2143 2144
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
2145 2146 2147
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2148 2149 2150 2151
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2152
		account_entity_dequeue(cfs_rq, se);
2153
	}
P
Peter Zijlstra 已提交
2154 2155 2156 2157 2158 2159 2160

	update_load_set(&se->load, weight);

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

2161 2162
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2163
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2164 2165 2166
{
	struct task_group *tg;
	struct sched_entity *se;
2167
	long shares;
P
Peter Zijlstra 已提交
2168 2169 2170

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2171
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2172
		return;
2173 2174 2175 2176
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2177
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2178 2179 2180 2181

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2182
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2183 2184 2185 2186
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2187
#ifdef CONFIG_SMP
2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215
/*
 * 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,
};

2216 2217 2218 2219 2220 2221
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233
	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
2234 2235
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
2236 2237 2238 2239 2240 2241
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
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 2268 2269 2270 2271 2272 2273 2274
	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];
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 2302 2303 2304 2305 2306 2307 2308
}

/*
 * 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)
{
2309 2310
	u64 delta, periods;
	u32 runnable_contrib;
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 2337 2338 2339 2340 2341 2342 2343
	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;
2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363
		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;
2364 2365 2366 2367 2368 2369 2370 2371 2372 2373
	}

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

	return decayed;
}

2374
/* Synchronize an entity's decay with its parenting cfs_rq.*/
2375
static inline u64 __synchronize_entity_decay(struct sched_entity *se)
2376 2377 2378 2379 2380 2381
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 decays = atomic64_read(&cfs_rq->decay_counter);

	decays -= se->avg.decay_count;
	if (!decays)
2382
		return 0;
2383 2384 2385

	se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
	se->avg.decay_count = 0;
2386 2387

	return decays;
2388 2389
}

2390 2391 2392 2393 2394
#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;
2395
	long tg_contrib;
2396 2397 2398 2399

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

2400 2401 2402
	if (!tg_contrib)
		return;

2403 2404
	if (force_update || abs(tg_contrib) > cfs_rq->tg_load_contrib / 8) {
		atomic_long_add(tg_contrib, &tg->load_avg);
2405 2406 2407
		cfs_rq->tg_load_contrib += tg_contrib;
	}
}
2408

2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419
/*
 * 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 */
2420
	contrib = div_u64((u64)sa->runnable_avg_sum << NICE_0_SHIFT,
2421 2422 2423 2424 2425 2426 2427 2428 2429
			  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;
	}
}

2430 2431 2432 2433
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;
2434 2435
	int runnable_avg;

2436 2437 2438
	u64 contrib;

	contrib = cfs_rq->tg_load_contrib * tg->shares;
2439 2440
	se->avg.load_avg_contrib = div_u64(contrib,
				     atomic_long_read(&tg->load_avg) + 1);
2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469

	/*
	 * 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;
	}
2470
}
2471 2472 2473 2474 2475 2476

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);
}
2477
#else /* CONFIG_FAIR_GROUP_SCHED */
2478 2479
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update) {}
2480 2481
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq) {}
2482
static inline void __update_group_entity_contrib(struct sched_entity *se) {}
2483
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2484
#endif /* CONFIG_FAIR_GROUP_SCHED */
2485

2486 2487 2488 2489 2490 2491 2492 2493 2494 2495
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);
}

2496 2497 2498 2499 2500
/* 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;

2501 2502 2503
	if (entity_is_task(se)) {
		__update_task_entity_contrib(se);
	} else {
2504
		__update_tg_runnable_avg(&se->avg, group_cfs_rq(se));
2505 2506
		__update_group_entity_contrib(se);
	}
2507 2508 2509 2510

	return se->avg.load_avg_contrib - old_contrib;
}

2511 2512 2513 2514 2515 2516 2517 2518 2519
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;
}

2520 2521
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);

2522
/* Update a sched_entity's runnable average */
2523 2524
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq)
2525
{
2526 2527
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	long contrib_delta;
2528
	u64 now;
2529

2530 2531 2532 2533 2534 2535 2536 2537 2538 2539
	/*
	 * 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))
2540 2541 2542
		return;

	contrib_delta = __update_entity_load_avg_contrib(se);
2543 2544 2545 2546

	if (!update_cfs_rq)
		return;

2547 2548
	if (se->on_rq)
		cfs_rq->runnable_load_avg += contrib_delta;
2549 2550 2551 2552 2553 2554 2555 2556
	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.
 */
2557
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq, int force_update)
2558
{
2559
	u64 now = cfs_rq_clock_task(cfs_rq) >> 20;
2560 2561 2562
	u64 decays;

	decays = now - cfs_rq->last_decay;
2563
	if (!decays && !force_update)
2564 2565
		return;

2566 2567 2568
	if (atomic_long_read(&cfs_rq->removed_load)) {
		unsigned long removed_load;
		removed_load = atomic_long_xchg(&cfs_rq->removed_load, 0);
2569 2570
		subtract_blocked_load_contrib(cfs_rq, removed_load);
	}
2571

2572 2573 2574 2575 2576 2577
	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;
	}
2578 2579

	__update_cfs_rq_tg_load_contrib(cfs_rq, force_update);
2580
}
2581

2582 2583
/* 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,
2584 2585
						  struct sched_entity *se,
						  int wakeup)
2586
{
2587 2588 2589 2590
	/*
	 * 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.
2591 2592 2593 2594
	 *
	 * 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.
2595 2596
	 */
	if (unlikely(se->avg.decay_count <= 0)) {
2597
		se->avg.last_runnable_update = rq_clock_task(rq_of(cfs_rq));
2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612
		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;
		}
2613 2614
		wakeup = 0;
	} else {
2615
		__synchronize_entity_decay(se);
2616 2617
	}

2618 2619
	/* migrated tasks did not contribute to our blocked load */
	if (wakeup) {
2620
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
2621 2622
		update_entity_load_avg(se, 0);
	}
2623

2624
	cfs_rq->runnable_load_avg += se->avg.load_avg_contrib;
2625 2626
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !wakeup);
2627 2628
}

2629 2630 2631 2632 2633
/*
 * 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.
 */
2634
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2635 2636
						  struct sched_entity *se,
						  int sleep)
2637
{
2638
	update_entity_load_avg(se, 1);
2639 2640
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !sleep);
2641

2642
	cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib;
2643 2644 2645 2646
	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 */
2647
}
2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668

/*
 * 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);
}

2669 2670
static int idle_balance(struct rq *this_rq);

2671 2672
#else /* CONFIG_SMP */

2673 2674
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq) {}
2675
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2676
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2677 2678
					   struct sched_entity *se,
					   int wakeup) {}
2679
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2680 2681
					   struct sched_entity *se,
					   int sleep) {}
2682 2683
static inline void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
					      int force_update) {}
2684 2685 2686 2687 2688 2689

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

2690
#endif /* CONFIG_SMP */
2691

2692
static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
2693 2694
{
#ifdef CONFIG_SCHEDSTATS
2695 2696 2697 2698 2699
	struct task_struct *tsk = NULL;

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

2700
	if (se->statistics.sleep_start) {
2701
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.sleep_start;
2702 2703 2704 2705

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

2706 2707
		if (unlikely(delta > se->statistics.sleep_max))
			se->statistics.sleep_max = delta;
2708

2709
		se->statistics.sleep_start = 0;
2710
		se->statistics.sum_sleep_runtime += delta;
A
Arjan van de Ven 已提交
2711

2712
		if (tsk) {
2713
			account_scheduler_latency(tsk, delta >> 10, 1);
2714 2715
			trace_sched_stat_sleep(tsk, delta);
		}
2716
	}
2717
	if (se->statistics.block_start) {
2718
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.block_start;
2719 2720 2721 2722

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

2723 2724
		if (unlikely(delta > se->statistics.block_max))
			se->statistics.block_max = delta;
2725

2726
		se->statistics.block_start = 0;
2727
		se->statistics.sum_sleep_runtime += delta;
I
Ingo Molnar 已提交
2728

2729
		if (tsk) {
2730
			if (tsk->in_iowait) {
2731 2732
				se->statistics.iowait_sum += delta;
				se->statistics.iowait_count++;
2733
				trace_sched_stat_iowait(tsk, delta);
2734 2735
			}

2736 2737
			trace_sched_stat_blocked(tsk, delta);

2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
			/*
			 * 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
2749
		}
2750 2751 2752 2753
	}
#endif
}

P
Peter Zijlstra 已提交
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766
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
}

2767 2768 2769
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
2770
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
2771

2772 2773 2774 2775 2776 2777
	/*
	 * 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
2778
	if (initial && sched_feat(START_DEBIT))
2779
		vruntime += sched_vslice(cfs_rq, se);
2780

2781
	/* sleeps up to a single latency don't count. */
2782
	if (!initial) {
2783
		unsigned long thresh = sysctl_sched_latency;
2784

2785 2786 2787 2788 2789 2790
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
2791

2792
		vruntime -= thresh;
2793 2794
	}

2795
	/* ensure we never gain time by being placed backwards. */
2796
	se->vruntime = max_vruntime(se->vruntime, vruntime);
2797 2798
}

2799 2800
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

2801
static void
2802
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2803
{
2804 2805
	/*
	 * Update the normalized vruntime before updating min_vruntime
2806
	 * through calling update_curr().
2807
	 */
2808
	if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
2809 2810
		se->vruntime += cfs_rq->min_vruntime;

2811
	/*
2812
	 * Update run-time statistics of the 'current'.
2813
	 */
2814
	update_curr(cfs_rq);
2815
	enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP);
2816 2817
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
2818

2819
	if (flags & ENQUEUE_WAKEUP) {
2820
		place_entity(cfs_rq, se, 0);
2821
		enqueue_sleeper(cfs_rq, se);
I
Ingo Molnar 已提交
2822
	}
2823

2824
	update_stats_enqueue(cfs_rq, se);
P
Peter Zijlstra 已提交
2825
	check_spread(cfs_rq, se);
2826 2827
	if (se != cfs_rq->curr)
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
2828
	se->on_rq = 1;
2829

2830
	if (cfs_rq->nr_running == 1) {
2831
		list_add_leaf_cfs_rq(cfs_rq);
2832 2833
		check_enqueue_throttle(cfs_rq);
	}
2834 2835
}

2836
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
2837
{
2838 2839
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2840
		if (cfs_rq->last != se)
2841
			break;
2842 2843

		cfs_rq->last = NULL;
2844 2845
	}
}
P
Peter Zijlstra 已提交
2846

2847 2848 2849 2850
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2851
		if (cfs_rq->next != se)
2852
			break;
2853 2854

		cfs_rq->next = NULL;
2855
	}
P
Peter Zijlstra 已提交
2856 2857
}

2858 2859 2860 2861
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2862
		if (cfs_rq->skip != se)
2863
			break;
2864 2865

		cfs_rq->skip = NULL;
2866 2867 2868
	}
}

P
Peter Zijlstra 已提交
2869 2870
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2871 2872 2873 2874 2875
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
2876 2877 2878

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
2879 2880
}

2881
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
2882

2883
static void
2884
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2885
{
2886 2887 2888 2889
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
2890
	dequeue_entity_load_avg(cfs_rq, se, flags & DEQUEUE_SLEEP);
2891

2892
	update_stats_dequeue(cfs_rq, se);
2893
	if (flags & DEQUEUE_SLEEP) {
P
Peter Zijlstra 已提交
2894
#ifdef CONFIG_SCHEDSTATS
2895 2896 2897 2898
		if (entity_is_task(se)) {
			struct task_struct *tsk = task_of(se);

			if (tsk->state & TASK_INTERRUPTIBLE)
2899
				se->statistics.sleep_start = rq_clock(rq_of(cfs_rq));
2900
			if (tsk->state & TASK_UNINTERRUPTIBLE)
2901
				se->statistics.block_start = rq_clock(rq_of(cfs_rq));
2902
		}
2903
#endif
P
Peter Zijlstra 已提交
2904 2905
	}

P
Peter Zijlstra 已提交
2906
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
2907

2908
	if (se != cfs_rq->curr)
2909
		__dequeue_entity(cfs_rq, se);
2910
	se->on_rq = 0;
2911
	account_entity_dequeue(cfs_rq, se);
2912 2913 2914 2915 2916 2917

	/*
	 * 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.
	 */
2918
	if (!(flags & DEQUEUE_SLEEP))
2919
		se->vruntime -= cfs_rq->min_vruntime;
2920

2921 2922 2923
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

2924
	update_min_vruntime(cfs_rq);
2925
	update_cfs_shares(cfs_rq);
2926 2927 2928 2929 2930
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
2931
static void
I
Ingo Molnar 已提交
2932
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
2933
{
2934
	unsigned long ideal_runtime, delta_exec;
2935 2936
	struct sched_entity *se;
	s64 delta;
2937

P
Peter Zijlstra 已提交
2938
	ideal_runtime = sched_slice(cfs_rq, curr);
2939
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
2940
	if (delta_exec > ideal_runtime) {
2941
		resched_curr(rq_of(cfs_rq));
2942 2943 2944 2945 2946
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957
		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;

2958 2959
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
2960

2961 2962
	if (delta < 0)
		return;
2963

2964
	if (delta > ideal_runtime)
2965
		resched_curr(rq_of(cfs_rq));
2966 2967
}

2968
static void
2969
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
2970
{
2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981
	/* '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);
	}

2982
	update_stats_curr_start(cfs_rq, se);
2983
	cfs_rq->curr = se;
I
Ingo Molnar 已提交
2984 2985 2986 2987 2988 2989
#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):
	 */
2990
	if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
2991
		se->statistics.slice_max = max(se->statistics.slice_max,
I
Ingo Molnar 已提交
2992 2993 2994
			se->sum_exec_runtime - se->prev_sum_exec_runtime);
	}
#endif
2995
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
2996 2997
}

2998 2999 3000
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

3001 3002 3003 3004 3005 3006 3007
/*
 * 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
 */
3008 3009
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3010
{
3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021
	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 */
3022

3023 3024 3025 3026 3027
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3028 3029 3030 3031 3032 3033 3034 3035 3036 3037
		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;
		}

3038 3039 3040
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3041

3042 3043 3044 3045 3046 3047
	/*
	 * 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;

3048 3049 3050 3051 3052 3053
	/*
	 * 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;

3054
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3055 3056

	return se;
3057 3058
}

3059
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3060

3061
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3062 3063 3064 3065 3066 3067
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3068
		update_curr(cfs_rq);
3069

3070 3071 3072
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

P
Peter Zijlstra 已提交
3073
	check_spread(cfs_rq, prev);
3074
	if (prev->on_rq) {
3075
		update_stats_wait_start(cfs_rq, prev);
3076 3077
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3078
		/* in !on_rq case, update occurred at dequeue */
3079
		update_entity_load_avg(prev, 1);
3080
	}
3081
	cfs_rq->curr = NULL;
3082 3083
}

P
Peter Zijlstra 已提交
3084 3085
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3086 3087
{
	/*
3088
	 * Update run-time statistics of the 'current'.
3089
	 */
3090
	update_curr(cfs_rq);
3091

3092 3093 3094
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3095
	update_entity_load_avg(curr, 1);
3096
	update_cfs_rq_blocked_load(cfs_rq, 1);
3097
	update_cfs_shares(cfs_rq);
3098

P
Peter Zijlstra 已提交
3099 3100 3101 3102 3103
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3104
	if (queued) {
3105
		resched_curr(rq_of(cfs_rq));
3106 3107
		return;
	}
P
Peter Zijlstra 已提交
3108 3109 3110 3111 3112 3113 3114 3115
	/*
	 * 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
3116
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3117
		check_preempt_tick(cfs_rq, curr);
3118 3119
}

3120 3121 3122 3123 3124 3125

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

#ifdef CONFIG_CFS_BANDWIDTH
3126 3127

#ifdef HAVE_JUMP_LABEL
3128
static struct static_key __cfs_bandwidth_used;
3129 3130 3131

static inline bool cfs_bandwidth_used(void)
{
3132
	return static_key_false(&__cfs_bandwidth_used);
3133 3134
}

3135
void cfs_bandwidth_usage_inc(void)
3136
{
3137 3138 3139 3140 3141 3142
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3143 3144 3145 3146 3147 3148 3149
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3150 3151
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3152 3153
#endif /* HAVE_JUMP_LABEL */

3154 3155 3156 3157 3158 3159 3160 3161
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3162 3163 3164 3165 3166 3167

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

P
Paul Turner 已提交
3168 3169 3170 3171 3172 3173 3174
/*
 * 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
 */
3175
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186
{
	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);
}

3187 3188 3189 3190 3191
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3192 3193 3194 3195 3196 3197
/* 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;

3198
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3199 3200
}

3201 3202
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3203 3204 3205
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3206
	u64 amount = 0, min_amount, expires;
3207 3208 3209 3210 3211 3212 3213

	/* 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;
3214
	else {
P
Paul Turner 已提交
3215 3216 3217 3218 3219 3220 3221 3222
		/*
		 * 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);
3223
			__start_cfs_bandwidth(cfs_b, false);
P
Paul Turner 已提交
3224
		}
3225 3226 3227 3228 3229 3230

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3231
	}
P
Paul Turner 已提交
3232
	expires = cfs_b->runtime_expires;
3233 3234 3235
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3236 3237 3238 3239 3240 3241 3242
	/*
	 * 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;
3243 3244

	return cfs_rq->runtime_remaining > 0;
3245 3246
}

P
Paul Turner 已提交
3247 3248 3249 3250 3251
/*
 * 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)
3252
{
P
Paul Turner 已提交
3253 3254 3255
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3256
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3257 3258
		return;

P
Paul Turner 已提交
3259 3260 3261 3262 3263 3264 3265 3266 3267
	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
3268 3269 3270
	 * 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
3271 3272
	 */

3273
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3274 3275 3276 3277 3278 3279 3280 3281
		/* 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;
	}
}

3282
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3283 3284
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3285
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3286 3287 3288
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3289 3290
		return;

3291 3292 3293 3294 3295
	/*
	 * 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))
3296
		resched_curr(rq_of(cfs_rq));
3297 3298
}

3299
static __always_inline
3300
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3301
{
3302
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3303 3304 3305 3306 3307
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3308 3309
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3310
	return cfs_bandwidth_used() && cfs_rq->throttled;
3311 3312
}

3313 3314 3315
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3316
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344
}

/*
 * 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) {
3345
		/* adjust cfs_rq_clock_task() */
3346
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3347
					     cfs_rq->throttled_clock_task;
3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358
	}
#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)];

3359 3360
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3361
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3362 3363 3364 3365 3366
	cfs_rq->throttle_count++;

	return 0;
}

3367
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3368 3369 3370 3371 3372 3373 3374 3375
{
	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))];

3376
	/* freeze hierarchy runnable averages while throttled */
3377 3378 3379
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396

	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)
3397
		sub_nr_running(rq, task_delta);
3398 3399

	cfs_rq->throttled = 1;
3400
	cfs_rq->throttled_clock = rq_clock(rq);
3401
	raw_spin_lock(&cfs_b->lock);
3402 3403 3404 3405 3406
	/*
	 * 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);
3407
	if (!cfs_b->timer_active)
3408
		__start_cfs_bandwidth(cfs_b, false);
3409 3410 3411
	raw_spin_unlock(&cfs_b->lock);
}

3412
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3413 3414 3415 3416 3417 3418 3419
{
	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;

3420
	se = cfs_rq->tg->se[cpu_of(rq)];
3421 3422

	cfs_rq->throttled = 0;
3423 3424 3425

	update_rq_clock(rq);

3426
	raw_spin_lock(&cfs_b->lock);
3427
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
3428 3429 3430
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

3431 3432 3433
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451
	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)
3452
		add_nr_running(rq, task_delta);
3453 3454 3455

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
3456
		resched_curr(rq);
3457 3458 3459 3460 3461 3462
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
3463 3464
	u64 runtime;
	u64 starting_runtime = remaining;
3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494

	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();

3495
	return starting_runtime - remaining;
3496 3497
}

3498 3499 3500 3501 3502 3503 3504 3505
/*
 * 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)
{
3506
	u64 runtime, runtime_expires;
3507
	int throttled;
3508 3509 3510

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
3511
		goto out_deactivate;
3512

3513
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
3514
	cfs_b->nr_periods += overrun;
3515

3516 3517 3518 3519 3520 3521
	/*
	 * 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
3522

3523 3524 3525 3526 3527 3528 3529
	/*
	 * 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
3530 3531
	__refill_cfs_bandwidth_runtime(cfs_b);

3532 3533 3534
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
3535
		return 0;
3536 3537
	}

3538 3539 3540
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

3541 3542 3543
	runtime_expires = cfs_b->runtime_expires;

	/*
3544 3545 3546 3547 3548
	 * 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.
3549
	 */
3550 3551
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
3552 3553 3554 3555 3556 3557 3558
		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);
3559 3560

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3561
	}
3562

3563 3564 3565 3566 3567 3568 3569
	/*
	 * 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;
3570

3571 3572 3573 3574 3575
	return 0;

out_deactivate:
	cfs_b->timer_active = 0;
	return 1;
3576
}
3577

3578 3579 3580 3581 3582 3583 3584
/* 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;

3585 3586 3587 3588 3589 3590 3591
/*
 * 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.
 */
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 3641 3642 3643 3644 3645 3646 3647
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)
{
3648 3649 3650
	if (!cfs_bandwidth_used())
		return;

3651
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666
		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 */
3667 3668 3669
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
3670
		return;
3671
	}
3672

3673
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
3674
		runtime = cfs_b->runtime;
3675

3676 3677 3678 3679 3680 3681 3682 3683 3684 3685
	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)
3686
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3687 3688 3689
	raw_spin_unlock(&cfs_b->lock);
}

3690 3691 3692 3693 3694 3695 3696
/*
 * 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)
{
3697 3698 3699
	if (!cfs_bandwidth_used())
		return;

3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714
	/* 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() */
3715
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3716
{
3717
	if (!cfs_bandwidth_used())
3718
		return false;
3719

3720
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
3721
		return false;
3722 3723 3724 3725 3726 3727

	/*
	 * 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))
3728
		return true;
3729 3730

	throttle_cfs_rq(cfs_rq);
3731
	return true;
3732
}
3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750

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;

3751
	raw_spin_lock(&cfs_b->lock);
3752 3753 3754 3755 3756 3757 3758 3759 3760
	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);
	}
3761
	raw_spin_unlock(&cfs_b->lock);
3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786

	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 */
3787
void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force)
3788 3789 3790 3791 3792 3793 3794
{
	/*
	 * 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
	 */
3795 3796 3797
	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 */
3798
		raw_spin_unlock(&cfs_b->lock);
3799
		cpu_relax();
3800 3801
		raw_spin_lock(&cfs_b->lock);
		/* if someone else restarted the timer then we're done */
3802
		if (!force && cfs_b->timer_active)
3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815
			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);
}

3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828
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);
	}
}

3829
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840
{
	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
		 */
3841
		cfs_rq->runtime_remaining = 1;
3842 3843 3844 3845 3846 3847
		/*
		 * 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;

3848 3849 3850 3851 3852 3853
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
3854 3855
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
3856
	return rq_clock_task(rq_of(cfs_rq));
3857 3858
}

3859
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
3860
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
3861
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
3862
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
3863 3864 3865 3866 3867

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878

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;
}
3879 3880 3881 3882 3883

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) {}
3884 3885
#endif

3886 3887 3888 3889 3890
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) {}
3891
static inline void update_runtime_enabled(struct rq *rq) {}
3892
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
3893 3894 3895

#endif /* CONFIG_CFS_BANDWIDTH */

3896 3897 3898 3899
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
3900 3901 3902 3903 3904 3905 3906 3907
#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);

3908
	if (cfs_rq->nr_running > 1) {
P
Peter Zijlstra 已提交
3909 3910 3911 3912 3913 3914
		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)
3915
				resched_curr(rq);
P
Peter Zijlstra 已提交
3916 3917
			return;
		}
3918
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
3919 3920
	}
}
3921 3922 3923 3924 3925 3926 3927 3928 3929 3930

/*
 * 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;

3931
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
3932 3933 3934 3935 3936
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
3937
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
3938 3939 3940 3941
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
3942 3943 3944 3945

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
3946 3947
#endif

3948 3949 3950 3951 3952
/*
 * 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:
 */
3953
static void
3954
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
3955 3956
{
	struct cfs_rq *cfs_rq;
3957
	struct sched_entity *se = &p->se;
3958 3959

	for_each_sched_entity(se) {
3960
		if (se->on_rq)
3961 3962
			break;
		cfs_rq = cfs_rq_of(se);
3963
		enqueue_entity(cfs_rq, se, flags);
3964 3965 3966 3967 3968 3969 3970 3971 3972

		/*
		 * 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;
3973
		cfs_rq->h_nr_running++;
3974

3975
		flags = ENQUEUE_WAKEUP;
3976
	}
P
Peter Zijlstra 已提交
3977

P
Peter Zijlstra 已提交
3978
	for_each_sched_entity(se) {
3979
		cfs_rq = cfs_rq_of(se);
3980
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
3981

3982 3983 3984
		if (cfs_rq_throttled(cfs_rq))
			break;

3985
		update_cfs_shares(cfs_rq);
3986
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
3987 3988
	}

3989 3990
	if (!se) {
		update_rq_runnable_avg(rq, rq->nr_running);
3991
		add_nr_running(rq, 1);
3992
	}
3993
	hrtick_update(rq);
3994 3995
}

3996 3997
static void set_next_buddy(struct sched_entity *se);

3998 3999 4000 4001 4002
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
4003
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4004 4005
{
	struct cfs_rq *cfs_rq;
4006
	struct sched_entity *se = &p->se;
4007
	int task_sleep = flags & DEQUEUE_SLEEP;
4008 4009 4010

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4011
		dequeue_entity(cfs_rq, se, flags);
4012 4013 4014 4015 4016 4017 4018 4019 4020

		/*
		 * 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;
4021
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4022

4023
		/* Don't dequeue parent if it has other entities besides us */
4024 4025 4026 4027 4028 4029 4030
		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));
4031 4032 4033

			/* avoid re-evaluating load for this entity */
			se = parent_entity(se);
4034
			break;
4035
		}
4036
		flags |= DEQUEUE_SLEEP;
4037
	}
P
Peter Zijlstra 已提交
4038

P
Peter Zijlstra 已提交
4039
	for_each_sched_entity(se) {
4040
		cfs_rq = cfs_rq_of(se);
4041
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4042

4043 4044 4045
		if (cfs_rq_throttled(cfs_rq))
			break;

4046
		update_cfs_shares(cfs_rq);
4047
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
4048 4049
	}

4050
	if (!se) {
4051
		sub_nr_running(rq, 1);
4052 4053
		update_rq_runnable_avg(rq, 1);
	}
4054
	hrtick_update(rq);
4055 4056
}

4057
#ifdef CONFIG_SMP
4058 4059 4060
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
4061
	return cpu_rq(cpu)->cfs.runnable_load_avg;
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 4090 4091 4092 4093 4094 4095 4096
}

/*
 * 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);
}

4097
static unsigned long capacity_of(int cpu)
4098
{
4099
	return cpu_rq(cpu)->cpu_capacity;
4100 4101 4102 4103 4104
}

static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
4105
	unsigned long nr_running = ACCESS_ONCE(rq->cfs.h_nr_running);
4106
	unsigned long load_avg = rq->cfs.runnable_load_avg;
4107 4108

	if (nr_running)
4109
		return load_avg / nr_running;
4110 4111 4112 4113

	return 0;
}

4114 4115 4116 4117 4118 4119 4120
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.
	 */
4121
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
4122
		current->wakee_flips >>= 1;
4123 4124 4125 4126 4127 4128 4129 4130
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}
4131

4132
static void task_waking_fair(struct task_struct *p)
4133 4134 4135
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
4136 4137 4138 4139
	u64 min_vruntime;

#ifndef CONFIG_64BIT
	u64 min_vruntime_copy;
4140

4141 4142 4143 4144 4145 4146 4147 4148
	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
4149

4150
	se->vruntime -= min_vruntime;
4151
	record_wakee(p);
4152 4153
}

4154
#ifdef CONFIG_FAIR_GROUP_SCHED
4155 4156 4157 4158 4159 4160
/*
 * 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.
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 4197 4198 4199 4200 4201 4202 4203
 *
 * 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.
4204
 */
P
Peter Zijlstra 已提交
4205
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
4206
{
P
Peter Zijlstra 已提交
4207
	struct sched_entity *se = tg->se[cpu];
4208

4209
	if (!tg->parent)	/* the trivial, non-cgroup case */
4210 4211
		return wl;

P
Peter Zijlstra 已提交
4212
	for_each_sched_entity(se) {
4213
		long w, W;
P
Peter Zijlstra 已提交
4214

4215
		tg = se->my_q->tg;
4216

4217 4218 4219 4220
		/*
		 * W = @wg + \Sum rw_j
		 */
		W = wg + calc_tg_weight(tg, se->my_q);
P
Peter Zijlstra 已提交
4221

4222 4223 4224 4225
		/*
		 * w = rw_i + @wl
		 */
		w = se->my_q->load.weight + wl;
4226

4227 4228 4229 4230 4231
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
			wl = (w * tg->shares) / W;
4232 4233
		else
			wl = tg->shares;
4234

4235 4236 4237 4238 4239
		/*
		 * 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().
		 */
4240 4241
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
4242 4243 4244 4245

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
4246
		wl -= se->load.weight;
4247 4248 4249 4250 4251 4252 4253 4254

		/*
		 * 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
4255 4256
		wg = 0;
	}
4257

P
Peter Zijlstra 已提交
4258
	return wl;
4259 4260
}
#else
P
Peter Zijlstra 已提交
4261

4262
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
4263
{
4264
	return wl;
4265
}
P
Peter Zijlstra 已提交
4266

4267 4268
#endif

4269 4270
static int wake_wide(struct task_struct *p)
{
4271
	int factor = this_cpu_read(sd_llc_size);
4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290

	/*
	 * 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;
}

4291
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
4292
{
4293
	s64 this_load, load;
4294
	s64 this_eff_load, prev_eff_load;
4295 4296
	int idx, this_cpu, prev_cpu;
	struct task_group *tg;
4297
	unsigned long weight;
4298
	int balanced;
4299

4300 4301 4302 4303 4304 4305 4306
	/*
	 * If we wake multiple tasks be careful to not bounce
	 * ourselves around too much.
	 */
	if (wake_wide(p))
		return 0;

4307 4308 4309 4310 4311
	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);
4312

4313 4314 4315 4316 4317
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
4318 4319 4320 4321
	if (sync) {
		tg = task_group(current);
		weight = current->se.load.weight;

4322
		this_load += effective_load(tg, this_cpu, -weight, -weight);
4323 4324
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
4325

4326 4327
	tg = task_group(p);
	weight = p->se.load.weight;
4328

4329 4330
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
4331 4332 4333
	 * 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.
4334 4335 4336 4337
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
4338 4339
	this_eff_load = 100;
	this_eff_load *= capacity_of(prev_cpu);
4340

4341 4342
	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
4343

4344
	if (this_load > 0) {
4345 4346 4347 4348
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
4349
	}
4350

4351
	balanced = this_eff_load <= prev_eff_load;
4352

4353
	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
4354

4355 4356
	if (!balanced)
		return 0;
4357

4358 4359 4360 4361
	schedstat_inc(sd, ttwu_move_affine);
	schedstat_inc(p, se.statistics.nr_wakeups_affine);

	return 1;
4362 4363
}

4364 4365 4366 4367 4368
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
4369
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
4370
		  int this_cpu, int sd_flag)
4371
{
4372
	struct sched_group *idlest = NULL, *group = sd->groups;
4373
	unsigned long min_load = ULONG_MAX, this_load = 0;
4374
	int load_idx = sd->forkexec_idx;
4375
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
4376

4377 4378 4379
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

4380 4381 4382 4383
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
4384

4385 4386
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
4387
					tsk_cpus_allowed(p)))
4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405
			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;
		}

4406
		/* Adjust by relative CPU capacity of the group */
4407
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428

		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;
4429 4430 4431 4432
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
4433 4434 4435
	int i;

	/* Traverse only the allowed CPUs */
4436
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464
		if (idle_cpu(i)) {
			struct rq *rq = cpu_rq(i);
			struct cpuidle_state *idle = idle_get_state(rq);
			if (idle && idle->exit_latency < min_exit_latency) {
				/*
				 * We give priority to a CPU whose idle state
				 * has the smallest exit latency irrespective
				 * of any idle timestamp.
				 */
				min_exit_latency = idle->exit_latency;
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			} else if ((!idle || idle->exit_latency == min_exit_latency) &&
				   rq->idle_stamp > latest_idle_timestamp) {
				/*
				 * If equal or no active idle state, then
				 * the most recently idled CPU might have
				 * a warmer cache.
				 */
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			}
		} else {
			load = weighted_cpuload(i);
			if (load < min_load || (load == min_load && i == this_cpu)) {
				min_load = load;
				least_loaded_cpu = i;
			}
4465 4466 4467
		}
	}

4468
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
4469
}
4470

4471 4472 4473
/*
 * Try and locate an idle CPU in the sched_domain.
 */
4474
static int select_idle_sibling(struct task_struct *p, int target)
4475
{
4476
	struct sched_domain *sd;
4477
	struct sched_group *sg;
4478
	int i = task_cpu(p);
4479

4480 4481
	if (idle_cpu(target))
		return target;
4482 4483

	/*
4484
	 * If the prevous cpu is cache affine and idle, don't be stupid.
4485
	 */
4486 4487
	if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
		return i;
4488 4489

	/*
4490
	 * Otherwise, iterate the domains and find an elegible idle cpu.
4491
	 */
4492
	sd = rcu_dereference(per_cpu(sd_llc, target));
4493
	for_each_lower_domain(sd) {
4494 4495 4496 4497 4498 4499 4500
		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)) {
4501
				if (i == target || !idle_cpu(i))
4502 4503
					goto next;
			}
4504

4505 4506 4507 4508 4509 4510 4511 4512
			target = cpumask_first_and(sched_group_cpus(sg),
					tsk_cpus_allowed(p));
			goto done;
next:
			sg = sg->next;
		} while (sg != sd->groups);
	}
done:
4513 4514 4515
	return target;
}

4516
/*
4517 4518 4519
 * 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.
4520
 *
4521 4522
 * 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.
4523
 *
4524
 * Returns the target cpu number.
4525 4526 4527
 *
 * preempt must be disabled.
 */
4528
static int
4529
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
4530
{
4531
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
4532 4533
	int cpu = smp_processor_id();
	int new_cpu = cpu;
4534
	int want_affine = 0;
4535
	int sync = wake_flags & WF_SYNC;
4536

4537
	if (p->nr_cpus_allowed == 1)
4538 4539
		return prev_cpu;

4540 4541
	if (sd_flag & SD_BALANCE_WAKE)
		want_affine = cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
4542

4543
	rcu_read_lock();
4544
	for_each_domain(cpu, tmp) {
4545 4546 4547
		if (!(tmp->flags & SD_LOAD_BALANCE))
			continue;

4548
		/*
4549 4550
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
4551
		 */
4552 4553 4554
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
4555
			break;
4556
		}
4557

4558
		if (tmp->flags & sd_flag)
4559 4560 4561
			sd = tmp;
	}

4562 4563
	if (affine_sd && cpu != prev_cpu && wake_affine(affine_sd, p, sync))
		prev_cpu = cpu;
4564

4565
	if (sd_flag & SD_BALANCE_WAKE) {
4566 4567
		new_cpu = select_idle_sibling(p, prev_cpu);
		goto unlock;
4568
	}
4569

4570 4571
	while (sd) {
		struct sched_group *group;
4572
		int weight;
4573

4574
		if (!(sd->flags & sd_flag)) {
4575 4576 4577
			sd = sd->child;
			continue;
		}
4578

4579
		group = find_idlest_group(sd, p, cpu, sd_flag);
4580 4581 4582 4583
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
4584

4585
		new_cpu = find_idlest_cpu(group, p, cpu);
4586 4587 4588 4589
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
4590
		}
4591 4592 4593

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
4594
		weight = sd->span_weight;
4595 4596
		sd = NULL;
		for_each_domain(cpu, tmp) {
4597
			if (weight <= tmp->span_weight)
4598
				break;
4599
			if (tmp->flags & sd_flag)
4600 4601 4602
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
4603
	}
4604 4605
unlock:
	rcu_read_unlock();
4606

4607
	return new_cpu;
4608
}
4609 4610 4611 4612 4613 4614 4615 4616 4617 4618

/*
 * 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)
{
4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629
	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);
4630 4631
		atomic_long_add(se->avg.load_avg_contrib,
						&cfs_rq->removed_load);
4632
	}
4633 4634 4635

	/* We have migrated, no longer consider this task hot */
	se->exec_start = 0;
4636
}
4637 4638
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
4639 4640
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
4641 4642 4643 4644
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
4645 4646
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
4647 4648 4649 4650 4651 4652 4653 4654 4655
	 *
	 * 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.
4656
	 */
4657
	return calc_delta_fair(gran, se);
4658 4659
}

4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681
/*
 * 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
4682
	gran = wakeup_gran(curr, se);
4683 4684 4685 4686 4687 4688
	if (vdiff > gran)
		return 1;

	return 0;
}

4689 4690
static void set_last_buddy(struct sched_entity *se)
{
4691 4692 4693 4694 4695
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
4696 4697 4698 4699
}

static void set_next_buddy(struct sched_entity *se)
{
4700 4701 4702 4703 4704
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
4705 4706
}

4707 4708
static void set_skip_buddy(struct sched_entity *se)
{
4709 4710
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
4711 4712
}

4713 4714 4715
/*
 * Preempt the current task with a newly woken task if needed:
 */
4716
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
4717 4718
{
	struct task_struct *curr = rq->curr;
4719
	struct sched_entity *se = &curr->se, *pse = &p->se;
4720
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
4721
	int scale = cfs_rq->nr_running >= sched_nr_latency;
4722
	int next_buddy_marked = 0;
4723

I
Ingo Molnar 已提交
4724 4725 4726
	if (unlikely(se == pse))
		return;

4727
	/*
4728
	 * This is possible from callers such as attach_tasks(), in which we
4729 4730 4731 4732 4733 4734 4735
	 * 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;

4736
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
4737
		set_next_buddy(pse);
4738 4739
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
4740

4741 4742 4743
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
4744 4745 4746 4747 4748 4749
	 *
	 * 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.
4750 4751 4752 4753
	 */
	if (test_tsk_need_resched(curr))
		return;

4754 4755 4756 4757 4758
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

4759
	/*
4760 4761
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
4762
	 */
4763
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
4764
		return;
4765

4766
	find_matching_se(&se, &pse);
4767
	update_curr(cfs_rq_of(se));
4768
	BUG_ON(!pse);
4769 4770 4771 4772 4773 4774 4775
	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);
4776
		goto preempt;
4777
	}
4778

4779
	return;
4780

4781
preempt:
4782
	resched_curr(rq);
4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796
	/*
	 * 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);
4797 4798
}

4799 4800
static struct task_struct *
pick_next_task_fair(struct rq *rq, struct task_struct *prev)
4801 4802 4803
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
4804
	struct task_struct *p;
4805
	int new_tasks;
4806

4807
again:
4808 4809
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
4810
		goto idle;
4811

4812
	if (prev->sched_class != &fair_sched_class)
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 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883
		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
4884

4885
	if (!cfs_rq->nr_running)
4886
		goto idle;
4887

4888
	put_prev_task(rq, prev);
4889

4890
	do {
4891
		se = pick_next_entity(cfs_rq, NULL);
4892
		set_next_entity(cfs_rq, se);
4893 4894 4895
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
4896
	p = task_of(se);
4897

4898 4899
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
4900 4901

	return p;
4902 4903

idle:
4904
	new_tasks = idle_balance(rq);
4905 4906 4907 4908 4909
	/*
	 * 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.
	 */
4910
	if (new_tasks < 0)
4911 4912
		return RETRY_TASK;

4913
	if (new_tasks > 0)
4914 4915 4916
		goto again;

	return NULL;
4917 4918 4919 4920 4921
}

/*
 * Account for a descheduled task:
 */
4922
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
4923 4924 4925 4926 4927 4928
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4929
		put_prev_entity(cfs_rq, se);
4930 4931 4932
	}
}

4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957
/*
 * 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);
4958 4959 4960 4961 4962 4963
		/*
		 * 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;
4964 4965 4966 4967 4968
	}

	set_skip_buddy(se);
}

4969 4970 4971 4972
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

4973 4974
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
4975 4976 4977 4978 4979 4980 4981 4982 4983 4984
		return false;

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

	yield_task_fair(rq);

	return true;
}

4985
#ifdef CONFIG_SMP
4986
/**************************************************
P
Peter Zijlstra 已提交
4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009
 * 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)
 *
5010
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
5011 5012 5013 5014 5015 5016
 * 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):
 *
5017
 *   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
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 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102
 *
 * 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.]
 */ 
5103

5104 5105
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

5106 5107
enum fbq_type { regular, remote, all };

5108
#define LBF_ALL_PINNED	0x01
5109
#define LBF_NEED_BREAK	0x02
5110 5111
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
5112 5113 5114 5115 5116

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
5117
	int			src_cpu;
5118 5119 5120 5121

	int			dst_cpu;
	struct rq		*dst_rq;

5122 5123
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
5124
	enum cpu_idle_type	idle;
5125
	long			imbalance;
5126 5127 5128
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

5129
	unsigned int		flags;
5130 5131 5132 5133

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
5134 5135

	enum fbq_type		fbq_type;
5136
	struct list_head	tasks;
5137 5138
};

5139 5140 5141
/*
 * Is this task likely cache-hot:
 */
5142
static int task_hot(struct task_struct *p, struct lb_env *env)
5143 5144 5145
{
	s64 delta;

5146 5147
	lockdep_assert_held(&env->src_rq->lock);

5148 5149 5150 5151 5152 5153 5154 5155 5156
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
5157
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
5158 5159 5160 5161 5162 5163 5164 5165 5166
			(&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;

5167
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
5168 5169 5170 5171

	return delta < (s64)sysctl_sched_migration_cost;
}

5172 5173 5174 5175
#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)
{
5176
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5177 5178
	int src_nid, dst_nid;

5179
	if (!sched_feat(NUMA_FAVOUR_HIGHER) || !p->numa_faults_memory ||
5180 5181 5182 5183 5184 5185 5186
	    !(env->sd->flags & SD_NUMA)) {
		return false;
	}

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

5187
	if (src_nid == dst_nid)
5188 5189
		return false;

5190 5191 5192 5193
	if (numa_group) {
		/* Task is already in the group's interleave set. */
		if (node_isset(src_nid, numa_group->active_nodes))
			return false;
5194

5195 5196 5197
		/* Task is moving into the group's interleave set. */
		if (node_isset(dst_nid, numa_group->active_nodes))
			return true;
5198

5199 5200 5201 5202 5203
		return group_faults(p, dst_nid) > group_faults(p, src_nid);
	}

	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
5204 5205
		return true;

5206
	return task_faults(p, dst_nid) > task_faults(p, src_nid);
5207
}
5208 5209 5210 5211


static bool migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
{
5212
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5213 5214 5215 5216 5217
	int src_nid, dst_nid;

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

5218
	if (!p->numa_faults_memory || !(env->sd->flags & SD_NUMA))
5219 5220 5221 5222 5223
		return false;

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

5224
	if (src_nid == dst_nid)
5225 5226
		return false;

5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238
	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);
	}

5239 5240 5241 5242
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid)
		return true;

5243
	return task_faults(p, dst_nid) < task_faults(p, src_nid);
5244 5245
}

5246 5247 5248 5249 5250 5251
#else
static inline bool migrate_improves_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5252 5253 5254 5255 5256 5257

static inline bool migrate_degrades_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5258 5259
#endif

5260 5261 5262 5263
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
5264
int can_migrate_task(struct task_struct *p, struct lb_env *env)
5265 5266
{
	int tsk_cache_hot = 0;
5267 5268 5269

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

5270 5271
	/*
	 * We do not migrate tasks that are:
5272
	 * 1) throttled_lb_pair, or
5273
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
5274 5275
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
5276
	 */
5277 5278 5279
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

5280
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
5281
		int cpu;
5282

5283
		schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
5284

5285 5286
		env->flags |= LBF_SOME_PINNED;

5287 5288 5289 5290 5291 5292 5293 5294
		/*
		 * 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.
		 */
5295
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
5296 5297
			return 0;

5298 5299 5300
		/* 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))) {
5301
				env->flags |= LBF_DST_PINNED;
5302 5303 5304
				env->new_dst_cpu = cpu;
				break;
			}
5305
		}
5306

5307 5308
		return 0;
	}
5309 5310

	/* Record that we found atleast one task that could run on dst_cpu */
5311
	env->flags &= ~LBF_ALL_PINNED;
5312

5313
	if (task_running(env->src_rq, p)) {
5314
		schedstat_inc(p, se.statistics.nr_failed_migrations_running);
5315 5316 5317 5318 5319
		return 0;
	}

	/*
	 * Aggressive migration if:
5320 5321 5322
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
5323
	 */
5324
	tsk_cache_hot = task_hot(p, env);
5325 5326
	if (!tsk_cache_hot)
		tsk_cache_hot = migrate_degrades_locality(p, env);
5327

5328 5329
	if (migrate_improves_locality(p, env) || !tsk_cache_hot ||
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
5330 5331 5332 5333
		if (tsk_cache_hot) {
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
			schedstat_inc(p, se.statistics.nr_forced_migrations);
		}
5334 5335 5336
		return 1;
	}

Z
Zhang Hang 已提交
5337 5338
	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
	return 0;
5339 5340
}

5341
/*
5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352
 * 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);
}

5353
/*
5354
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
5355 5356
 * part of active balancing operations within "domain".
 *
5357
 * Returns a task if successful and NULL otherwise.
5358
 */
5359
static struct task_struct *detach_one_task(struct lb_env *env)
5360 5361 5362
{
	struct task_struct *p, *n;

5363 5364
	lockdep_assert_held(&env->src_rq->lock);

5365 5366 5367
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
5368

5369
		detach_task(p, env);
5370

5371
		/*
5372
		 * Right now, this is only the second place where
5373
		 * lb_gained[env->idle] is updated (other is detach_tasks)
5374
		 * so we can safely collect stats here rather than
5375
		 * inside detach_tasks().
5376 5377
		 */
		schedstat_inc(env->sd, lb_gained[env->idle]);
5378
		return p;
5379
	}
5380
	return NULL;
5381 5382
}

5383 5384
static const unsigned int sched_nr_migrate_break = 32;

5385
/*
5386 5387
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
5388
 *
5389
 * Returns number of detached tasks if successful and 0 otherwise.
5390
 */
5391
static int detach_tasks(struct lb_env *env)
5392
{
5393 5394
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
5395
	unsigned long load;
5396 5397 5398
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
5399

5400
	if (env->imbalance <= 0)
5401
		return 0;
5402

5403 5404
	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
5405

5406 5407
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
5408
		if (env->loop > env->loop_max)
5409
			break;
5410 5411

		/* take a breather every nr_migrate tasks */
5412
		if (env->loop > env->loop_break) {
5413
			env->loop_break += sched_nr_migrate_break;
5414
			env->flags |= LBF_NEED_BREAK;
5415
			break;
5416
		}
5417

5418
		if (!can_migrate_task(p, env))
5419 5420 5421
			goto next;

		load = task_h_load(p);
5422

5423
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
5424 5425
			goto next;

5426
		if ((load / 2) > env->imbalance)
5427
			goto next;
5428

5429 5430 5431 5432
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
5433
		env->imbalance -= load;
5434 5435

#ifdef CONFIG_PREEMPT
5436 5437
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
5438
		 * kernels will stop after the first task is detached to minimize
5439 5440
		 * the critical section.
		 */
5441
		if (env->idle == CPU_NEWLY_IDLE)
5442
			break;
5443 5444
#endif

5445 5446 5447 5448
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
5449
		if (env->imbalance <= 0)
5450
			break;
5451 5452 5453

		continue;
next:
5454
		list_move_tail(&p->se.group_node, tasks);
5455
	}
5456

5457
	/*
5458 5459 5460
	 * 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().
5461
	 */
5462
	schedstat_add(env->sd, lb_gained[env->idle], detached);
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
	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);
5505

5506 5507 5508 5509
		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
5510 5511
}

P
Peter Zijlstra 已提交
5512
#ifdef CONFIG_FAIR_GROUP_SCHED
5513 5514 5515
/*
 * update tg->load_weight by folding this cpu's load_avg
 */
5516
static void __update_blocked_averages_cpu(struct task_group *tg, int cpu)
5517
{
5518 5519
	struct sched_entity *se = tg->se[cpu];
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
5520

5521 5522 5523
	/* throttled entities do not contribute to load */
	if (throttled_hierarchy(cfs_rq))
		return;
5524

5525
	update_cfs_rq_blocked_load(cfs_rq, 1);
5526

5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540
	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 {
5541
		struct rq *rq = rq_of(cfs_rq);
5542 5543
		update_rq_runnable_avg(rq, rq->nr_running);
	}
5544 5545
}

5546
static void update_blocked_averages(int cpu)
5547 5548
{
	struct rq *rq = cpu_rq(cpu);
5549 5550
	struct cfs_rq *cfs_rq;
	unsigned long flags;
5551

5552 5553
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
5554 5555 5556 5557
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
5558
	for_each_leaf_cfs_rq(rq, cfs_rq) {
5559 5560 5561 5562 5563 5564
		/*
		 * 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);
5565
	}
5566 5567

	raw_spin_unlock_irqrestore(&rq->lock, flags);
5568 5569
}

5570
/*
5571
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
5572 5573 5574
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
5575
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
5576
{
5577 5578
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
5579
	unsigned long now = jiffies;
5580
	unsigned long load;
5581

5582
	if (cfs_rq->last_h_load_update == now)
5583 5584
		return;

5585 5586 5587 5588 5589 5590 5591
	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;
	}
5592

5593
	if (!se) {
5594
		cfs_rq->h_load = cfs_rq->runnable_load_avg;
5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605
		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;
	}
5606 5607
}

5608
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
5609
{
5610
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
5611

5612
	update_cfs_rq_h_load(cfs_rq);
5613 5614
	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
5615 5616
}
#else
5617
static inline void update_blocked_averages(int cpu)
5618 5619 5620
{
}

5621
static unsigned long task_h_load(struct task_struct *p)
5622
{
5623
	return p->se.avg.load_avg_contrib;
5624
}
P
Peter Zijlstra 已提交
5625
#endif
5626 5627

/********** Helpers for find_busiest_group ************************/
5628 5629 5630 5631 5632 5633 5634

enum group_type {
	group_other = 0,
	group_imbalanced,
	group_overloaded,
};

5635 5636 5637 5638 5639 5640 5641
/*
 * 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
5642
	unsigned long load_per_task;
5643
	unsigned long group_capacity;
5644
	unsigned int sum_nr_running; /* Nr tasks running in the group */
5645
	unsigned int group_capacity_factor;
5646 5647
	unsigned int idle_cpus;
	unsigned int group_weight;
5648
	enum group_type group_type;
5649
	int group_has_free_capacity;
5650 5651 5652 5653
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
5654 5655
};

J
Joonsoo Kim 已提交
5656 5657 5658 5659 5660 5661 5662 5663
/*
 * 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 */
5664
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
5665 5666 5667
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
5668
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
5669 5670
};

5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682
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,
5683
		.total_capacity = 0UL,
5684 5685
		.busiest_stat = {
			.avg_load = 0UL,
5686 5687
			.sum_nr_running = 0,
			.group_type = group_other,
5688 5689 5690 5691
		},
	};
}

5692 5693 5694
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
5695
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
5696 5697
 *
 * Return: The load index.
5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719
 */
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;
}

5720
static unsigned long default_scale_capacity(struct sched_domain *sd, int cpu)
5721
{
5722
	return SCHED_CAPACITY_SCALE;
5723 5724
}

5725
unsigned long __weak arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
5726
{
5727
	return default_scale_capacity(sd, cpu);
5728 5729
}

5730
static unsigned long default_scale_cpu_capacity(struct sched_domain *sd, int cpu)
5731
{
5732 5733
	if ((sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
		return sd->smt_gain / sd->span_weight;
5734

5735
	return SCHED_CAPACITY_SCALE;
5736 5737
}

5738
unsigned long __weak arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
5739
{
5740
	return default_scale_cpu_capacity(sd, cpu);
5741 5742
}

5743
static unsigned long scale_rt_capacity(int cpu)
5744 5745
{
	struct rq *rq = cpu_rq(cpu);
5746
	u64 total, available, age_stamp, avg;
5747
	s64 delta;
5748

5749 5750 5751 5752 5753 5754 5755
	/*
	 * 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);

5756 5757 5758 5759 5760
	delta = rq_clock(rq) - age_stamp;
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
5761

5762
	if (unlikely(total < avg)) {
5763
		/* Ensures that capacity won't end up being negative */
5764 5765
		available = 0;
	} else {
5766
		available = total - avg;
5767
	}
5768

5769 5770
	if (unlikely((s64)total < SCHED_CAPACITY_SCALE))
		total = SCHED_CAPACITY_SCALE;
5771

5772
	total >>= SCHED_CAPACITY_SHIFT;
5773 5774 5775 5776

	return div_u64(available, total);
}

5777
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
5778
{
5779
	unsigned long capacity = SCHED_CAPACITY_SCALE;
5780 5781
	struct sched_group *sdg = sd->groups;

5782 5783 5784 5785
	if (sched_feat(ARCH_CAPACITY))
		capacity *= arch_scale_cpu_capacity(sd, cpu);
	else
		capacity *= default_scale_cpu_capacity(sd, cpu);
5786

5787
	capacity >>= SCHED_CAPACITY_SHIFT;
5788

5789
	sdg->sgc->capacity_orig = capacity;
5790

5791
	if (sched_feat(ARCH_CAPACITY))
5792
		capacity *= arch_scale_freq_capacity(sd, cpu);
5793
	else
5794
		capacity *= default_scale_capacity(sd, cpu);
5795

5796
	capacity >>= SCHED_CAPACITY_SHIFT;
5797

5798
	capacity *= scale_rt_capacity(cpu);
5799
	capacity >>= SCHED_CAPACITY_SHIFT;
5800

5801 5802
	if (!capacity)
		capacity = 1;
5803

5804 5805
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
5806 5807
}

5808
void update_group_capacity(struct sched_domain *sd, int cpu)
5809 5810 5811
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
5812
	unsigned long capacity, capacity_orig;
5813 5814 5815 5816
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
5817
	sdg->sgc->next_update = jiffies + interval;
5818 5819

	if (!child) {
5820
		update_cpu_capacity(sd, cpu);
5821 5822 5823
		return;
	}

5824
	capacity_orig = capacity = 0;
5825

P
Peter Zijlstra 已提交
5826 5827 5828 5829 5830 5831
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

5832
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
5833
			struct sched_group_capacity *sgc;
5834
			struct rq *rq = cpu_rq(cpu);
5835

5836
			/*
5837
			 * build_sched_domains() -> init_sched_groups_capacity()
5838 5839 5840
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
5841 5842
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
5843
			 *
5844
			 * This avoids capacity/capacity_orig from being 0 and
5845 5846
			 * causing divide-by-zero issues on boot.
			 *
5847
			 * Runtime updates will correct capacity_orig.
5848 5849
			 */
			if (unlikely(!rq->sd)) {
5850 5851
				capacity_orig += capacity_of(cpu);
				capacity += capacity_of(cpu);
5852 5853
				continue;
			}
5854

5855 5856 5857
			sgc = rq->sd->groups->sgc;
			capacity_orig += sgc->capacity_orig;
			capacity += sgc->capacity;
5858
		}
P
Peter Zijlstra 已提交
5859 5860 5861 5862 5863 5864 5865 5866
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
		 */ 

		group = child->groups;
		do {
5867 5868
			capacity_orig += group->sgc->capacity_orig;
			capacity += group->sgc->capacity;
P
Peter Zijlstra 已提交
5869 5870 5871
			group = group->next;
		} while (group != child->groups);
	}
5872

5873 5874
	sdg->sgc->capacity_orig = capacity_orig;
	sdg->sgc->capacity = capacity;
5875 5876
}

5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887
/*
 * 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)
{
	/*
5888
	 * Only siblings can have significantly less than SCHED_CAPACITY_SCALE
5889
	 */
5890
	if (!(sd->flags & SD_SHARE_CPUCAPACITY))
5891 5892 5893
		return 0;

	/*
5894
	 * If ~90% of the cpu_capacity is still there, we're good.
5895
	 */
5896
	if (group->sgc->capacity * 32 > group->sgc->capacity_orig * 29)
5897 5898 5899 5900 5901
		return 1;

	return 0;
}

5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917
/*
 * 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
5918 5919
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
5920 5921
 *
 * When this is so detected; this group becomes a candidate for busiest; see
5922
 * update_sd_pick_busiest(). And calculate_imbalance() and
5923
 * find_busiest_group() avoid some of the usual balance conditions to allow it
5924 5925 5926 5927 5928 5929 5930
 * 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.
 */

5931
static inline int sg_imbalanced(struct sched_group *group)
5932
{
5933
	return group->sgc->imbalance;
5934 5935
}

5936
/*
5937
 * Compute the group capacity factor.
5938
 *
5939
 * Avoid the issue where N*frac(smt_capacity) >= 1 creates 'phantom' cores by
5940
 * first dividing out the smt factor and computing the actual number of cores
5941
 * and limit unit capacity with that.
5942
 */
5943
static inline int sg_capacity_factor(struct lb_env *env, struct sched_group *group)
5944
{
5945
	unsigned int capacity_factor, smt, cpus;
5946
	unsigned int capacity, capacity_orig;
5947

5948 5949
	capacity = group->sgc->capacity;
	capacity_orig = group->sgc->capacity_orig;
5950
	cpus = group->group_weight;
5951

5952
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_capacity < 2 */
5953
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, capacity_orig);
5954
	capacity_factor = cpus / smt; /* cores */
5955

5956
	capacity_factor = min_t(unsigned,
5957
		capacity_factor, DIV_ROUND_CLOSEST(capacity, SCHED_CAPACITY_SCALE));
5958 5959
	if (!capacity_factor)
		capacity_factor = fix_small_capacity(env->sd, group);
5960

5961
	return capacity_factor;
5962 5963
}

5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975
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;
}

5976 5977
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
5978
 * @env: The load balancing environment.
5979 5980 5981 5982
 * @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.
5983
 * @overload: Indicate more than one runnable task for any CPU.
5984
 */
5985 5986
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
5987 5988
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
5989
{
5990
	unsigned long load;
5991
	int i;
5992

5993 5994
	memset(sgs, 0, sizeof(*sgs));

5995
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
5996 5997 5998
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
5999
		if (local_group)
6000
			load = target_load(i, load_idx);
6001
		else
6002 6003 6004
			load = source_load(i, load_idx);

		sgs->group_load += load;
6005
		sgs->sum_nr_running += rq->cfs.h_nr_running;
6006 6007 6008 6009

		if (rq->nr_running > 1)
			*overload = true;

6010 6011 6012 6013
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
6014
		sgs->sum_weighted_load += weighted_cpuload(i);
6015 6016
		if (idle_cpu(i))
			sgs->idle_cpus++;
6017 6018
	}

6019 6020
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
6021
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
6022

6023
	if (sgs->sum_nr_running)
6024
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
6025

6026
	sgs->group_weight = group->group_weight;
6027
	sgs->group_capacity_factor = sg_capacity_factor(env, group);
6028
	sgs->group_type = group_classify(group, sgs);
6029

6030
	if (sgs->group_capacity_factor > sgs->sum_nr_running)
6031
		sgs->group_has_free_capacity = 1;
6032 6033
}

6034 6035
/**
 * update_sd_pick_busiest - return 1 on busiest group
6036
 * @env: The load balancing environment.
6037 6038
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
6039
 * @sgs: sched_group statistics
6040 6041 6042
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
6043 6044 6045
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
6046
 */
6047
static bool update_sd_pick_busiest(struct lb_env *env,
6048 6049
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
6050
				   struct sg_lb_stats *sgs)
6051
{
6052
	struct sg_lb_stats *busiest = &sds->busiest_stat;
6053

6054
	if (sgs->group_type > busiest->group_type)
6055 6056
		return true;

6057 6058 6059 6060 6061 6062 6063 6064
	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))
6065 6066 6067 6068 6069 6070 6071
		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.
	 */
6072
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
6073 6074 6075 6076 6077 6078 6079 6080 6081 6082
		if (!sds->busiest)
			return true;

		if (group_first_cpu(sds->busiest) > group_first_cpu(sg))
			return true;
	}

	return false;
}

6083 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
#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 */

6113
/**
6114
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
6115
 * @env: The load balancing environment.
6116 6117
 * @sds: variable to hold the statistics for this sched_domain.
 */
6118
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
6119
{
6120 6121
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
6122
	struct sg_lb_stats tmp_sgs;
6123
	int load_idx, prefer_sibling = 0;
6124
	bool overload = false;
6125 6126 6127 6128

	if (child && child->flags & SD_PREFER_SIBLING)
		prefer_sibling = 1;

6129
	load_idx = get_sd_load_idx(env->sd, env->idle);
6130 6131

	do {
J
Joonsoo Kim 已提交
6132
		struct sg_lb_stats *sgs = &tmp_sgs;
6133 6134
		int local_group;

6135
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
6136 6137 6138
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
6139 6140

			if (env->idle != CPU_NEWLY_IDLE ||
6141 6142
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
6143
		}
6144

6145 6146
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
6147

6148 6149 6150
		if (local_group)
			goto next_group;

6151 6152
		/*
		 * In case the child domain prefers tasks go to siblings
6153
		 * first, lower the sg capacity factor to one so that we'll try
6154 6155
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
6156
		 * these excess tasks, i.e. nr_running < group_capacity_factor. The
6157 6158 6159
		 * 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).
6160
		 */
6161
		if (prefer_sibling && sds->local &&
6162
		    sds->local_stat.group_has_free_capacity)
6163
			sgs->group_capacity_factor = min(sgs->group_capacity_factor, 1U);
6164

6165
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
6166
			sds->busiest = sg;
J
Joonsoo Kim 已提交
6167
			sds->busiest_stat = *sgs;
6168 6169
		}

6170 6171 6172
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
6173
		sds->total_capacity += sgs->group_capacity;
6174

6175
		sg = sg->next;
6176
	} while (sg != env->sd->groups);
6177 6178 6179

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
6180 6181 6182 6183 6184 6185 6186

	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;
	}

6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205
}

/**
 * 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.
 *
6206
 * Return: 1 when packing is required and a task should be moved to
6207 6208
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
6209
 * @env: The load balancing environment.
6210 6211
 * @sds: Statistics of the sched_domain which is to be packed
 */
6212
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
6213 6214 6215
{
	int busiest_cpu;

6216
	if (!(env->sd->flags & SD_ASYM_PACKING))
6217 6218 6219 6220 6221 6222
		return 0;

	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
6223
	if (env->dst_cpu > busiest_cpu)
6224 6225
		return 0;

6226
	env->imbalance = DIV_ROUND_CLOSEST(
6227
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
6228
		SCHED_CAPACITY_SCALE);
6229

6230
	return 1;
6231 6232 6233 6234 6235 6236
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
6237
 * @env: The load balancing environment.
6238 6239
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
6240 6241
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6242
{
6243
	unsigned long tmp, capa_now = 0, capa_move = 0;
6244
	unsigned int imbn = 2;
6245
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
6246
	struct sg_lb_stats *local, *busiest;
6247

J
Joonsoo Kim 已提交
6248 6249
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6250

J
Joonsoo Kim 已提交
6251 6252 6253 6254
	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;
6255

J
Joonsoo Kim 已提交
6256
	scaled_busy_load_per_task =
6257
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6258
		busiest->group_capacity;
J
Joonsoo Kim 已提交
6259

6260 6261
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
6262
		env->imbalance = busiest->load_per_task;
6263 6264 6265 6266 6267
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
6268
	 * however we may be able to increase total CPU capacity used by
6269 6270 6271
	 * moving them.
	 */

6272
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
6273
			min(busiest->load_per_task, busiest->avg_load);
6274
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
6275
			min(local->load_per_task, local->avg_load);
6276
	capa_now /= SCHED_CAPACITY_SCALE;
6277 6278

	/* Amount of load we'd subtract */
6279
	if (busiest->avg_load > scaled_busy_load_per_task) {
6280
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
6281
			    min(busiest->load_per_task,
6282
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
6283
	}
6284 6285

	/* Amount of load we'd add */
6286
	if (busiest->avg_load * busiest->group_capacity <
6287
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
6288 6289
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
6290
	} else {
6291
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6292
		      local->group_capacity;
J
Joonsoo Kim 已提交
6293
	}
6294
	capa_move += local->group_capacity *
6295
		    min(local->load_per_task, local->avg_load + tmp);
6296
	capa_move /= SCHED_CAPACITY_SCALE;
6297 6298

	/* Move if we gain throughput */
6299
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
6300
		env->imbalance = busiest->load_per_task;
6301 6302 6303 6304 6305
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
6306
 * @env: load balance environment
6307 6308
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
6309
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6310
{
6311
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
6312 6313 6314 6315
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6316

6317
	if (busiest->group_type == group_imbalanced) {
6318 6319 6320 6321
		/*
		 * 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
6322 6323
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
6324 6325
	}

6326 6327 6328
	/*
	 * 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
6329
	 * its cpu_capacity, while calculating max_load..)
6330
	 */
6331 6332
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
6333 6334
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
6335 6336
	}

6337 6338 6339 6340 6341
	/*
	 * 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
6342
		load_above_capacity =
6343
			(busiest->sum_nr_running - busiest->group_capacity_factor);
6344

6345
		load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_CAPACITY_SCALE);
6346
		load_above_capacity /= busiest->group_capacity;
6347 6348 6349 6350 6351 6352 6353 6354 6355 6356
	}

	/*
	 * 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.
	 */
6357
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
6358 6359

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
6360
	env->imbalance = min(
6361 6362
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
6363
	) / SCHED_CAPACITY_SCALE;
6364 6365 6366

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
6367
	 * there is no guarantee that any tasks will be moved so we'll have
6368 6369 6370
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
6371
	if (env->imbalance < busiest->load_per_task)
6372
		return fix_small_imbalance(env, sds);
6373
}
6374

6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386
/******* 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.
 *
6387
 * @env: The load balancing environment.
6388
 *
6389
 * Return:	- The busiest group if imbalance exists.
6390 6391 6392 6393
 *		- 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
6394
static struct sched_group *find_busiest_group(struct lb_env *env)
6395
{
J
Joonsoo Kim 已提交
6396
	struct sg_lb_stats *local, *busiest;
6397 6398
	struct sd_lb_stats sds;

6399
	init_sd_lb_stats(&sds);
6400 6401 6402 6403 6404

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
6405
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
6406 6407
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
6408

6409 6410
	if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
	    check_asym_packing(env, &sds))
6411 6412
		return sds.busiest;

6413
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
6414
	if (!sds.busiest || busiest->sum_nr_running == 0)
6415 6416
		goto out_balanced;

6417 6418
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
6419

P
Peter Zijlstra 已提交
6420 6421
	/*
	 * If the busiest group is imbalanced the below checks don't
6422
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
6423 6424
	 * isn't true due to cpus_allowed constraints and the like.
	 */
6425
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
6426 6427
		goto force_balance;

6428
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
6429 6430
	if (env->idle == CPU_NEWLY_IDLE && local->group_has_free_capacity &&
	    !busiest->group_has_free_capacity)
6431 6432
		goto force_balance;

6433
	/*
6434
	 * If the local group is busier than the selected busiest group
6435 6436
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
6437
	if (local->avg_load >= busiest->avg_load)
6438 6439
		goto out_balanced;

6440 6441 6442 6443
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
6444
	if (local->avg_load >= sds.avg_load)
6445 6446
		goto out_balanced;

6447
	if (env->idle == CPU_IDLE) {
6448
		/*
6449 6450 6451 6452 6453
		 * This cpu is idle. If the busiest group is not overloaded
		 * and there is no imbalance between this and busiest group
		 * wrt idle cpus, it is balanced. The imbalance becomes
		 * significant if the diff is greater than 1 otherwise we
		 * might end up to just move the imbalance on another group
6454
		 */
6455 6456
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
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 = this_cpu_cpumask_var_ptr(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
		cur_ld_moved = detach_tasks(&env);
6696 6697

		/*
6698 6699 6700 6701 6702
		 * 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.
6703
		 */
6704 6705 6706 6707 6708 6709 6710 6711

		raw_spin_unlock(&busiest->lock);

		if (cur_ld_moved) {
			attach_tasks(&env);
			ld_moved += cur_ld_moved;
		}

6712
		local_irq_restore(flags);
6713

6714 6715 6716 6717 6718
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737
		/*
		 * 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.
		 */
6738
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
6739

6740 6741 6742
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

6743
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
6744
			env.dst_cpu	 = env.new_dst_cpu;
6745
			env.flags	&= ~LBF_DST_PINNED;
6746 6747
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
6748

6749 6750 6751 6752 6753 6754
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
6755

6756 6757 6758 6759
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
6760
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
6761

6762
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
6763 6764 6765
				*group_imbalance = 1;
		}

6766
		/* All tasks on this runqueue were pinned by CPU affinity */
6767
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
6768
			cpumask_clear_cpu(cpu_of(busiest), cpus);
6769 6770 6771
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
6772
				goto redo;
6773
			}
6774
			goto out_all_pinned;
6775 6776 6777 6778 6779
		}
	}

	if (!ld_moved) {
		schedstat_inc(sd, lb_failed[idle]);
6780 6781 6782 6783 6784 6785 6786 6787
		/*
		 * 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++;
6788

6789
		if (need_active_balance(&env)) {
6790 6791
			raw_spin_lock_irqsave(&busiest->lock, flags);

6792 6793 6794
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
6795 6796
			 */
			if (!cpumask_test_cpu(this_cpu,
6797
					tsk_cpus_allowed(busiest->curr))) {
6798 6799
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
6800
				env.flags |= LBF_ALL_PINNED;
6801 6802 6803
				goto out_one_pinned;
			}

6804 6805 6806 6807 6808
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
6809 6810 6811 6812 6813 6814
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
6815

6816
			if (active_balance) {
6817 6818 6819
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
6820
			}
6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838

			/*
			 * 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
6839
		 * detach_tasks).
6840 6841 6842 6843 6844 6845 6846 6847
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864
	/*
	 * 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.
	 */
6865 6866 6867 6868 6869 6870
	schedstat_inc(sd, lb_balanced[idle]);

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
6871
	if (((env.flags & LBF_ALL_PINNED) &&
6872
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
6873 6874 6875
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

6876
	ld_moved = 0;
6877 6878 6879 6880
out:
	return ld_moved;
}

6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907
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;
}

6908 6909 6910 6911
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
6912
static int idle_balance(struct rq *this_rq)
6913
{
6914 6915
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
6916 6917
	struct sched_domain *sd;
	int pulled_task = 0;
6918
	u64 curr_cost = 0;
6919

6920
	idle_enter_fair(this_rq);
6921

6922 6923 6924 6925 6926 6927
	/*
	 * 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);

6928 6929
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
6930 6931 6932 6933 6934 6935
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, 0, &next_balance);
		rcu_read_unlock();

6936
		goto out;
6937
	}
6938

6939 6940 6941 6942 6943
	/*
	 * Drop the rq->lock, but keep IRQ/preempt disabled.
	 */
	raw_spin_unlock(&this_rq->lock);

6944
	update_blocked_averages(this_cpu);
6945
	rcu_read_lock();
6946
	for_each_domain(this_cpu, sd) {
6947
		int continue_balancing = 1;
6948
		u64 t0, domain_cost;
6949 6950 6951 6952

		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

6953 6954
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, 0, &next_balance);
6955
			break;
6956
		}
6957

6958
		if (sd->flags & SD_BALANCE_NEWIDLE) {
6959 6960
			t0 = sched_clock_cpu(this_cpu);

6961
			pulled_task = load_balance(this_cpu, this_rq,
6962 6963
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
6964 6965 6966 6967 6968 6969

			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;
6970
		}
6971

6972
		update_next_balance(sd, 0, &next_balance);
6973 6974 6975 6976 6977 6978

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
6979 6980
			break;
	}
6981
	rcu_read_unlock();
6982 6983 6984

	raw_spin_lock(&this_rq->lock);

6985 6986 6987
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

6988
	/*
6989 6990 6991
	 * 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.
6992
	 */
6993
	if (this_rq->cfs.h_nr_running && !pulled_task)
6994
		pulled_task = 1;
6995

6996 6997 6998
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
6999
		this_rq->next_balance = next_balance;
7000

7001
	/* Is there a task of a high priority class? */
7002
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
7003 7004 7005 7006
		pulled_task = -1;

	if (pulled_task) {
		idle_exit_fair(this_rq);
7007
		this_rq->idle_stamp = 0;
7008
	}
7009

7010
	return pulled_task;
7011 7012 7013
}

/*
7014 7015 7016 7017
 * 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.
7018
 */
7019
static int active_load_balance_cpu_stop(void *data)
7020
{
7021 7022
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
7023
	int target_cpu = busiest_rq->push_cpu;
7024
	struct rq *target_rq = cpu_rq(target_cpu);
7025
	struct sched_domain *sd;
7026
	struct task_struct *p = NULL;
7027 7028 7029 7030 7031 7032 7033

	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;
7034 7035 7036

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
7037
		goto out_unlock;
7038 7039 7040 7041 7042 7043 7044 7045 7046

	/*
	 * 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. */
7047
	rcu_read_lock();
7048 7049 7050 7051 7052 7053 7054
	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)) {
7055 7056
		struct lb_env env = {
			.sd		= sd,
7057 7058 7059 7060
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
7061 7062 7063
			.idle		= CPU_IDLE,
		};

7064 7065
		schedstat_inc(sd, alb_count);

7066 7067
		p = detach_one_task(&env);
		if (p)
7068 7069 7070 7071
			schedstat_inc(sd, alb_pushed);
		else
			schedstat_inc(sd, alb_failed);
	}
7072
	rcu_read_unlock();
7073 7074
out_unlock:
	busiest_rq->active_balance = 0;
7075 7076 7077 7078 7079 7080 7081
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

7082
	return 0;
7083 7084
}

7085 7086 7087 7088 7089
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

7090
#ifdef CONFIG_NO_HZ_COMMON
7091 7092 7093 7094 7095 7096
/*
 * 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.
 */
7097
static struct {
7098
	cpumask_var_t idle_cpus_mask;
7099
	atomic_t nr_cpus;
7100 7101
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
7102

7103
static inline int find_new_ilb(void)
7104
{
7105
	int ilb = cpumask_first(nohz.idle_cpus_mask);
7106

7107 7108 7109 7110
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
7111 7112
}

7113 7114 7115 7116 7117
/*
 * 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).
 */
7118
static void nohz_balancer_kick(void)
7119 7120 7121 7122 7123
{
	int ilb_cpu;

	nohz.next_balance++;

7124
	ilb_cpu = find_new_ilb();
7125

7126 7127
	if (ilb_cpu >= nr_cpu_ids)
		return;
7128

7129
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
7130 7131 7132 7133 7134 7135 7136 7137
		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);
7138 7139 7140
	return;
}

7141
static inline void nohz_balance_exit_idle(int cpu)
7142 7143
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
7144 7145 7146 7147 7148 7149 7150
		/*
		 * 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);
		}
7151 7152 7153 7154
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

7155 7156 7157
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
7158
	int cpu = smp_processor_id();
7159 7160

	rcu_read_lock();
7161
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7162 7163 7164 7165 7166

	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

7167
	atomic_inc(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7168
unlock:
7169 7170 7171 7172 7173 7174
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
7175
	int cpu = smp_processor_id();
7176 7177

	rcu_read_lock();
7178
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7179 7180 7181 7182 7183

	if (!sd || sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 1;

7184
	atomic_dec(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7185
unlock:
7186 7187 7188
	rcu_read_unlock();
}

7189
/*
7190
 * This routine will record that the cpu is going idle with tick stopped.
7191
 * This info will be used in performing idle load balancing in the future.
7192
 */
7193
void nohz_balance_enter_idle(int cpu)
7194
{
7195 7196 7197 7198 7199 7200
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

7201 7202
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
7203

7204 7205 7206 7207 7208 7209
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

7210 7211 7212
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
7213
}
7214

7215
static int sched_ilb_notifier(struct notifier_block *nfb,
7216 7217 7218 7219
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DYING:
7220
		nohz_balance_exit_idle(smp_processor_id());
7221 7222 7223 7224 7225
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}
7226 7227 7228 7229
#endif

static DEFINE_SPINLOCK(balancing);

7230 7231 7232 7233
/*
 * 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.
 */
7234
void update_max_interval(void)
7235 7236 7237 7238
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

7239 7240 7241 7242
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
7243
 * Balancing parameters are set up in init_sched_domains.
7244
 */
7245
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
7246
{
7247
	int continue_balancing = 1;
7248
	int cpu = rq->cpu;
7249
	unsigned long interval;
7250
	struct sched_domain *sd;
7251 7252 7253
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
7254 7255
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
7256

7257
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
7258

7259
	rcu_read_lock();
7260
	for_each_domain(cpu, sd) {
7261 7262 7263 7264 7265 7266 7267 7268 7269 7270 7271 7272
		/*
		 * 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;

7273 7274 7275
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

7276 7277 7278 7279 7280 7281 7282 7283 7284 7285 7286
		/*
		 * 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;
		}

7287
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7288 7289 7290 7291 7292 7293 7294 7295

		need_serialize = sd->flags & SD_SERIALIZE;
		if (need_serialize) {
			if (!spin_trylock(&balancing))
				goto out;
		}

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
7296
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
7297
				/*
7298
				 * The LBF_DST_PINNED logic could have changed
7299 7300
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
7301
				 */
7302
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
7303 7304
			}
			sd->last_balance = jiffies;
7305
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7306 7307 7308 7309 7310 7311 7312 7313
		}
		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;
		}
7314 7315
	}
	if (need_decay) {
7316
		/*
7317 7318
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
7319
		 */
7320 7321
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
7322
	}
7323
	rcu_read_unlock();
7324 7325 7326 7327 7328 7329 7330 7331 7332 7333

	/*
	 * 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;
}

7334
#ifdef CONFIG_NO_HZ_COMMON
7335
/*
7336
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
7337 7338
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
7339
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
7340
{
7341
	int this_cpu = this_rq->cpu;
7342 7343 7344
	struct rq *rq;
	int balance_cpu;

7345 7346 7347
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
7348 7349

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
7350
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
7351 7352 7353 7354 7355 7356 7357
			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.
		 */
7358
		if (need_resched())
7359 7360
			break;

V
Vincent Guittot 已提交
7361 7362
		rq = cpu_rq(balance_cpu);

7363 7364 7365 7366 7367 7368 7369 7370 7371 7372 7373
		/*
		 * 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);
		}
7374 7375 7376 7377 7378

		if (time_after(this_rq->next_balance, rq->next_balance))
			this_rq->next_balance = rq->next_balance;
	}
	nohz.next_balance = this_rq->next_balance;
7379 7380
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
7381 7382 7383
}

/*
7384 7385 7386 7387
 * 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
7388
 *     busy cpu's exceeding the group's capacity.
7389 7390
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
7391
 */
7392
static inline int nohz_kick_needed(struct rq *rq)
7393 7394
{
	unsigned long now = jiffies;
7395
	struct sched_domain *sd;
7396
	struct sched_group_capacity *sgc;
7397
	int nr_busy, cpu = rq->cpu;
7398

7399
	if (unlikely(rq->idle_balance))
7400 7401
		return 0;

7402 7403 7404 7405
       /*
	* 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.
	*/
7406
	set_cpu_sd_state_busy();
7407
	nohz_balance_exit_idle(cpu);
7408 7409 7410 7411 7412 7413 7414

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
		return 0;
7415 7416

	if (time_before(now, nohz.next_balance))
7417 7418
		return 0;

7419 7420
	if (rq->nr_running >= 2)
		goto need_kick;
7421

7422
	rcu_read_lock();
7423
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
7424

7425
	if (sd) {
7426 7427
		sgc = sd->groups->sgc;
		nr_busy = atomic_read(&sgc->nr_busy_cpus);
7428

7429
		if (nr_busy > 1)
7430
			goto need_kick_unlock;
7431
	}
7432 7433 7434 7435 7436 7437 7438

	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;

7439
	rcu_read_unlock();
7440
	return 0;
7441 7442 7443

need_kick_unlock:
	rcu_read_unlock();
7444 7445
need_kick:
	return 1;
7446 7447
}
#else
7448
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
7449 7450 7451 7452 7453 7454
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
7455 7456
static void run_rebalance_domains(struct softirq_action *h)
{
7457
	struct rq *this_rq = this_rq();
7458
	enum cpu_idle_type idle = this_rq->idle_balance ?
7459 7460
						CPU_IDLE : CPU_NOT_IDLE;

7461
	rebalance_domains(this_rq, idle);
7462 7463

	/*
7464
	 * If this cpu has a pending nohz_balance_kick, then do the
7465 7466 7467
	 * balancing on behalf of the other idle cpus whose ticks are
	 * stopped.
	 */
7468
	nohz_idle_balance(this_rq, idle);
7469 7470 7471 7472 7473
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
7474
void trigger_load_balance(struct rq *rq)
7475 7476
{
	/* Don't need to rebalance while attached to NULL domain */
7477 7478 7479 7480
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
7481
		raise_softirq(SCHED_SOFTIRQ);
7482
#ifdef CONFIG_NO_HZ_COMMON
7483
	if (nohz_kick_needed(rq))
7484
		nohz_balancer_kick();
7485
#endif
7486 7487
}

7488 7489 7490
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
7491 7492

	update_runtime_enabled(rq);
7493 7494 7495 7496 7497
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
7498 7499 7500

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
7501 7502
}

7503
#endif /* CONFIG_SMP */
7504

7505 7506 7507
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
7508
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
7509 7510 7511 7512 7513 7514
{
	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
7515
		entity_tick(cfs_rq, se, queued);
7516
	}
7517

7518
	if (numabalancing_enabled)
7519
		task_tick_numa(rq, curr);
7520

7521
	update_rq_runnable_avg(rq, 1);
7522 7523 7524
}

/*
P
Peter Zijlstra 已提交
7525 7526 7527
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
7528
 */
P
Peter Zijlstra 已提交
7529
static void task_fork_fair(struct task_struct *p)
7530
{
7531 7532
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
7533
	int this_cpu = smp_processor_id();
P
Peter Zijlstra 已提交
7534 7535 7536
	struct rq *rq = this_rq();
	unsigned long flags;

7537
	raw_spin_lock_irqsave(&rq->lock, flags);
7538

7539 7540
	update_rq_clock(rq);

7541 7542 7543
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;

7544 7545 7546 7547 7548 7549 7550 7551 7552
	/*
	 * 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();
7553

7554
	update_curr(cfs_rq);
P
Peter Zijlstra 已提交
7555

7556 7557
	if (curr)
		se->vruntime = curr->vruntime;
7558
	place_entity(cfs_rq, se, 1);
7559

P
Peter Zijlstra 已提交
7560
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
7561
		/*
7562 7563 7564
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
7565
		swap(curr->vruntime, se->vruntime);
7566
		resched_curr(rq);
7567
	}
7568

7569 7570
	se->vruntime -= cfs_rq->min_vruntime;

7571
	raw_spin_unlock_irqrestore(&rq->lock, flags);
7572 7573
}

7574 7575 7576 7577
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
7578 7579
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
7580
{
7581
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7582 7583
		return;

7584 7585 7586 7587 7588
	/*
	 * 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
7589
	if (rq->curr == p) {
7590
		if (p->prio > oldprio)
7591
			resched_curr(rq);
7592
	} else
7593
		check_preempt_curr(rq, p, 0);
7594 7595
}

P
Peter Zijlstra 已提交
7596 7597 7598 7599 7600 7601
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);

	/*
7602
	 * Ensure the task's vruntime is normalized, so that when it's
P
Peter Zijlstra 已提交
7603 7604 7605
	 * switched back to the fair class the enqueue_entity(.flags=0) will
	 * do the right thing.
	 *
7606 7607
	 * 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
7608 7609
	 * the task is sleeping will it still have non-normalized vruntime.
	 */
7610
	if (!task_on_rq_queued(p) && p->state != TASK_RUNNING) {
P
Peter Zijlstra 已提交
7611 7612 7613 7614 7615 7616 7617
		/*
		 * 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;
	}
7618

7619
#ifdef CONFIG_SMP
7620 7621 7622 7623 7624
	/*
	* 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.
	*/
7625 7626 7627
	if (se->avg.decay_count) {
		__synchronize_entity_decay(se);
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
7628 7629
	}
#endif
P
Peter Zijlstra 已提交
7630 7631
}

7632 7633 7634
/*
 * We switched to the sched_fair class.
 */
P
Peter Zijlstra 已提交
7635
static void switched_to_fair(struct rq *rq, struct task_struct *p)
7636
{
7637
#ifdef CONFIG_FAIR_GROUP_SCHED
7638
	struct sched_entity *se = &p->se;
7639 7640 7641 7642 7643 7644
	/*
	 * 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
7645
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7646 7647
		return;

7648 7649 7650 7651 7652
	/*
	 * 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
7653
	if (rq->curr == p)
7654
		resched_curr(rq);
7655
	else
7656
		check_preempt_curr(rq, p, 0);
7657 7658
}

7659 7660 7661 7662 7663 7664 7665 7666 7667
/* 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;

7668 7669 7670 7671 7672 7673 7674
	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);
	}
7675 7676
}

7677 7678 7679 7680 7681 7682 7683
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
7684
#ifdef CONFIG_SMP
7685
	atomic64_set(&cfs_rq->decay_counter, 1);
7686
	atomic_long_set(&cfs_rq->removed_load, 0);
7687
#endif
7688 7689
}

P
Peter Zijlstra 已提交
7690
#ifdef CONFIG_FAIR_GROUP_SCHED
7691
static void task_move_group_fair(struct task_struct *p, int queued)
P
Peter Zijlstra 已提交
7692
{
P
Peter Zijlstra 已提交
7693
	struct sched_entity *se = &p->se;
7694
	struct cfs_rq *cfs_rq;
P
Peter Zijlstra 已提交
7695

7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708
	/*
	 * 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.
	 */
7709
	/*
7710
	 * When !queued, vruntime of the task has usually NOT been normalized.
7711 7712 7713 7714
	 * 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().
7715 7716
	 * - Moving a task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
7717 7718 7719 7720
	 *
	 * To prevent boost or penalty in the new cfs_rq caused by delta
	 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
	 */
7721 7722
	if (!queued && (!se->sum_exec_runtime || p->state == TASK_WAKING))
		queued = 1;
7723

7724
	if (!queued)
P
Peter Zijlstra 已提交
7725
		se->vruntime -= cfs_rq_of(se)->min_vruntime;
7726
	set_task_rq(p, task_cpu(p));
P
Peter Zijlstra 已提交
7727
	se->depth = se->parent ? se->parent->depth + 1 : 0;
7728
	if (!queued) {
P
Peter Zijlstra 已提交
7729 7730
		cfs_rq = cfs_rq_of(se);
		se->vruntime += cfs_rq->min_vruntime;
7731 7732 7733 7734 7735 7736
#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
7737 7738
		se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
		cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
7739 7740
#endif
	}
P
Peter Zijlstra 已提交
7741
}
7742 7743 7744 7745 7746 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

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
7834
	if (!parent) {
7835
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
7836 7837
		se->depth = 0;
	} else {
7838
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
7839 7840
		se->depth = parent->depth + 1;
	}
7841 7842

	se->my_q = cfs_rq;
7843 7844
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874
	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);
7875 7876 7877

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
7878
		for_each_sched_entity(se)
7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897 7898 7899
			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
7900

7901
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
7902 7903 7904 7905 7906 7907 7908 7909 7910
{
	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)
7911
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
7912 7913 7914 7915

	return rr_interval;
}

7916 7917 7918
/*
 * All the scheduling class methods:
 */
7919
const struct sched_class fair_sched_class = {
7920
	.next			= &idle_sched_class,
7921 7922 7923
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
7924
	.yield_to_task		= yield_to_task_fair,
7925

I
Ingo Molnar 已提交
7926
	.check_preempt_curr	= check_preempt_wakeup,
7927 7928 7929 7930

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

7931
#ifdef CONFIG_SMP
L
Li Zefan 已提交
7932
	.select_task_rq		= select_task_rq_fair,
7933
	.migrate_task_rq	= migrate_task_rq_fair,
7934

7935 7936
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
7937 7938

	.task_waking		= task_waking_fair,
7939
#endif
7940

7941
	.set_curr_task          = set_curr_task_fair,
7942
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
7943
	.task_fork		= task_fork_fair,
7944 7945

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
7946
	.switched_from		= switched_from_fair,
7947
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
7948

7949 7950
	.get_rr_interval	= get_rr_interval_fair,

P
Peter Zijlstra 已提交
7951
#ifdef CONFIG_FAIR_GROUP_SCHED
7952
	.task_move_group	= task_move_group_fair,
P
Peter Zijlstra 已提交
7953
#endif
7954 7955 7956
};

#ifdef CONFIG_SCHED_DEBUG
7957
void print_cfs_stats(struct seq_file *m, int cpu)
7958 7959 7960
{
	struct cfs_rq *cfs_rq;

7961
	rcu_read_lock();
7962
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
7963
		print_cfs_rq(m, cpu, cfs_rq);
7964
	rcu_read_unlock();
7965 7966
}
#endif
7967 7968 7969 7970 7971 7972

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

7973
#ifdef CONFIG_NO_HZ_COMMON
7974
	nohz.next_balance = jiffies;
7975
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
7976
	cpu_notifier(sched_ilb_notifier, 0);
7977 7978 7979 7980
#endif
#endif /* SMP */

}