fair.c 211.5 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
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)
{
831
	unsigned int scan_size = ACCESS_ONCE(sysctl_numa_balancing_scan_size);
832 833 834
	unsigned int scan, floor;
	unsigned int windows = 1;

835 836
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
	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);
}

853 854 855 856 857 858 859 860 861 862 863 864
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));
}

865 866 867 868 869
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
870
	pid_t gid;
871 872 873
	struct list_head task_list;

	struct rcu_head rcu;
874
	nodemask_t active_nodes;
875
	unsigned long total_faults;
876 877 878 879 880
	/*
	 * 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.
	 */
881
	unsigned long *faults_cpu;
882
	unsigned long faults[0];
883 884
};

885 886 887 888 889 890 891 892 893
/* 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)

894 895 896 897 898
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

899 900
static inline int task_faults_idx(int nid, int priv)
{
901
	return NR_NUMA_HINT_FAULT_TYPES * nid + priv;
902 903 904 905
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
906
	if (!p->numa_faults_memory)
907 908
		return 0;

909 910
	return p->numa_faults_memory[task_faults_idx(nid, 0)] +
		p->numa_faults_memory[task_faults_idx(nid, 1)];
911 912
}

913 914 915 916 917
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

918 919
	return p->numa_group->faults[task_faults_idx(nid, 0)] +
		p->numa_group->faults[task_faults_idx(nid, 1)];
920 921
}

922 923 924 925 926 927
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)];
}

928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992
/* Handle placement on systems where not all nodes are directly connected. */
static unsigned long score_nearby_nodes(struct task_struct *p, int nid,
					int maxdist, bool task)
{
	unsigned long score = 0;
	int node;

	/*
	 * All nodes are directly connected, and the same distance
	 * from each other. No need for fancy placement algorithms.
	 */
	if (sched_numa_topology_type == NUMA_DIRECT)
		return 0;

	/*
	 * This code is called for each node, introducing N^2 complexity,
	 * which should be ok given the number of nodes rarely exceeds 8.
	 */
	for_each_online_node(node) {
		unsigned long faults;
		int dist = node_distance(nid, node);

		/*
		 * The furthest away nodes in the system are not interesting
		 * for placement; nid was already counted.
		 */
		if (dist == sched_max_numa_distance || node == nid)
			continue;

		/*
		 * On systems with a backplane NUMA topology, compare groups
		 * of nodes, and move tasks towards the group with the most
		 * memory accesses. When comparing two nodes at distance
		 * "hoplimit", only nodes closer by than "hoplimit" are part
		 * of each group. Skip other nodes.
		 */
		if (sched_numa_topology_type == NUMA_BACKPLANE &&
					dist > maxdist)
			continue;

		/* Add up the faults from nearby nodes. */
		if (task)
			faults = task_faults(p, node);
		else
			faults = group_faults(p, node);

		/*
		 * On systems with a glueless mesh NUMA topology, there are
		 * no fixed "groups of nodes". Instead, nodes that are not
		 * directly connected bounce traffic through intermediate
		 * nodes; a numa_group can occupy any set of nodes.
		 * The further away a node is, the less the faults count.
		 * This seems to result in good task placement.
		 */
		if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
			faults *= (sched_max_numa_distance - dist);
			faults /= (sched_max_numa_distance - LOCAL_DISTANCE);
		}

		score += faults;
	}

	return score;
}

993 994 995 996 997 998
/*
 * 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.
 */
999 1000
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1001
{
1002
	unsigned long faults, total_faults;
1003

1004
	if (!p->numa_faults_memory)
1005 1006 1007 1008 1009 1010 1011
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1012
	faults = task_faults(p, nid);
1013 1014
	faults += score_nearby_nodes(p, nid, dist, true);

1015
	return 1000 * faults / total_faults;
1016 1017
}

1018 1019
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1020
{
1021 1022 1023 1024 1025 1026 1027 1028
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1029 1030
		return 0;

1031
	faults = group_faults(p, nid);
1032 1033
	faults += score_nearby_nodes(p, nid, dist, false);

1034
	return 1000 * faults / total_faults;
1035 1036
}

1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
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);
}

1100
static unsigned long weighted_cpuload(const int cpu);
1101 1102
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1103
static unsigned long capacity_of(int cpu);
1104 1105
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1106
/* Cached statistics for all CPUs within a node */
1107
struct numa_stats {
1108
	unsigned long nr_running;
1109
	unsigned long load;
1110 1111

	/* Total compute capacity of CPUs on a node */
1112
	unsigned long compute_capacity;
1113 1114

	/* Approximate capacity in terms of runnable tasks on a node */
1115
	unsigned long task_capacity;
1116
	int has_free_capacity;
1117
};
1118

1119 1120 1121 1122 1123
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1124 1125
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1126 1127 1128 1129 1130 1131 1132

	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);
1133
		ns->compute_capacity += capacity_of(cpu);
1134 1135

		cpus++;
1136 1137
	}

1138 1139 1140 1141 1142
	/*
	 * 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.
	 *
1143 1144
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1145 1146 1147 1148
	 */
	if (!cpus)
		return;

1149 1150 1151 1152 1153 1154
	/* 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));
1155
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1156 1157
}

1158 1159
struct task_numa_env {
	struct task_struct *p;
1160

1161 1162
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1163

1164
	struct numa_stats src_stats, dst_stats;
1165

1166
	int imbalance_pct;
1167
	int dist;
1168 1169 1170

	struct task_struct *best_task;
	long best_imp;
1171 1172 1173
	int best_cpu;
};

1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186
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;
}

1187
static bool load_too_imbalanced(long src_load, long dst_load,
1188 1189 1190
				struct task_numa_env *env)
{
	long imb, old_imb;
1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202
	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;
1203 1204 1205 1206 1207 1208

	/* 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? */
1209 1210
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1211 1212 1213 1214 1215 1216 1217
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
	 * Compare it with the old imbalance.
	 */
1218 1219 1220
	orig_src_load = env->src_stats.load;
	orig_dst_load = env->dst_stats.load;

1221 1222 1223
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);

1224 1225
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;
1226 1227

	/* Would this change make things worse? */
1228
	return (imb > old_imb);
1229 1230
}

1231 1232 1233 1234 1235 1236
/*
 * 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
 */
1237 1238
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1239 1240 1241 1242
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1243
	long src_load, dst_load;
1244
	long load;
1245
	long imp = env->p->numa_group ? groupimp : taskimp;
1246
	long moveimp = imp;
1247
	int dist = env->dist;
1248 1249

	rcu_read_lock();
1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260

	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))
1261
		cur = NULL;
1262
	raw_spin_unlock_irq(&dst_rq->lock);
1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275

	/*
	 * "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;

1276 1277
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1278
		 * in any group then look only at task weights.
1279
		 */
1280
		if (cur->numa_group == env->p->numa_group) {
1281 1282
			imp = taskimp + task_weight(cur, env->src_nid, dist) -
			      task_weight(cur, env->dst_nid, dist);
1283 1284 1285 1286 1287 1288
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1289
		} else {
1290 1291 1292 1293 1294 1295
			/*
			 * 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)
1296 1297
				imp += group_weight(cur, env->src_nid, dist) -
				       group_weight(cur, env->dst_nid, dist);
1298
			else
1299 1300
				imp += task_weight(cur, env->src_nid, dist) -
				       task_weight(cur, env->dst_nid, dist);
1301
		}
1302 1303
	}

1304
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1305 1306 1307 1308
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1309
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1310
		    !env->dst_stats.has_free_capacity)
1311 1312 1313 1314 1315 1316
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1317 1318
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1319 1320 1321 1322 1323 1324
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1325 1326 1327
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1328

1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345
	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;

1346
	if (cur) {
1347 1348 1349
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1350 1351
	}

1352
	if (load_too_imbalanced(src_load, dst_load, env))
1353 1354
		goto unlock;

1355 1356 1357 1358 1359 1360 1361
	/*
	 * 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);

1362 1363 1364 1365 1366 1367
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1368 1369
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1370 1371 1372 1373 1374 1375 1376 1377 1378
{
	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;
1379
		task_numa_compare(env, taskimp, groupimp);
1380 1381 1382
	}
}

1383 1384 1385 1386
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1387

1388
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1389
		.src_nid = task_node(p),
1390 1391 1392 1393 1394 1395

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
		.best_cpu = -1
1396 1397
	};
	struct sched_domain *sd;
1398
	unsigned long taskweight, groupweight;
1399
	int nid, ret, dist;
1400
	long taskimp, groupimp;
1401

1402
	/*
1403 1404 1405 1406 1407 1408
	 * 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.
1409 1410
	 */
	rcu_read_lock();
1411
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1412 1413
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1414 1415
	rcu_read_unlock();

1416 1417 1418 1419 1420 1421 1422
	/*
	 * 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)) {
1423
		p->numa_preferred_nid = task_node(p);
1424 1425 1426
		return -EINVAL;
	}

1427
	env.dst_nid = p->numa_preferred_nid;
1428 1429 1430 1431 1432 1433
	dist = env.dist = node_distance(env.src_nid, env.dst_nid);
	taskweight = task_weight(p, env.src_nid, dist);
	groupweight = group_weight(p, env.src_nid, dist);
	update_numa_stats(&env.src_stats, env.src_nid);
	taskimp = task_weight(p, env.dst_nid, dist) - taskweight;
	groupimp = group_weight(p, env.dst_nid, dist) - groupweight;
1434
	update_numa_stats(&env.dst_stats, env.dst_nid);
1435

1436 1437
	/* Try to find a spot on the preferred nid. */
	task_numa_find_cpu(&env, taskimp, groupimp);
1438

1439 1440 1441 1442 1443 1444 1445 1446 1447
	/*
	 * Look at other nodes in these cases:
	 * - there is no space available on the preferred_nid
	 * - the task is part of a numa_group that is interleaved across
	 *   multiple NUMA nodes; in order to better consolidate the group,
	 *   we need to check other locations.
	 */
	if (env.best_cpu == -1 || (p->numa_group &&
			nodes_weight(p->numa_group->active_nodes) > 1)) {
1448 1449 1450
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1451

1452
			dist = node_distance(env.src_nid, env.dst_nid);
1453 1454 1455 1456 1457
			if (sched_numa_topology_type == NUMA_BACKPLANE &&
						dist != env.dist) {
				taskweight = task_weight(p, env.src_nid, dist);
				groupweight = group_weight(p, env.src_nid, dist);
			}
1458

1459
			/* Only consider nodes where both task and groups benefit */
1460 1461
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1462
			if (taskimp < 0 && groupimp < 0)
1463 1464
				continue;

1465
			env.dist = dist;
1466 1467
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1468
			task_numa_find_cpu(&env, taskimp, groupimp);
1469 1470 1471
		}
	}

1472 1473 1474 1475 1476 1477 1478 1479
	/*
	 * 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.
	 */
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492
	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;
1493

1494 1495 1496 1497 1498 1499
	/*
	 * 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);

1500
	if (env.best_task == NULL) {
1501 1502 1503
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1504 1505 1506 1507
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1508 1509
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1510 1511
	put_task_struct(env.best_task);
	return ret;
1512 1513
}

1514 1515 1516
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1517 1518
	unsigned long interval = HZ;

1519
	/* This task has no NUMA fault statistics yet */
1520
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults_memory))
1521 1522
		return;

1523
	/* Periodically retry migrating the task to the preferred node */
1524 1525
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1526 1527

	/* Success if task is already running on preferred CPU */
1528
	if (task_node(p) == p->numa_preferred_nid)
1529 1530 1531
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1532
	task_numa_migrate(p);
1533 1534
}

1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566
/*
 * 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);
	}
}

1567 1568 1569
/*
 * 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
1570 1571 1572
 * 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.
1573 1574
 */
#define NUMA_PERIOD_SLOTS 10
1575
#define NUMA_PERIOD_THRESHOLD 7
1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631

/*
 * 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
		 */
1632
		ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared + 1));
1633 1634 1635 1636 1637 1638 1639 1640
		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));
}

1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668
/*
 * 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;
}

1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754
/*
 * Determine the preferred nid for a task in a numa_group. This needs to
 * be done in a way that produces consistent results with group_weight,
 * otherwise workloads might not converge.
 */
static int preferred_group_nid(struct task_struct *p, int nid)
{
	nodemask_t nodes;
	int dist;

	/* Direct connections between all NUMA nodes. */
	if (sched_numa_topology_type == NUMA_DIRECT)
		return nid;

	/*
	 * On a system with glueless mesh NUMA topology, group_weight
	 * scores nodes according to the number of NUMA hinting faults on
	 * both the node itself, and on nearby nodes.
	 */
	if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
		unsigned long score, max_score = 0;
		int node, max_node = nid;

		dist = sched_max_numa_distance;

		for_each_online_node(node) {
			score = group_weight(p, node, dist);
			if (score > max_score) {
				max_score = score;
				max_node = node;
			}
		}
		return max_node;
	}

	/*
	 * Finding the preferred nid in a system with NUMA backplane
	 * interconnect topology is more involved. The goal is to locate
	 * tasks from numa_groups near each other in the system, and
	 * untangle workloads from different sides of the system. This requires
	 * searching down the hierarchy of node groups, recursively searching
	 * inside the highest scoring group of nodes. The nodemask tricks
	 * keep the complexity of the search down.
	 */
	nodes = node_online_map;
	for (dist = sched_max_numa_distance; dist > LOCAL_DISTANCE; dist--) {
		unsigned long max_faults = 0;
		nodemask_t max_group;
		int a, b;

		/* Are there nodes at this distance from each other? */
		if (!find_numa_distance(dist))
			continue;

		for_each_node_mask(a, nodes) {
			unsigned long faults = 0;
			nodemask_t this_group;
			nodes_clear(this_group);

			/* Sum group's NUMA faults; includes a==b case. */
			for_each_node_mask(b, nodes) {
				if (node_distance(a, b) < dist) {
					faults += group_faults(p, b);
					node_set(b, this_group);
					node_clear(b, nodes);
				}
			}

			/* Remember the top group. */
			if (faults > max_faults) {
				max_faults = faults;
				max_group = this_group;
				/*
				 * subtle: at the smallest distance there is
				 * just one node left in each "group", the
				 * winner is the preferred nid.
				 */
				nid = a;
			}
		}
		/* Next round, evaluate the nodes within max_group. */
		nodes = max_group;
	}
	return nid;
}

1755 1756
static void task_numa_placement(struct task_struct *p)
{
1757 1758
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
1759
	unsigned long fault_types[2] = { 0, 0 };
1760 1761
	unsigned long total_faults;
	u64 runtime, period;
1762
	spinlock_t *group_lock = NULL;
1763

1764
	seq = ACCESS_ONCE(p->mm->numa_scan_seq);
1765 1766 1767
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
1768
	p->numa_scan_period_max = task_scan_max(p);
1769

1770 1771 1772 1773
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

1774 1775 1776
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
1777
		spin_lock_irq(group_lock);
1778 1779
	}

1780 1781
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
1782
		unsigned long faults = 0, group_faults = 0;
1783
		int priv, i;
1784

1785
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
1786
			long diff, f_diff, f_weight;
1787

1788
			i = task_faults_idx(nid, priv);
1789

1790
			/* Decay existing window, copy faults since last scan */
1791
			diff = p->numa_faults_buffer_memory[i] - p->numa_faults_memory[i] / 2;
1792 1793
			fault_types[priv] += p->numa_faults_buffer_memory[i];
			p->numa_faults_buffer_memory[i] = 0;
1794

1795 1796 1797 1798 1799 1800 1801 1802 1803 1804
			/*
			 * 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);
1805
			f_diff = f_weight - p->numa_faults_cpu[i] / 2;
1806 1807
			p->numa_faults_buffer_cpu[i] = 0;

1808 1809
			p->numa_faults_memory[i] += diff;
			p->numa_faults_cpu[i] += f_diff;
1810
			faults += p->numa_faults_memory[i];
1811
			p->total_numa_faults += diff;
1812 1813
			if (p->numa_group) {
				/* safe because we can only change our own group */
1814
				p->numa_group->faults[i] += diff;
1815
				p->numa_group->faults_cpu[i] += f_diff;
1816 1817
				p->numa_group->total_faults += diff;
				group_faults += p->numa_group->faults[i];
1818
			}
1819 1820
		}

1821 1822 1823 1824
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
1825 1826 1827 1828 1829 1830 1831

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

1832 1833
	update_task_scan_period(p, fault_types[0], fault_types[1]);

1834
	if (p->numa_group) {
1835
		update_numa_active_node_mask(p->numa_group);
1836
		spin_unlock_irq(group_lock);
1837
		max_nid = preferred_group_nid(p, max_group_nid);
1838 1839
	}

1840 1841 1842 1843 1844 1845 1846
	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);
1847
	}
1848 1849
}

1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860
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);
}

1861 1862
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
1863 1864 1865 1866 1867 1868 1869 1870 1871
{
	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) +
1872
				    4*nr_node_ids*sizeof(unsigned long);
1873 1874 1875 1876 1877 1878 1879 1880

		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);
1881
		grp->gid = p->pid;
1882
		/* Second half of the array tracks nids where faults happen */
1883 1884
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
1885

1886 1887
		node_set(task_node(current), grp->active_nodes);

1888
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1889
			grp->faults[i] = p->numa_faults_memory[i];
1890

1891
		grp->total_faults = p->total_numa_faults;
1892

1893 1894 1895 1896 1897 1898 1899 1900 1901
		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))
1902
		goto no_join;
1903 1904 1905

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
1906
		goto no_join;
1907 1908 1909

	my_grp = p->numa_group;
	if (grp == my_grp)
1910
		goto no_join;
1911 1912 1913 1914 1915 1916

	/*
	 * 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)
1917
		goto no_join;
1918 1919 1920 1921 1922

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
1923
		goto no_join;
1924

1925 1926 1927 1928 1929 1930 1931
	/* 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;
1932

1933 1934 1935
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

1936
	if (join && !get_numa_group(grp))
1937
		goto no_join;
1938 1939 1940 1941 1942 1943

	rcu_read_unlock();

	if (!join)
		return;

1944 1945
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
1946

1947
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
1948 1949
		my_grp->faults[i] -= p->numa_faults_memory[i];
		grp->faults[i] += p->numa_faults_memory[i];
1950
	}
1951 1952
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
1953 1954 1955 1956 1957 1958

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

	spin_unlock(&my_grp->lock);
1959
	spin_unlock_irq(&grp->lock);
1960 1961 1962 1963

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
1964 1965 1966 1967 1968
	return;

no_join:
	rcu_read_unlock();
	return;
1969 1970 1971 1972 1973
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
1974
	void *numa_faults = p->numa_faults_memory;
1975 1976
	unsigned long flags;
	int i;
1977 1978

	if (grp) {
1979
		spin_lock_irqsave(&grp->lock, flags);
1980
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1981
			grp->faults[i] -= p->numa_faults_memory[i];
1982
		grp->total_faults -= p->total_numa_faults;
1983

1984 1985
		list_del(&p->numa_entry);
		grp->nr_tasks--;
1986
		spin_unlock_irqrestore(&grp->lock, flags);
1987
		RCU_INIT_POINTER(p->numa_group, NULL);
1988 1989 1990
		put_numa_group(grp);
	}

1991 1992
	p->numa_faults_memory = NULL;
	p->numa_faults_buffer_memory = NULL;
1993 1994
	p->numa_faults_cpu= NULL;
	p->numa_faults_buffer_cpu = NULL;
1995
	kfree(numa_faults);
1996 1997
}

1998 1999 2000
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2001
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2002 2003
{
	struct task_struct *p = current;
2004
	bool migrated = flags & TNF_MIGRATED;
2005
	int cpu_node = task_node(current);
2006
	int local = !!(flags & TNF_FAULT_LOCAL);
2007
	int priv;
2008

2009
	if (!numabalancing_enabled)
2010 2011
		return;

2012 2013 2014 2015
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2016
	/* Allocate buffer to track faults on a per-node basis */
2017
	if (unlikely(!p->numa_faults_memory)) {
2018 2019
		int size = sizeof(*p->numa_faults_memory) *
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2020

2021
		p->numa_faults_memory = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
2022
		if (!p->numa_faults_memory)
2023
			return;
2024

2025
		BUG_ON(p->numa_faults_buffer_memory);
2026 2027 2028 2029 2030 2031
		/*
		 * 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.
		 */
2032 2033 2034
		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);
2035
		p->total_numa_faults = 0;
2036
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2037
	}
2038

2039 2040 2041 2042 2043 2044 2045 2046
	/*
	 * 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);
2047
		if (!priv && !(flags & TNF_NO_GROUP))
2048
			task_numa_group(p, last_cpupid, flags, &priv);
2049 2050
	}

2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061
	/*
	 * 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;

2062
	task_numa_placement(p);
2063

2064 2065 2066 2067 2068
	/*
	 * 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))
2069 2070
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
2071 2072 2073
	if (migrated)
		p->numa_pages_migrated += pages;

2074 2075
	p->numa_faults_buffer_memory[task_faults_idx(mem_node, priv)] += pages;
	p->numa_faults_buffer_cpu[task_faults_idx(cpu_node, priv)] += pages;
2076
	p->numa_faults_locality[local] += pages;
2077 2078
}

2079 2080 2081 2082 2083 2084
static void reset_ptenuma_scan(struct task_struct *p)
{
	ACCESS_ONCE(p->mm->numa_scan_seq)++;
	p->mm->numa_scan_offset = 0;
}

2085 2086 2087 2088 2089 2090 2091 2092 2093
/*
 * 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;
2094
	struct vm_area_struct *vma;
2095
	unsigned long start, end;
2096
	unsigned long nr_pte_updates = 0;
2097
	long pages;
2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112

	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;

2113
	if (!mm->numa_next_scan) {
2114 2115
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2116 2117
	}

2118 2119 2120 2121 2122 2123 2124
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2125 2126 2127 2128
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
2129

2130
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2131 2132 2133
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2134 2135 2136 2137 2138 2139
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2140 2141 2142 2143 2144
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
	if (!pages)
		return;
2145

2146
	down_read(&mm->mmap_sem);
2147
	vma = find_vma(mm, start);
2148 2149
	if (!vma) {
		reset_ptenuma_scan(p);
2150
		start = 0;
2151 2152
		vma = mm->mmap;
	}
2153
	for (; vma; vma = vma->vm_next) {
2154
		if (!vma_migratable(vma) || !vma_policy_mof(vma))
2155 2156
			continue;

2157 2158 2159 2160 2161 2162 2163 2164 2165 2166
		/*
		 * 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
2167 2168 2169 2170 2171 2172
		/*
		 * 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;
2173

2174 2175 2176 2177
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2178 2179 2180 2181 2182 2183 2184 2185 2186
			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;
2187

2188 2189 2190
			start = end;
			if (pages <= 0)
				goto out;
2191 2192

			cond_resched();
2193
		} while (end != vma->vm_end);
2194
	}
2195

2196
out:
2197
	/*
P
Peter Zijlstra 已提交
2198 2199 2200 2201
	 * 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.
2202 2203
	 */
	if (vma)
2204
		mm->numa_scan_offset = start;
2205 2206 2207
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233
}

/*
 * 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) {
2234
		if (!curr->node_stamp)
2235
			curr->numa_scan_period = task_scan_min(curr);
2236
		curr->node_stamp += period;
2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247

		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)
{
}
2248 2249 2250 2251 2252 2253 2254 2255

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)
{
}
2256 2257
#endif /* CONFIG_NUMA_BALANCING */

2258 2259 2260 2261
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2262
	if (!parent_entity(se))
2263
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2264
#ifdef CONFIG_SMP
2265 2266 2267 2268 2269 2270
	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);
	}
2271
#endif
2272 2273 2274 2275 2276 2277 2278
	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);
2279
	if (!parent_entity(se))
2280
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2281 2282
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2283
		list_del_init(&se->group_node);
2284
	}
2285 2286 2287
	cfs_rq->nr_running--;
}

2288 2289
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2290 2291 2292 2293 2294 2295 2296 2297 2298
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().
	 */
2299
	tg_weight = atomic_long_read(&tg->load_avg);
2300
	tg_weight -= cfs_rq->tg_load_contrib;
2301 2302 2303 2304 2305
	tg_weight += cfs_rq->load.weight;

	return tg_weight;
}

2306
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2307
{
2308
	long tg_weight, load, shares;
2309

2310
	tg_weight = calc_tg_weight(tg, cfs_rq);
2311
	load = cfs_rq->load.weight;
2312 2313

	shares = (tg->shares * load);
2314 2315
	if (tg_weight)
		shares /= tg_weight;
2316 2317 2318 2319 2320 2321 2322 2323 2324

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

	return shares;
}
# else /* CONFIG_SMP */
2325
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2326 2327 2328 2329
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
2330 2331 2332
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2333 2334 2335 2336
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2337
		account_entity_dequeue(cfs_rq, se);
2338
	}
P
Peter Zijlstra 已提交
2339 2340 2341 2342 2343 2344 2345

	update_load_set(&se->load, weight);

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

2346 2347
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2348
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2349 2350 2351
{
	struct task_group *tg;
	struct sched_entity *se;
2352
	long shares;
P
Peter Zijlstra 已提交
2353 2354 2355

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2356
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2357
		return;
2358 2359 2360 2361
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2362
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2363 2364 2365 2366

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2367
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2368 2369 2370 2371
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2372
#ifdef CONFIG_SMP
2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400
/*
 * 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,
};

2401 2402 2403 2404 2405 2406
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418
	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
2419 2420
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
2421 2422 2423 2424 2425 2426
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
2427 2428
	}

2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459
	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];
2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493
}

/*
 * 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)
{
2494 2495
	u64 delta, periods;
	u32 runnable_contrib;
2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528
	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;
2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548
		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;
2549 2550 2551 2552 2553 2554 2555 2556 2557 2558
	}

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

	return decayed;
}

2559
/* Synchronize an entity's decay with its parenting cfs_rq.*/
2560
static inline u64 __synchronize_entity_decay(struct sched_entity *se)
2561 2562 2563 2564 2565 2566
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 decays = atomic64_read(&cfs_rq->decay_counter);

	decays -= se->avg.decay_count;
	if (!decays)
2567
		return 0;
2568 2569 2570

	se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
	se->avg.decay_count = 0;
2571 2572

	return decays;
2573 2574
}

2575 2576 2577 2578 2579
#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;
2580
	long tg_contrib;
2581 2582 2583 2584

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

2585 2586 2587
	if (!tg_contrib)
		return;

2588 2589
	if (force_update || abs(tg_contrib) > cfs_rq->tg_load_contrib / 8) {
		atomic_long_add(tg_contrib, &tg->load_avg);
2590 2591 2592
		cfs_rq->tg_load_contrib += tg_contrib;
	}
}
2593

2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604
/*
 * 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 */
2605
	contrib = div_u64((u64)sa->runnable_avg_sum << NICE_0_SHIFT,
2606 2607 2608 2609 2610 2611 2612 2613 2614
			  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;
	}
}

2615 2616 2617 2618
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;
2619 2620
	int runnable_avg;

2621 2622 2623
	u64 contrib;

	contrib = cfs_rq->tg_load_contrib * tg->shares;
2624 2625
	se->avg.load_avg_contrib = div_u64(contrib,
				     atomic_long_read(&tg->load_avg) + 1);
2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654

	/*
	 * 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;
	}
2655
}
2656 2657 2658 2659 2660 2661

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);
}
2662
#else /* CONFIG_FAIR_GROUP_SCHED */
2663 2664
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update) {}
2665 2666
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq) {}
2667
static inline void __update_group_entity_contrib(struct sched_entity *se) {}
2668
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2669
#endif /* CONFIG_FAIR_GROUP_SCHED */
2670

2671 2672 2673 2674 2675 2676 2677 2678 2679 2680
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);
}

2681 2682 2683 2684 2685
/* 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;

2686 2687 2688
	if (entity_is_task(se)) {
		__update_task_entity_contrib(se);
	} else {
2689
		__update_tg_runnable_avg(&se->avg, group_cfs_rq(se));
2690 2691
		__update_group_entity_contrib(se);
	}
2692 2693 2694 2695

	return se->avg.load_avg_contrib - old_contrib;
}

2696 2697 2698 2699 2700 2701 2702 2703 2704
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;
}

2705 2706
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);

2707
/* Update a sched_entity's runnable average */
2708 2709
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq)
2710
{
2711 2712
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	long contrib_delta;
2713
	u64 now;
2714

2715 2716 2717 2718 2719 2720 2721 2722 2723 2724
	/*
	 * 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))
2725 2726 2727
		return;

	contrib_delta = __update_entity_load_avg_contrib(se);
2728 2729 2730 2731

	if (!update_cfs_rq)
		return;

2732 2733
	if (se->on_rq)
		cfs_rq->runnable_load_avg += contrib_delta;
2734 2735 2736 2737 2738 2739 2740 2741
	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.
 */
2742
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq, int force_update)
2743
{
2744
	u64 now = cfs_rq_clock_task(cfs_rq) >> 20;
2745 2746 2747
	u64 decays;

	decays = now - cfs_rq->last_decay;
2748
	if (!decays && !force_update)
2749 2750
		return;

2751 2752 2753
	if (atomic_long_read(&cfs_rq->removed_load)) {
		unsigned long removed_load;
		removed_load = atomic_long_xchg(&cfs_rq->removed_load, 0);
2754 2755
		subtract_blocked_load_contrib(cfs_rq, removed_load);
	}
2756

2757 2758 2759 2760 2761 2762
	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;
	}
2763 2764

	__update_cfs_rq_tg_load_contrib(cfs_rq, force_update);
2765
}
2766

2767 2768
/* 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,
2769 2770
						  struct sched_entity *se,
						  int wakeup)
2771
{
2772 2773 2774 2775
	/*
	 * 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.
2776 2777 2778 2779
	 *
	 * 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.
2780 2781
	 */
	if (unlikely(se->avg.decay_count <= 0)) {
2782
		se->avg.last_runnable_update = rq_clock_task(rq_of(cfs_rq));
2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797
		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;
		}
2798 2799
		wakeup = 0;
	} else {
2800
		__synchronize_entity_decay(se);
2801 2802
	}

2803 2804
	/* migrated tasks did not contribute to our blocked load */
	if (wakeup) {
2805
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
2806 2807
		update_entity_load_avg(se, 0);
	}
2808

2809
	cfs_rq->runnable_load_avg += se->avg.load_avg_contrib;
2810 2811
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !wakeup);
2812 2813
}

2814 2815 2816 2817 2818
/*
 * 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.
 */
2819
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2820 2821
						  struct sched_entity *se,
						  int sleep)
2822
{
2823
	update_entity_load_avg(se, 1);
2824 2825
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !sleep);
2826

2827
	cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib;
2828 2829 2830 2831
	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 */
2832
}
2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853

/*
 * 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);
}

2854 2855
static int idle_balance(struct rq *this_rq);

2856 2857
#else /* CONFIG_SMP */

2858 2859
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq) {}
2860
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2861
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2862 2863
					   struct sched_entity *se,
					   int wakeup) {}
2864
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2865 2866
					   struct sched_entity *se,
					   int sleep) {}
2867 2868
static inline void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
					      int force_update) {}
2869 2870 2871 2872 2873 2874

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

2875
#endif /* CONFIG_SMP */
2876

2877
static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
2878 2879
{
#ifdef CONFIG_SCHEDSTATS
2880 2881 2882 2883 2884
	struct task_struct *tsk = NULL;

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

2885
	if (se->statistics.sleep_start) {
2886
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.sleep_start;
2887 2888 2889 2890

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

2891 2892
		if (unlikely(delta > se->statistics.sleep_max))
			se->statistics.sleep_max = delta;
2893

2894
		se->statistics.sleep_start = 0;
2895
		se->statistics.sum_sleep_runtime += delta;
A
Arjan van de Ven 已提交
2896

2897
		if (tsk) {
2898
			account_scheduler_latency(tsk, delta >> 10, 1);
2899 2900
			trace_sched_stat_sleep(tsk, delta);
		}
2901
	}
2902
	if (se->statistics.block_start) {
2903
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.block_start;
2904 2905 2906 2907

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

2908 2909
		if (unlikely(delta > se->statistics.block_max))
			se->statistics.block_max = delta;
2910

2911
		se->statistics.block_start = 0;
2912
		se->statistics.sum_sleep_runtime += delta;
I
Ingo Molnar 已提交
2913

2914
		if (tsk) {
2915
			if (tsk->in_iowait) {
2916 2917
				se->statistics.iowait_sum += delta;
				se->statistics.iowait_count++;
2918
				trace_sched_stat_iowait(tsk, delta);
2919 2920
			}

2921 2922
			trace_sched_stat_blocked(tsk, delta);

2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933
			/*
			 * 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
2934
		}
2935 2936 2937 2938
	}
#endif
}

P
Peter Zijlstra 已提交
2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951
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
}

2952 2953 2954
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
2955
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
2956

2957 2958 2959 2960 2961 2962
	/*
	 * 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
2963
	if (initial && sched_feat(START_DEBIT))
2964
		vruntime += sched_vslice(cfs_rq, se);
2965

2966
	/* sleeps up to a single latency don't count. */
2967
	if (!initial) {
2968
		unsigned long thresh = sysctl_sched_latency;
2969

2970 2971 2972 2973 2974 2975
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
2976

2977
		vruntime -= thresh;
2978 2979
	}

2980
	/* ensure we never gain time by being placed backwards. */
2981
	se->vruntime = max_vruntime(se->vruntime, vruntime);
2982 2983
}

2984 2985
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

2986
static void
2987
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2988
{
2989 2990
	/*
	 * Update the normalized vruntime before updating min_vruntime
2991
	 * through calling update_curr().
2992
	 */
2993
	if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
2994 2995
		se->vruntime += cfs_rq->min_vruntime;

2996
	/*
2997
	 * Update run-time statistics of the 'current'.
2998
	 */
2999
	update_curr(cfs_rq);
3000
	enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP);
3001 3002
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
3003

3004
	if (flags & ENQUEUE_WAKEUP) {
3005
		place_entity(cfs_rq, se, 0);
3006
		enqueue_sleeper(cfs_rq, se);
I
Ingo Molnar 已提交
3007
	}
3008

3009
	update_stats_enqueue(cfs_rq, se);
P
Peter Zijlstra 已提交
3010
	check_spread(cfs_rq, se);
3011 3012
	if (se != cfs_rq->curr)
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3013
	se->on_rq = 1;
3014

3015
	if (cfs_rq->nr_running == 1) {
3016
		list_add_leaf_cfs_rq(cfs_rq);
3017 3018
		check_enqueue_throttle(cfs_rq);
	}
3019 3020
}

3021
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3022
{
3023 3024
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3025
		if (cfs_rq->last != se)
3026
			break;
3027 3028

		cfs_rq->last = NULL;
3029 3030
	}
}
P
Peter Zijlstra 已提交
3031

3032 3033 3034 3035
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3036
		if (cfs_rq->next != se)
3037
			break;
3038 3039

		cfs_rq->next = NULL;
3040
	}
P
Peter Zijlstra 已提交
3041 3042
}

3043 3044 3045 3046
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3047
		if (cfs_rq->skip != se)
3048
			break;
3049 3050

		cfs_rq->skip = NULL;
3051 3052 3053
	}
}

P
Peter Zijlstra 已提交
3054 3055
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3056 3057 3058 3059 3060
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
3061 3062 3063

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
3064 3065
}

3066
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3067

3068
static void
3069
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3070
{
3071 3072 3073 3074
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
3075
	dequeue_entity_load_avg(cfs_rq, se, flags & DEQUEUE_SLEEP);
3076

3077
	update_stats_dequeue(cfs_rq, se);
3078
	if (flags & DEQUEUE_SLEEP) {
P
Peter Zijlstra 已提交
3079
#ifdef CONFIG_SCHEDSTATS
3080 3081 3082 3083
		if (entity_is_task(se)) {
			struct task_struct *tsk = task_of(se);

			if (tsk->state & TASK_INTERRUPTIBLE)
3084
				se->statistics.sleep_start = rq_clock(rq_of(cfs_rq));
3085
			if (tsk->state & TASK_UNINTERRUPTIBLE)
3086
				se->statistics.block_start = rq_clock(rq_of(cfs_rq));
3087
		}
3088
#endif
P
Peter Zijlstra 已提交
3089 3090
	}

P
Peter Zijlstra 已提交
3091
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3092

3093
	if (se != cfs_rq->curr)
3094
		__dequeue_entity(cfs_rq, se);
3095
	se->on_rq = 0;
3096
	account_entity_dequeue(cfs_rq, se);
3097 3098 3099 3100 3101 3102

	/*
	 * 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.
	 */
3103
	if (!(flags & DEQUEUE_SLEEP))
3104
		se->vruntime -= cfs_rq->min_vruntime;
3105

3106 3107 3108
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

3109
	update_min_vruntime(cfs_rq);
3110
	update_cfs_shares(cfs_rq);
3111 3112 3113 3114 3115
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
3116
static void
I
Ingo Molnar 已提交
3117
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3118
{
3119
	unsigned long ideal_runtime, delta_exec;
3120 3121
	struct sched_entity *se;
	s64 delta;
3122

P
Peter Zijlstra 已提交
3123
	ideal_runtime = sched_slice(cfs_rq, curr);
3124
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
3125
	if (delta_exec > ideal_runtime) {
3126
		resched_curr(rq_of(cfs_rq));
3127 3128 3129 3130 3131
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142
		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;

3143 3144
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
3145

3146 3147
	if (delta < 0)
		return;
3148

3149
	if (delta > ideal_runtime)
3150
		resched_curr(rq_of(cfs_rq));
3151 3152
}

3153
static void
3154
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
3155
{
3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166
	/* '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);
	}

3167
	update_stats_curr_start(cfs_rq, se);
3168
	cfs_rq->curr = se;
I
Ingo Molnar 已提交
3169 3170 3171 3172 3173 3174
#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):
	 */
3175
	if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
3176
		se->statistics.slice_max = max(se->statistics.slice_max,
I
Ingo Molnar 已提交
3177 3178 3179
			se->sum_exec_runtime - se->prev_sum_exec_runtime);
	}
#endif
3180
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
3181 3182
}

3183 3184 3185
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

3186 3187 3188 3189 3190 3191 3192
/*
 * 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
 */
3193 3194
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3195
{
3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206
	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 */
3207

3208 3209 3210 3211 3212
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3213 3214 3215 3216 3217 3218 3219 3220 3221 3222
		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;
		}

3223 3224 3225
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3226

3227 3228 3229 3230 3231 3232
	/*
	 * 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;

3233 3234 3235 3236 3237 3238
	/*
	 * 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;

3239
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3240 3241

	return se;
3242 3243
}

3244
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3245

3246
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3247 3248 3249 3250 3251 3252
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3253
		update_curr(cfs_rq);
3254

3255 3256 3257
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

P
Peter Zijlstra 已提交
3258
	check_spread(cfs_rq, prev);
3259
	if (prev->on_rq) {
3260
		update_stats_wait_start(cfs_rq, prev);
3261 3262
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3263
		/* in !on_rq case, update occurred at dequeue */
3264
		update_entity_load_avg(prev, 1);
3265
	}
3266
	cfs_rq->curr = NULL;
3267 3268
}

P
Peter Zijlstra 已提交
3269 3270
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3271 3272
{
	/*
3273
	 * Update run-time statistics of the 'current'.
3274
	 */
3275
	update_curr(cfs_rq);
3276

3277 3278 3279
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3280
	update_entity_load_avg(curr, 1);
3281
	update_cfs_rq_blocked_load(cfs_rq, 1);
3282
	update_cfs_shares(cfs_rq);
3283

P
Peter Zijlstra 已提交
3284 3285 3286 3287 3288
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3289
	if (queued) {
3290
		resched_curr(rq_of(cfs_rq));
3291 3292
		return;
	}
P
Peter Zijlstra 已提交
3293 3294 3295 3296 3297 3298 3299 3300
	/*
	 * 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
3301
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3302
		check_preempt_tick(cfs_rq, curr);
3303 3304
}

3305 3306 3307 3308 3309 3310

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

#ifdef CONFIG_CFS_BANDWIDTH
3311 3312

#ifdef HAVE_JUMP_LABEL
3313
static struct static_key __cfs_bandwidth_used;
3314 3315 3316

static inline bool cfs_bandwidth_used(void)
{
3317
	return static_key_false(&__cfs_bandwidth_used);
3318 3319
}

3320
void cfs_bandwidth_usage_inc(void)
3321
{
3322 3323 3324 3325 3326 3327
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3328 3329 3330 3331 3332 3333 3334
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3335 3336
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3337 3338
#endif /* HAVE_JUMP_LABEL */

3339 3340 3341 3342 3343 3344 3345 3346
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3347 3348 3349 3350 3351 3352

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

P
Paul Turner 已提交
3353 3354 3355 3356 3357 3358 3359
/*
 * 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
 */
3360
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371
{
	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);
}

3372 3373 3374 3375 3376
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3377 3378 3379 3380 3381 3382
/* 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;

3383
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3384 3385
}

3386 3387
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3388 3389 3390
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3391
	u64 amount = 0, min_amount, expires;
3392 3393 3394 3395 3396 3397 3398

	/* 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;
3399
	else {
P
Paul Turner 已提交
3400 3401 3402 3403 3404 3405 3406 3407
		/*
		 * 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);
3408
			__start_cfs_bandwidth(cfs_b, false);
P
Paul Turner 已提交
3409
		}
3410 3411 3412 3413 3414 3415

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3416
	}
P
Paul Turner 已提交
3417
	expires = cfs_b->runtime_expires;
3418 3419 3420
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3421 3422 3423 3424 3425 3426 3427
	/*
	 * 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;
3428 3429

	return cfs_rq->runtime_remaining > 0;
3430 3431
}

P
Paul Turner 已提交
3432 3433 3434 3435 3436
/*
 * 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)
3437
{
P
Paul Turner 已提交
3438 3439 3440
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3441
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3442 3443
		return;

P
Paul Turner 已提交
3444 3445 3446 3447 3448 3449 3450 3451 3452
	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
3453 3454 3455
	 * 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
3456 3457
	 */

3458
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3459 3460 3461 3462 3463 3464 3465 3466
		/* 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;
	}
}

3467
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3468 3469
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3470
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3471 3472 3473
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3474 3475
		return;

3476 3477 3478 3479 3480
	/*
	 * 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))
3481
		resched_curr(rq_of(cfs_rq));
3482 3483
}

3484
static __always_inline
3485
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3486
{
3487
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3488 3489 3490 3491 3492
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3493 3494
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3495
	return cfs_bandwidth_used() && cfs_rq->throttled;
3496 3497
}

3498 3499 3500
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3501
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529
}

/*
 * 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) {
3530
		/* adjust cfs_rq_clock_task() */
3531
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3532
					     cfs_rq->throttled_clock_task;
3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543
	}
#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)];

3544 3545
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3546
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3547 3548 3549 3550 3551
	cfs_rq->throttle_count++;

	return 0;
}

3552
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3553 3554 3555 3556 3557 3558 3559 3560
{
	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))];

3561
	/* freeze hierarchy runnable averages while throttled */
3562 3563 3564
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581

	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)
3582
		sub_nr_running(rq, task_delta);
3583 3584

	cfs_rq->throttled = 1;
3585
	cfs_rq->throttled_clock = rq_clock(rq);
3586
	raw_spin_lock(&cfs_b->lock);
3587 3588 3589 3590 3591
	/*
	 * 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);
3592
	if (!cfs_b->timer_active)
3593
		__start_cfs_bandwidth(cfs_b, false);
3594 3595 3596
	raw_spin_unlock(&cfs_b->lock);
}

3597
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3598 3599 3600 3601 3602 3603 3604
{
	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;

3605
	se = cfs_rq->tg->se[cpu_of(rq)];
3606 3607

	cfs_rq->throttled = 0;
3608 3609 3610

	update_rq_clock(rq);

3611
	raw_spin_lock(&cfs_b->lock);
3612
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
3613 3614 3615
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

3616 3617 3618
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636
	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)
3637
		add_nr_running(rq, task_delta);
3638 3639 3640

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
3641
		resched_curr(rq);
3642 3643 3644 3645 3646 3647
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
3648 3649
	u64 runtime;
	u64 starting_runtime = remaining;
3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679

	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();

3680
	return starting_runtime - remaining;
3681 3682
}

3683 3684 3685 3686 3687 3688 3689 3690
/*
 * 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)
{
3691
	u64 runtime, runtime_expires;
3692
	int throttled;
3693 3694 3695

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
3696
		goto out_deactivate;
3697

3698
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
3699
	cfs_b->nr_periods += overrun;
3700

3701 3702 3703 3704 3705 3706
	/*
	 * 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
3707

3708 3709 3710 3711 3712 3713 3714
	/*
	 * 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
3715 3716
	__refill_cfs_bandwidth_runtime(cfs_b);

3717 3718 3719
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
3720
		return 0;
3721 3722
	}

3723 3724 3725
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

3726 3727 3728
	runtime_expires = cfs_b->runtime_expires;

	/*
3729 3730 3731 3732 3733
	 * 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.
3734
	 */
3735 3736
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
3737 3738 3739 3740 3741 3742 3743
		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);
3744 3745

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3746
	}
3747

3748 3749 3750 3751 3752 3753 3754
	/*
	 * 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;
3755

3756 3757 3758 3759 3760
	return 0;

out_deactivate:
	cfs_b->timer_active = 0;
	return 1;
3761
}
3762

3763 3764 3765 3766 3767 3768 3769
/* 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;

3770 3771 3772 3773 3774 3775 3776
/*
 * 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.
 */
3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832
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)
{
3833 3834 3835
	if (!cfs_bandwidth_used())
		return;

3836
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851
		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 */
3852 3853 3854
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
3855
		return;
3856
	}
3857

3858
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
3859
		runtime = cfs_b->runtime;
3860

3861 3862 3863 3864 3865 3866 3867 3868 3869 3870
	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)
3871
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3872 3873 3874
	raw_spin_unlock(&cfs_b->lock);
}

3875 3876 3877 3878 3879 3880 3881
/*
 * 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)
{
3882 3883 3884
	if (!cfs_bandwidth_used())
		return;

3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899
	/* 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() */
3900
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3901
{
3902
	if (!cfs_bandwidth_used())
3903
		return false;
3904

3905
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
3906
		return false;
3907 3908 3909 3910 3911 3912

	/*
	 * 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))
3913
		return true;
3914 3915

	throttle_cfs_rq(cfs_rq);
3916
	return true;
3917
}
3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935

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;

3936
	raw_spin_lock(&cfs_b->lock);
3937 3938 3939 3940 3941 3942 3943 3944 3945
	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);
	}
3946
	raw_spin_unlock(&cfs_b->lock);
3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971

	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 */
3972
void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force)
3973 3974 3975 3976 3977 3978 3979
{
	/*
	 * 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
	 */
3980 3981 3982
	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 */
3983
		raw_spin_unlock(&cfs_b->lock);
3984
		cpu_relax();
3985 3986
		raw_spin_lock(&cfs_b->lock);
		/* if someone else restarted the timer then we're done */
3987
		if (!force && cfs_b->timer_active)
3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000
			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);
}

4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013
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);
	}
}

4014
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025
{
	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
		 */
4026
		cfs_rq->runtime_remaining = 1;
4027 4028 4029 4030 4031 4032
		/*
		 * 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;

4033 4034 4035 4036 4037 4038
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
4039 4040
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
4041
	return rq_clock_task(rq_of(cfs_rq));
4042 4043
}

4044
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
4045
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
4046
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
4047
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
4048 4049 4050 4051 4052

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063

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;
}
4064 4065 4066 4067 4068

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) {}
4069 4070
#endif

4071 4072 4073 4074 4075
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) {}
4076
static inline void update_runtime_enabled(struct rq *rq) {}
4077
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
4078 4079 4080

#endif /* CONFIG_CFS_BANDWIDTH */

4081 4082 4083 4084
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
4085 4086 4087 4088 4089 4090 4091 4092
#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);

4093
	if (cfs_rq->nr_running > 1) {
P
Peter Zijlstra 已提交
4094 4095 4096 4097 4098 4099
		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)
4100
				resched_curr(rq);
P
Peter Zijlstra 已提交
4101 4102
			return;
		}
4103
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
4104 4105
	}
}
4106 4107 4108 4109 4110 4111 4112 4113 4114 4115

/*
 * 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;

4116
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
4117 4118 4119 4120 4121
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
4122
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
4123 4124 4125 4126
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
4127 4128 4129 4130

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
4131 4132
#endif

4133 4134 4135 4136 4137
/*
 * 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:
 */
4138
static void
4139
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4140 4141
{
	struct cfs_rq *cfs_rq;
4142
	struct sched_entity *se = &p->se;
4143 4144

	for_each_sched_entity(se) {
4145
		if (se->on_rq)
4146 4147
			break;
		cfs_rq = cfs_rq_of(se);
4148
		enqueue_entity(cfs_rq, se, flags);
4149 4150 4151 4152 4153 4154 4155 4156 4157

		/*
		 * 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;
4158
		cfs_rq->h_nr_running++;
4159

4160
		flags = ENQUEUE_WAKEUP;
4161
	}
P
Peter Zijlstra 已提交
4162

P
Peter Zijlstra 已提交
4163
	for_each_sched_entity(se) {
4164
		cfs_rq = cfs_rq_of(se);
4165
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
4166

4167 4168 4169
		if (cfs_rq_throttled(cfs_rq))
			break;

4170
		update_cfs_shares(cfs_rq);
4171
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
4172 4173
	}

4174 4175
	if (!se) {
		update_rq_runnable_avg(rq, rq->nr_running);
4176
		add_nr_running(rq, 1);
4177
	}
4178
	hrtick_update(rq);
4179 4180
}

4181 4182
static void set_next_buddy(struct sched_entity *se);

4183 4184 4185 4186 4187
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
4188
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4189 4190
{
	struct cfs_rq *cfs_rq;
4191
	struct sched_entity *se = &p->se;
4192
	int task_sleep = flags & DEQUEUE_SLEEP;
4193 4194 4195

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4196
		dequeue_entity(cfs_rq, se, flags);
4197 4198 4199 4200 4201 4202 4203 4204 4205

		/*
		 * 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;
4206
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4207

4208
		/* Don't dequeue parent if it has other entities besides us */
4209 4210 4211 4212 4213 4214 4215
		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));
4216 4217 4218

			/* avoid re-evaluating load for this entity */
			se = parent_entity(se);
4219
			break;
4220
		}
4221
		flags |= DEQUEUE_SLEEP;
4222
	}
P
Peter Zijlstra 已提交
4223

P
Peter Zijlstra 已提交
4224
	for_each_sched_entity(se) {
4225
		cfs_rq = cfs_rq_of(se);
4226
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4227

4228 4229 4230
		if (cfs_rq_throttled(cfs_rq))
			break;

4231
		update_cfs_shares(cfs_rq);
4232
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
4233 4234
	}

4235
	if (!se) {
4236
		sub_nr_running(rq, 1);
4237 4238
		update_rq_runnable_avg(rq, 1);
	}
4239
	hrtick_update(rq);
4240 4241
}

4242
#ifdef CONFIG_SMP
4243 4244 4245
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
4246
	return cpu_rq(cpu)->cfs.runnable_load_avg;
4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281
}

/*
 * 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);
}

4282
static unsigned long capacity_of(int cpu)
4283
{
4284
	return cpu_rq(cpu)->cpu_capacity;
4285 4286 4287 4288 4289
}

static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
4290
	unsigned long nr_running = ACCESS_ONCE(rq->cfs.h_nr_running);
4291
	unsigned long load_avg = rq->cfs.runnable_load_avg;
4292 4293

	if (nr_running)
4294
		return load_avg / nr_running;
4295 4296 4297 4298

	return 0;
}

4299 4300 4301 4302 4303 4304 4305
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.
	 */
4306
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
4307
		current->wakee_flips >>= 1;
4308 4309 4310 4311 4312 4313 4314 4315
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}
4316

4317
static void task_waking_fair(struct task_struct *p)
4318 4319 4320
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
4321 4322 4323 4324
	u64 min_vruntime;

#ifndef CONFIG_64BIT
	u64 min_vruntime_copy;
4325

4326 4327 4328 4329 4330 4331 4332 4333
	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
4334

4335
	se->vruntime -= min_vruntime;
4336
	record_wakee(p);
4337 4338
}

4339
#ifdef CONFIG_FAIR_GROUP_SCHED
4340 4341 4342 4343 4344 4345
/*
 * 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.
4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388
 *
 * 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.
4389
 */
P
Peter Zijlstra 已提交
4390
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
4391
{
P
Peter Zijlstra 已提交
4392
	struct sched_entity *se = tg->se[cpu];
4393

4394
	if (!tg->parent)	/* the trivial, non-cgroup case */
4395 4396
		return wl;

P
Peter Zijlstra 已提交
4397
	for_each_sched_entity(se) {
4398
		long w, W;
P
Peter Zijlstra 已提交
4399

4400
		tg = se->my_q->tg;
4401

4402 4403 4404 4405
		/*
		 * W = @wg + \Sum rw_j
		 */
		W = wg + calc_tg_weight(tg, se->my_q);
P
Peter Zijlstra 已提交
4406

4407 4408 4409 4410
		/*
		 * w = rw_i + @wl
		 */
		w = se->my_q->load.weight + wl;
4411

4412 4413 4414 4415 4416
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
			wl = (w * tg->shares) / W;
4417 4418
		else
			wl = tg->shares;
4419

4420 4421 4422 4423 4424
		/*
		 * 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().
		 */
4425 4426
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
4427 4428 4429 4430

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
4431
		wl -= se->load.weight;
4432 4433 4434 4435 4436 4437 4438 4439

		/*
		 * 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
4440 4441
		wg = 0;
	}
4442

P
Peter Zijlstra 已提交
4443
	return wl;
4444 4445
}
#else
P
Peter Zijlstra 已提交
4446

4447
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
4448
{
4449
	return wl;
4450
}
P
Peter Zijlstra 已提交
4451

4452 4453
#endif

4454 4455
static int wake_wide(struct task_struct *p)
{
4456
	int factor = this_cpu_read(sd_llc_size);
4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475

	/*
	 * 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;
}

4476
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
4477
{
4478
	s64 this_load, load;
4479
	s64 this_eff_load, prev_eff_load;
4480 4481
	int idx, this_cpu, prev_cpu;
	struct task_group *tg;
4482
	unsigned long weight;
4483
	int balanced;
4484

4485 4486 4487 4488 4489 4490 4491
	/*
	 * If we wake multiple tasks be careful to not bounce
	 * ourselves around too much.
	 */
	if (wake_wide(p))
		return 0;

4492 4493 4494 4495 4496
	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);
4497

4498 4499 4500 4501 4502
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
4503 4504 4505 4506
	if (sync) {
		tg = task_group(current);
		weight = current->se.load.weight;

4507
		this_load += effective_load(tg, this_cpu, -weight, -weight);
4508 4509
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
4510

4511 4512
	tg = task_group(p);
	weight = p->se.load.weight;
4513

4514 4515
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
4516 4517 4518
	 * 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.
4519 4520 4521 4522
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
4523 4524
	this_eff_load = 100;
	this_eff_load *= capacity_of(prev_cpu);
4525

4526 4527
	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
4528

4529
	if (this_load > 0) {
4530 4531 4532 4533
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
4534
	}
4535

4536
	balanced = this_eff_load <= prev_eff_load;
4537

4538
	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
4539

4540 4541
	if (!balanced)
		return 0;
4542

4543 4544 4545 4546
	schedstat_inc(sd, ttwu_move_affine);
	schedstat_inc(p, se.statistics.nr_wakeups_affine);

	return 1;
4547 4548
}

4549 4550 4551 4552 4553
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
4554
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
4555
		  int this_cpu, int sd_flag)
4556
{
4557
	struct sched_group *idlest = NULL, *group = sd->groups;
4558
	unsigned long min_load = ULONG_MAX, this_load = 0;
4559
	int load_idx = sd->forkexec_idx;
4560
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
4561

4562 4563 4564
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

4565 4566 4567 4568
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
4569

4570 4571
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
4572
					tsk_cpus_allowed(p)))
4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590
			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;
		}

4591
		/* Adjust by relative CPU capacity of the group */
4592
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613

		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;
4614 4615 4616 4617
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
4618 4619 4620
	int i;

	/* Traverse only the allowed CPUs */
4621
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649
		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;
			}
4650 4651 4652
		}
	}

4653
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
4654
}
4655

4656 4657 4658
/*
 * Try and locate an idle CPU in the sched_domain.
 */
4659
static int select_idle_sibling(struct task_struct *p, int target)
4660
{
4661
	struct sched_domain *sd;
4662
	struct sched_group *sg;
4663
	int i = task_cpu(p);
4664

4665 4666
	if (idle_cpu(target))
		return target;
4667 4668

	/*
4669
	 * If the prevous cpu is cache affine and idle, don't be stupid.
4670
	 */
4671 4672
	if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
		return i;
4673 4674

	/*
4675
	 * Otherwise, iterate the domains and find an elegible idle cpu.
4676
	 */
4677
	sd = rcu_dereference(per_cpu(sd_llc, target));
4678
	for_each_lower_domain(sd) {
4679 4680 4681 4682 4683 4684 4685
		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)) {
4686
				if (i == target || !idle_cpu(i))
4687 4688
					goto next;
			}
4689

4690 4691 4692 4693 4694 4695 4696 4697
			target = cpumask_first_and(sched_group_cpus(sg),
					tsk_cpus_allowed(p));
			goto done;
next:
			sg = sg->next;
		} while (sg != sd->groups);
	}
done:
4698 4699 4700
	return target;
}

4701
/*
4702 4703 4704
 * 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.
4705
 *
4706 4707
 * 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.
4708
 *
4709
 * Returns the target cpu number.
4710 4711 4712
 *
 * preempt must be disabled.
 */
4713
static int
4714
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
4715
{
4716
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
4717 4718
	int cpu = smp_processor_id();
	int new_cpu = cpu;
4719
	int want_affine = 0;
4720
	int sync = wake_flags & WF_SYNC;
4721

4722
	if (p->nr_cpus_allowed == 1)
4723 4724
		return prev_cpu;

4725 4726
	if (sd_flag & SD_BALANCE_WAKE)
		want_affine = cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
4727

4728
	rcu_read_lock();
4729
	for_each_domain(cpu, tmp) {
4730 4731 4732
		if (!(tmp->flags & SD_LOAD_BALANCE))
			continue;

4733
		/*
4734 4735
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
4736
		 */
4737 4738 4739
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
4740
			break;
4741
		}
4742

4743
		if (tmp->flags & sd_flag)
4744 4745 4746
			sd = tmp;
	}

4747 4748
	if (affine_sd && cpu != prev_cpu && wake_affine(affine_sd, p, sync))
		prev_cpu = cpu;
4749

4750
	if (sd_flag & SD_BALANCE_WAKE) {
4751 4752
		new_cpu = select_idle_sibling(p, prev_cpu);
		goto unlock;
4753
	}
4754

4755 4756
	while (sd) {
		struct sched_group *group;
4757
		int weight;
4758

4759
		if (!(sd->flags & sd_flag)) {
4760 4761 4762
			sd = sd->child;
			continue;
		}
4763

4764
		group = find_idlest_group(sd, p, cpu, sd_flag);
4765 4766 4767 4768
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
4769

4770
		new_cpu = find_idlest_cpu(group, p, cpu);
4771 4772 4773 4774
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
4775
		}
4776 4777 4778

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
4779
		weight = sd->span_weight;
4780 4781
		sd = NULL;
		for_each_domain(cpu, tmp) {
4782
			if (weight <= tmp->span_weight)
4783
				break;
4784
			if (tmp->flags & sd_flag)
4785 4786 4787
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
4788
	}
4789 4790
unlock:
	rcu_read_unlock();
4791

4792
	return new_cpu;
4793
}
4794 4795 4796 4797 4798 4799 4800 4801 4802 4803

/*
 * 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)
{
4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814
	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);
4815 4816
		atomic_long_add(se->avg.load_avg_contrib,
						&cfs_rq->removed_load);
4817
	}
4818 4819 4820

	/* We have migrated, no longer consider this task hot */
	se->exec_start = 0;
4821
}
4822 4823
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
4824 4825
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
4826 4827 4828 4829
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
4830 4831
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
4832 4833 4834 4835 4836 4837 4838 4839 4840
	 *
	 * 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.
4841
	 */
4842
	return calc_delta_fair(gran, se);
4843 4844
}

4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866
/*
 * 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
4867
	gran = wakeup_gran(curr, se);
4868 4869 4870 4871 4872 4873
	if (vdiff > gran)
		return 1;

	return 0;
}

4874 4875
static void set_last_buddy(struct sched_entity *se)
{
4876 4877 4878 4879 4880
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
4881 4882 4883 4884
}

static void set_next_buddy(struct sched_entity *se)
{
4885 4886 4887 4888 4889
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
4890 4891
}

4892 4893
static void set_skip_buddy(struct sched_entity *se)
{
4894 4895
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
4896 4897
}

4898 4899 4900
/*
 * Preempt the current task with a newly woken task if needed:
 */
4901
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
4902 4903
{
	struct task_struct *curr = rq->curr;
4904
	struct sched_entity *se = &curr->se, *pse = &p->se;
4905
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
4906
	int scale = cfs_rq->nr_running >= sched_nr_latency;
4907
	int next_buddy_marked = 0;
4908

I
Ingo Molnar 已提交
4909 4910 4911
	if (unlikely(se == pse))
		return;

4912
	/*
4913
	 * This is possible from callers such as attach_tasks(), in which we
4914 4915 4916 4917 4918 4919 4920
	 * 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;

4921
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
4922
		set_next_buddy(pse);
4923 4924
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
4925

4926 4927 4928
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
4929 4930 4931 4932 4933 4934
	 *
	 * 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.
4935 4936 4937 4938
	 */
	if (test_tsk_need_resched(curr))
		return;

4939 4940 4941 4942 4943
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

4944
	/*
4945 4946
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
4947
	 */
4948
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
4949
		return;
4950

4951
	find_matching_se(&se, &pse);
4952
	update_curr(cfs_rq_of(se));
4953
	BUG_ON(!pse);
4954 4955 4956 4957 4958 4959 4960
	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);
4961
		goto preempt;
4962
	}
4963

4964
	return;
4965

4966
preempt:
4967
	resched_curr(rq);
4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981
	/*
	 * 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);
4982 4983
}

4984 4985
static struct task_struct *
pick_next_task_fair(struct rq *rq, struct task_struct *prev)
4986 4987 4988
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
4989
	struct task_struct *p;
4990
	int new_tasks;
4991

4992
again:
4993 4994
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
4995
		goto idle;
4996

4997
	if (prev->sched_class != &fair_sched_class)
4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068
		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
5069

5070
	if (!cfs_rq->nr_running)
5071
		goto idle;
5072

5073
	put_prev_task(rq, prev);
5074

5075
	do {
5076
		se = pick_next_entity(cfs_rq, NULL);
5077
		set_next_entity(cfs_rq, se);
5078 5079 5080
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
5081
	p = task_of(se);
5082

5083 5084
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
5085 5086

	return p;
5087 5088

idle:
5089
	new_tasks = idle_balance(rq);
5090 5091 5092 5093 5094
	/*
	 * 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.
	 */
5095
	if (new_tasks < 0)
5096 5097
		return RETRY_TASK;

5098
	if (new_tasks > 0)
5099 5100 5101
		goto again;

	return NULL;
5102 5103 5104 5105 5106
}

/*
 * Account for a descheduled task:
 */
5107
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
5108 5109 5110 5111 5112 5113
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5114
		put_prev_entity(cfs_rq, se);
5115 5116 5117
	}
}

5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142
/*
 * 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);
5143 5144 5145 5146 5147 5148
		/*
		 * 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;
5149 5150 5151 5152 5153
	}

	set_skip_buddy(se);
}

5154 5155 5156 5157
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

5158 5159
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
5160 5161 5162 5163 5164 5165 5166 5167 5168 5169
		return false;

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

	yield_task_fair(rq);

	return true;
}

5170
#ifdef CONFIG_SMP
5171
/**************************************************
P
Peter Zijlstra 已提交
5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194
 * 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)
 *
5195
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
5196 5197 5198 5199 5200 5201
 * 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):
 *
5202
 *   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
5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287
 *
 * 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.]
 */ 
5288

5289 5290
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

5291 5292
enum fbq_type { regular, remote, all };

5293
#define LBF_ALL_PINNED	0x01
5294
#define LBF_NEED_BREAK	0x02
5295 5296
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
5297 5298 5299 5300 5301

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
5302
	int			src_cpu;
5303 5304 5305 5306

	int			dst_cpu;
	struct rq		*dst_rq;

5307 5308
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
5309
	enum cpu_idle_type	idle;
5310
	long			imbalance;
5311 5312 5313
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

5314
	unsigned int		flags;
5315 5316 5317 5318

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
5319 5320

	enum fbq_type		fbq_type;
5321
	struct list_head	tasks;
5322 5323
};

5324 5325 5326
/*
 * Is this task likely cache-hot:
 */
5327
static int task_hot(struct task_struct *p, struct lb_env *env)
5328 5329 5330
{
	s64 delta;

5331 5332
	lockdep_assert_held(&env->src_rq->lock);

5333 5334 5335 5336 5337 5338 5339 5340 5341
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
5342
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
5343 5344 5345 5346 5347 5348 5349 5350 5351
			(&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;

5352
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
5353 5354 5355 5356

	return delta < (s64)sysctl_sched_migration_cost;
}

5357 5358 5359 5360
#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)
{
5361
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5362 5363
	int src_nid, dst_nid;

5364
	if (!sched_feat(NUMA_FAVOUR_HIGHER) || !p->numa_faults_memory ||
5365 5366 5367 5368 5369 5370 5371
	    !(env->sd->flags & SD_NUMA)) {
		return false;
	}

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

5372
	if (src_nid == dst_nid)
5373 5374
		return false;

5375 5376 5377 5378
	if (numa_group) {
		/* Task is already in the group's interleave set. */
		if (node_isset(src_nid, numa_group->active_nodes))
			return false;
5379

5380 5381 5382
		/* Task is moving into the group's interleave set. */
		if (node_isset(dst_nid, numa_group->active_nodes))
			return true;
5383

5384 5385 5386 5387 5388
		return group_faults(p, dst_nid) > group_faults(p, src_nid);
	}

	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
5389 5390
		return true;

5391
	return task_faults(p, dst_nid) > task_faults(p, src_nid);
5392
}
5393 5394 5395 5396


static bool migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
{
5397
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5398 5399 5400 5401 5402
	int src_nid, dst_nid;

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

5403
	if (!p->numa_faults_memory || !(env->sd->flags & SD_NUMA))
5404 5405 5406 5407 5408
		return false;

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

5409
	if (src_nid == dst_nid)
5410 5411
		return false;

5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423
	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);
	}

5424 5425 5426 5427
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid)
		return true;

5428
	return task_faults(p, dst_nid) < task_faults(p, src_nid);
5429 5430
}

5431 5432 5433 5434 5435 5436
#else
static inline bool migrate_improves_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5437 5438 5439 5440 5441 5442

static inline bool migrate_degrades_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5443 5444
#endif

5445 5446 5447 5448
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
5449
int can_migrate_task(struct task_struct *p, struct lb_env *env)
5450 5451
{
	int tsk_cache_hot = 0;
5452 5453 5454

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

5455 5456
	/*
	 * We do not migrate tasks that are:
5457
	 * 1) throttled_lb_pair, or
5458
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
5459 5460
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
5461
	 */
5462 5463 5464
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

5465
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
5466
		int cpu;
5467

5468
		schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
5469

5470 5471
		env->flags |= LBF_SOME_PINNED;

5472 5473 5474 5475 5476 5477 5478 5479
		/*
		 * 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.
		 */
5480
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
5481 5482
			return 0;

5483 5484 5485
		/* 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))) {
5486
				env->flags |= LBF_DST_PINNED;
5487 5488 5489
				env->new_dst_cpu = cpu;
				break;
			}
5490
		}
5491

5492 5493
		return 0;
	}
5494 5495

	/* Record that we found atleast one task that could run on dst_cpu */
5496
	env->flags &= ~LBF_ALL_PINNED;
5497

5498
	if (task_running(env->src_rq, p)) {
5499
		schedstat_inc(p, se.statistics.nr_failed_migrations_running);
5500 5501 5502 5503 5504
		return 0;
	}

	/*
	 * Aggressive migration if:
5505 5506 5507
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
5508
	 */
5509
	tsk_cache_hot = task_hot(p, env);
5510 5511
	if (!tsk_cache_hot)
		tsk_cache_hot = migrate_degrades_locality(p, env);
5512

5513 5514
	if (migrate_improves_locality(p, env) || !tsk_cache_hot ||
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
5515 5516 5517 5518
		if (tsk_cache_hot) {
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
			schedstat_inc(p, se.statistics.nr_forced_migrations);
		}
5519 5520 5521
		return 1;
	}

Z
Zhang Hang 已提交
5522 5523
	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
	return 0;
5524 5525
}

5526
/*
5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537
 * 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);
}

5538
/*
5539
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
5540 5541
 * part of active balancing operations within "domain".
 *
5542
 * Returns a task if successful and NULL otherwise.
5543
 */
5544
static struct task_struct *detach_one_task(struct lb_env *env)
5545 5546 5547
{
	struct task_struct *p, *n;

5548 5549
	lockdep_assert_held(&env->src_rq->lock);

5550 5551 5552
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
5553

5554
		detach_task(p, env);
5555

5556
		/*
5557
		 * Right now, this is only the second place where
5558
		 * lb_gained[env->idle] is updated (other is detach_tasks)
5559
		 * so we can safely collect stats here rather than
5560
		 * inside detach_tasks().
5561 5562
		 */
		schedstat_inc(env->sd, lb_gained[env->idle]);
5563
		return p;
5564
	}
5565
	return NULL;
5566 5567
}

5568 5569
static const unsigned int sched_nr_migrate_break = 32;

5570
/*
5571 5572
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
5573
 *
5574
 * Returns number of detached tasks if successful and 0 otherwise.
5575
 */
5576
static int detach_tasks(struct lb_env *env)
5577
{
5578 5579
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
5580
	unsigned long load;
5581 5582 5583
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
5584

5585
	if (env->imbalance <= 0)
5586
		return 0;
5587

5588 5589
	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
5590

5591 5592
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
5593
		if (env->loop > env->loop_max)
5594
			break;
5595 5596

		/* take a breather every nr_migrate tasks */
5597
		if (env->loop > env->loop_break) {
5598
			env->loop_break += sched_nr_migrate_break;
5599
			env->flags |= LBF_NEED_BREAK;
5600
			break;
5601
		}
5602

5603
		if (!can_migrate_task(p, env))
5604 5605 5606
			goto next;

		load = task_h_load(p);
5607

5608
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
5609 5610
			goto next;

5611
		if ((load / 2) > env->imbalance)
5612
			goto next;
5613

5614 5615 5616 5617
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
5618
		env->imbalance -= load;
5619 5620

#ifdef CONFIG_PREEMPT
5621 5622
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
5623
		 * kernels will stop after the first task is detached to minimize
5624 5625
		 * the critical section.
		 */
5626
		if (env->idle == CPU_NEWLY_IDLE)
5627
			break;
5628 5629
#endif

5630 5631 5632 5633
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
5634
		if (env->imbalance <= 0)
5635
			break;
5636 5637 5638

		continue;
next:
5639
		list_move_tail(&p->se.group_node, tasks);
5640
	}
5641

5642
	/*
5643 5644 5645
	 * 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().
5646
	 */
5647
	schedstat_add(env->sd, lb_gained[env->idle], detached);
5648

5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689
	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);
5690

5691 5692 5693 5694
		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
5695 5696
}

P
Peter Zijlstra 已提交
5697
#ifdef CONFIG_FAIR_GROUP_SCHED
5698 5699 5700
/*
 * update tg->load_weight by folding this cpu's load_avg
 */
5701
static void __update_blocked_averages_cpu(struct task_group *tg, int cpu)
5702
{
5703 5704
	struct sched_entity *se = tg->se[cpu];
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
5705

5706 5707 5708
	/* throttled entities do not contribute to load */
	if (throttled_hierarchy(cfs_rq))
		return;
5709

5710
	update_cfs_rq_blocked_load(cfs_rq, 1);
5711

5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725
	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 {
5726
		struct rq *rq = rq_of(cfs_rq);
5727 5728
		update_rq_runnable_avg(rq, rq->nr_running);
	}
5729 5730
}

5731
static void update_blocked_averages(int cpu)
5732 5733
{
	struct rq *rq = cpu_rq(cpu);
5734 5735
	struct cfs_rq *cfs_rq;
	unsigned long flags;
5736

5737 5738
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
5739 5740 5741 5742
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
5743
	for_each_leaf_cfs_rq(rq, cfs_rq) {
5744 5745 5746 5747 5748 5749
		/*
		 * 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);
5750
	}
5751 5752

	raw_spin_unlock_irqrestore(&rq->lock, flags);
5753 5754
}

5755
/*
5756
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
5757 5758 5759
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
5760
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
5761
{
5762 5763
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
5764
	unsigned long now = jiffies;
5765
	unsigned long load;
5766

5767
	if (cfs_rq->last_h_load_update == now)
5768 5769
		return;

5770 5771 5772 5773 5774 5775 5776
	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;
	}
5777

5778
	if (!se) {
5779
		cfs_rq->h_load = cfs_rq->runnable_load_avg;
5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790
		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;
	}
5791 5792
}

5793
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
5794
{
5795
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
5796

5797
	update_cfs_rq_h_load(cfs_rq);
5798 5799
	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
5800 5801
}
#else
5802
static inline void update_blocked_averages(int cpu)
5803 5804 5805
{
}

5806
static unsigned long task_h_load(struct task_struct *p)
5807
{
5808
	return p->se.avg.load_avg_contrib;
5809
}
P
Peter Zijlstra 已提交
5810
#endif
5811 5812

/********** Helpers for find_busiest_group ************************/
5813 5814 5815 5816 5817 5818 5819

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

5820 5821 5822 5823 5824 5825 5826
/*
 * 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
5827
	unsigned long load_per_task;
5828
	unsigned long group_capacity;
5829
	unsigned int sum_nr_running; /* Nr tasks running in the group */
5830
	unsigned int group_capacity_factor;
5831 5832
	unsigned int idle_cpus;
	unsigned int group_weight;
5833
	enum group_type group_type;
5834
	int group_has_free_capacity;
5835 5836 5837 5838
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
5839 5840
};

J
Joonsoo Kim 已提交
5841 5842 5843 5844 5845 5846 5847 5848
/*
 * 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 */
5849
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
5850 5851 5852
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
5853
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
5854 5855
};

5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867
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,
5868
		.total_capacity = 0UL,
5869 5870
		.busiest_stat = {
			.avg_load = 0UL,
5871 5872
			.sum_nr_running = 0,
			.group_type = group_other,
5873 5874 5875 5876
		},
	};
}

5877 5878 5879
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
5880
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
5881 5882
 *
 * Return: The load index.
5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904
 */
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;
}

5905
static unsigned long default_scale_capacity(struct sched_domain *sd, int cpu)
5906
{
5907
	return SCHED_CAPACITY_SCALE;
5908 5909
}

5910
unsigned long __weak arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
5911
{
5912
	return default_scale_capacity(sd, cpu);
5913 5914
}

5915
static unsigned long default_scale_cpu_capacity(struct sched_domain *sd, int cpu)
5916
{
5917 5918
	if ((sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
		return sd->smt_gain / sd->span_weight;
5919

5920
	return SCHED_CAPACITY_SCALE;
5921 5922
}

5923
unsigned long __weak arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
5924
{
5925
	return default_scale_cpu_capacity(sd, cpu);
5926 5927
}

5928
static unsigned long scale_rt_capacity(int cpu)
5929 5930
{
	struct rq *rq = cpu_rq(cpu);
5931
	u64 total, available, age_stamp, avg;
5932
	s64 delta;
5933

5934 5935 5936 5937 5938 5939 5940
	/*
	 * 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);

5941 5942 5943 5944 5945
	delta = rq_clock(rq) - age_stamp;
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
5946

5947
	if (unlikely(total < avg)) {
5948
		/* Ensures that capacity won't end up being negative */
5949 5950
		available = 0;
	} else {
5951
		available = total - avg;
5952
	}
5953

5954 5955
	if (unlikely((s64)total < SCHED_CAPACITY_SCALE))
		total = SCHED_CAPACITY_SCALE;
5956

5957
	total >>= SCHED_CAPACITY_SHIFT;
5958 5959 5960 5961

	return div_u64(available, total);
}

5962
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
5963
{
5964
	unsigned long capacity = SCHED_CAPACITY_SCALE;
5965 5966
	struct sched_group *sdg = sd->groups;

5967 5968 5969 5970
	if (sched_feat(ARCH_CAPACITY))
		capacity *= arch_scale_cpu_capacity(sd, cpu);
	else
		capacity *= default_scale_cpu_capacity(sd, cpu);
5971

5972
	capacity >>= SCHED_CAPACITY_SHIFT;
5973

5974
	sdg->sgc->capacity_orig = capacity;
5975

5976
	if (sched_feat(ARCH_CAPACITY))
5977
		capacity *= arch_scale_freq_capacity(sd, cpu);
5978
	else
5979
		capacity *= default_scale_capacity(sd, cpu);
5980

5981
	capacity >>= SCHED_CAPACITY_SHIFT;
5982

5983
	capacity *= scale_rt_capacity(cpu);
5984
	capacity >>= SCHED_CAPACITY_SHIFT;
5985

5986 5987
	if (!capacity)
		capacity = 1;
5988

5989 5990
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
5991 5992
}

5993
void update_group_capacity(struct sched_domain *sd, int cpu)
5994 5995 5996
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
5997
	unsigned long capacity, capacity_orig;
5998 5999 6000 6001
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
6002
	sdg->sgc->next_update = jiffies + interval;
6003 6004

	if (!child) {
6005
		update_cpu_capacity(sd, cpu);
6006 6007 6008
		return;
	}

6009
	capacity_orig = capacity = 0;
6010

P
Peter Zijlstra 已提交
6011 6012 6013 6014 6015 6016
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

6017
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
6018
			struct sched_group_capacity *sgc;
6019
			struct rq *rq = cpu_rq(cpu);
6020

6021
			/*
6022
			 * build_sched_domains() -> init_sched_groups_capacity()
6023 6024 6025
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
6026 6027
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
6028
			 *
6029
			 * This avoids capacity/capacity_orig from being 0 and
6030 6031
			 * causing divide-by-zero issues on boot.
			 *
6032
			 * Runtime updates will correct capacity_orig.
6033 6034
			 */
			if (unlikely(!rq->sd)) {
6035 6036
				capacity_orig += capacity_of(cpu);
				capacity += capacity_of(cpu);
6037 6038
				continue;
			}
6039

6040 6041 6042
			sgc = rq->sd->groups->sgc;
			capacity_orig += sgc->capacity_orig;
			capacity += sgc->capacity;
6043
		}
P
Peter Zijlstra 已提交
6044 6045 6046 6047 6048 6049 6050 6051
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
		 */ 

		group = child->groups;
		do {
6052 6053
			capacity_orig += group->sgc->capacity_orig;
			capacity += group->sgc->capacity;
P
Peter Zijlstra 已提交
6054 6055 6056
			group = group->next;
		} while (group != child->groups);
	}
6057

6058 6059
	sdg->sgc->capacity_orig = capacity_orig;
	sdg->sgc->capacity = capacity;
6060 6061
}

6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072
/*
 * 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)
{
	/*
6073
	 * Only siblings can have significantly less than SCHED_CAPACITY_SCALE
6074
	 */
6075
	if (!(sd->flags & SD_SHARE_CPUCAPACITY))
6076 6077 6078
		return 0;

	/*
6079
	 * If ~90% of the cpu_capacity is still there, we're good.
6080
	 */
6081
	if (group->sgc->capacity * 32 > group->sgc->capacity_orig * 29)
6082 6083 6084 6085 6086
		return 1;

	return 0;
}

6087 6088 6089 6090 6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102
/*
 * 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
6103 6104
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
6105 6106
 *
 * When this is so detected; this group becomes a candidate for busiest; see
6107
 * update_sd_pick_busiest(). And calculate_imbalance() and
6108
 * find_busiest_group() avoid some of the usual balance conditions to allow it
6109 6110 6111 6112 6113 6114 6115
 * 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.
 */

6116
static inline int sg_imbalanced(struct sched_group *group)
6117
{
6118
	return group->sgc->imbalance;
6119 6120
}

6121
/*
6122
 * Compute the group capacity factor.
6123
 *
6124
 * Avoid the issue where N*frac(smt_capacity) >= 1 creates 'phantom' cores by
6125
 * first dividing out the smt factor and computing the actual number of cores
6126
 * and limit unit capacity with that.
6127
 */
6128
static inline int sg_capacity_factor(struct lb_env *env, struct sched_group *group)
6129
{
6130
	unsigned int capacity_factor, smt, cpus;
6131
	unsigned int capacity, capacity_orig;
6132

6133 6134
	capacity = group->sgc->capacity;
	capacity_orig = group->sgc->capacity_orig;
6135
	cpus = group->group_weight;
6136

6137
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_capacity < 2 */
6138
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, capacity_orig);
6139
	capacity_factor = cpus / smt; /* cores */
6140

6141
	capacity_factor = min_t(unsigned,
6142
		capacity_factor, DIV_ROUND_CLOSEST(capacity, SCHED_CAPACITY_SCALE));
6143 6144
	if (!capacity_factor)
		capacity_factor = fix_small_capacity(env->sd, group);
6145

6146
	return capacity_factor;
6147 6148
}

6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160
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;
}

6161 6162
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
6163
 * @env: The load balancing environment.
6164 6165 6166 6167
 * @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.
6168
 * @overload: Indicate more than one runnable task for any CPU.
6169
 */
6170 6171
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
6172 6173
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
6174
{
6175
	unsigned long load;
6176
	int i;
6177

6178 6179
	memset(sgs, 0, sizeof(*sgs));

6180
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6181 6182 6183
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
6184
		if (local_group)
6185
			load = target_load(i, load_idx);
6186
		else
6187 6188 6189
			load = source_load(i, load_idx);

		sgs->group_load += load;
6190
		sgs->sum_nr_running += rq->cfs.h_nr_running;
6191 6192 6193 6194

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

6195 6196 6197 6198
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
6199
		sgs->sum_weighted_load += weighted_cpuload(i);
6200 6201
		if (idle_cpu(i))
			sgs->idle_cpus++;
6202 6203
	}

6204 6205
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
6206
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
6207

6208
	if (sgs->sum_nr_running)
6209
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
6210

6211
	sgs->group_weight = group->group_weight;
6212
	sgs->group_capacity_factor = sg_capacity_factor(env, group);
6213
	sgs->group_type = group_classify(group, sgs);
6214

6215
	if (sgs->group_capacity_factor > sgs->sum_nr_running)
6216
		sgs->group_has_free_capacity = 1;
6217 6218
}

6219 6220
/**
 * update_sd_pick_busiest - return 1 on busiest group
6221
 * @env: The load balancing environment.
6222 6223
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
6224
 * @sgs: sched_group statistics
6225 6226 6227
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
6228 6229 6230
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
6231
 */
6232
static bool update_sd_pick_busiest(struct lb_env *env,
6233 6234
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
6235
				   struct sg_lb_stats *sgs)
6236
{
6237
	struct sg_lb_stats *busiest = &sds->busiest_stat;
6238

6239
	if (sgs->group_type > busiest->group_type)
6240 6241
		return true;

6242 6243 6244 6245 6246 6247 6248 6249
	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))
6250 6251 6252 6253 6254 6255 6256
		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.
	 */
6257
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
6258 6259 6260 6261 6262 6263 6264 6265 6266 6267
		if (!sds->busiest)
			return true;

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

	return false;
}

6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297
#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 */

6298
/**
6299
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
6300
 * @env: The load balancing environment.
6301 6302
 * @sds: variable to hold the statistics for this sched_domain.
 */
6303
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
6304
{
6305 6306
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
6307
	struct sg_lb_stats tmp_sgs;
6308
	int load_idx, prefer_sibling = 0;
6309
	bool overload = false;
6310 6311 6312 6313

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

6314
	load_idx = get_sd_load_idx(env->sd, env->idle);
6315 6316

	do {
J
Joonsoo Kim 已提交
6317
		struct sg_lb_stats *sgs = &tmp_sgs;
6318 6319
		int local_group;

6320
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
6321 6322 6323
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
6324 6325

			if (env->idle != CPU_NEWLY_IDLE ||
6326 6327
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
6328
		}
6329

6330 6331
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
6332

6333 6334 6335
		if (local_group)
			goto next_group;

6336 6337
		/*
		 * In case the child domain prefers tasks go to siblings
6338
		 * first, lower the sg capacity factor to one so that we'll try
6339 6340
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
6341
		 * these excess tasks, i.e. nr_running < group_capacity_factor. The
6342 6343 6344
		 * 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).
6345
		 */
6346
		if (prefer_sibling && sds->local &&
6347
		    sds->local_stat.group_has_free_capacity)
6348
			sgs->group_capacity_factor = min(sgs->group_capacity_factor, 1U);
6349

6350
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
6351
			sds->busiest = sg;
J
Joonsoo Kim 已提交
6352
			sds->busiest_stat = *sgs;
6353 6354
		}

6355 6356 6357
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
6358
		sds->total_capacity += sgs->group_capacity;
6359

6360
		sg = sg->next;
6361
	} while (sg != env->sd->groups);
6362 6363 6364

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
6365 6366 6367 6368 6369 6370 6371

	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;
	}

6372 6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390
}

/**
 * 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.
 *
6391
 * Return: 1 when packing is required and a task should be moved to
6392 6393
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
6394
 * @env: The load balancing environment.
6395 6396
 * @sds: Statistics of the sched_domain which is to be packed
 */
6397
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
6398 6399 6400
{
	int busiest_cpu;

6401
	if (!(env->sd->flags & SD_ASYM_PACKING))
6402 6403 6404 6405 6406 6407
		return 0;

	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
6408
	if (env->dst_cpu > busiest_cpu)
6409 6410
		return 0;

6411
	env->imbalance = DIV_ROUND_CLOSEST(
6412
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
6413
		SCHED_CAPACITY_SCALE);
6414

6415
	return 1;
6416 6417 6418 6419 6420 6421
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
6422
 * @env: The load balancing environment.
6423 6424
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
6425 6426
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6427
{
6428
	unsigned long tmp, capa_now = 0, capa_move = 0;
6429
	unsigned int imbn = 2;
6430
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
6431
	struct sg_lb_stats *local, *busiest;
6432

J
Joonsoo Kim 已提交
6433 6434
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6435

J
Joonsoo Kim 已提交
6436 6437 6438 6439
	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;
6440

J
Joonsoo Kim 已提交
6441
	scaled_busy_load_per_task =
6442
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6443
		busiest->group_capacity;
J
Joonsoo Kim 已提交
6444

6445 6446
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
6447
		env->imbalance = busiest->load_per_task;
6448 6449 6450 6451 6452
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
6453
	 * however we may be able to increase total CPU capacity used by
6454 6455 6456
	 * moving them.
	 */

6457
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
6458
			min(busiest->load_per_task, busiest->avg_load);
6459
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
6460
			min(local->load_per_task, local->avg_load);
6461
	capa_now /= SCHED_CAPACITY_SCALE;
6462 6463

	/* Amount of load we'd subtract */
6464
	if (busiest->avg_load > scaled_busy_load_per_task) {
6465
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
6466
			    min(busiest->load_per_task,
6467
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
6468
	}
6469 6470

	/* Amount of load we'd add */
6471
	if (busiest->avg_load * busiest->group_capacity <
6472
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
6473 6474
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
6475
	} else {
6476
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6477
		      local->group_capacity;
J
Joonsoo Kim 已提交
6478
	}
6479
	capa_move += local->group_capacity *
6480
		    min(local->load_per_task, local->avg_load + tmp);
6481
	capa_move /= SCHED_CAPACITY_SCALE;
6482 6483

	/* Move if we gain throughput */
6484
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
6485
		env->imbalance = busiest->load_per_task;
6486 6487 6488 6489 6490
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
6491
 * @env: load balance environment
6492 6493
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
6494
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6495
{
6496
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
6497 6498 6499 6500
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6501

6502
	if (busiest->group_type == group_imbalanced) {
6503 6504 6505 6506
		/*
		 * 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
6507 6508
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
6509 6510
	}

6511 6512 6513
	/*
	 * 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
6514
	 * its cpu_capacity, while calculating max_load..)
6515
	 */
6516 6517
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
6518 6519
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
6520 6521
	}

6522 6523 6524 6525 6526
	/*
	 * 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
6527
		load_above_capacity =
6528
			(busiest->sum_nr_running - busiest->group_capacity_factor);
6529

6530
		load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_CAPACITY_SCALE);
6531
		load_above_capacity /= busiest->group_capacity;
6532 6533 6534 6535 6536 6537 6538 6539 6540 6541
	}

	/*
	 * 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.
	 */
6542
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
6543 6544

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
6545
	env->imbalance = min(
6546 6547
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
6548
	) / SCHED_CAPACITY_SCALE;
6549 6550 6551

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
6552
	 * there is no guarantee that any tasks will be moved so we'll have
6553 6554 6555
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
6556
	if (env->imbalance < busiest->load_per_task)
6557
		return fix_small_imbalance(env, sds);
6558
}
6559

6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571
/******* 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.
 *
6572
 * @env: The load balancing environment.
6573
 *
6574
 * Return:	- The busiest group if imbalance exists.
6575 6576 6577 6578
 *		- 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
6579
static struct sched_group *find_busiest_group(struct lb_env *env)
6580
{
J
Joonsoo Kim 已提交
6581
	struct sg_lb_stats *local, *busiest;
6582 6583
	struct sd_lb_stats sds;

6584
	init_sd_lb_stats(&sds);
6585 6586 6587 6588 6589

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
6590
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
6591 6592
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
6593

6594 6595
	if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
	    check_asym_packing(env, &sds))
6596 6597
		return sds.busiest;

6598
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
6599
	if (!sds.busiest || busiest->sum_nr_running == 0)
6600 6601
		goto out_balanced;

6602 6603
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
6604

P
Peter Zijlstra 已提交
6605 6606
	/*
	 * If the busiest group is imbalanced the below checks don't
6607
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
6608 6609
	 * isn't true due to cpus_allowed constraints and the like.
	 */
6610
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
6611 6612
		goto force_balance;

6613
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
6614 6615
	if (env->idle == CPU_NEWLY_IDLE && local->group_has_free_capacity &&
	    !busiest->group_has_free_capacity)
6616 6617
		goto force_balance;

6618
	/*
6619
	 * If the local group is busier than the selected busiest group
6620 6621
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
6622
	if (local->avg_load >= busiest->avg_load)
6623 6624
		goto out_balanced;

6625 6626 6627 6628
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
6629
	if (local->avg_load >= sds.avg_load)
6630 6631
		goto out_balanced;

6632
	if (env->idle == CPU_IDLE) {
6633
		/*
6634 6635 6636 6637 6638
		 * 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
6639
		 */
6640 6641
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
6642
			goto out_balanced;
6643 6644 6645 6646 6647
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
6648 6649
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
6650
			goto out_balanced;
6651
	}
6652

6653
force_balance:
6654
	/* Looks like there is an imbalance. Compute it */
6655
	calculate_imbalance(env, &sds);
6656 6657 6658
	return sds.busiest;

out_balanced:
6659
	env->imbalance = 0;
6660 6661 6662 6663 6664 6665
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
6666
static struct rq *find_busiest_queue(struct lb_env *env,
6667
				     struct sched_group *group)
6668 6669
{
	struct rq *busiest = NULL, *rq;
6670
	unsigned long busiest_load = 0, busiest_capacity = 1;
6671 6672
	int i;

6673
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6674
		unsigned long capacity, capacity_factor, wl;
6675 6676 6677 6678
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
6679

6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701
		/*
		 * 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;

6702
		capacity = capacity_of(i);
6703
		capacity_factor = DIV_ROUND_CLOSEST(capacity, SCHED_CAPACITY_SCALE);
6704 6705
		if (!capacity_factor)
			capacity_factor = fix_small_capacity(env->sd, group);
6706

6707
		wl = weighted_cpuload(i);
6708

6709 6710
		/*
		 * When comparing with imbalance, use weighted_cpuload()
6711
		 * which is not scaled with the cpu capacity.
6712
		 */
6713
		if (capacity_factor && rq->nr_running == 1 && wl > env->imbalance)
6714 6715
			continue;

6716 6717
		/*
		 * For the load comparisons with the other cpu's, consider
6718 6719 6720
		 * 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.
6721
		 *
6722
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
6723
		 * multiplication to rid ourselves of the division works out
6724 6725
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
6726
		 */
6727
		if (wl * busiest_capacity > busiest_load * capacity) {
6728
			busiest_load = wl;
6729
			busiest_capacity = capacity;
6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743
			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. */
6744
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
6745

6746
static int need_active_balance(struct lb_env *env)
6747
{
6748 6749 6750
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
6751 6752 6753 6754 6755 6756

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
6757
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
6758
			return 1;
6759 6760 6761 6762 6763
	}

	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

6764 6765
static int active_load_balance_cpu_stop(void *data);

6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796
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.
	 */
6797
	return balance_cpu == env->dst_cpu;
6798 6799
}

6800 6801 6802 6803 6804 6805
/*
 * 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,
6806
			int *continue_balancing)
6807
{
6808
	int ld_moved, cur_ld_moved, active_balance = 0;
6809
	struct sched_domain *sd_parent = sd->parent;
6810 6811 6812
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
6813
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
6814

6815 6816
	struct lb_env env = {
		.sd		= sd,
6817 6818
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
6819
		.dst_grpmask    = sched_group_cpus(sd->groups),
6820
		.idle		= idle,
6821
		.loop_break	= sched_nr_migrate_break,
6822
		.cpus		= cpus,
6823
		.fbq_type	= all,
6824
		.tasks		= LIST_HEAD_INIT(env.tasks),
6825 6826
	};

6827 6828 6829 6830
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
6831
	if (idle == CPU_NEWLY_IDLE)
6832 6833
		env.dst_grpmask = NULL;

6834 6835 6836 6837 6838
	cpumask_copy(cpus, cpu_active_mask);

	schedstat_inc(sd, lb_count[idle]);

redo:
6839 6840
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
6841
		goto out_balanced;
6842
	}
6843

6844
	group = find_busiest_group(&env);
6845 6846 6847 6848 6849
	if (!group) {
		schedstat_inc(sd, lb_nobusyg[idle]);
		goto out_balanced;
	}

6850
	busiest = find_busiest_queue(&env, group);
6851 6852 6853 6854 6855
	if (!busiest) {
		schedstat_inc(sd, lb_nobusyq[idle]);
		goto out_balanced;
	}

6856
	BUG_ON(busiest == env.dst_rq);
6857

6858
	schedstat_add(sd, lb_imbalance[idle], env.imbalance);
6859 6860 6861 6862 6863 6864 6865 6866 6867

	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.
		 */
6868
		env.flags |= LBF_ALL_PINNED;
6869 6870 6871
		env.src_cpu   = busiest->cpu;
		env.src_rq    = busiest;
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
6872

6873
more_balance:
6874
		raw_spin_lock_irqsave(&busiest->lock, flags);
6875 6876 6877 6878 6879

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
6880
		cur_ld_moved = detach_tasks(&env);
6881 6882

		/*
6883 6884 6885 6886 6887
		 * 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.
6888
		 */
6889 6890 6891 6892 6893 6894 6895 6896

		raw_spin_unlock(&busiest->lock);

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

6897
		local_irq_restore(flags);
6898

6899 6900 6901 6902 6903
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

6904 6905 6906 6907 6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922
		/*
		 * 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.
		 */
6923
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
6924

6925 6926 6927
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

6928
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
6929
			env.dst_cpu	 = env.new_dst_cpu;
6930
			env.flags	&= ~LBF_DST_PINNED;
6931 6932
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
6933

6934 6935 6936 6937 6938 6939
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
6940

6941 6942 6943 6944
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
6945
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
6946

6947
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
6948 6949 6950
				*group_imbalance = 1;
		}

6951
		/* All tasks on this runqueue were pinned by CPU affinity */
6952
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
6953
			cpumask_clear_cpu(cpu_of(busiest), cpus);
6954 6955 6956
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
6957
				goto redo;
6958
			}
6959
			goto out_all_pinned;
6960 6961 6962 6963 6964
		}
	}

	if (!ld_moved) {
		schedstat_inc(sd, lb_failed[idle]);
6965 6966 6967 6968 6969 6970 6971 6972
		/*
		 * 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++;
6973

6974
		if (need_active_balance(&env)) {
6975 6976
			raw_spin_lock_irqsave(&busiest->lock, flags);

6977 6978 6979
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
6980 6981
			 */
			if (!cpumask_test_cpu(this_cpu,
6982
					tsk_cpus_allowed(busiest->curr))) {
6983 6984
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
6985
				env.flags |= LBF_ALL_PINNED;
6986 6987 6988
				goto out_one_pinned;
			}

6989 6990 6991 6992 6993
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
6994 6995 6996 6997 6998 6999
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
7000

7001
			if (active_balance) {
7002 7003 7004
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
7005
			}
7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023

			/*
			 * 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
7024
		 * detach_tasks).
7025 7026 7027 7028 7029 7030 7031 7032
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049
	/*
	 * 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.
	 */
7050 7051 7052 7053 7054 7055
	schedstat_inc(sd, lb_balanced[idle]);

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
7056
	if (((env.flags & LBF_ALL_PINNED) &&
7057
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
7058 7059 7060
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

7061
	ld_moved = 0;
7062 7063 7064 7065
out:
	return ld_moved;
}

7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092
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;
}

7093 7094 7095 7096
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
7097
static int idle_balance(struct rq *this_rq)
7098
{
7099 7100
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
7101 7102
	struct sched_domain *sd;
	int pulled_task = 0;
7103
	u64 curr_cost = 0;
7104

7105
	idle_enter_fair(this_rq);
7106

7107 7108 7109 7110 7111 7112
	/*
	 * 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);

7113 7114
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
7115 7116 7117 7118 7119 7120
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, 0, &next_balance);
		rcu_read_unlock();

7121
		goto out;
7122
	}
7123

7124 7125 7126 7127 7128
	/*
	 * Drop the rq->lock, but keep IRQ/preempt disabled.
	 */
	raw_spin_unlock(&this_rq->lock);

7129
	update_blocked_averages(this_cpu);
7130
	rcu_read_lock();
7131
	for_each_domain(this_cpu, sd) {
7132
		int continue_balancing = 1;
7133
		u64 t0, domain_cost;
7134 7135 7136 7137

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

7138 7139
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, 0, &next_balance);
7140
			break;
7141
		}
7142

7143
		if (sd->flags & SD_BALANCE_NEWIDLE) {
7144 7145
			t0 = sched_clock_cpu(this_cpu);

7146
			pulled_task = load_balance(this_cpu, this_rq,
7147 7148
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
7149 7150 7151 7152 7153 7154

			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;
7155
		}
7156

7157
		update_next_balance(sd, 0, &next_balance);
7158 7159 7160 7161 7162 7163

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
7164 7165
			break;
	}
7166
	rcu_read_unlock();
7167 7168 7169

	raw_spin_lock(&this_rq->lock);

7170 7171 7172
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

7173
	/*
7174 7175 7176
	 * 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.
7177
	 */
7178
	if (this_rq->cfs.h_nr_running && !pulled_task)
7179
		pulled_task = 1;
7180

7181 7182 7183
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
7184
		this_rq->next_balance = next_balance;
7185

7186
	/* Is there a task of a high priority class? */
7187
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
7188 7189 7190 7191
		pulled_task = -1;

	if (pulled_task) {
		idle_exit_fair(this_rq);
7192
		this_rq->idle_stamp = 0;
7193
	}
7194

7195
	return pulled_task;
7196 7197 7198
}

/*
7199 7200 7201 7202
 * 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.
7203
 */
7204
static int active_load_balance_cpu_stop(void *data)
7205
{
7206 7207
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
7208
	int target_cpu = busiest_rq->push_cpu;
7209
	struct rq *target_rq = cpu_rq(target_cpu);
7210
	struct sched_domain *sd;
7211
	struct task_struct *p = NULL;
7212 7213 7214 7215 7216 7217 7218

	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;
7219 7220 7221

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
7222
		goto out_unlock;
7223 7224 7225 7226 7227 7228 7229 7230 7231

	/*
	 * 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. */
7232
	rcu_read_lock();
7233 7234 7235 7236 7237 7238 7239
	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)) {
7240 7241
		struct lb_env env = {
			.sd		= sd,
7242 7243 7244 7245
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
7246 7247 7248
			.idle		= CPU_IDLE,
		};

7249 7250
		schedstat_inc(sd, alb_count);

7251 7252
		p = detach_one_task(&env);
		if (p)
7253 7254 7255 7256
			schedstat_inc(sd, alb_pushed);
		else
			schedstat_inc(sd, alb_failed);
	}
7257
	rcu_read_unlock();
7258 7259
out_unlock:
	busiest_rq->active_balance = 0;
7260 7261 7262 7263 7264 7265 7266
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

7267
	return 0;
7268 7269
}

7270 7271 7272 7273 7274
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

7275
#ifdef CONFIG_NO_HZ_COMMON
7276 7277 7278 7279 7280 7281
/*
 * 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.
 */
7282
static struct {
7283
	cpumask_var_t idle_cpus_mask;
7284
	atomic_t nr_cpus;
7285 7286
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
7287

7288
static inline int find_new_ilb(void)
7289
{
7290
	int ilb = cpumask_first(nohz.idle_cpus_mask);
7291

7292 7293 7294 7295
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
7296 7297
}

7298 7299 7300 7301 7302
/*
 * 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).
 */
7303
static void nohz_balancer_kick(void)
7304 7305 7306 7307 7308
{
	int ilb_cpu;

	nohz.next_balance++;

7309
	ilb_cpu = find_new_ilb();
7310

7311 7312
	if (ilb_cpu >= nr_cpu_ids)
		return;
7313

7314
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
7315 7316 7317 7318 7319 7320 7321 7322
		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);
7323 7324 7325
	return;
}

7326
static inline void nohz_balance_exit_idle(int cpu)
7327 7328
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
7329 7330 7331 7332 7333 7334 7335
		/*
		 * 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);
		}
7336 7337 7338 7339
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

7340 7341 7342
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
7343
	int cpu = smp_processor_id();
7344 7345

	rcu_read_lock();
7346
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7347 7348 7349 7350 7351

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

7352
	atomic_inc(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7353
unlock:
7354 7355 7356 7357 7358 7359
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
7360
	int cpu = smp_processor_id();
7361 7362

	rcu_read_lock();
7363
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7364 7365 7366 7367 7368

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

7369
	atomic_dec(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7370
unlock:
7371 7372 7373
	rcu_read_unlock();
}

7374
/*
7375
 * This routine will record that the cpu is going idle with tick stopped.
7376
 * This info will be used in performing idle load balancing in the future.
7377
 */
7378
void nohz_balance_enter_idle(int cpu)
7379
{
7380 7381 7382 7383 7384 7385
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

7386 7387
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
7388

7389 7390 7391 7392 7393 7394
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

7395 7396 7397
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
7398
}
7399

7400
static int sched_ilb_notifier(struct notifier_block *nfb,
7401 7402 7403 7404
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DYING:
7405
		nohz_balance_exit_idle(smp_processor_id());
7406 7407 7408 7409 7410
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}
7411 7412 7413 7414
#endif

static DEFINE_SPINLOCK(balancing);

7415 7416 7417 7418
/*
 * 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.
 */
7419
void update_max_interval(void)
7420 7421 7422 7423
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

7424 7425 7426 7427
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
7428
 * Balancing parameters are set up in init_sched_domains.
7429
 */
7430
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
7431
{
7432
	int continue_balancing = 1;
7433
	int cpu = rq->cpu;
7434
	unsigned long interval;
7435
	struct sched_domain *sd;
7436 7437 7438
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
7439 7440
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
7441

7442
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
7443

7444
	rcu_read_lock();
7445
	for_each_domain(cpu, sd) {
7446 7447 7448 7449 7450 7451 7452 7453 7454 7455 7456 7457
		/*
		 * 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;

7458 7459 7460
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

7461 7462 7463 7464 7465 7466 7467 7468 7469 7470 7471
		/*
		 * 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;
		}

7472
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7473 7474 7475 7476 7477 7478 7479 7480

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

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
7481
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
7482
				/*
7483
				 * The LBF_DST_PINNED logic could have changed
7484 7485
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
7486
				 */
7487
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
7488 7489
			}
			sd->last_balance = jiffies;
7490
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7491 7492 7493 7494 7495 7496 7497 7498
		}
		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;
		}
7499 7500
	}
	if (need_decay) {
7501
		/*
7502 7503
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
7504
		 */
7505 7506
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
7507
	}
7508
	rcu_read_unlock();
7509 7510 7511 7512 7513 7514 7515 7516 7517 7518

	/*
	 * 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;
}

7519
#ifdef CONFIG_NO_HZ_COMMON
7520
/*
7521
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
7522 7523
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
7524
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
7525
{
7526
	int this_cpu = this_rq->cpu;
7527 7528 7529
	struct rq *rq;
	int balance_cpu;

7530 7531 7532
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
7533 7534

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
7535
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
7536 7537 7538 7539 7540 7541 7542
			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.
		 */
7543
		if (need_resched())
7544 7545
			break;

V
Vincent Guittot 已提交
7546 7547
		rq = cpu_rq(balance_cpu);

7548 7549 7550 7551 7552 7553 7554 7555 7556 7557 7558
		/*
		 * 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);
		}
7559 7560 7561 7562 7563

		if (time_after(this_rq->next_balance, rq->next_balance))
			this_rq->next_balance = rq->next_balance;
	}
	nohz.next_balance = this_rq->next_balance;
7564 7565
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
7566 7567 7568
}

/*
7569 7570 7571 7572
 * 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
7573
 *     busy cpu's exceeding the group's capacity.
7574 7575
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
7576
 */
7577
static inline int nohz_kick_needed(struct rq *rq)
7578 7579
{
	unsigned long now = jiffies;
7580
	struct sched_domain *sd;
7581
	struct sched_group_capacity *sgc;
7582
	int nr_busy, cpu = rq->cpu;
7583

7584
	if (unlikely(rq->idle_balance))
7585 7586
		return 0;

7587 7588 7589 7590
       /*
	* 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.
	*/
7591
	set_cpu_sd_state_busy();
7592
	nohz_balance_exit_idle(cpu);
7593 7594 7595 7596 7597 7598 7599

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
		return 0;
7600 7601

	if (time_before(now, nohz.next_balance))
7602 7603
		return 0;

7604 7605
	if (rq->nr_running >= 2)
		goto need_kick;
7606

7607
	rcu_read_lock();
7608
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
7609

7610
	if (sd) {
7611 7612
		sgc = sd->groups->sgc;
		nr_busy = atomic_read(&sgc->nr_busy_cpus);
7613

7614
		if (nr_busy > 1)
7615
			goto need_kick_unlock;
7616
	}
7617 7618 7619 7620 7621 7622 7623

	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;

7624
	rcu_read_unlock();
7625
	return 0;
7626 7627 7628

need_kick_unlock:
	rcu_read_unlock();
7629 7630
need_kick:
	return 1;
7631 7632
}
#else
7633
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
7634 7635 7636 7637 7638 7639
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
7640 7641
static void run_rebalance_domains(struct softirq_action *h)
{
7642
	struct rq *this_rq = this_rq();
7643
	enum cpu_idle_type idle = this_rq->idle_balance ?
7644 7645
						CPU_IDLE : CPU_NOT_IDLE;

7646
	rebalance_domains(this_rq, idle);
7647 7648

	/*
7649
	 * If this cpu has a pending nohz_balance_kick, then do the
7650 7651 7652
	 * balancing on behalf of the other idle cpus whose ticks are
	 * stopped.
	 */
7653
	nohz_idle_balance(this_rq, idle);
7654 7655 7656 7657 7658
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
7659
void trigger_load_balance(struct rq *rq)
7660 7661
{
	/* Don't need to rebalance while attached to NULL domain */
7662 7663 7664 7665
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
7666
		raise_softirq(SCHED_SOFTIRQ);
7667
#ifdef CONFIG_NO_HZ_COMMON
7668
	if (nohz_kick_needed(rq))
7669
		nohz_balancer_kick();
7670
#endif
7671 7672
}

7673 7674 7675
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
7676 7677

	update_runtime_enabled(rq);
7678 7679 7680 7681 7682
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
7683 7684 7685

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
7686 7687
}

7688
#endif /* CONFIG_SMP */
7689

7690 7691 7692
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
7693
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
7694 7695 7696 7697 7698 7699
{
	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
7700
		entity_tick(cfs_rq, se, queued);
7701
	}
7702

7703
	if (numabalancing_enabled)
7704
		task_tick_numa(rq, curr);
7705

7706
	update_rq_runnable_avg(rq, 1);
7707 7708 7709
}

/*
P
Peter Zijlstra 已提交
7710 7711 7712
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
7713
 */
P
Peter Zijlstra 已提交
7714
static void task_fork_fair(struct task_struct *p)
7715
{
7716 7717
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
7718
	int this_cpu = smp_processor_id();
P
Peter Zijlstra 已提交
7719 7720 7721
	struct rq *rq = this_rq();
	unsigned long flags;

7722
	raw_spin_lock_irqsave(&rq->lock, flags);
7723

7724 7725
	update_rq_clock(rq);

7726 7727 7728
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;

7729 7730 7731 7732 7733 7734 7735 7736 7737
	/*
	 * 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();
7738

7739
	update_curr(cfs_rq);
P
Peter Zijlstra 已提交
7740

7741 7742
	if (curr)
		se->vruntime = curr->vruntime;
7743
	place_entity(cfs_rq, se, 1);
7744

P
Peter Zijlstra 已提交
7745
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
7746
		/*
7747 7748 7749
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
7750
		swap(curr->vruntime, se->vruntime);
7751
		resched_curr(rq);
7752
	}
7753

7754 7755
	se->vruntime -= cfs_rq->min_vruntime;

7756
	raw_spin_unlock_irqrestore(&rq->lock, flags);
7757 7758
}

7759 7760 7761 7762
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
7763 7764
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
7765
{
7766
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7767 7768
		return;

7769 7770 7771 7772 7773
	/*
	 * 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
7774
	if (rq->curr == p) {
7775
		if (p->prio > oldprio)
7776
			resched_curr(rq);
7777
	} else
7778
		check_preempt_curr(rq, p, 0);
7779 7780
}

P
Peter Zijlstra 已提交
7781 7782 7783 7784 7785 7786
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);

	/*
7787
	 * Ensure the task's vruntime is normalized, so that when it's
P
Peter Zijlstra 已提交
7788 7789 7790
	 * switched back to the fair class the enqueue_entity(.flags=0) will
	 * do the right thing.
	 *
7791 7792
	 * 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
7793 7794
	 * the task is sleeping will it still have non-normalized vruntime.
	 */
7795
	if (!task_on_rq_queued(p) && p->state != TASK_RUNNING) {
P
Peter Zijlstra 已提交
7796 7797 7798 7799 7800 7801 7802
		/*
		 * 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;
	}
7803

7804
#ifdef CONFIG_SMP
7805 7806 7807 7808 7809
	/*
	* 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.
	*/
7810 7811 7812
	if (se->avg.decay_count) {
		__synchronize_entity_decay(se);
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
7813 7814
	}
#endif
P
Peter Zijlstra 已提交
7815 7816
}

7817 7818 7819
/*
 * We switched to the sched_fair class.
 */
P
Peter Zijlstra 已提交
7820
static void switched_to_fair(struct rq *rq, struct task_struct *p)
7821
{
7822
#ifdef CONFIG_FAIR_GROUP_SCHED
7823
	struct sched_entity *se = &p->se;
7824 7825 7826 7827 7828 7829
	/*
	 * 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
7830
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7831 7832
		return;

7833 7834 7835 7836 7837
	/*
	 * 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
7838
	if (rq->curr == p)
7839
		resched_curr(rq);
7840
	else
7841
		check_preempt_curr(rq, p, 0);
7842 7843
}

7844 7845 7846 7847 7848 7849 7850 7851 7852
/* 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;

7853 7854 7855 7856 7857 7858 7859
	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);
	}
7860 7861
}

7862 7863 7864 7865 7866 7867 7868
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
7869
#ifdef CONFIG_SMP
7870
	atomic64_set(&cfs_rq->decay_counter, 1);
7871
	atomic_long_set(&cfs_rq->removed_load, 0);
7872
#endif
7873 7874
}

P
Peter Zijlstra 已提交
7875
#ifdef CONFIG_FAIR_GROUP_SCHED
7876
static void task_move_group_fair(struct task_struct *p, int queued)
P
Peter Zijlstra 已提交
7877
{
P
Peter Zijlstra 已提交
7878
	struct sched_entity *se = &p->se;
7879
	struct cfs_rq *cfs_rq;
P
Peter Zijlstra 已提交
7880

7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893
	/*
	 * 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.
	 */
7894
	/*
7895
	 * When !queued, vruntime of the task has usually NOT been normalized.
7896 7897 7898 7899
	 * 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().
7900 7901
	 * - Moving a task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
7902 7903 7904 7905
	 *
	 * To prevent boost or penalty in the new cfs_rq caused by delta
	 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
	 */
7906 7907
	if (!queued && (!se->sum_exec_runtime || p->state == TASK_WAKING))
		queued = 1;
7908

7909
	if (!queued)
P
Peter Zijlstra 已提交
7910
		se->vruntime -= cfs_rq_of(se)->min_vruntime;
7911
	set_task_rq(p, task_cpu(p));
P
Peter Zijlstra 已提交
7912
	se->depth = se->parent ? se->parent->depth + 1 : 0;
7913
	if (!queued) {
P
Peter Zijlstra 已提交
7914 7915
		cfs_rq = cfs_rq_of(se);
		se->vruntime += cfs_rq->min_vruntime;
7916 7917 7918 7919 7920 7921
#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
7922 7923
		se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
		cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
7924 7925
#endif
	}
P
Peter Zijlstra 已提交
7926
}
7927 7928 7929 7930 7931 7932 7933 7934 7935 7936 7937 7938 7939 7940 7941 7942 7943 7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960 7961 7962 7963 7964 7965 7966 7967 7968 7969 7970 7971 7972 7973 7974 7975 7976 7977 7978 7979 7980 7981 7982 7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018

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
8019
	if (!parent) {
8020
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
8021 8022
		se->depth = 0;
	} else {
8023
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
8024 8025
		se->depth = parent->depth + 1;
	}
8026 8027

	se->my_q = cfs_rq;
8028 8029
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
8030 8031 8032 8033 8034 8035 8036 8037 8038 8039 8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059
	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);
8060 8061 8062

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
8063
		for_each_sched_entity(se)
8064 8065 8066 8067 8068 8069 8070 8071 8072 8073 8074 8075 8076 8077 8078 8079 8080 8081 8082 8083 8084
			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
8085

8086
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
8087 8088 8089 8090 8091 8092 8093 8094 8095
{
	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)
8096
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
8097 8098 8099 8100

	return rr_interval;
}

8101 8102 8103
/*
 * All the scheduling class methods:
 */
8104
const struct sched_class fair_sched_class = {
8105
	.next			= &idle_sched_class,
8106 8107 8108
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
8109
	.yield_to_task		= yield_to_task_fair,
8110

I
Ingo Molnar 已提交
8111
	.check_preempt_curr	= check_preempt_wakeup,
8112 8113 8114 8115

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

8116
#ifdef CONFIG_SMP
L
Li Zefan 已提交
8117
	.select_task_rq		= select_task_rq_fair,
8118
	.migrate_task_rq	= migrate_task_rq_fair,
8119

8120 8121
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
8122 8123

	.task_waking		= task_waking_fair,
8124
#endif
8125

8126
	.set_curr_task          = set_curr_task_fair,
8127
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
8128
	.task_fork		= task_fork_fair,
8129 8130

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
8131
	.switched_from		= switched_from_fair,
8132
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
8133

8134 8135
	.get_rr_interval	= get_rr_interval_fair,

P
Peter Zijlstra 已提交
8136
#ifdef CONFIG_FAIR_GROUP_SCHED
8137
	.task_move_group	= task_move_group_fair,
P
Peter Zijlstra 已提交
8138
#endif
8139 8140 8141
};

#ifdef CONFIG_SCHED_DEBUG
8142
void print_cfs_stats(struct seq_file *m, int cpu)
8143 8144 8145
{
	struct cfs_rq *cfs_rq;

8146
	rcu_read_lock();
8147
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
8148
		print_cfs_rq(m, cpu, cfs_rq);
8149
	rcu_read_unlock();
8150 8151
}
#endif
8152 8153 8154 8155 8156 8157

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

8158
#ifdef CONFIG_NO_HZ_COMMON
8159
	nohz.next_balance = jiffies;
8160
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
8161
	cpu_notifier(sched_ilb_notifier, 0);
8162 8163 8164 8165
#endif
#endif /* SMP */

}