fair.c 206.5 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
/*
 * These return the fraction of accesses done by a particular task, or
 * task group, on a particular numa node.  The group weight is given a
 * larger multiplier, in order to group tasks together that are almost
 * evenly spread out between numa nodes.
 */
static inline unsigned long task_weight(struct task_struct *p, int nid)
{
	unsigned long total_faults;

938
	if (!p->numa_faults_memory)
939 940 941 942 943 944 945 946 947 948 949 950
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

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

static inline unsigned long group_weight(struct task_struct *p, int nid)
{
951
	if (!p->numa_group || !p->numa_group->total_faults)
952 953
		return 0;

954
	return 1000 * group_faults(p, nid) / p->numa_group->total_faults;
955 956
}

957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019
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);
}

1020
static unsigned long weighted_cpuload(const int cpu);
1021 1022
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1023
static unsigned long capacity_of(int cpu);
1024 1025
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1026
/* Cached statistics for all CPUs within a node */
1027
struct numa_stats {
1028
	unsigned long nr_running;
1029
	unsigned long load;
1030 1031

	/* Total compute capacity of CPUs on a node */
1032
	unsigned long compute_capacity;
1033 1034

	/* Approximate capacity in terms of runnable tasks on a node */
1035
	unsigned long task_capacity;
1036
	int has_free_capacity;
1037
};
1038

1039 1040 1041 1042 1043
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1044 1045
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1046 1047 1048 1049 1050 1051 1052

	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);
1053
		ns->compute_capacity += capacity_of(cpu);
1054 1055

		cpus++;
1056 1057
	}

1058 1059 1060 1061 1062
	/*
	 * 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.
	 *
1063 1064
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1065 1066 1067 1068
	 */
	if (!cpus)
		return;

1069 1070 1071 1072 1073 1074
	/* 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));
1075
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1076 1077
}

1078 1079
struct task_numa_env {
	struct task_struct *p;
1080

1081 1082
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1083

1084
	struct numa_stats src_stats, dst_stats;
1085

1086
	int imbalance_pct;
1087 1088 1089

	struct task_struct *best_task;
	long best_imp;
1090 1091 1092
	int best_cpu;
};

1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
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;
}

1106
static bool load_too_imbalanced(long src_load, long dst_load,
1107 1108 1109
				struct task_numa_env *env)
{
	long imb, old_imb;
1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121
	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;
1122 1123 1124 1125 1126 1127

	/* 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? */
1128 1129
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1130 1131 1132 1133 1134 1135 1136
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
	 * Compare it with the old imbalance.
	 */
1137 1138 1139
	orig_src_load = env->src_stats.load;
	orig_dst_load = env->dst_stats.load;

1140 1141 1142
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);

1143 1144
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;
1145 1146

	/* Would this change make things worse? */
1147
	return (imb > old_imb);
1148 1149
}

1150 1151 1152 1153 1154 1155
/*
 * 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
 */
1156 1157
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1158 1159 1160 1161
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1162
	long src_load, dst_load;
1163
	long load;
1164
	long imp = env->p->numa_group ? groupimp : taskimp;
1165
	long moveimp = imp;
1166 1167

	rcu_read_lock();
1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178

	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))
1179
		cur = NULL;
1180
	raw_spin_unlock_irq(&dst_rq->lock);
1181

1182 1183 1184 1185 1186 1187 1188
	/*
	 * Because we have preemption enabled we can get migrated around and
	 * end try selecting ourselves (current == env->p) as a swap candidate.
	 */
	if (cur == env->p)
		goto unlock;

1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200
	/*
	 * "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;

1201 1202
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1203
		 * in any group then look only at task weights.
1204
		 */
1205
		if (cur->numa_group == env->p->numa_group) {
1206 1207
			imp = taskimp + task_weight(cur, env->src_nid) -
			      task_weight(cur, env->dst_nid);
1208 1209 1210 1211 1212 1213
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1214
		} else {
1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225
			/*
			 * Compare the group weights. If a task is all by
			 * itself (not part of a group), use the task weight
			 * instead.
			 */
			if (cur->numa_group)
				imp += group_weight(cur, env->src_nid) -
				       group_weight(cur, env->dst_nid);
			else
				imp += task_weight(cur, env->src_nid) -
				       task_weight(cur, env->dst_nid);
1226
		}
1227 1228
	}

1229
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1230 1231 1232 1233
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1234
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1235
		    !env->dst_stats.has_free_capacity)
1236 1237 1238 1239 1240 1241
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1242 1243
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1244 1245 1246 1247 1248 1249
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1250 1251 1252
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1253

1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270
	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;

1271
	if (cur) {
1272 1273 1274
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1275 1276
	}

1277
	if (load_too_imbalanced(src_load, dst_load, env))
1278 1279
		goto unlock;

1280 1281 1282 1283 1284 1285 1286
	/*
	 * 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);

1287 1288 1289 1290 1291 1292
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1293 1294
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1295 1296 1297 1298 1299 1300 1301 1302 1303
{
	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;
1304
		task_numa_compare(env, taskimp, groupimp);
1305 1306 1307
	}
}

1308 1309 1310 1311
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1312

1313
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1314
		.src_nid = task_node(p),
1315 1316 1317 1318 1319 1320

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
		.best_cpu = -1
1321 1322
	};
	struct sched_domain *sd;
1323
	unsigned long taskweight, groupweight;
1324
	int nid, ret;
1325
	long taskimp, groupimp;
1326

1327
	/*
1328 1329 1330 1331 1332 1333
	 * 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.
1334 1335
	 */
	rcu_read_lock();
1336
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1337 1338
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1339 1340
	rcu_read_unlock();

1341 1342 1343 1344 1345 1346 1347
	/*
	 * 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)) {
1348
		p->numa_preferred_nid = task_node(p);
1349 1350 1351
		return -EINVAL;
	}

1352 1353
	taskweight = task_weight(p, env.src_nid);
	groupweight = group_weight(p, env.src_nid);
1354
	update_numa_stats(&env.src_stats, env.src_nid);
1355
	env.dst_nid = p->numa_preferred_nid;
1356 1357
	taskimp = task_weight(p, env.dst_nid) - taskweight;
	groupimp = group_weight(p, env.dst_nid) - groupweight;
1358
	update_numa_stats(&env.dst_stats, env.dst_nid);
1359

1360 1361
	/* Try to find a spot on the preferred nid. */
	task_numa_find_cpu(&env, taskimp, groupimp);
1362 1363 1364

	/* No space available on the preferred nid. Look elsewhere. */
	if (env.best_cpu == -1) {
1365 1366 1367
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1368

1369
			/* Only consider nodes where both task and groups benefit */
1370 1371 1372
			taskimp = task_weight(p, nid) - taskweight;
			groupimp = group_weight(p, nid) - groupweight;
			if (taskimp < 0 && groupimp < 0)
1373 1374
				continue;

1375 1376
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1377
			task_numa_find_cpu(&env, taskimp, groupimp);
1378 1379 1380
		}
	}

1381 1382 1383 1384 1385 1386 1387 1388
	/*
	 * 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.
	 */
1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401
	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;
1402

1403 1404 1405 1406 1407 1408
	/*
	 * 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);

1409
	if (env.best_task == NULL) {
1410 1411 1412
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1413 1414 1415 1416
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1417 1418
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1419 1420
	put_task_struct(env.best_task);
	return ret;
1421 1422
}

1423 1424 1425
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1426 1427
	unsigned long interval = HZ;

1428
	/* This task has no NUMA fault statistics yet */
1429
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults_memory))
1430 1431
		return;

1432
	/* Periodically retry migrating the task to the preferred node */
1433 1434
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1435 1436

	/* Success if task is already running on preferred CPU */
1437
	if (task_node(p) == p->numa_preferred_nid)
1438 1439 1440
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1441
	task_numa_migrate(p);
1442 1443
}

1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475
/*
 * 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);
	}
}

1476 1477 1478
/*
 * 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
1479 1480 1481
 * 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.
1482 1483
 */
#define NUMA_PERIOD_SLOTS 10
1484
#define NUMA_PERIOD_THRESHOLD 7
1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540

/*
 * 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
		 */
1541
		ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared + 1));
1542 1543 1544 1545 1546 1547 1548 1549
		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));
}

1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577
/*
 * 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;
}

1578 1579
static void task_numa_placement(struct task_struct *p)
{
1580 1581
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
1582
	unsigned long fault_types[2] = { 0, 0 };
1583 1584
	unsigned long total_faults;
	u64 runtime, period;
1585
	spinlock_t *group_lock = NULL;
1586

1587
	seq = ACCESS_ONCE(p->mm->numa_scan_seq);
1588 1589 1590
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
1591
	p->numa_scan_period_max = task_scan_max(p);
1592

1593 1594 1595 1596
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

1597 1598 1599
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
1600
		spin_lock_irq(group_lock);
1601 1602
	}

1603 1604
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
1605
		unsigned long faults = 0, group_faults = 0;
1606
		int priv, i;
1607

1608
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
1609
			long diff, f_diff, f_weight;
1610

1611
			i = task_faults_idx(nid, priv);
1612

1613
			/* Decay existing window, copy faults since last scan */
1614
			diff = p->numa_faults_buffer_memory[i] - p->numa_faults_memory[i] / 2;
1615 1616
			fault_types[priv] += p->numa_faults_buffer_memory[i];
			p->numa_faults_buffer_memory[i] = 0;
1617

1618 1619 1620 1621 1622 1623 1624 1625 1626 1627
			/*
			 * 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);
1628
			f_diff = f_weight - p->numa_faults_cpu[i] / 2;
1629 1630
			p->numa_faults_buffer_cpu[i] = 0;

1631 1632
			p->numa_faults_memory[i] += diff;
			p->numa_faults_cpu[i] += f_diff;
1633
			faults += p->numa_faults_memory[i];
1634
			p->total_numa_faults += diff;
1635 1636
			if (p->numa_group) {
				/* safe because we can only change our own group */
1637
				p->numa_group->faults[i] += diff;
1638
				p->numa_group->faults_cpu[i] += f_diff;
1639 1640
				p->numa_group->total_faults += diff;
				group_faults += p->numa_group->faults[i];
1641
			}
1642 1643
		}

1644 1645 1646 1647
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
1648 1649 1650 1651 1652 1653 1654

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

1655 1656
	update_task_scan_period(p, fault_types[0], fault_types[1]);

1657
	if (p->numa_group) {
1658
		update_numa_active_node_mask(p->numa_group);
1659
		spin_unlock_irq(group_lock);
1660
		max_nid = max_group_nid;
1661 1662
	}

1663 1664 1665 1666 1667 1668 1669
	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);
1670
	}
1671 1672
}

1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683
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);
}

1684 1685
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
1686 1687 1688 1689 1690 1691 1692 1693 1694
{
	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) +
1695
				    4*nr_node_ids*sizeof(unsigned long);
1696 1697 1698 1699 1700 1701 1702 1703

		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);
1704
		grp->gid = p->pid;
1705
		/* Second half of the array tracks nids where faults happen */
1706 1707
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
1708

1709 1710
		node_set(task_node(current), grp->active_nodes);

1711
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1712
			grp->faults[i] = p->numa_faults_memory[i];
1713

1714
		grp->total_faults = p->total_numa_faults;
1715

1716 1717 1718 1719 1720 1721 1722 1723 1724
		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))
1725
		goto no_join;
1726 1727 1728

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
1729
		goto no_join;
1730 1731 1732

	my_grp = p->numa_group;
	if (grp == my_grp)
1733
		goto no_join;
1734 1735 1736 1737 1738 1739

	/*
	 * 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)
1740
		goto no_join;
1741 1742 1743 1744 1745

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
1746
		goto no_join;
1747

1748 1749 1750 1751 1752 1753 1754
	/* 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;
1755

1756 1757 1758
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

1759
	if (join && !get_numa_group(grp))
1760
		goto no_join;
1761 1762 1763 1764 1765 1766

	rcu_read_unlock();

	if (!join)
		return;

1767 1768
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
1769

1770
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
1771 1772
		my_grp->faults[i] -= p->numa_faults_memory[i];
		grp->faults[i] += p->numa_faults_memory[i];
1773
	}
1774 1775
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
1776 1777 1778 1779 1780 1781

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

	spin_unlock(&my_grp->lock);
1782
	spin_unlock_irq(&grp->lock);
1783 1784 1785 1786

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
1787 1788 1789 1790 1791
	return;

no_join:
	rcu_read_unlock();
	return;
1792 1793 1794 1795 1796
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
1797
	void *numa_faults = p->numa_faults_memory;
1798 1799
	unsigned long flags;
	int i;
1800 1801

	if (grp) {
1802
		spin_lock_irqsave(&grp->lock, flags);
1803
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1804
			grp->faults[i] -= p->numa_faults_memory[i];
1805
		grp->total_faults -= p->total_numa_faults;
1806

1807 1808
		list_del(&p->numa_entry);
		grp->nr_tasks--;
1809
		spin_unlock_irqrestore(&grp->lock, flags);
1810
		RCU_INIT_POINTER(p->numa_group, NULL);
1811 1812 1813
		put_numa_group(grp);
	}

1814 1815
	p->numa_faults_memory = NULL;
	p->numa_faults_buffer_memory = NULL;
1816 1817
	p->numa_faults_cpu= NULL;
	p->numa_faults_buffer_cpu = NULL;
1818
	kfree(numa_faults);
1819 1820
}

1821 1822 1823
/*
 * Got a PROT_NONE fault for a page on @node.
 */
1824
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
1825 1826
{
	struct task_struct *p = current;
1827
	bool migrated = flags & TNF_MIGRATED;
1828
	int cpu_node = task_node(current);
1829
	int local = !!(flags & TNF_FAULT_LOCAL);
1830
	int priv;
1831

1832
	if (!numabalancing_enabled)
1833 1834
		return;

1835 1836 1837 1838
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

1839
	/* Allocate buffer to track faults on a per-node basis */
1840
	if (unlikely(!p->numa_faults_memory)) {
1841 1842
		int size = sizeof(*p->numa_faults_memory) *
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
1843

1844
		p->numa_faults_memory = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
1845
		if (!p->numa_faults_memory)
1846
			return;
1847

1848
		BUG_ON(p->numa_faults_buffer_memory);
1849 1850 1851 1852 1853 1854
		/*
		 * 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.
		 */
1855 1856 1857
		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);
1858
		p->total_numa_faults = 0;
1859
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
1860
	}
1861

1862 1863 1864 1865 1866 1867 1868 1869
	/*
	 * 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);
1870
		if (!priv && !(flags & TNF_NO_GROUP))
1871
			task_numa_group(p, last_cpupid, flags, &priv);
1872 1873
	}

1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884
	/*
	 * 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;

1885
	task_numa_placement(p);
1886

1887 1888 1889 1890 1891
	/*
	 * 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))
1892 1893
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
1894 1895 1896
	if (migrated)
		p->numa_pages_migrated += pages;

1897 1898
	p->numa_faults_buffer_memory[task_faults_idx(mem_node, priv)] += pages;
	p->numa_faults_buffer_cpu[task_faults_idx(cpu_node, priv)] += pages;
1899
	p->numa_faults_locality[local] += pages;
1900 1901
}

1902 1903 1904 1905 1906 1907
static void reset_ptenuma_scan(struct task_struct *p)
{
	ACCESS_ONCE(p->mm->numa_scan_seq)++;
	p->mm->numa_scan_offset = 0;
}

1908 1909 1910 1911 1912 1913 1914 1915 1916
/*
 * 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;
1917
	struct vm_area_struct *vma;
1918
	unsigned long start, end;
1919
	unsigned long nr_pte_updates = 0;
1920
	long pages;
1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935

	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;

1936
	if (!mm->numa_next_scan) {
1937 1938
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
1939 1940
	}

1941 1942 1943 1944 1945 1946 1947
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

1948 1949 1950 1951
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
1952

1953
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
1954 1955 1956
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

1957 1958 1959 1960 1961 1962
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

1963 1964 1965 1966 1967
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
	if (!pages)
		return;
1968

1969
	down_read(&mm->mmap_sem);
1970
	vma = find_vma(mm, start);
1971 1972
	if (!vma) {
		reset_ptenuma_scan(p);
1973
		start = 0;
1974 1975
		vma = mm->mmap;
	}
1976
	for (; vma; vma = vma->vm_next) {
1977
		if (!vma_migratable(vma) || !vma_policy_mof(vma))
1978 1979
			continue;

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
		/*
		 * 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
1990 1991 1992 1993 1994 1995
		/*
		 * 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;
1996

1997 1998 1999 2000
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2001 2002 2003 2004 2005 2006 2007 2008 2009
			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;
2010

2011 2012 2013
			start = end;
			if (pages <= 0)
				goto out;
2014 2015

			cond_resched();
2016
		} while (end != vma->vm_end);
2017
	}
2018

2019
out:
2020
	/*
P
Peter Zijlstra 已提交
2021 2022 2023 2024
	 * 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.
2025 2026
	 */
	if (vma)
2027
		mm->numa_scan_offset = start;
2028 2029 2030
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056
}

/*
 * 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) {
2057
		if (!curr->node_stamp)
2058
			curr->numa_scan_period = task_scan_min(curr);
2059
		curr->node_stamp += period;
2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070

		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)
{
}
2071 2072 2073 2074 2075 2076 2077 2078

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)
{
}
2079 2080
#endif /* CONFIG_NUMA_BALANCING */

2081 2082 2083 2084
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2085
	if (!parent_entity(se))
2086
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2087
#ifdef CONFIG_SMP
2088 2089 2090 2091 2092 2093
	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);
	}
2094
#endif
2095 2096 2097 2098 2099 2100 2101
	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);
2102
	if (!parent_entity(se))
2103
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2104 2105
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2106
		list_del_init(&se->group_node);
2107
	}
2108 2109 2110
	cfs_rq->nr_running--;
}

2111 2112
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2113 2114 2115 2116 2117 2118 2119 2120 2121
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().
	 */
2122
	tg_weight = atomic_long_read(&tg->load_avg);
2123
	tg_weight -= cfs_rq->tg_load_contrib;
2124 2125 2126 2127 2128
	tg_weight += cfs_rq->load.weight;

	return tg_weight;
}

2129
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2130
{
2131
	long tg_weight, load, shares;
2132

2133
	tg_weight = calc_tg_weight(tg, cfs_rq);
2134
	load = cfs_rq->load.weight;
2135 2136

	shares = (tg->shares * load);
2137 2138
	if (tg_weight)
		shares /= tg_weight;
2139 2140 2141 2142 2143 2144 2145 2146 2147

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

	return shares;
}
# else /* CONFIG_SMP */
2148
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2149 2150 2151 2152
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
2153 2154 2155
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2156 2157 2158 2159
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2160
		account_entity_dequeue(cfs_rq, se);
2161
	}
P
Peter Zijlstra 已提交
2162 2163 2164 2165 2166 2167 2168

	update_load_set(&se->load, weight);

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

2169 2170
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2171
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2172 2173 2174
{
	struct task_group *tg;
	struct sched_entity *se;
2175
	long shares;
P
Peter Zijlstra 已提交
2176 2177 2178

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2179
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2180
		return;
2181 2182 2183 2184
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2185
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2186 2187 2188 2189

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2190
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2191 2192 2193 2194
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2195
#ifdef CONFIG_SMP
2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223
/*
 * 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,
};

2224 2225 2226 2227 2228 2229
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241
	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
2242 2243
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
2244 2245 2246 2247 2248 2249
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
2250 2251
	}

2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282
	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];
2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316
}

/*
 * 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)
{
2317 2318
	u64 delta, periods;
	u32 runnable_contrib;
2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351
	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;
2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371
		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;
2372 2373 2374 2375 2376 2377 2378 2379 2380 2381
	}

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

	return decayed;
}

2382
/* Synchronize an entity's decay with its parenting cfs_rq.*/
2383
static inline u64 __synchronize_entity_decay(struct sched_entity *se)
2384 2385 2386 2387 2388 2389
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 decays = atomic64_read(&cfs_rq->decay_counter);

	decays -= se->avg.decay_count;
	if (!decays)
2390
		return 0;
2391 2392 2393

	se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
	se->avg.decay_count = 0;
2394 2395

	return decays;
2396 2397
}

2398 2399 2400 2401 2402
#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;
2403
	long tg_contrib;
2404 2405 2406 2407

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

2408 2409 2410
	if (!tg_contrib)
		return;

2411 2412
	if (force_update || abs(tg_contrib) > cfs_rq->tg_load_contrib / 8) {
		atomic_long_add(tg_contrib, &tg->load_avg);
2413 2414 2415
		cfs_rq->tg_load_contrib += tg_contrib;
	}
}
2416

2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427
/*
 * 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 */
2428
	contrib = div_u64((u64)sa->runnable_avg_sum << NICE_0_SHIFT,
2429 2430 2431 2432 2433 2434 2435 2436 2437
			  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;
	}
}

2438 2439 2440 2441
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;
2442 2443
	int runnable_avg;

2444 2445 2446
	u64 contrib;

	contrib = cfs_rq->tg_load_contrib * tg->shares;
2447 2448
	se->avg.load_avg_contrib = div_u64(contrib,
				     atomic_long_read(&tg->load_avg) + 1);
2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477

	/*
	 * 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;
	}
2478
}
2479 2480 2481 2482 2483 2484

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);
}
2485
#else /* CONFIG_FAIR_GROUP_SCHED */
2486 2487
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update) {}
2488 2489
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq) {}
2490
static inline void __update_group_entity_contrib(struct sched_entity *se) {}
2491
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2492
#endif /* CONFIG_FAIR_GROUP_SCHED */
2493

2494 2495 2496 2497 2498 2499 2500 2501 2502 2503
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);
}

2504 2505 2506 2507 2508
/* 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;

2509 2510 2511
	if (entity_is_task(se)) {
		__update_task_entity_contrib(se);
	} else {
2512
		__update_tg_runnable_avg(&se->avg, group_cfs_rq(se));
2513 2514
		__update_group_entity_contrib(se);
	}
2515 2516 2517 2518

	return se->avg.load_avg_contrib - old_contrib;
}

2519 2520 2521 2522 2523 2524 2525 2526 2527
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;
}

2528 2529
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);

2530
/* Update a sched_entity's runnable average */
2531 2532
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq)
2533
{
2534 2535
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	long contrib_delta;
2536
	u64 now;
2537

2538 2539 2540 2541 2542 2543 2544 2545 2546 2547
	/*
	 * 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))
2548 2549 2550
		return;

	contrib_delta = __update_entity_load_avg_contrib(se);
2551 2552 2553 2554

	if (!update_cfs_rq)
		return;

2555 2556
	if (se->on_rq)
		cfs_rq->runnable_load_avg += contrib_delta;
2557 2558 2559 2560 2561 2562 2563 2564
	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.
 */
2565
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq, int force_update)
2566
{
2567
	u64 now = cfs_rq_clock_task(cfs_rq) >> 20;
2568 2569 2570
	u64 decays;

	decays = now - cfs_rq->last_decay;
2571
	if (!decays && !force_update)
2572 2573
		return;

2574 2575 2576
	if (atomic_long_read(&cfs_rq->removed_load)) {
		unsigned long removed_load;
		removed_load = atomic_long_xchg(&cfs_rq->removed_load, 0);
2577 2578
		subtract_blocked_load_contrib(cfs_rq, removed_load);
	}
2579

2580 2581 2582 2583 2584 2585
	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;
	}
2586 2587

	__update_cfs_rq_tg_load_contrib(cfs_rq, force_update);
2588
}
2589

2590 2591
/* 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,
2592 2593
						  struct sched_entity *se,
						  int wakeup)
2594
{
2595 2596 2597 2598
	/*
	 * 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.
2599 2600 2601 2602
	 *
	 * 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.
2603 2604
	 */
	if (unlikely(se->avg.decay_count <= 0)) {
2605
		se->avg.last_runnable_update = rq_clock_task(rq_of(cfs_rq));
2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620
		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;
		}
2621 2622
		wakeup = 0;
	} else {
2623
		__synchronize_entity_decay(se);
2624 2625
	}

2626 2627
	/* migrated tasks did not contribute to our blocked load */
	if (wakeup) {
2628
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
2629 2630
		update_entity_load_avg(se, 0);
	}
2631

2632
	cfs_rq->runnable_load_avg += se->avg.load_avg_contrib;
2633 2634
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !wakeup);
2635 2636
}

2637 2638 2639 2640 2641
/*
 * 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.
 */
2642
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2643 2644
						  struct sched_entity *se,
						  int sleep)
2645
{
2646
	update_entity_load_avg(se, 1);
2647 2648
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !sleep);
2649

2650
	cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib;
2651 2652 2653 2654
	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 */
2655
}
2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676

/*
 * 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);
}

2677 2678
static int idle_balance(struct rq *this_rq);

2679 2680
#else /* CONFIG_SMP */

2681 2682
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq) {}
2683
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2684
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2685 2686
					   struct sched_entity *se,
					   int wakeup) {}
2687
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2688 2689
					   struct sched_entity *se,
					   int sleep) {}
2690 2691
static inline void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
					      int force_update) {}
2692 2693 2694 2695 2696 2697

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

2698
#endif /* CONFIG_SMP */
2699

2700
static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
2701 2702
{
#ifdef CONFIG_SCHEDSTATS
2703 2704 2705 2706 2707
	struct task_struct *tsk = NULL;

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

2708
	if (se->statistics.sleep_start) {
2709
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.sleep_start;
2710 2711 2712 2713

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

2714 2715
		if (unlikely(delta > se->statistics.sleep_max))
			se->statistics.sleep_max = delta;
2716

2717
		se->statistics.sleep_start = 0;
2718
		se->statistics.sum_sleep_runtime += delta;
A
Arjan van de Ven 已提交
2719

2720
		if (tsk) {
2721
			account_scheduler_latency(tsk, delta >> 10, 1);
2722 2723
			trace_sched_stat_sleep(tsk, delta);
		}
2724
	}
2725
	if (se->statistics.block_start) {
2726
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.block_start;
2727 2728 2729 2730

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

2731 2732
		if (unlikely(delta > se->statistics.block_max))
			se->statistics.block_max = delta;
2733

2734
		se->statistics.block_start = 0;
2735
		se->statistics.sum_sleep_runtime += delta;
I
Ingo Molnar 已提交
2736

2737
		if (tsk) {
2738
			if (tsk->in_iowait) {
2739 2740
				se->statistics.iowait_sum += delta;
				se->statistics.iowait_count++;
2741
				trace_sched_stat_iowait(tsk, delta);
2742 2743
			}

2744 2745
			trace_sched_stat_blocked(tsk, delta);

2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756
			/*
			 * 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
2757
		}
2758 2759 2760 2761
	}
#endif
}

P
Peter Zijlstra 已提交
2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774
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
}

2775 2776 2777
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
2778
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
2779

2780 2781 2782 2783 2784 2785
	/*
	 * 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
2786
	if (initial && sched_feat(START_DEBIT))
2787
		vruntime += sched_vslice(cfs_rq, se);
2788

2789
	/* sleeps up to a single latency don't count. */
2790
	if (!initial) {
2791
		unsigned long thresh = sysctl_sched_latency;
2792

2793 2794 2795 2796 2797 2798
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
2799

2800
		vruntime -= thresh;
2801 2802
	}

2803
	/* ensure we never gain time by being placed backwards. */
2804
	se->vruntime = max_vruntime(se->vruntime, vruntime);
2805 2806
}

2807 2808
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

2809
static void
2810
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2811
{
2812 2813
	/*
	 * Update the normalized vruntime before updating min_vruntime
2814
	 * through calling update_curr().
2815
	 */
2816
	if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
2817 2818
		se->vruntime += cfs_rq->min_vruntime;

2819
	/*
2820
	 * Update run-time statistics of the 'current'.
2821
	 */
2822
	update_curr(cfs_rq);
2823
	enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP);
2824 2825
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
2826

2827
	if (flags & ENQUEUE_WAKEUP) {
2828
		place_entity(cfs_rq, se, 0);
2829
		enqueue_sleeper(cfs_rq, se);
I
Ingo Molnar 已提交
2830
	}
2831

2832
	update_stats_enqueue(cfs_rq, se);
P
Peter Zijlstra 已提交
2833
	check_spread(cfs_rq, se);
2834 2835
	if (se != cfs_rq->curr)
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
2836
	se->on_rq = 1;
2837

2838
	if (cfs_rq->nr_running == 1) {
2839
		list_add_leaf_cfs_rq(cfs_rq);
2840 2841
		check_enqueue_throttle(cfs_rq);
	}
2842 2843
}

2844
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
2845
{
2846 2847
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2848
		if (cfs_rq->last != se)
2849
			break;
2850 2851

		cfs_rq->last = NULL;
2852 2853
	}
}
P
Peter Zijlstra 已提交
2854

2855 2856 2857 2858
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2859
		if (cfs_rq->next != se)
2860
			break;
2861 2862

		cfs_rq->next = NULL;
2863
	}
P
Peter Zijlstra 已提交
2864 2865
}

2866 2867 2868 2869
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
2870
		if (cfs_rq->skip != se)
2871
			break;
2872 2873

		cfs_rq->skip = NULL;
2874 2875 2876
	}
}

P
Peter Zijlstra 已提交
2877 2878
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2879 2880 2881 2882 2883
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
2884 2885 2886

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
2887 2888
}

2889
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
2890

2891
static void
2892
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2893
{
2894 2895 2896 2897
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
2898
	dequeue_entity_load_avg(cfs_rq, se, flags & DEQUEUE_SLEEP);
2899

2900
	update_stats_dequeue(cfs_rq, se);
2901
	if (flags & DEQUEUE_SLEEP) {
P
Peter Zijlstra 已提交
2902
#ifdef CONFIG_SCHEDSTATS
2903 2904 2905 2906
		if (entity_is_task(se)) {
			struct task_struct *tsk = task_of(se);

			if (tsk->state & TASK_INTERRUPTIBLE)
2907
				se->statistics.sleep_start = rq_clock(rq_of(cfs_rq));
2908
			if (tsk->state & TASK_UNINTERRUPTIBLE)
2909
				se->statistics.block_start = rq_clock(rq_of(cfs_rq));
2910
		}
2911
#endif
P
Peter Zijlstra 已提交
2912 2913
	}

P
Peter Zijlstra 已提交
2914
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
2915

2916
	if (se != cfs_rq->curr)
2917
		__dequeue_entity(cfs_rq, se);
2918
	se->on_rq = 0;
2919
	account_entity_dequeue(cfs_rq, se);
2920 2921 2922 2923 2924 2925

	/*
	 * 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.
	 */
2926
	if (!(flags & DEQUEUE_SLEEP))
2927
		se->vruntime -= cfs_rq->min_vruntime;
2928

2929 2930 2931
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

2932
	update_min_vruntime(cfs_rq);
2933
	update_cfs_shares(cfs_rq);
2934 2935 2936 2937 2938
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
2939
static void
I
Ingo Molnar 已提交
2940
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
2941
{
2942
	unsigned long ideal_runtime, delta_exec;
2943 2944
	struct sched_entity *se;
	s64 delta;
2945

P
Peter Zijlstra 已提交
2946
	ideal_runtime = sched_slice(cfs_rq, curr);
2947
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
2948
	if (delta_exec > ideal_runtime) {
2949
		resched_curr(rq_of(cfs_rq));
2950 2951 2952 2953 2954
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965
		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;

2966 2967
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
2968

2969 2970
	if (delta < 0)
		return;
2971

2972
	if (delta > ideal_runtime)
2973
		resched_curr(rq_of(cfs_rq));
2974 2975
}

2976
static void
2977
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
2978
{
2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989
	/* '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);
	}

2990
	update_stats_curr_start(cfs_rq, se);
2991
	cfs_rq->curr = se;
I
Ingo Molnar 已提交
2992 2993 2994 2995 2996 2997
#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):
	 */
2998
	if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
2999
		se->statistics.slice_max = max(se->statistics.slice_max,
I
Ingo Molnar 已提交
3000 3001 3002
			se->sum_exec_runtime - se->prev_sum_exec_runtime);
	}
#endif
3003
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
3004 3005
}

3006 3007 3008
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

3009 3010 3011 3012 3013 3014 3015
/*
 * 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
 */
3016 3017
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3018
{
3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029
	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 */
3030

3031 3032 3033 3034 3035
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3036 3037 3038 3039 3040 3041 3042 3043 3044 3045
		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;
		}

3046 3047 3048
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3049

3050 3051 3052 3053 3054 3055
	/*
	 * 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;

3056 3057 3058 3059 3060 3061
	/*
	 * 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;

3062
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3063 3064

	return se;
3065 3066
}

3067
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3068

3069
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3070 3071 3072 3073 3074 3075
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3076
		update_curr(cfs_rq);
3077

3078 3079 3080
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

P
Peter Zijlstra 已提交
3081
	check_spread(cfs_rq, prev);
3082
	if (prev->on_rq) {
3083
		update_stats_wait_start(cfs_rq, prev);
3084 3085
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3086
		/* in !on_rq case, update occurred at dequeue */
3087
		update_entity_load_avg(prev, 1);
3088
	}
3089
	cfs_rq->curr = NULL;
3090 3091
}

P
Peter Zijlstra 已提交
3092 3093
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3094 3095
{
	/*
3096
	 * Update run-time statistics of the 'current'.
3097
	 */
3098
	update_curr(cfs_rq);
3099

3100 3101 3102
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3103
	update_entity_load_avg(curr, 1);
3104
	update_cfs_rq_blocked_load(cfs_rq, 1);
3105
	update_cfs_shares(cfs_rq);
3106

P
Peter Zijlstra 已提交
3107 3108 3109 3110 3111
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3112
	if (queued) {
3113
		resched_curr(rq_of(cfs_rq));
3114 3115
		return;
	}
P
Peter Zijlstra 已提交
3116 3117 3118 3119 3120 3121 3122 3123
	/*
	 * 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
3124
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3125
		check_preempt_tick(cfs_rq, curr);
3126 3127
}

3128 3129 3130 3131 3132 3133

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

#ifdef CONFIG_CFS_BANDWIDTH
3134 3135

#ifdef HAVE_JUMP_LABEL
3136
static struct static_key __cfs_bandwidth_used;
3137 3138 3139

static inline bool cfs_bandwidth_used(void)
{
3140
	return static_key_false(&__cfs_bandwidth_used);
3141 3142
}

3143
void cfs_bandwidth_usage_inc(void)
3144
{
3145 3146 3147 3148 3149 3150
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3151 3152 3153 3154 3155 3156 3157
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3158 3159
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3160 3161
#endif /* HAVE_JUMP_LABEL */

3162 3163 3164 3165 3166 3167 3168 3169
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3170 3171 3172 3173 3174 3175

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

P
Paul Turner 已提交
3176 3177 3178 3179 3180 3181 3182
/*
 * 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
 */
3183
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194
{
	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);
}

3195 3196 3197 3198 3199
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3200 3201 3202 3203 3204 3205
/* 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;

3206
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3207 3208
}

3209 3210
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3211 3212 3213
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3214
	u64 amount = 0, min_amount, expires;
3215 3216 3217 3218 3219 3220 3221

	/* 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;
3222
	else {
P
Paul Turner 已提交
3223 3224 3225 3226 3227 3228 3229 3230
		/*
		 * 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);
3231
			__start_cfs_bandwidth(cfs_b, false);
P
Paul Turner 已提交
3232
		}
3233 3234 3235 3236 3237 3238

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3239
	}
P
Paul Turner 已提交
3240
	expires = cfs_b->runtime_expires;
3241 3242 3243
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3244 3245 3246 3247 3248 3249 3250
	/*
	 * 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;
3251 3252

	return cfs_rq->runtime_remaining > 0;
3253 3254
}

P
Paul Turner 已提交
3255 3256 3257 3258 3259
/*
 * 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)
3260
{
P
Paul Turner 已提交
3261 3262 3263
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3264
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3265 3266
		return;

P
Paul Turner 已提交
3267 3268 3269 3270 3271 3272 3273 3274 3275
	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
3276 3277 3278
	 * 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
3279 3280
	 */

3281
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3282 3283 3284 3285 3286 3287 3288 3289
		/* 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;
	}
}

3290
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3291 3292
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3293
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3294 3295 3296
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3297 3298
		return;

3299 3300 3301 3302 3303
	/*
	 * 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))
3304
		resched_curr(rq_of(cfs_rq));
3305 3306
}

3307
static __always_inline
3308
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3309
{
3310
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3311 3312 3313 3314 3315
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3316 3317
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3318
	return cfs_bandwidth_used() && cfs_rq->throttled;
3319 3320
}

3321 3322 3323
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3324
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352
}

/*
 * 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) {
3353
		/* adjust cfs_rq_clock_task() */
3354
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3355
					     cfs_rq->throttled_clock_task;
3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366
	}
#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)];

3367 3368
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3369
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3370 3371 3372 3373 3374
	cfs_rq->throttle_count++;

	return 0;
}

3375
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3376 3377 3378 3379 3380 3381 3382 3383
{
	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))];

3384
	/* freeze hierarchy runnable averages while throttled */
3385 3386 3387
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404

	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)
3405
		sub_nr_running(rq, task_delta);
3406 3407

	cfs_rq->throttled = 1;
3408
	cfs_rq->throttled_clock = rq_clock(rq);
3409
	raw_spin_lock(&cfs_b->lock);
3410 3411 3412 3413 3414
	/*
	 * 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);
3415
	if (!cfs_b->timer_active)
3416
		__start_cfs_bandwidth(cfs_b, false);
3417 3418 3419
	raw_spin_unlock(&cfs_b->lock);
}

3420
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3421 3422 3423 3424 3425 3426 3427
{
	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;

3428
	se = cfs_rq->tg->se[cpu_of(rq)];
3429 3430

	cfs_rq->throttled = 0;
3431 3432 3433

	update_rq_clock(rq);

3434
	raw_spin_lock(&cfs_b->lock);
3435
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
3436 3437 3438
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

3439 3440 3441
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459
	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)
3460
		add_nr_running(rq, task_delta);
3461 3462 3463

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
3464
		resched_curr(rq);
3465 3466 3467 3468 3469 3470
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
3471 3472
	u64 runtime;
	u64 starting_runtime = remaining;
3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502

	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();

3503
	return starting_runtime - remaining;
3504 3505
}

3506 3507 3508 3509 3510 3511 3512 3513
/*
 * 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)
{
3514
	u64 runtime, runtime_expires;
3515
	int throttled;
3516 3517 3518

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
3519
		goto out_deactivate;
3520

3521
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
3522
	cfs_b->nr_periods += overrun;
3523

3524 3525 3526 3527 3528 3529
	/*
	 * 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
3530

3531 3532 3533 3534 3535 3536 3537
	/*
	 * 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
3538 3539
	__refill_cfs_bandwidth_runtime(cfs_b);

3540 3541 3542
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
3543
		return 0;
3544 3545
	}

3546 3547 3548
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

3549 3550 3551
	runtime_expires = cfs_b->runtime_expires;

	/*
3552 3553 3554 3555 3556
	 * 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.
3557
	 */
3558 3559
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
3560 3561 3562 3563 3564 3565 3566
		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);
3567 3568

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3569
	}
3570

3571 3572 3573 3574 3575 3576 3577
	/*
	 * 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;
3578

3579 3580 3581 3582 3583
	return 0;

out_deactivate:
	cfs_b->timer_active = 0;
	return 1;
3584
}
3585

3586 3587 3588 3589 3590 3591 3592
/* 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;

3593 3594 3595 3596 3597 3598 3599
/*
 * 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.
 */
3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655
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)
{
3656 3657 3658
	if (!cfs_bandwidth_used())
		return;

3659
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674
		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 */
3675 3676 3677
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
3678
		return;
3679
	}
3680

3681
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
3682
		runtime = cfs_b->runtime;
3683

3684 3685 3686 3687 3688 3689 3690 3691 3692 3693
	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)
3694
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3695 3696 3697
	raw_spin_unlock(&cfs_b->lock);
}

3698 3699 3700 3701 3702 3703 3704
/*
 * 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)
{
3705 3706 3707
	if (!cfs_bandwidth_used())
		return;

3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722
	/* 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() */
3723
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3724
{
3725
	if (!cfs_bandwidth_used())
3726
		return false;
3727

3728
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
3729
		return false;
3730 3731 3732 3733 3734 3735

	/*
	 * 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))
3736
		return true;
3737 3738

	throttle_cfs_rq(cfs_rq);
3739
	return true;
3740
}
3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758

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;

3759
	raw_spin_lock(&cfs_b->lock);
3760 3761 3762 3763 3764 3765 3766 3767 3768
	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);
	}
3769
	raw_spin_unlock(&cfs_b->lock);
3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794

	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 */
3795
void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force)
3796 3797 3798 3799 3800 3801 3802
{
	/*
	 * 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
	 */
3803 3804 3805
	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 */
3806
		raw_spin_unlock(&cfs_b->lock);
3807
		cpu_relax();
3808 3809
		raw_spin_lock(&cfs_b->lock);
		/* if someone else restarted the timer then we're done */
3810
		if (!force && cfs_b->timer_active)
3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823
			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);
}

3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836
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);
	}
}

3837
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848
{
	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
		 */
3849
		cfs_rq->runtime_remaining = 1;
3850 3851 3852 3853 3854 3855
		/*
		 * 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;

3856 3857 3858 3859 3860 3861
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
3862 3863
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
3864
	return rq_clock_task(rq_of(cfs_rq));
3865 3866
}

3867
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
3868
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
3869
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
3870
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
3871 3872 3873 3874 3875

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886

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;
}
3887 3888 3889 3890 3891

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) {}
3892 3893
#endif

3894 3895 3896 3897 3898
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) {}
3899
static inline void update_runtime_enabled(struct rq *rq) {}
3900
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
3901 3902 3903

#endif /* CONFIG_CFS_BANDWIDTH */

3904 3905 3906 3907
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
3908 3909 3910 3911 3912 3913 3914 3915
#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);

3916
	if (cfs_rq->nr_running > 1) {
P
Peter Zijlstra 已提交
3917 3918 3919 3920 3921 3922
		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)
3923
				resched_curr(rq);
P
Peter Zijlstra 已提交
3924 3925
			return;
		}
3926
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
3927 3928
	}
}
3929 3930 3931 3932 3933 3934 3935 3936 3937 3938

/*
 * 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;

3939
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
3940 3941 3942 3943 3944
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
3945
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
3946 3947 3948 3949
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
3950 3951 3952 3953

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
3954 3955
#endif

3956 3957 3958 3959 3960
/*
 * 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:
 */
3961
static void
3962
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
3963 3964
{
	struct cfs_rq *cfs_rq;
3965
	struct sched_entity *se = &p->se;
3966 3967

	for_each_sched_entity(se) {
3968
		if (se->on_rq)
3969 3970
			break;
		cfs_rq = cfs_rq_of(se);
3971
		enqueue_entity(cfs_rq, se, flags);
3972 3973 3974 3975 3976 3977 3978 3979 3980

		/*
		 * 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;
3981
		cfs_rq->h_nr_running++;
3982

3983
		flags = ENQUEUE_WAKEUP;
3984
	}
P
Peter Zijlstra 已提交
3985

P
Peter Zijlstra 已提交
3986
	for_each_sched_entity(se) {
3987
		cfs_rq = cfs_rq_of(se);
3988
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
3989

3990 3991 3992
		if (cfs_rq_throttled(cfs_rq))
			break;

3993
		update_cfs_shares(cfs_rq);
3994
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
3995 3996
	}

3997 3998
	if (!se) {
		update_rq_runnable_avg(rq, rq->nr_running);
3999
		add_nr_running(rq, 1);
4000
	}
4001
	hrtick_update(rq);
4002 4003
}

4004 4005
static void set_next_buddy(struct sched_entity *se);

4006 4007 4008 4009 4010
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
4011
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4012 4013
{
	struct cfs_rq *cfs_rq;
4014
	struct sched_entity *se = &p->se;
4015
	int task_sleep = flags & DEQUEUE_SLEEP;
4016 4017 4018

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4019
		dequeue_entity(cfs_rq, se, flags);
4020 4021 4022 4023 4024 4025 4026 4027 4028

		/*
		 * 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;
4029
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4030

4031
		/* Don't dequeue parent if it has other entities besides us */
4032 4033 4034 4035 4036 4037 4038
		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));
4039 4040 4041

			/* avoid re-evaluating load for this entity */
			se = parent_entity(se);
4042
			break;
4043
		}
4044
		flags |= DEQUEUE_SLEEP;
4045
	}
P
Peter Zijlstra 已提交
4046

P
Peter Zijlstra 已提交
4047
	for_each_sched_entity(se) {
4048
		cfs_rq = cfs_rq_of(se);
4049
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4050

4051 4052 4053
		if (cfs_rq_throttled(cfs_rq))
			break;

4054
		update_cfs_shares(cfs_rq);
4055
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
4056 4057
	}

4058
	if (!se) {
4059
		sub_nr_running(rq, 1);
4060 4061
		update_rq_runnable_avg(rq, 1);
	}
4062
	hrtick_update(rq);
4063 4064
}

4065
#ifdef CONFIG_SMP
4066 4067 4068
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
4069
	return cpu_rq(cpu)->cfs.runnable_load_avg;
4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104
}

/*
 * 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);
}

4105
static unsigned long capacity_of(int cpu)
4106
{
4107
	return cpu_rq(cpu)->cpu_capacity;
4108 4109 4110 4111 4112
}

static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
4113
	unsigned long nr_running = ACCESS_ONCE(rq->cfs.h_nr_running);
4114
	unsigned long load_avg = rq->cfs.runnable_load_avg;
4115 4116

	if (nr_running)
4117
		return load_avg / nr_running;
4118 4119 4120 4121

	return 0;
}

4122 4123 4124 4125 4126 4127 4128
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.
	 */
4129
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
4130
		current->wakee_flips >>= 1;
4131 4132 4133 4134 4135 4136 4137 4138
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}
4139

4140
static void task_waking_fair(struct task_struct *p)
4141 4142 4143
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
4144 4145 4146 4147
	u64 min_vruntime;

#ifndef CONFIG_64BIT
	u64 min_vruntime_copy;
4148

4149 4150 4151 4152 4153 4154 4155 4156
	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
4157

4158
	se->vruntime -= min_vruntime;
4159
	record_wakee(p);
4160 4161
}

4162
#ifdef CONFIG_FAIR_GROUP_SCHED
4163 4164 4165 4166 4167 4168
/*
 * 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.
4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211
 *
 * 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.
4212
 */
P
Peter Zijlstra 已提交
4213
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
4214
{
P
Peter Zijlstra 已提交
4215
	struct sched_entity *se = tg->se[cpu];
4216

4217
	if (!tg->parent)	/* the trivial, non-cgroup case */
4218 4219
		return wl;

P
Peter Zijlstra 已提交
4220
	for_each_sched_entity(se) {
4221
		long w, W;
P
Peter Zijlstra 已提交
4222

4223
		tg = se->my_q->tg;
4224

4225 4226 4227 4228
		/*
		 * W = @wg + \Sum rw_j
		 */
		W = wg + calc_tg_weight(tg, se->my_q);
P
Peter Zijlstra 已提交
4229

4230 4231 4232 4233
		/*
		 * w = rw_i + @wl
		 */
		w = se->my_q->load.weight + wl;
4234

4235 4236 4237 4238 4239
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
			wl = (w * tg->shares) / W;
4240 4241
		else
			wl = tg->shares;
4242

4243 4244 4245 4246 4247
		/*
		 * 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().
		 */
4248 4249
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
4250 4251 4252 4253

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
4254
		wl -= se->load.weight;
4255 4256 4257 4258 4259 4260 4261 4262

		/*
		 * 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
4263 4264
		wg = 0;
	}
4265

P
Peter Zijlstra 已提交
4266
	return wl;
4267 4268
}
#else
P
Peter Zijlstra 已提交
4269

4270
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
4271
{
4272
	return wl;
4273
}
P
Peter Zijlstra 已提交
4274

4275 4276
#endif

4277 4278
static int wake_wide(struct task_struct *p)
{
4279
	int factor = this_cpu_read(sd_llc_size);
4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298

	/*
	 * 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;
}

4299
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
4300
{
4301
	s64 this_load, load;
4302
	s64 this_eff_load, prev_eff_load;
4303 4304
	int idx, this_cpu, prev_cpu;
	struct task_group *tg;
4305
	unsigned long weight;
4306
	int balanced;
4307

4308 4309 4310 4311 4312 4313 4314
	/*
	 * If we wake multiple tasks be careful to not bounce
	 * ourselves around too much.
	 */
	if (wake_wide(p))
		return 0;

4315 4316 4317 4318 4319
	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);
4320

4321 4322 4323 4324 4325
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
4326 4327 4328 4329
	if (sync) {
		tg = task_group(current);
		weight = current->se.load.weight;

4330
		this_load += effective_load(tg, this_cpu, -weight, -weight);
4331 4332
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
4333

4334 4335
	tg = task_group(p);
	weight = p->se.load.weight;
4336

4337 4338
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
4339 4340 4341
	 * 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.
4342 4343 4344 4345
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
4346 4347
	this_eff_load = 100;
	this_eff_load *= capacity_of(prev_cpu);
4348

4349 4350
	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
4351

4352
	if (this_load > 0) {
4353 4354 4355 4356
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
4357
	}
4358

4359
	balanced = this_eff_load <= prev_eff_load;
4360

4361
	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
4362

4363 4364
	if (!balanced)
		return 0;
4365

4366 4367 4368 4369
	schedstat_inc(sd, ttwu_move_affine);
	schedstat_inc(p, se.statistics.nr_wakeups_affine);

	return 1;
4370 4371
}

4372 4373 4374 4375 4376
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
4377
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
4378
		  int this_cpu, int sd_flag)
4379
{
4380
	struct sched_group *idlest = NULL, *group = sd->groups;
4381
	unsigned long min_load = ULONG_MAX, this_load = 0;
4382
	int load_idx = sd->forkexec_idx;
4383
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
4384

4385 4386 4387
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

4388 4389 4390 4391
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
4392

4393 4394
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
4395
					tsk_cpus_allowed(p)))
4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413
			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;
		}

4414
		/* Adjust by relative CPU capacity of the group */
4415
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436

		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;
4437 4438 4439 4440
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
4441 4442 4443
	int i;

	/* Traverse only the allowed CPUs */
4444
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472
		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;
			}
4473 4474 4475
		}
	}

4476
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
4477
}
4478

4479 4480 4481
/*
 * Try and locate an idle CPU in the sched_domain.
 */
4482
static int select_idle_sibling(struct task_struct *p, int target)
4483
{
4484
	struct sched_domain *sd;
4485
	struct sched_group *sg;
4486
	int i = task_cpu(p);
4487

4488 4489
	if (idle_cpu(target))
		return target;
4490 4491

	/*
4492
	 * If the prevous cpu is cache affine and idle, don't be stupid.
4493
	 */
4494 4495
	if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
		return i;
4496 4497

	/*
4498
	 * Otherwise, iterate the domains and find an elegible idle cpu.
4499
	 */
4500
	sd = rcu_dereference(per_cpu(sd_llc, target));
4501
	for_each_lower_domain(sd) {
4502 4503 4504 4505 4506 4507 4508
		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)) {
4509
				if (i == target || !idle_cpu(i))
4510 4511
					goto next;
			}
4512

4513 4514 4515 4516 4517 4518 4519 4520
			target = cpumask_first_and(sched_group_cpus(sg),
					tsk_cpus_allowed(p));
			goto done;
next:
			sg = sg->next;
		} while (sg != sd->groups);
	}
done:
4521 4522 4523
	return target;
}

4524
/*
4525 4526 4527
 * 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.
4528
 *
4529 4530
 * 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.
4531
 *
4532
 * Returns the target cpu number.
4533 4534 4535
 *
 * preempt must be disabled.
 */
4536
static int
4537
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
4538
{
4539
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
4540 4541
	int cpu = smp_processor_id();
	int new_cpu = cpu;
4542
	int want_affine = 0;
4543
	int sync = wake_flags & WF_SYNC;
4544

4545
	if (p->nr_cpus_allowed == 1)
4546 4547
		return prev_cpu;

4548 4549
	if (sd_flag & SD_BALANCE_WAKE)
		want_affine = cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
4550

4551
	rcu_read_lock();
4552
	for_each_domain(cpu, tmp) {
4553 4554 4555
		if (!(tmp->flags & SD_LOAD_BALANCE))
			continue;

4556
		/*
4557 4558
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
4559
		 */
4560 4561 4562
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
4563
			break;
4564
		}
4565

4566
		if (tmp->flags & sd_flag)
4567 4568 4569
			sd = tmp;
	}

4570 4571
	if (affine_sd && cpu != prev_cpu && wake_affine(affine_sd, p, sync))
		prev_cpu = cpu;
4572

4573
	if (sd_flag & SD_BALANCE_WAKE) {
4574 4575
		new_cpu = select_idle_sibling(p, prev_cpu);
		goto unlock;
4576
	}
4577

4578 4579
	while (sd) {
		struct sched_group *group;
4580
		int weight;
4581

4582
		if (!(sd->flags & sd_flag)) {
4583 4584 4585
			sd = sd->child;
			continue;
		}
4586

4587
		group = find_idlest_group(sd, p, cpu, sd_flag);
4588 4589 4590 4591
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
4592

4593
		new_cpu = find_idlest_cpu(group, p, cpu);
4594 4595 4596 4597
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
4598
		}
4599 4600 4601

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
4602
		weight = sd->span_weight;
4603 4604
		sd = NULL;
		for_each_domain(cpu, tmp) {
4605
			if (weight <= tmp->span_weight)
4606
				break;
4607
			if (tmp->flags & sd_flag)
4608 4609 4610
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
4611
	}
4612 4613
unlock:
	rcu_read_unlock();
4614

4615
	return new_cpu;
4616
}
4617 4618 4619 4620 4621 4622 4623 4624 4625 4626

/*
 * 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)
{
4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637
	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);
4638 4639
		atomic_long_add(se->avg.load_avg_contrib,
						&cfs_rq->removed_load);
4640
	}
4641 4642 4643

	/* We have migrated, no longer consider this task hot */
	se->exec_start = 0;
4644
}
4645 4646
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
4647 4648
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
4649 4650 4651 4652
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
4653 4654
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
4655 4656 4657 4658 4659 4660 4661 4662 4663
	 *
	 * 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.
4664
	 */
4665
	return calc_delta_fair(gran, se);
4666 4667
}

4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689
/*
 * 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
4690
	gran = wakeup_gran(curr, se);
4691 4692 4693 4694 4695 4696
	if (vdiff > gran)
		return 1;

	return 0;
}

4697 4698
static void set_last_buddy(struct sched_entity *se)
{
4699 4700 4701 4702 4703
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
4704 4705 4706 4707
}

static void set_next_buddy(struct sched_entity *se)
{
4708 4709 4710 4711 4712
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
4713 4714
}

4715 4716
static void set_skip_buddy(struct sched_entity *se)
{
4717 4718
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
4719 4720
}

4721 4722 4723
/*
 * Preempt the current task with a newly woken task if needed:
 */
4724
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
4725 4726
{
	struct task_struct *curr = rq->curr;
4727
	struct sched_entity *se = &curr->se, *pse = &p->se;
4728
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
4729
	int scale = cfs_rq->nr_running >= sched_nr_latency;
4730
	int next_buddy_marked = 0;
4731

I
Ingo Molnar 已提交
4732 4733 4734
	if (unlikely(se == pse))
		return;

4735
	/*
4736
	 * This is possible from callers such as attach_tasks(), in which we
4737 4738 4739 4740 4741 4742 4743
	 * 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;

4744
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
4745
		set_next_buddy(pse);
4746 4747
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
4748

4749 4750 4751
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
4752 4753 4754 4755 4756 4757
	 *
	 * 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.
4758 4759 4760 4761
	 */
	if (test_tsk_need_resched(curr))
		return;

4762 4763 4764 4765 4766
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

4767
	/*
4768 4769
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
4770
	 */
4771
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
4772
		return;
4773

4774
	find_matching_se(&se, &pse);
4775
	update_curr(cfs_rq_of(se));
4776
	BUG_ON(!pse);
4777 4778 4779 4780 4781 4782 4783
	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);
4784
		goto preempt;
4785
	}
4786

4787
	return;
4788

4789
preempt:
4790
	resched_curr(rq);
4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804
	/*
	 * 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);
4805 4806
}

4807 4808
static struct task_struct *
pick_next_task_fair(struct rq *rq, struct task_struct *prev)
4809 4810 4811
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
4812
	struct task_struct *p;
4813
	int new_tasks;
4814

4815
again:
4816 4817
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
4818
		goto idle;
4819

4820
	if (prev->sched_class != &fair_sched_class)
4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891
		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
4892

4893
	if (!cfs_rq->nr_running)
4894
		goto idle;
4895

4896
	put_prev_task(rq, prev);
4897

4898
	do {
4899
		se = pick_next_entity(cfs_rq, NULL);
4900
		set_next_entity(cfs_rq, se);
4901 4902 4903
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
4904
	p = task_of(se);
4905

4906 4907
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
4908 4909

	return p;
4910 4911

idle:
4912
	new_tasks = idle_balance(rq);
4913 4914 4915 4916 4917
	/*
	 * 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.
	 */
4918
	if (new_tasks < 0)
4919 4920
		return RETRY_TASK;

4921
	if (new_tasks > 0)
4922 4923 4924
		goto again;

	return NULL;
4925 4926 4927 4928 4929
}

/*
 * Account for a descheduled task:
 */
4930
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
4931 4932 4933 4934 4935 4936
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4937
		put_prev_entity(cfs_rq, se);
4938 4939 4940
	}
}

4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965
/*
 * 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);
4966 4967 4968 4969 4970 4971
		/*
		 * 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;
4972 4973 4974 4975 4976
	}

	set_skip_buddy(se);
}

4977 4978 4979 4980
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

4981 4982
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
4983 4984 4985 4986 4987 4988 4989 4990 4991 4992
		return false;

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

	yield_task_fair(rq);

	return true;
}

4993
#ifdef CONFIG_SMP
4994
/**************************************************
P
Peter Zijlstra 已提交
4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017
 * 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)
 *
5018
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
5019 5020 5021 5022 5023 5024
 * 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):
 *
5025
 *   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
5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110
 *
 * 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.]
 */ 
5111

5112 5113
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

5114 5115
enum fbq_type { regular, remote, all };

5116
#define LBF_ALL_PINNED	0x01
5117
#define LBF_NEED_BREAK	0x02
5118 5119
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
5120 5121 5122 5123 5124

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
5125
	int			src_cpu;
5126 5127 5128 5129

	int			dst_cpu;
	struct rq		*dst_rq;

5130 5131
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
5132
	enum cpu_idle_type	idle;
5133
	long			imbalance;
5134 5135 5136
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

5137
	unsigned int		flags;
5138 5139 5140 5141

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
5142 5143

	enum fbq_type		fbq_type;
5144
	struct list_head	tasks;
5145 5146
};

5147 5148 5149
/*
 * Is this task likely cache-hot:
 */
5150
static int task_hot(struct task_struct *p, struct lb_env *env)
5151 5152 5153
{
	s64 delta;

5154 5155
	lockdep_assert_held(&env->src_rq->lock);

5156 5157 5158 5159 5160 5161 5162 5163 5164
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
5165
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
5166 5167 5168 5169 5170 5171 5172 5173 5174
			(&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;

5175
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
5176 5177 5178 5179

	return delta < (s64)sysctl_sched_migration_cost;
}

5180 5181 5182 5183
#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)
{
5184
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5185 5186
	int src_nid, dst_nid;

5187
	if (!sched_feat(NUMA_FAVOUR_HIGHER) || !p->numa_faults_memory ||
5188 5189 5190 5191 5192 5193 5194
	    !(env->sd->flags & SD_NUMA)) {
		return false;
	}

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

5195
	if (src_nid == dst_nid)
5196 5197
		return false;

5198 5199 5200 5201
	if (numa_group) {
		/* Task is already in the group's interleave set. */
		if (node_isset(src_nid, numa_group->active_nodes))
			return false;
5202

5203 5204 5205
		/* Task is moving into the group's interleave set. */
		if (node_isset(dst_nid, numa_group->active_nodes))
			return true;
5206

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

	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
5212 5213
		return true;

5214
	return task_faults(p, dst_nid) > task_faults(p, src_nid);
5215
}
5216 5217 5218 5219


static bool migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
{
5220
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5221 5222 5223 5224 5225
	int src_nid, dst_nid;

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

5226
	if (!p->numa_faults_memory || !(env->sd->flags & SD_NUMA))
5227 5228 5229 5230 5231
		return false;

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

5232
	if (src_nid == dst_nid)
5233 5234
		return false;

5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246
	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);
	}

5247 5248 5249 5250
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid)
		return true;

5251
	return task_faults(p, dst_nid) < task_faults(p, src_nid);
5252 5253
}

5254 5255 5256 5257 5258 5259
#else
static inline bool migrate_improves_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5260 5261 5262 5263 5264 5265

static inline bool migrate_degrades_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5266 5267
#endif

5268 5269 5270 5271
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
5272
int can_migrate_task(struct task_struct *p, struct lb_env *env)
5273 5274
{
	int tsk_cache_hot = 0;
5275 5276 5277

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

5278 5279
	/*
	 * We do not migrate tasks that are:
5280
	 * 1) throttled_lb_pair, or
5281
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
5282 5283
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
5284
	 */
5285 5286 5287
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

5288
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
5289
		int cpu;
5290

5291
		schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
5292

5293 5294
		env->flags |= LBF_SOME_PINNED;

5295 5296 5297 5298 5299 5300 5301 5302
		/*
		 * 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.
		 */
5303
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
5304 5305
			return 0;

5306 5307 5308
		/* 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))) {
5309
				env->flags |= LBF_DST_PINNED;
5310 5311 5312
				env->new_dst_cpu = cpu;
				break;
			}
5313
		}
5314

5315 5316
		return 0;
	}
5317 5318

	/* Record that we found atleast one task that could run on dst_cpu */
5319
	env->flags &= ~LBF_ALL_PINNED;
5320

5321
	if (task_running(env->src_rq, p)) {
5322
		schedstat_inc(p, se.statistics.nr_failed_migrations_running);
5323 5324 5325 5326 5327
		return 0;
	}

	/*
	 * Aggressive migration if:
5328 5329 5330
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
5331
	 */
5332
	tsk_cache_hot = task_hot(p, env);
5333 5334
	if (!tsk_cache_hot)
		tsk_cache_hot = migrate_degrades_locality(p, env);
5335

5336 5337
	if (migrate_improves_locality(p, env) || !tsk_cache_hot ||
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
5338 5339 5340 5341
		if (tsk_cache_hot) {
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
			schedstat_inc(p, se.statistics.nr_forced_migrations);
		}
5342 5343 5344
		return 1;
	}

Z
Zhang Hang 已提交
5345 5346
	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
	return 0;
5347 5348
}

5349
/*
5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360
 * 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);
}

5361
/*
5362
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
5363 5364
 * part of active balancing operations within "domain".
 *
5365
 * Returns a task if successful and NULL otherwise.
5366
 */
5367
static struct task_struct *detach_one_task(struct lb_env *env)
5368 5369 5370
{
	struct task_struct *p, *n;

5371 5372
	lockdep_assert_held(&env->src_rq->lock);

5373 5374 5375
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
5376

5377
		detach_task(p, env);
5378

5379
		/*
5380
		 * Right now, this is only the second place where
5381
		 * lb_gained[env->idle] is updated (other is detach_tasks)
5382
		 * so we can safely collect stats here rather than
5383
		 * inside detach_tasks().
5384 5385
		 */
		schedstat_inc(env->sd, lb_gained[env->idle]);
5386
		return p;
5387
	}
5388
	return NULL;
5389 5390
}

5391 5392
static const unsigned int sched_nr_migrate_break = 32;

5393
/*
5394 5395
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
5396
 *
5397
 * Returns number of detached tasks if successful and 0 otherwise.
5398
 */
5399
static int detach_tasks(struct lb_env *env)
5400
{
5401 5402
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
5403
	unsigned long load;
5404 5405 5406
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
5407

5408
	if (env->imbalance <= 0)
5409
		return 0;
5410

5411 5412
	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
5413

5414 5415
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
5416
		if (env->loop > env->loop_max)
5417
			break;
5418 5419

		/* take a breather every nr_migrate tasks */
5420
		if (env->loop > env->loop_break) {
5421
			env->loop_break += sched_nr_migrate_break;
5422
			env->flags |= LBF_NEED_BREAK;
5423
			break;
5424
		}
5425

5426
		if (!can_migrate_task(p, env))
5427 5428 5429
			goto next;

		load = task_h_load(p);
5430

5431
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
5432 5433
			goto next;

5434
		if ((load / 2) > env->imbalance)
5435
			goto next;
5436

5437 5438 5439 5440
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
5441
		env->imbalance -= load;
5442 5443

#ifdef CONFIG_PREEMPT
5444 5445
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
5446
		 * kernels will stop after the first task is detached to minimize
5447 5448
		 * the critical section.
		 */
5449
		if (env->idle == CPU_NEWLY_IDLE)
5450
			break;
5451 5452
#endif

5453 5454 5455 5456
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
5457
		if (env->imbalance <= 0)
5458
			break;
5459 5460 5461

		continue;
next:
5462
		list_move_tail(&p->se.group_node, tasks);
5463
	}
5464

5465
	/*
5466 5467 5468
	 * 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().
5469
	 */
5470
	schedstat_add(env->sd, lb_gained[env->idle], detached);
5471

5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512
	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);
5513

5514 5515 5516 5517
		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
5518 5519
}

P
Peter Zijlstra 已提交
5520
#ifdef CONFIG_FAIR_GROUP_SCHED
5521 5522 5523
/*
 * update tg->load_weight by folding this cpu's load_avg
 */
5524
static void __update_blocked_averages_cpu(struct task_group *tg, int cpu)
5525
{
5526 5527
	struct sched_entity *se = tg->se[cpu];
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
5528

5529 5530 5531
	/* throttled entities do not contribute to load */
	if (throttled_hierarchy(cfs_rq))
		return;
5532

5533
	update_cfs_rq_blocked_load(cfs_rq, 1);
5534

5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548
	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 {
5549
		struct rq *rq = rq_of(cfs_rq);
5550 5551
		update_rq_runnable_avg(rq, rq->nr_running);
	}
5552 5553
}

5554
static void update_blocked_averages(int cpu)
5555 5556
{
	struct rq *rq = cpu_rq(cpu);
5557 5558
	struct cfs_rq *cfs_rq;
	unsigned long flags;
5559

5560 5561
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
5562 5563 5564 5565
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
5566
	for_each_leaf_cfs_rq(rq, cfs_rq) {
5567 5568 5569 5570 5571 5572
		/*
		 * 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);
5573
	}
5574 5575

	raw_spin_unlock_irqrestore(&rq->lock, flags);
5576 5577
}

5578
/*
5579
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
5580 5581 5582
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
5583
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
5584
{
5585 5586
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
5587
	unsigned long now = jiffies;
5588
	unsigned long load;
5589

5590
	if (cfs_rq->last_h_load_update == now)
5591 5592
		return;

5593 5594 5595 5596 5597 5598 5599
	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;
	}
5600

5601
	if (!se) {
5602
		cfs_rq->h_load = cfs_rq->runnable_load_avg;
5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613
		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;
	}
5614 5615
}

5616
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
5617
{
5618
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
5619

5620
	update_cfs_rq_h_load(cfs_rq);
5621 5622
	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
5623 5624
}
#else
5625
static inline void update_blocked_averages(int cpu)
5626 5627 5628
{
}

5629
static unsigned long task_h_load(struct task_struct *p)
5630
{
5631
	return p->se.avg.load_avg_contrib;
5632
}
P
Peter Zijlstra 已提交
5633
#endif
5634 5635

/********** Helpers for find_busiest_group ************************/
5636 5637 5638 5639 5640 5641 5642

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

5643 5644 5645 5646 5647 5648 5649
/*
 * 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
5650
	unsigned long load_per_task;
5651
	unsigned long group_capacity;
5652
	unsigned int sum_nr_running; /* Nr tasks running in the group */
5653
	unsigned int group_capacity_factor;
5654 5655
	unsigned int idle_cpus;
	unsigned int group_weight;
5656
	enum group_type group_type;
5657
	int group_has_free_capacity;
5658 5659 5660 5661
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
5662 5663
};

J
Joonsoo Kim 已提交
5664 5665 5666 5667 5668 5669 5670 5671
/*
 * 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 */
5672
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
5673 5674 5675
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
5676
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
5677 5678
};

5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690
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,
5691
		.total_capacity = 0UL,
5692 5693
		.busiest_stat = {
			.avg_load = 0UL,
5694 5695
			.sum_nr_running = 0,
			.group_type = group_other,
5696 5697 5698 5699
		},
	};
}

5700 5701 5702
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
5703
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
5704 5705
 *
 * Return: The load index.
5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727
 */
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;
}

5728
static unsigned long default_scale_capacity(struct sched_domain *sd, int cpu)
5729
{
5730
	return SCHED_CAPACITY_SCALE;
5731 5732
}

5733
unsigned long __weak arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
5734
{
5735
	return default_scale_capacity(sd, cpu);
5736 5737
}

5738
static unsigned long default_scale_cpu_capacity(struct sched_domain *sd, int cpu)
5739
{
5740 5741
	if ((sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
		return sd->smt_gain / sd->span_weight;
5742

5743
	return SCHED_CAPACITY_SCALE;
5744 5745
}

5746
unsigned long __weak arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
5747
{
5748
	return default_scale_cpu_capacity(sd, cpu);
5749 5750
}

5751
static unsigned long scale_rt_capacity(int cpu)
5752 5753
{
	struct rq *rq = cpu_rq(cpu);
5754
	u64 total, available, age_stamp, avg;
5755
	s64 delta;
5756

5757 5758 5759 5760 5761 5762 5763
	/*
	 * 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);

5764 5765 5766 5767 5768
	delta = rq_clock(rq) - age_stamp;
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
5769

5770
	if (unlikely(total < avg)) {
5771
		/* Ensures that capacity won't end up being negative */
5772 5773
		available = 0;
	} else {
5774
		available = total - avg;
5775
	}
5776

5777 5778
	if (unlikely((s64)total < SCHED_CAPACITY_SCALE))
		total = SCHED_CAPACITY_SCALE;
5779

5780
	total >>= SCHED_CAPACITY_SHIFT;
5781 5782 5783 5784

	return div_u64(available, total);
}

5785
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
5786
{
5787
	unsigned long capacity = SCHED_CAPACITY_SCALE;
5788 5789
	struct sched_group *sdg = sd->groups;

5790 5791 5792 5793
	if (sched_feat(ARCH_CAPACITY))
		capacity *= arch_scale_cpu_capacity(sd, cpu);
	else
		capacity *= default_scale_cpu_capacity(sd, cpu);
5794

5795
	capacity >>= SCHED_CAPACITY_SHIFT;
5796

5797
	sdg->sgc->capacity_orig = capacity;
5798

5799
	if (sched_feat(ARCH_CAPACITY))
5800
		capacity *= arch_scale_freq_capacity(sd, cpu);
5801
	else
5802
		capacity *= default_scale_capacity(sd, cpu);
5803

5804
	capacity >>= SCHED_CAPACITY_SHIFT;
5805

5806
	capacity *= scale_rt_capacity(cpu);
5807
	capacity >>= SCHED_CAPACITY_SHIFT;
5808

5809 5810
	if (!capacity)
		capacity = 1;
5811

5812 5813
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
5814 5815
}

5816
void update_group_capacity(struct sched_domain *sd, int cpu)
5817 5818 5819
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
5820
	unsigned long capacity, capacity_orig;
5821 5822 5823 5824
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
5825
	sdg->sgc->next_update = jiffies + interval;
5826 5827

	if (!child) {
5828
		update_cpu_capacity(sd, cpu);
5829 5830 5831
		return;
	}

5832
	capacity_orig = capacity = 0;
5833

P
Peter Zijlstra 已提交
5834 5835 5836 5837 5838 5839
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

5840
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
5841
			struct sched_group_capacity *sgc;
5842
			struct rq *rq = cpu_rq(cpu);
5843

5844
			/*
5845
			 * build_sched_domains() -> init_sched_groups_capacity()
5846 5847 5848
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
5849 5850
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
5851
			 *
5852
			 * This avoids capacity/capacity_orig from being 0 and
5853 5854
			 * causing divide-by-zero issues on boot.
			 *
5855
			 * Runtime updates will correct capacity_orig.
5856 5857
			 */
			if (unlikely(!rq->sd)) {
5858 5859
				capacity_orig += capacity_of(cpu);
				capacity += capacity_of(cpu);
5860 5861
				continue;
			}
5862

5863 5864 5865
			sgc = rq->sd->groups->sgc;
			capacity_orig += sgc->capacity_orig;
			capacity += sgc->capacity;
5866
		}
P
Peter Zijlstra 已提交
5867 5868 5869 5870 5871 5872 5873 5874
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
		 */ 

		group = child->groups;
		do {
5875 5876
			capacity_orig += group->sgc->capacity_orig;
			capacity += group->sgc->capacity;
P
Peter Zijlstra 已提交
5877 5878 5879
			group = group->next;
		} while (group != child->groups);
	}
5880

5881 5882
	sdg->sgc->capacity_orig = capacity_orig;
	sdg->sgc->capacity = capacity;
5883 5884
}

5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895
/*
 * 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)
{
	/*
5896
	 * Only siblings can have significantly less than SCHED_CAPACITY_SCALE
5897
	 */
5898
	if (!(sd->flags & SD_SHARE_CPUCAPACITY))
5899 5900 5901
		return 0;

	/*
5902
	 * If ~90% of the cpu_capacity is still there, we're good.
5903
	 */
5904
	if (group->sgc->capacity * 32 > group->sgc->capacity_orig * 29)
5905 5906 5907 5908 5909
		return 1;

	return 0;
}

5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925
/*
 * 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
5926 5927
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
5928 5929
 *
 * When this is so detected; this group becomes a candidate for busiest; see
5930
 * update_sd_pick_busiest(). And calculate_imbalance() and
5931
 * find_busiest_group() avoid some of the usual balance conditions to allow it
5932 5933 5934 5935 5936 5937 5938
 * 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.
 */

5939
static inline int sg_imbalanced(struct sched_group *group)
5940
{
5941
	return group->sgc->imbalance;
5942 5943
}

5944
/*
5945
 * Compute the group capacity factor.
5946
 *
5947
 * Avoid the issue where N*frac(smt_capacity) >= 1 creates 'phantom' cores by
5948
 * first dividing out the smt factor and computing the actual number of cores
5949
 * and limit unit capacity with that.
5950
 */
5951
static inline int sg_capacity_factor(struct lb_env *env, struct sched_group *group)
5952
{
5953
	unsigned int capacity_factor, smt, cpus;
5954
	unsigned int capacity, capacity_orig;
5955

5956 5957
	capacity = group->sgc->capacity;
	capacity_orig = group->sgc->capacity_orig;
5958
	cpus = group->group_weight;
5959

5960
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_capacity < 2 */
5961
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, capacity_orig);
5962
	capacity_factor = cpus / smt; /* cores */
5963

5964
	capacity_factor = min_t(unsigned,
5965
		capacity_factor, DIV_ROUND_CLOSEST(capacity, SCHED_CAPACITY_SCALE));
5966 5967
	if (!capacity_factor)
		capacity_factor = fix_small_capacity(env->sd, group);
5968

5969
	return capacity_factor;
5970 5971
}

5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983
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;
}

5984 5985
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
5986
 * @env: The load balancing environment.
5987 5988 5989 5990
 * @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.
5991
 * @overload: Indicate more than one runnable task for any CPU.
5992
 */
5993 5994
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
5995 5996
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
5997
{
5998
	unsigned long load;
5999
	int i;
6000

6001 6002
	memset(sgs, 0, sizeof(*sgs));

6003
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6004 6005 6006
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
6007
		if (local_group)
6008
			load = target_load(i, load_idx);
6009
		else
6010 6011 6012
			load = source_load(i, load_idx);

		sgs->group_load += load;
6013
		sgs->sum_nr_running += rq->cfs.h_nr_running;
6014 6015 6016 6017

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

6018 6019 6020 6021
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
6022
		sgs->sum_weighted_load += weighted_cpuload(i);
6023 6024
		if (idle_cpu(i))
			sgs->idle_cpus++;
6025 6026
	}

6027 6028
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
6029
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
6030

6031
	if (sgs->sum_nr_running)
6032
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
6033

6034
	sgs->group_weight = group->group_weight;
6035
	sgs->group_capacity_factor = sg_capacity_factor(env, group);
6036
	sgs->group_type = group_classify(group, sgs);
6037

6038
	if (sgs->group_capacity_factor > sgs->sum_nr_running)
6039
		sgs->group_has_free_capacity = 1;
6040 6041
}

6042 6043
/**
 * update_sd_pick_busiest - return 1 on busiest group
6044
 * @env: The load balancing environment.
6045 6046
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
6047
 * @sgs: sched_group statistics
6048 6049 6050
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
6051 6052 6053
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
6054
 */
6055
static bool update_sd_pick_busiest(struct lb_env *env,
6056 6057
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
6058
				   struct sg_lb_stats *sgs)
6059
{
6060
	struct sg_lb_stats *busiest = &sds->busiest_stat;
6061

6062
	if (sgs->group_type > busiest->group_type)
6063 6064
		return true;

6065 6066 6067 6068 6069 6070 6071 6072
	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))
6073 6074 6075 6076 6077 6078 6079
		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.
	 */
6080
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
6081 6082 6083 6084 6085 6086 6087 6088 6089 6090
		if (!sds->busiest)
			return true;

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

	return false;
}

6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120
#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 */

6121
/**
6122
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
6123
 * @env: The load balancing environment.
6124 6125
 * @sds: variable to hold the statistics for this sched_domain.
 */
6126
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
6127
{
6128 6129
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
6130
	struct sg_lb_stats tmp_sgs;
6131
	int load_idx, prefer_sibling = 0;
6132
	bool overload = false;
6133 6134 6135 6136

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

6137
	load_idx = get_sd_load_idx(env->sd, env->idle);
6138 6139

	do {
J
Joonsoo Kim 已提交
6140
		struct sg_lb_stats *sgs = &tmp_sgs;
6141 6142
		int local_group;

6143
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
6144 6145 6146
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
6147 6148

			if (env->idle != CPU_NEWLY_IDLE ||
6149 6150
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
6151
		}
6152

6153 6154
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
6155

6156 6157 6158
		if (local_group)
			goto next_group;

6159 6160
		/*
		 * In case the child domain prefers tasks go to siblings
6161
		 * first, lower the sg capacity factor to one so that we'll try
6162 6163
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
6164
		 * these excess tasks, i.e. nr_running < group_capacity_factor. The
6165 6166 6167
		 * 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).
6168
		 */
6169
		if (prefer_sibling && sds->local &&
6170
		    sds->local_stat.group_has_free_capacity)
6171
			sgs->group_capacity_factor = min(sgs->group_capacity_factor, 1U);
6172

6173
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
6174
			sds->busiest = sg;
J
Joonsoo Kim 已提交
6175
			sds->busiest_stat = *sgs;
6176 6177
		}

6178 6179 6180
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
6181
		sds->total_capacity += sgs->group_capacity;
6182

6183
		sg = sg->next;
6184
	} while (sg != env->sd->groups);
6185 6186 6187

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
6188 6189 6190 6191 6192 6193 6194

	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;
	}

6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213
}

/**
 * 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.
 *
6214
 * Return: 1 when packing is required and a task should be moved to
6215 6216
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
6217
 * @env: The load balancing environment.
6218 6219
 * @sds: Statistics of the sched_domain which is to be packed
 */
6220
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
6221 6222 6223
{
	int busiest_cpu;

6224
	if (!(env->sd->flags & SD_ASYM_PACKING))
6225 6226 6227 6228 6229 6230
		return 0;

	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
6231
	if (env->dst_cpu > busiest_cpu)
6232 6233
		return 0;

6234
	env->imbalance = DIV_ROUND_CLOSEST(
6235
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
6236
		SCHED_CAPACITY_SCALE);
6237

6238
	return 1;
6239 6240 6241 6242 6243 6244
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
6245
 * @env: The load balancing environment.
6246 6247
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
6248 6249
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6250
{
6251
	unsigned long tmp, capa_now = 0, capa_move = 0;
6252
	unsigned int imbn = 2;
6253
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
6254
	struct sg_lb_stats *local, *busiest;
6255

J
Joonsoo Kim 已提交
6256 6257
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6258

J
Joonsoo Kim 已提交
6259 6260 6261 6262
	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;
6263

J
Joonsoo Kim 已提交
6264
	scaled_busy_load_per_task =
6265
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6266
		busiest->group_capacity;
J
Joonsoo Kim 已提交
6267

6268 6269
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
6270
		env->imbalance = busiest->load_per_task;
6271 6272 6273 6274 6275
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
6276
	 * however we may be able to increase total CPU capacity used by
6277 6278 6279
	 * moving them.
	 */

6280
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
6281
			min(busiest->load_per_task, busiest->avg_load);
6282
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
6283
			min(local->load_per_task, local->avg_load);
6284
	capa_now /= SCHED_CAPACITY_SCALE;
6285 6286

	/* Amount of load we'd subtract */
6287
	if (busiest->avg_load > scaled_busy_load_per_task) {
6288
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
6289
			    min(busiest->load_per_task,
6290
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
6291
	}
6292 6293

	/* Amount of load we'd add */
6294
	if (busiest->avg_load * busiest->group_capacity <
6295
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
6296 6297
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
6298
	} else {
6299
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6300
		      local->group_capacity;
J
Joonsoo Kim 已提交
6301
	}
6302
	capa_move += local->group_capacity *
6303
		    min(local->load_per_task, local->avg_load + tmp);
6304
	capa_move /= SCHED_CAPACITY_SCALE;
6305 6306

	/* Move if we gain throughput */
6307
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
6308
		env->imbalance = busiest->load_per_task;
6309 6310 6311 6312 6313
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
6314
 * @env: load balance environment
6315 6316
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
6317
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6318
{
6319
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
6320 6321 6322 6323
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6324

6325
	if (busiest->group_type == group_imbalanced) {
6326 6327 6328 6329
		/*
		 * 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
6330 6331
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
6332 6333
	}

6334 6335 6336
	/*
	 * 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
6337
	 * its cpu_capacity, while calculating max_load..)
6338
	 */
6339 6340
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
6341 6342
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
6343 6344
	}

6345 6346 6347 6348 6349
	/*
	 * 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
6350
		load_above_capacity =
6351
			(busiest->sum_nr_running - busiest->group_capacity_factor);
6352

6353
		load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_CAPACITY_SCALE);
6354
		load_above_capacity /= busiest->group_capacity;
6355 6356 6357 6358 6359 6360 6361 6362 6363 6364
	}

	/*
	 * 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.
	 */
6365
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
6366 6367

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
6368
	env->imbalance = min(
6369 6370
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
6371
	) / SCHED_CAPACITY_SCALE;
6372 6373 6374

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
6375
	 * there is no guarantee that any tasks will be moved so we'll have
6376 6377 6378
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
6379
	if (env->imbalance < busiest->load_per_task)
6380
		return fix_small_imbalance(env, sds);
6381
}
6382

6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394
/******* 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.
 *
6395
 * @env: The load balancing environment.
6396
 *
6397
 * Return:	- The busiest group if imbalance exists.
6398 6399 6400 6401
 *		- 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
6402
static struct sched_group *find_busiest_group(struct lb_env *env)
6403
{
J
Joonsoo Kim 已提交
6404
	struct sg_lb_stats *local, *busiest;
6405 6406
	struct sd_lb_stats sds;

6407
	init_sd_lb_stats(&sds);
6408 6409 6410 6411 6412

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
6413
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
6414 6415
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
6416

6417 6418
	if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
	    check_asym_packing(env, &sds))
6419 6420
		return sds.busiest;

6421
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
6422
	if (!sds.busiest || busiest->sum_nr_running == 0)
6423 6424
		goto out_balanced;

6425 6426
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
6427

P
Peter Zijlstra 已提交
6428 6429
	/*
	 * If the busiest group is imbalanced the below checks don't
6430
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
6431 6432
	 * isn't true due to cpus_allowed constraints and the like.
	 */
6433
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
6434 6435
		goto force_balance;

6436
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
6437 6438
	if (env->idle == CPU_NEWLY_IDLE && local->group_has_free_capacity &&
	    !busiest->group_has_free_capacity)
6439 6440
		goto force_balance;

6441
	/*
6442
	 * If the local group is busier than the selected busiest group
6443 6444
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
6445
	if (local->avg_load >= busiest->avg_load)
6446 6447
		goto out_balanced;

6448 6449 6450 6451
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
6452
	if (local->avg_load >= sds.avg_load)
6453 6454
		goto out_balanced;

6455
	if (env->idle == CPU_IDLE) {
6456
		/*
6457 6458 6459 6460 6461
		 * 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
6462
		 */
6463 6464
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
6465
			goto out_balanced;
6466 6467 6468 6469 6470
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
6471 6472
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
6473
			goto out_balanced;
6474
	}
6475

6476
force_balance:
6477
	/* Looks like there is an imbalance. Compute it */
6478
	calculate_imbalance(env, &sds);
6479 6480 6481
	return sds.busiest;

out_balanced:
6482
	env->imbalance = 0;
6483 6484 6485 6486 6487 6488
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
6489
static struct rq *find_busiest_queue(struct lb_env *env,
6490
				     struct sched_group *group)
6491 6492
{
	struct rq *busiest = NULL, *rq;
6493
	unsigned long busiest_load = 0, busiest_capacity = 1;
6494 6495
	int i;

6496
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6497
		unsigned long capacity, capacity_factor, wl;
6498 6499 6500 6501
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
6502

6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524
		/*
		 * 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;

6525
		capacity = capacity_of(i);
6526
		capacity_factor = DIV_ROUND_CLOSEST(capacity, SCHED_CAPACITY_SCALE);
6527 6528
		if (!capacity_factor)
			capacity_factor = fix_small_capacity(env->sd, group);
6529

6530
		wl = weighted_cpuload(i);
6531

6532 6533
		/*
		 * When comparing with imbalance, use weighted_cpuload()
6534
		 * which is not scaled with the cpu capacity.
6535
		 */
6536
		if (capacity_factor && rq->nr_running == 1 && wl > env->imbalance)
6537 6538
			continue;

6539 6540
		/*
		 * For the load comparisons with the other cpu's, consider
6541 6542 6543
		 * 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.
6544
		 *
6545
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
6546
		 * multiplication to rid ourselves of the division works out
6547 6548
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
6549
		 */
6550
		if (wl * busiest_capacity > busiest_load * capacity) {
6551
			busiest_load = wl;
6552
			busiest_capacity = capacity;
6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566
			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. */
6567
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
6568

6569
static int need_active_balance(struct lb_env *env)
6570
{
6571 6572 6573
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
6574 6575 6576 6577 6578 6579

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
6580
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
6581
			return 1;
6582 6583 6584 6585 6586
	}

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

6587 6588
static int active_load_balance_cpu_stop(void *data);

6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619
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.
	 */
6620
	return balance_cpu == env->dst_cpu;
6621 6622
}

6623 6624 6625 6626 6627 6628
/*
 * 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,
6629
			int *continue_balancing)
6630
{
6631
	int ld_moved, cur_ld_moved, active_balance = 0;
6632
	struct sched_domain *sd_parent = sd->parent;
6633 6634 6635
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
6636
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
6637

6638 6639
	struct lb_env env = {
		.sd		= sd,
6640 6641
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
6642
		.dst_grpmask    = sched_group_cpus(sd->groups),
6643
		.idle		= idle,
6644
		.loop_break	= sched_nr_migrate_break,
6645
		.cpus		= cpus,
6646
		.fbq_type	= all,
6647
		.tasks		= LIST_HEAD_INIT(env.tasks),
6648 6649
	};

6650 6651 6652 6653
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
6654
	if (idle == CPU_NEWLY_IDLE)
6655 6656
		env.dst_grpmask = NULL;

6657 6658 6659 6660 6661
	cpumask_copy(cpus, cpu_active_mask);

	schedstat_inc(sd, lb_count[idle]);

redo:
6662 6663
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
6664
		goto out_balanced;
6665
	}
6666

6667
	group = find_busiest_group(&env);
6668 6669 6670 6671 6672
	if (!group) {
		schedstat_inc(sd, lb_nobusyg[idle]);
		goto out_balanced;
	}

6673
	busiest = find_busiest_queue(&env, group);
6674 6675 6676 6677 6678
	if (!busiest) {
		schedstat_inc(sd, lb_nobusyq[idle]);
		goto out_balanced;
	}

6679
	BUG_ON(busiest == env.dst_rq);
6680

6681
	schedstat_add(sd, lb_imbalance[idle], env.imbalance);
6682 6683 6684 6685 6686 6687 6688 6689 6690

	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.
		 */
6691
		env.flags |= LBF_ALL_PINNED;
6692 6693 6694
		env.src_cpu   = busiest->cpu;
		env.src_rq    = busiest;
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
6695

6696
more_balance:
6697
		raw_spin_lock_irqsave(&busiest->lock, flags);
6698 6699 6700 6701 6702

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
6703
		cur_ld_moved = detach_tasks(&env);
6704 6705

		/*
6706 6707 6708 6709 6710
		 * 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.
6711
		 */
6712 6713 6714 6715 6716 6717 6718 6719

		raw_spin_unlock(&busiest->lock);

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

6720
		local_irq_restore(flags);
6721

6722 6723 6724 6725 6726
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745
		/*
		 * 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.
		 */
6746
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
6747

6748 6749 6750
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

6751
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
6752
			env.dst_cpu	 = env.new_dst_cpu;
6753
			env.flags	&= ~LBF_DST_PINNED;
6754 6755
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
6756

6757 6758 6759 6760 6761 6762
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
6763

6764 6765 6766 6767
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
6768
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
6769

6770
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
6771 6772 6773
				*group_imbalance = 1;
		}

6774
		/* All tasks on this runqueue were pinned by CPU affinity */
6775
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
6776
			cpumask_clear_cpu(cpu_of(busiest), cpus);
6777 6778 6779
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
6780
				goto redo;
6781
			}
6782
			goto out_all_pinned;
6783 6784 6785 6786 6787
		}
	}

	if (!ld_moved) {
		schedstat_inc(sd, lb_failed[idle]);
6788 6789 6790 6791 6792 6793 6794 6795
		/*
		 * 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++;
6796

6797
		if (need_active_balance(&env)) {
6798 6799
			raw_spin_lock_irqsave(&busiest->lock, flags);

6800 6801 6802
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
6803 6804
			 */
			if (!cpumask_test_cpu(this_cpu,
6805
					tsk_cpus_allowed(busiest->curr))) {
6806 6807
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
6808
				env.flags |= LBF_ALL_PINNED;
6809 6810 6811
				goto out_one_pinned;
			}

6812 6813 6814 6815 6816
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
6817 6818 6819 6820 6821 6822
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
6823

6824
			if (active_balance) {
6825 6826 6827
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
6828
			}
6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846

			/*
			 * 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
6847
		 * detach_tasks).
6848 6849 6850 6851 6852 6853 6854 6855
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872
	/*
	 * 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.
	 */
6873 6874 6875 6876 6877 6878
	schedstat_inc(sd, lb_balanced[idle]);

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
6879
	if (((env.flags & LBF_ALL_PINNED) &&
6880
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
6881 6882 6883
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

6884
	ld_moved = 0;
6885 6886 6887 6888
out:
	return ld_moved;
}

6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909 6910 6911 6912 6913 6914 6915
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;
}

6916 6917 6918 6919
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
6920
static int idle_balance(struct rq *this_rq)
6921
{
6922 6923
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
6924 6925
	struct sched_domain *sd;
	int pulled_task = 0;
6926
	u64 curr_cost = 0;
6927

6928
	idle_enter_fair(this_rq);
6929

6930 6931 6932 6933 6934 6935
	/*
	 * 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);

6936 6937
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
6938 6939 6940 6941 6942 6943
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, 0, &next_balance);
		rcu_read_unlock();

6944
		goto out;
6945
	}
6946

6947 6948 6949 6950 6951
	/*
	 * Drop the rq->lock, but keep IRQ/preempt disabled.
	 */
	raw_spin_unlock(&this_rq->lock);

6952
	update_blocked_averages(this_cpu);
6953
	rcu_read_lock();
6954
	for_each_domain(this_cpu, sd) {
6955
		int continue_balancing = 1;
6956
		u64 t0, domain_cost;
6957 6958 6959 6960

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

6961 6962
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, 0, &next_balance);
6963
			break;
6964
		}
6965

6966
		if (sd->flags & SD_BALANCE_NEWIDLE) {
6967 6968
			t0 = sched_clock_cpu(this_cpu);

6969
			pulled_task = load_balance(this_cpu, this_rq,
6970 6971
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
6972 6973 6974 6975 6976 6977

			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;
6978
		}
6979

6980
		update_next_balance(sd, 0, &next_balance);
6981 6982 6983 6984 6985 6986

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
6987 6988
			break;
	}
6989
	rcu_read_unlock();
6990 6991 6992

	raw_spin_lock(&this_rq->lock);

6993 6994 6995
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

6996
	/*
6997 6998 6999
	 * 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.
7000
	 */
7001
	if (this_rq->cfs.h_nr_running && !pulled_task)
7002
		pulled_task = 1;
7003

7004 7005 7006
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
7007
		this_rq->next_balance = next_balance;
7008

7009
	/* Is there a task of a high priority class? */
7010
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
7011 7012 7013 7014
		pulled_task = -1;

	if (pulled_task) {
		idle_exit_fair(this_rq);
7015
		this_rq->idle_stamp = 0;
7016
	}
7017

7018
	return pulled_task;
7019 7020 7021
}

/*
7022 7023 7024 7025
 * 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.
7026
 */
7027
static int active_load_balance_cpu_stop(void *data)
7028
{
7029 7030
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
7031
	int target_cpu = busiest_rq->push_cpu;
7032
	struct rq *target_rq = cpu_rq(target_cpu);
7033
	struct sched_domain *sd;
7034
	struct task_struct *p = NULL;
7035 7036 7037 7038 7039 7040 7041

	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;
7042 7043 7044

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
7045
		goto out_unlock;
7046 7047 7048 7049 7050 7051 7052 7053 7054

	/*
	 * 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. */
7055
	rcu_read_lock();
7056 7057 7058 7059 7060 7061 7062
	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)) {
7063 7064
		struct lb_env env = {
			.sd		= sd,
7065 7066 7067 7068
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
7069 7070 7071
			.idle		= CPU_IDLE,
		};

7072 7073
		schedstat_inc(sd, alb_count);

7074 7075
		p = detach_one_task(&env);
		if (p)
7076 7077 7078 7079
			schedstat_inc(sd, alb_pushed);
		else
			schedstat_inc(sd, alb_failed);
	}
7080
	rcu_read_unlock();
7081 7082
out_unlock:
	busiest_rq->active_balance = 0;
7083 7084 7085 7086 7087 7088 7089
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

7090
	return 0;
7091 7092
}

7093 7094 7095 7096 7097
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

7098
#ifdef CONFIG_NO_HZ_COMMON
7099 7100 7101 7102 7103 7104
/*
 * 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.
 */
7105
static struct {
7106
	cpumask_var_t idle_cpus_mask;
7107
	atomic_t nr_cpus;
7108 7109
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
7110

7111
static inline int find_new_ilb(void)
7112
{
7113
	int ilb = cpumask_first(nohz.idle_cpus_mask);
7114

7115 7116 7117 7118
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
7119 7120
}

7121 7122 7123 7124 7125
/*
 * 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).
 */
7126
static void nohz_balancer_kick(void)
7127 7128 7129 7130 7131
{
	int ilb_cpu;

	nohz.next_balance++;

7132
	ilb_cpu = find_new_ilb();
7133

7134 7135
	if (ilb_cpu >= nr_cpu_ids)
		return;
7136

7137
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
7138 7139 7140 7141 7142 7143 7144 7145
		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);
7146 7147 7148
	return;
}

7149
static inline void nohz_balance_exit_idle(int cpu)
7150 7151
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
7152 7153 7154 7155 7156 7157 7158
		/*
		 * 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);
		}
7159 7160 7161 7162
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

7163 7164 7165
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
7166
	int cpu = smp_processor_id();
7167 7168

	rcu_read_lock();
7169
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7170 7171 7172 7173 7174

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

7175
	atomic_inc(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7176
unlock:
7177 7178 7179 7180 7181 7182
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
7183
	int cpu = smp_processor_id();
7184 7185

	rcu_read_lock();
7186
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7187 7188 7189 7190 7191

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

7192
	atomic_dec(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7193
unlock:
7194 7195 7196
	rcu_read_unlock();
}

7197
/*
7198
 * This routine will record that the cpu is going idle with tick stopped.
7199
 * This info will be used in performing idle load balancing in the future.
7200
 */
7201
void nohz_balance_enter_idle(int cpu)
7202
{
7203 7204 7205 7206 7207 7208
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

7209 7210
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
7211

7212 7213 7214 7215 7216 7217
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

7218 7219 7220
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
7221
}
7222

7223
static int sched_ilb_notifier(struct notifier_block *nfb,
7224 7225 7226 7227
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DYING:
7228
		nohz_balance_exit_idle(smp_processor_id());
7229 7230 7231 7232 7233
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}
7234 7235 7236 7237
#endif

static DEFINE_SPINLOCK(balancing);

7238 7239 7240 7241
/*
 * 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.
 */
7242
void update_max_interval(void)
7243 7244 7245 7246
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

7247 7248 7249 7250
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
7251
 * Balancing parameters are set up in init_sched_domains.
7252
 */
7253
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
7254
{
7255
	int continue_balancing = 1;
7256
	int cpu = rq->cpu;
7257
	unsigned long interval;
7258
	struct sched_domain *sd;
7259 7260 7261
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
7262 7263
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
7264

7265
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
7266

7267
	rcu_read_lock();
7268
	for_each_domain(cpu, sd) {
7269 7270 7271 7272 7273 7274 7275 7276 7277 7278 7279 7280
		/*
		 * 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;

7281 7282 7283
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294
		/*
		 * 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;
		}

7295
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7296 7297 7298 7299 7300 7301 7302 7303

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

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
7304
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
7305
				/*
7306
				 * The LBF_DST_PINNED logic could have changed
7307 7308
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
7309
				 */
7310
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
7311 7312
			}
			sd->last_balance = jiffies;
7313
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7314 7315 7316 7317 7318 7319 7320 7321
		}
		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;
		}
7322 7323
	}
	if (need_decay) {
7324
		/*
7325 7326
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
7327
		 */
7328 7329
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
7330
	}
7331
	rcu_read_unlock();
7332 7333 7334 7335 7336 7337 7338 7339 7340 7341

	/*
	 * 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;
}

7342
#ifdef CONFIG_NO_HZ_COMMON
7343
/*
7344
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
7345 7346
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
7347
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
7348
{
7349
	int this_cpu = this_rq->cpu;
7350 7351 7352
	struct rq *rq;
	int balance_cpu;

7353 7354 7355
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
7356 7357

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
7358
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
7359 7360 7361 7362 7363 7364 7365
			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.
		 */
7366
		if (need_resched())
7367 7368
			break;

V
Vincent Guittot 已提交
7369 7370
		rq = cpu_rq(balance_cpu);

7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381
		/*
		 * 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);
		}
7382 7383 7384 7385 7386

		if (time_after(this_rq->next_balance, rq->next_balance))
			this_rq->next_balance = rq->next_balance;
	}
	nohz.next_balance = this_rq->next_balance;
7387 7388
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
7389 7390 7391
}

/*
7392 7393 7394 7395
 * 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
7396
 *     busy cpu's exceeding the group's capacity.
7397 7398
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
7399
 */
7400
static inline int nohz_kick_needed(struct rq *rq)
7401 7402
{
	unsigned long now = jiffies;
7403
	struct sched_domain *sd;
7404
	struct sched_group_capacity *sgc;
7405
	int nr_busy, cpu = rq->cpu;
7406

7407
	if (unlikely(rq->idle_balance))
7408 7409
		return 0;

7410 7411 7412 7413
       /*
	* 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.
	*/
7414
	set_cpu_sd_state_busy();
7415
	nohz_balance_exit_idle(cpu);
7416 7417 7418 7419 7420 7421 7422

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
		return 0;
7423 7424

	if (time_before(now, nohz.next_balance))
7425 7426
		return 0;

7427 7428
	if (rq->nr_running >= 2)
		goto need_kick;
7429

7430
	rcu_read_lock();
7431
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
7432

7433
	if (sd) {
7434 7435
		sgc = sd->groups->sgc;
		nr_busy = atomic_read(&sgc->nr_busy_cpus);
7436

7437
		if (nr_busy > 1)
7438
			goto need_kick_unlock;
7439
	}
7440 7441 7442 7443 7444 7445 7446

	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;

7447
	rcu_read_unlock();
7448
	return 0;
7449 7450 7451

need_kick_unlock:
	rcu_read_unlock();
7452 7453
need_kick:
	return 1;
7454 7455
}
#else
7456
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
7457 7458 7459 7460 7461 7462
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
7463 7464
static void run_rebalance_domains(struct softirq_action *h)
{
7465
	struct rq *this_rq = this_rq();
7466
	enum cpu_idle_type idle = this_rq->idle_balance ?
7467 7468
						CPU_IDLE : CPU_NOT_IDLE;

7469
	rebalance_domains(this_rq, idle);
7470 7471

	/*
7472
	 * If this cpu has a pending nohz_balance_kick, then do the
7473 7474 7475
	 * balancing on behalf of the other idle cpus whose ticks are
	 * stopped.
	 */
7476
	nohz_idle_balance(this_rq, idle);
7477 7478 7479 7480 7481
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
7482
void trigger_load_balance(struct rq *rq)
7483 7484
{
	/* Don't need to rebalance while attached to NULL domain */
7485 7486 7487 7488
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
7489
		raise_softirq(SCHED_SOFTIRQ);
7490
#ifdef CONFIG_NO_HZ_COMMON
7491
	if (nohz_kick_needed(rq))
7492
		nohz_balancer_kick();
7493
#endif
7494 7495
}

7496 7497 7498
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
7499 7500

	update_runtime_enabled(rq);
7501 7502 7503 7504 7505
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
7506 7507 7508

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
7509 7510
}

7511
#endif /* CONFIG_SMP */
7512

7513 7514 7515
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
7516
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
7517 7518 7519 7520 7521 7522
{
	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
7523
		entity_tick(cfs_rq, se, queued);
7524
	}
7525

7526
	if (numabalancing_enabled)
7527
		task_tick_numa(rq, curr);
7528

7529
	update_rq_runnable_avg(rq, 1);
7530 7531 7532
}

/*
P
Peter Zijlstra 已提交
7533 7534 7535
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
7536
 */
P
Peter Zijlstra 已提交
7537
static void task_fork_fair(struct task_struct *p)
7538
{
7539 7540
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
7541
	int this_cpu = smp_processor_id();
P
Peter Zijlstra 已提交
7542 7543 7544
	struct rq *rq = this_rq();
	unsigned long flags;

7545
	raw_spin_lock_irqsave(&rq->lock, flags);
7546

7547 7548
	update_rq_clock(rq);

7549 7550 7551
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;

7552 7553 7554 7555 7556 7557 7558 7559 7560
	/*
	 * 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();
7561

7562
	update_curr(cfs_rq);
P
Peter Zijlstra 已提交
7563

7564 7565
	if (curr)
		se->vruntime = curr->vruntime;
7566
	place_entity(cfs_rq, se, 1);
7567

P
Peter Zijlstra 已提交
7568
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
7569
		/*
7570 7571 7572
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
7573
		swap(curr->vruntime, se->vruntime);
7574
		resched_curr(rq);
7575
	}
7576

7577 7578
	se->vruntime -= cfs_rq->min_vruntime;

7579
	raw_spin_unlock_irqrestore(&rq->lock, flags);
7580 7581
}

7582 7583 7584 7585
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
7586 7587
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
7588
{
7589
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7590 7591
		return;

7592 7593 7594 7595 7596
	/*
	 * 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
7597
	if (rq->curr == p) {
7598
		if (p->prio > oldprio)
7599
			resched_curr(rq);
7600
	} else
7601
		check_preempt_curr(rq, p, 0);
7602 7603
}

P
Peter Zijlstra 已提交
7604 7605 7606 7607 7608 7609
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);

	/*
7610
	 * Ensure the task's vruntime is normalized, so that when it's
P
Peter Zijlstra 已提交
7611 7612 7613
	 * switched back to the fair class the enqueue_entity(.flags=0) will
	 * do the right thing.
	 *
7614 7615
	 * 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
7616 7617
	 * the task is sleeping will it still have non-normalized vruntime.
	 */
7618
	if (!task_on_rq_queued(p) && p->state != TASK_RUNNING) {
P
Peter Zijlstra 已提交
7619 7620 7621 7622 7623 7624 7625
		/*
		 * 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;
	}
7626

7627
#ifdef CONFIG_SMP
7628 7629 7630 7631 7632
	/*
	* 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.
	*/
7633 7634 7635
	if (se->avg.decay_count) {
		__synchronize_entity_decay(se);
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
7636 7637
	}
#endif
P
Peter Zijlstra 已提交
7638 7639
}

7640 7641 7642
/*
 * We switched to the sched_fair class.
 */
P
Peter Zijlstra 已提交
7643
static void switched_to_fair(struct rq *rq, struct task_struct *p)
7644
{
7645
#ifdef CONFIG_FAIR_GROUP_SCHED
7646
	struct sched_entity *se = &p->se;
7647 7648 7649 7650 7651 7652
	/*
	 * 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
7653
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
7654 7655
		return;

7656 7657 7658 7659 7660
	/*
	 * 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
7661
	if (rq->curr == p)
7662
		resched_curr(rq);
7663
	else
7664
		check_preempt_curr(rq, p, 0);
7665 7666
}

7667 7668 7669 7670 7671 7672 7673 7674 7675
/* 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;

7676 7677 7678 7679 7680 7681 7682
	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);
	}
7683 7684
}

7685 7686 7687 7688 7689 7690 7691
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
7692
#ifdef CONFIG_SMP
7693
	atomic64_set(&cfs_rq->decay_counter, 1);
7694
	atomic_long_set(&cfs_rq->removed_load, 0);
7695
#endif
7696 7697
}

P
Peter Zijlstra 已提交
7698
#ifdef CONFIG_FAIR_GROUP_SCHED
7699
static void task_move_group_fair(struct task_struct *p, int queued)
P
Peter Zijlstra 已提交
7700
{
P
Peter Zijlstra 已提交
7701
	struct sched_entity *se = &p->se;
7702
	struct cfs_rq *cfs_rq;
P
Peter Zijlstra 已提交
7703

7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716
	/*
	 * 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.
	 */
7717
	/*
7718
	 * When !queued, vruntime of the task has usually NOT been normalized.
7719 7720 7721 7722
	 * 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().
7723 7724
	 * - Moving a task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
7725 7726 7727 7728
	 *
	 * To prevent boost or penalty in the new cfs_rq caused by delta
	 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
	 */
7729 7730
	if (!queued && (!se->sum_exec_runtime || p->state == TASK_WAKING))
		queued = 1;
7731

7732
	if (!queued)
P
Peter Zijlstra 已提交
7733
		se->vruntime -= cfs_rq_of(se)->min_vruntime;
7734
	set_task_rq(p, task_cpu(p));
P
Peter Zijlstra 已提交
7735
	se->depth = se->parent ? se->parent->depth + 1 : 0;
7736
	if (!queued) {
P
Peter Zijlstra 已提交
7737 7738
		cfs_rq = cfs_rq_of(se);
		se->vruntime += cfs_rq->min_vruntime;
7739 7740 7741 7742 7743 7744
#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
7745 7746
		se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
		cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
7747 7748
#endif
	}
P
Peter Zijlstra 已提交
7749
}
7750 7751 7752 7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767 7768 7769 7770 7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800 7801 7802 7803 7804 7805 7806 7807 7808 7809 7810 7811 7812 7813 7814 7815 7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839 7840 7841

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
7842
	if (!parent) {
7843
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
7844 7845
		se->depth = 0;
	} else {
7846
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
7847 7848
		se->depth = parent->depth + 1;
	}
7849 7850

	se->my_q = cfs_rq;
7851 7852
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880 7881 7882
	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);
7883 7884 7885

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
7886
		for_each_sched_entity(se)
7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897 7898 7899 7900 7901 7902 7903 7904 7905 7906 7907
			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
7908

7909
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
7910 7911 7912 7913 7914 7915 7916 7917 7918
{
	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)
7919
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
7920 7921 7922 7923

	return rr_interval;
}

7924 7925 7926
/*
 * All the scheduling class methods:
 */
7927
const struct sched_class fair_sched_class = {
7928
	.next			= &idle_sched_class,
7929 7930 7931
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
7932
	.yield_to_task		= yield_to_task_fair,
7933

I
Ingo Molnar 已提交
7934
	.check_preempt_curr	= check_preempt_wakeup,
7935 7936 7937 7938

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

7939
#ifdef CONFIG_SMP
L
Li Zefan 已提交
7940
	.select_task_rq		= select_task_rq_fair,
7941
	.migrate_task_rq	= migrate_task_rq_fair,
7942

7943 7944
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
7945 7946

	.task_waking		= task_waking_fair,
7947
#endif
7948

7949
	.set_curr_task          = set_curr_task_fair,
7950
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
7951
	.task_fork		= task_fork_fair,
7952 7953

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
7954
	.switched_from		= switched_from_fair,
7955
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
7956

7957 7958
	.get_rr_interval	= get_rr_interval_fair,

P
Peter Zijlstra 已提交
7959
#ifdef CONFIG_FAIR_GROUP_SCHED
7960
	.task_move_group	= task_move_group_fair,
P
Peter Zijlstra 已提交
7961
#endif
7962 7963 7964
};

#ifdef CONFIG_SCHED_DEBUG
7965
void print_cfs_stats(struct seq_file *m, int cpu)
7966 7967 7968
{
	struct cfs_rq *cfs_rq;

7969
	rcu_read_lock();
7970
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
7971
		print_cfs_rq(m, cpu, cfs_rq);
7972
	rcu_read_unlock();
7973 7974
}
#endif
7975 7976 7977 7978 7979 7980

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

7981
#ifdef CONFIG_NO_HZ_COMMON
7982
	nohz.next_balance = jiffies;
7983
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
7984
	cpu_notifier(sched_ilb_notifier, 0);
7985 7986 7987 7988
#endif
#endif /* SMP */

}