fair.c 222.0 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
/*
 * 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:
 */
144
static unsigned int get_update_sysctl_factor(void)
145
{
146
	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
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
	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
	unsigned 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
static inline void __update_task_entity_contrib(struct sched_entity *se);
673
static inline void __update_task_entity_utilization(struct sched_entity *se);
674 675 676 677 678 679 680

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

	slice = sched_slice(task_cfs_rq(p), &p->se) >> 10;
681 682
	p->se.avg.runnable_avg_sum = p->se.avg.running_avg_sum = slice;
	p->se.avg.avg_period = slice;
683
	__update_task_entity_contrib(&p->se);
684
	__update_task_entity_utilization(&p->se);
685 686 687 688 689 690 691
}
#else
void init_task_runnable_average(struct task_struct *p)
{
}
#endif

692
/*
693
 * Update the current task's runtime statistics.
694
 */
695
static void update_curr(struct cfs_rq *cfs_rq)
696
{
697
	struct sched_entity *curr = cfs_rq->curr;
698
	u64 now = rq_clock_task(rq_of(cfs_rq));
699
	u64 delta_exec;
700 701 702 703

	if (unlikely(!curr))
		return;

704 705
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
706
		return;
707

I
Ingo Molnar 已提交
708
	curr->exec_start = now;
709

710 711 712 713 714 715 716 717 718
	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);

719 720 721
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

722
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
723
		cpuacct_charge(curtask, delta_exec);
724
		account_group_exec_runtime(curtask, delta_exec);
725
	}
726 727

	account_cfs_rq_runtime(cfs_rq, delta_exec);
728 729
}

730 731 732 733 734
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

735
static inline void
736
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
737
{
738
	schedstat_set(se->statistics.wait_start, rq_clock(rq_of(cfs_rq)));
739 740 741 742 743
}

/*
 * Task is being enqueued - update stats:
 */
744
static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
745 746 747 748 749
{
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
750
	if (se != cfs_rq->curr)
751
		update_stats_wait_start(cfs_rq, se);
752 753 754
}

static void
755
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
756
{
757
	schedstat_set(se->statistics.wait_max, max(se->statistics.wait_max,
758
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start));
759 760
	schedstat_set(se->statistics.wait_count, se->statistics.wait_count + 1);
	schedstat_set(se->statistics.wait_sum, se->statistics.wait_sum +
761
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
762 763 764
#ifdef CONFIG_SCHEDSTATS
	if (entity_is_task(se)) {
		trace_sched_stat_wait(task_of(se),
765
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
766 767
	}
#endif
768
	schedstat_set(se->statistics.wait_start, 0);
769 770 771
}

static inline void
772
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
773 774 775 776 777
{
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
778
	if (se != cfs_rq->curr)
779
		update_stats_wait_end(cfs_rq, se);
780 781 782 783 784 785
}

/*
 * We are picking a new current task - update its stats:
 */
static inline void
786
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
787 788 789 790
{
	/*
	 * We are starting a new run period:
	 */
791
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
792 793 794 795 796 797
}

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

798 799
#ifdef CONFIG_NUMA_BALANCING
/*
800 801 802
 * 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.
803
 */
804 805
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
806 807 808

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
809

810 811 812
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836
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)
{
837
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
838 839 840
	unsigned int scan, floor;
	unsigned int windows = 1;

841 842
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858
	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);
}

859 860 861 862 863 864 865 866 867 868 869 870
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));
}

871 872 873 874 875
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
876
	pid_t gid;
877 878

	struct rcu_head rcu;
879
	nodemask_t active_nodes;
880
	unsigned long total_faults;
881 882 883 884 885
	/*
	 * 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.
	 */
886
	unsigned long *faults_cpu;
887
	unsigned long faults[0];
888 889
};

890 891 892 893 894 895 896 897 898
/* 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)

899 900 901 902 903
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

904 905 906 907 908 909 910
/*
 * 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.
 */
static inline int task_faults_idx(enum numa_faults_stats s, int nid, int priv)
911
{
912
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
913 914 915 916
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
917
	if (!p->numa_faults)
918 919
		return 0;

920 921
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
922 923
}

924 925 926 927 928
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

929 930
	return p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 1)];
931 932
}

933 934
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
935 936
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
937 938
}

939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003
/* Handle placement on systems where not all nodes are directly connected. */
static unsigned long score_nearby_nodes(struct task_struct *p, int nid,
					int maxdist, bool task)
{
	unsigned long score = 0;
	int node;

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

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

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

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

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

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

		score += faults;
	}

	return score;
}

1004 1005 1006 1007 1008 1009
/*
 * 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.
 */
1010 1011
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1012
{
1013
	unsigned long faults, total_faults;
1014

1015
	if (!p->numa_faults)
1016 1017 1018 1019 1020 1021 1022
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1023
	faults = task_faults(p, nid);
1024 1025
	faults += score_nearby_nodes(p, nid, dist, true);

1026
	return 1000 * faults / total_faults;
1027 1028
}

1029 1030
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1031
{
1032 1033 1034 1035 1036 1037 1038 1039
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1040 1041
		return 0;

1042
	faults = group_faults(p, nid);
1043 1044
	faults += score_nearby_nodes(p, nid, dist, false);

1045
	return 1000 * faults / total_faults;
1046 1047
}

1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110
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);
}

1111
static unsigned long weighted_cpuload(const int cpu);
1112 1113
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1114
static unsigned long capacity_of(int cpu);
1115 1116
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1117
/* Cached statistics for all CPUs within a node */
1118
struct numa_stats {
1119
	unsigned long nr_running;
1120
	unsigned long load;
1121 1122

	/* Total compute capacity of CPUs on a node */
1123
	unsigned long compute_capacity;
1124 1125

	/* Approximate capacity in terms of runnable tasks on a node */
1126
	unsigned long task_capacity;
1127
	int has_free_capacity;
1128
};
1129

1130 1131 1132 1133 1134
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1135 1136
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1137 1138 1139 1140 1141 1142 1143

	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);
1144
		ns->compute_capacity += capacity_of(cpu);
1145 1146

		cpus++;
1147 1148
	}

1149 1150 1151 1152 1153
	/*
	 * 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.
	 *
1154 1155
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1156 1157 1158 1159
	 */
	if (!cpus)
		return;

1160 1161 1162 1163 1164 1165
	/* 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));
1166
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1167 1168
}

1169 1170
struct task_numa_env {
	struct task_struct *p;
1171

1172 1173
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1174

1175
	struct numa_stats src_stats, dst_stats;
1176

1177
	int imbalance_pct;
1178
	int dist;
1179 1180 1181

	struct task_struct *best_task;
	long best_imp;
1182 1183 1184
	int best_cpu;
};

1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
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;
}

1198
static bool load_too_imbalanced(long src_load, long dst_load,
1199 1200
				struct task_numa_env *env)
{
1201 1202
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
	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;
1214 1215

	/* We care about the slope of the imbalance, not the direction. */
1216 1217
	if (dst_load < src_load)
		swap(dst_load, src_load);
1218 1219

	/* Is the difference below the threshold? */
1220 1221
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1222 1223 1224 1225 1226
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
1227
	 * Compare it with the old imbalance.
1228
	 */
1229
	orig_src_load = env->src_stats.load;
1230
	orig_dst_load = env->dst_stats.load;
1231

1232 1233
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);
1234

1235 1236 1237 1238 1239
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;

	/* Would this change make things worse? */
	return (imb > old_imb);
1240 1241
}

1242 1243 1244 1245 1246 1247
/*
 * 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
 */
1248 1249
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1250 1251 1252 1253
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1254
	long src_load, dst_load;
1255
	long load;
1256
	long imp = env->p->numa_group ? groupimp : taskimp;
1257
	long moveimp = imp;
1258
	int dist = env->dist;
1259 1260

	rcu_read_lock();
1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271

	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))
1272
		cur = NULL;
1273
	raw_spin_unlock_irq(&dst_rq->lock);
1274

1275 1276 1277 1278 1279 1280 1281
	/*
	 * 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;

1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293
	/*
	 * "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;

1294 1295
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1296
		 * in any group then look only at task weights.
1297
		 */
1298
		if (cur->numa_group == env->p->numa_group) {
1299 1300
			imp = taskimp + task_weight(cur, env->src_nid, dist) -
			      task_weight(cur, env->dst_nid, dist);
1301 1302 1303 1304 1305 1306
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1307
		} else {
1308 1309 1310 1311 1312 1313
			/*
			 * 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)
1314 1315
				imp += group_weight(cur, env->src_nid, dist) -
				       group_weight(cur, env->dst_nid, dist);
1316
			else
1317 1318
				imp += task_weight(cur, env->src_nid, dist) -
				       task_weight(cur, env->dst_nid, dist);
1319
		}
1320 1321
	}

1322
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1323 1324 1325 1326
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1327
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1328
		    !env->dst_stats.has_free_capacity)
1329 1330 1331 1332 1333 1334
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1335 1336
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1337 1338 1339 1340 1341 1342
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1343 1344 1345
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1346

1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363
	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;

1364
	if (cur) {
1365 1366 1367
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1368 1369
	}

1370
	if (load_too_imbalanced(src_load, dst_load, env))
1371 1372
		goto unlock;

1373 1374 1375 1376 1377 1378 1379
	/*
	 * 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);

1380 1381 1382 1383 1384 1385
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1386 1387
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1388 1389 1390 1391 1392 1393 1394 1395 1396
{
	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;
1397
		task_numa_compare(env, taskimp, groupimp);
1398 1399 1400
	}
}

1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424
/* Only move tasks to a NUMA node less busy than the current node. */
static bool numa_has_capacity(struct task_numa_env *env)
{
	struct numa_stats *src = &env->src_stats;
	struct numa_stats *dst = &env->dst_stats;

	if (src->has_free_capacity && !dst->has_free_capacity)
		return false;

	/*
	 * Only consider a task move if the source has a higher load
	 * than the destination, corrected for CPU capacity on each node.
	 *
	 *      src->load                dst->load
	 * --------------------- vs ---------------------
	 * src->compute_capacity    dst->compute_capacity
	 */
	if (src->load * dst->compute_capacity >
	    dst->load * src->compute_capacity)
		return true;

	return false;
}

1425 1426 1427 1428
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1429

1430
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1431
		.src_nid = task_node(p),
1432 1433 1434 1435 1436 1437

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
		.best_cpu = -1
1438 1439
	};
	struct sched_domain *sd;
1440
	unsigned long taskweight, groupweight;
1441
	int nid, ret, dist;
1442
	long taskimp, groupimp;
1443

1444
	/*
1445 1446 1447 1448 1449 1450
	 * 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.
1451 1452
	 */
	rcu_read_lock();
1453
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1454 1455
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1456 1457
	rcu_read_unlock();

1458 1459 1460 1461 1462 1463 1464
	/*
	 * 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)) {
1465
		p->numa_preferred_nid = task_node(p);
1466 1467 1468
		return -EINVAL;
	}

1469
	env.dst_nid = p->numa_preferred_nid;
1470 1471 1472 1473 1474 1475
	dist = env.dist = node_distance(env.src_nid, env.dst_nid);
	taskweight = task_weight(p, env.src_nid, dist);
	groupweight = group_weight(p, env.src_nid, dist);
	update_numa_stats(&env.src_stats, env.src_nid);
	taskimp = task_weight(p, env.dst_nid, dist) - taskweight;
	groupimp = group_weight(p, env.dst_nid, dist) - groupweight;
1476
	update_numa_stats(&env.dst_stats, env.dst_nid);
1477

1478
	/* Try to find a spot on the preferred nid. */
1479 1480
	if (numa_has_capacity(&env))
		task_numa_find_cpu(&env, taskimp, groupimp);
1481

1482 1483 1484 1485 1486 1487 1488 1489 1490
	/*
	 * Look at other nodes in these cases:
	 * - there is no space available on the preferred_nid
	 * - the task is part of a numa_group that is interleaved across
	 *   multiple NUMA nodes; in order to better consolidate the group,
	 *   we need to check other locations.
	 */
	if (env.best_cpu == -1 || (p->numa_group &&
			nodes_weight(p->numa_group->active_nodes) > 1)) {
1491 1492 1493
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1494

1495
			dist = node_distance(env.src_nid, env.dst_nid);
1496 1497 1498 1499 1500
			if (sched_numa_topology_type == NUMA_BACKPLANE &&
						dist != env.dist) {
				taskweight = task_weight(p, env.src_nid, dist);
				groupweight = group_weight(p, env.src_nid, dist);
			}
1501

1502
			/* Only consider nodes where both task and groups benefit */
1503 1504
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1505
			if (taskimp < 0 && groupimp < 0)
1506 1507
				continue;

1508
			env.dist = dist;
1509 1510
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1511 1512
			if (numa_has_capacity(&env))
				task_numa_find_cpu(&env, taskimp, groupimp);
1513 1514 1515
		}
	}

1516 1517 1518 1519 1520 1521 1522 1523
	/*
	 * 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.
	 */
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536
	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;
1537

1538 1539 1540 1541 1542 1543
	/*
	 * 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);

1544
	if (env.best_task == NULL) {
1545 1546 1547
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1548 1549 1550 1551
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1552 1553
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1554 1555
	put_task_struct(env.best_task);
	return ret;
1556 1557
}

1558 1559 1560
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1561 1562
	unsigned long interval = HZ;

1563
	/* This task has no NUMA fault statistics yet */
1564
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1565 1566
		return;

1567
	/* Periodically retry migrating the task to the preferred node */
1568 1569
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1570 1571

	/* Success if task is already running on preferred CPU */
1572
	if (task_node(p) == p->numa_preferred_nid)
1573 1574 1575
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1576
	task_numa_migrate(p);
1577 1578
}

1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610
/*
 * 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);
	}
}

1611 1612 1613
/*
 * 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
1614 1615 1616
 * 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.
1617 1618
 */
#define NUMA_PERIOD_SLOTS 10
1619
#define NUMA_PERIOD_THRESHOLD 7
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639

/*
 * 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
1640 1641 1642
	 * to automatic numa balancing. Related to that, if there were failed
	 * migration then it implies we are migrating too quickly or the local
	 * node is overloaded. In either case, scan slower
1643
	 */
1644
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677
		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
		 */
1678
		ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared + 1));
1679 1680 1681 1682 1683 1684 1685 1686
		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));
}

1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
/*
 * 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;
1706
		*period = p->se.avg.avg_period;
1707 1708 1709 1710 1711 1712 1713 1714
	}

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

	return delta;
}

1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761
/*
 * Determine the preferred nid for a task in a numa_group. This needs to
 * be done in a way that produces consistent results with group_weight,
 * otherwise workloads might not converge.
 */
static int preferred_group_nid(struct task_struct *p, int nid)
{
	nodemask_t nodes;
	int dist;

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

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

		dist = sched_max_numa_distance;

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

	/*
	 * Finding the preferred nid in a system with NUMA backplane
	 * interconnect topology is more involved. The goal is to locate
	 * tasks from numa_groups near each other in the system, and
	 * untangle workloads from different sides of the system. This requires
	 * searching down the hierarchy of node groups, recursively searching
	 * inside the highest scoring group of nodes. The nodemask tricks
	 * keep the complexity of the search down.
	 */
	nodes = node_online_map;
	for (dist = sched_max_numa_distance; dist > LOCAL_DISTANCE; dist--) {
		unsigned long max_faults = 0;
1762
		nodemask_t max_group = NODE_MASK_NONE;
1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795
		int a, b;

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

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

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

			/* Remember the top group. */
			if (faults > max_faults) {
				max_faults = faults;
				max_group = this_group;
				/*
				 * subtle: at the smallest distance there is
				 * just one node left in each "group", the
				 * winner is the preferred nid.
				 */
				nid = a;
			}
		}
		/* Next round, evaluate the nodes within max_group. */
1796 1797
		if (!max_faults)
			break;
1798 1799 1800 1801 1802
		nodes = max_group;
	}
	return nid;
}

1803 1804
static void task_numa_placement(struct task_struct *p)
{
1805 1806
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
1807
	unsigned long fault_types[2] = { 0, 0 };
1808 1809
	unsigned long total_faults;
	u64 runtime, period;
1810
	spinlock_t *group_lock = NULL;
1811

1812 1813 1814 1815 1816
	/*
	 * The p->mm->numa_scan_seq field gets updated without
	 * exclusive access. Use READ_ONCE() here to ensure
	 * that the field is read in a single access:
	 */
1817
	seq = READ_ONCE(p->mm->numa_scan_seq);
1818 1819 1820
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
1821
	p->numa_scan_period_max = task_scan_max(p);
1822

1823 1824 1825 1826
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

1827 1828 1829
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
1830
		spin_lock_irq(group_lock);
1831 1832
	}

1833 1834
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
1835 1836
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
1837
		unsigned long faults = 0, group_faults = 0;
1838
		int priv;
1839

1840
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
1841
			long diff, f_diff, f_weight;
1842

1843 1844 1845 1846
			mem_idx = task_faults_idx(NUMA_MEM, nid, priv);
			membuf_idx = task_faults_idx(NUMA_MEMBUF, nid, priv);
			cpu_idx = task_faults_idx(NUMA_CPU, nid, priv);
			cpubuf_idx = task_faults_idx(NUMA_CPUBUF, nid, priv);
1847

1848
			/* Decay existing window, copy faults since last scan */
1849 1850 1851
			diff = p->numa_faults[membuf_idx] - p->numa_faults[mem_idx] / 2;
			fault_types[priv] += p->numa_faults[membuf_idx];
			p->numa_faults[membuf_idx] = 0;
1852

1853 1854 1855 1856 1857 1858 1859 1860
			/*
			 * 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);
1861
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
1862
				   (total_faults + 1);
1863 1864
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
1865

1866 1867 1868
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
1869
			p->total_numa_faults += diff;
1870
			if (p->numa_group) {
1871 1872 1873 1874 1875 1876 1877 1878 1879
				/*
				 * safe because we can only change our own group
				 *
				 * mem_idx represents the offset for a given
				 * nid and priv in a specific region because it
				 * is at the beginning of the numa_faults array.
				 */
				p->numa_group->faults[mem_idx] += diff;
				p->numa_group->faults_cpu[mem_idx] += f_diff;
1880
				p->numa_group->total_faults += diff;
1881
				group_faults += p->numa_group->faults[mem_idx];
1882
			}
1883 1884
		}

1885 1886 1887 1888
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
1889 1890 1891 1892 1893 1894 1895

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

1896 1897
	update_task_scan_period(p, fault_types[0], fault_types[1]);

1898
	if (p->numa_group) {
1899
		update_numa_active_node_mask(p->numa_group);
1900
		spin_unlock_irq(group_lock);
1901
		max_nid = preferred_group_nid(p, max_group_nid);
1902 1903
	}

1904 1905 1906 1907 1908 1909 1910
	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);
1911
	}
1912 1913
}

1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924
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);
}

1925 1926
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
1927 1928 1929 1930 1931 1932 1933 1934 1935
{
	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) +
1936
				    4*nr_node_ids*sizeof(unsigned long);
1937 1938 1939 1940 1941 1942 1943

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

		atomic_set(&grp->refcount, 1);
		spin_lock_init(&grp->lock);
1944
		grp->gid = p->pid;
1945
		/* Second half of the array tracks nids where faults happen */
1946 1947
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
1948

1949 1950
		node_set(task_node(current), grp->active_nodes);

1951
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
1952
			grp->faults[i] = p->numa_faults[i];
1953

1954
		grp->total_faults = p->total_numa_faults;
1955

1956 1957 1958 1959 1960
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
1961
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
1962 1963

	if (!cpupid_match_pid(tsk, cpupid))
1964
		goto no_join;
1965 1966 1967

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
1968
		goto no_join;
1969 1970 1971

	my_grp = p->numa_group;
	if (grp == my_grp)
1972
		goto no_join;
1973 1974 1975 1976 1977 1978

	/*
	 * 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)
1979
		goto no_join;
1980 1981 1982 1983 1984

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
1985
		goto no_join;
1986

1987 1988 1989 1990 1991 1992 1993
	/* 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;
1994

1995 1996 1997
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

1998
	if (join && !get_numa_group(grp))
1999
		goto no_join;
2000 2001 2002 2003 2004 2005

	rcu_read_unlock();

	if (!join)
		return;

2006 2007
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2008

2009
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2010 2011
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2012
	}
2013 2014
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2015 2016 2017 2018 2019

	my_grp->nr_tasks--;
	grp->nr_tasks++;

	spin_unlock(&my_grp->lock);
2020
	spin_unlock_irq(&grp->lock);
2021 2022 2023 2024

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2025 2026 2027 2028 2029
	return;

no_join:
	rcu_read_unlock();
	return;
2030 2031 2032 2033 2034
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2035
	void *numa_faults = p->numa_faults;
2036 2037
	unsigned long flags;
	int i;
2038 2039

	if (grp) {
2040
		spin_lock_irqsave(&grp->lock, flags);
2041
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2042
			grp->faults[i] -= p->numa_faults[i];
2043
		grp->total_faults -= p->total_numa_faults;
2044

2045
		grp->nr_tasks--;
2046
		spin_unlock_irqrestore(&grp->lock, flags);
2047
		RCU_INIT_POINTER(p->numa_group, NULL);
2048 2049 2050
		put_numa_group(grp);
	}

2051
	p->numa_faults = NULL;
2052
	kfree(numa_faults);
2053 2054
}

2055 2056 2057
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2058
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2059 2060
{
	struct task_struct *p = current;
2061
	bool migrated = flags & TNF_MIGRATED;
2062
	int cpu_node = task_node(current);
2063
	int local = !!(flags & TNF_FAULT_LOCAL);
2064
	int priv;
2065

2066
	if (!numabalancing_enabled)
2067 2068
		return;

2069 2070 2071 2072
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2073
	/* Allocate buffer to track faults on a per-node basis */
2074 2075
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2076
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2077

2078 2079
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2080
			return;
2081

2082
		p->total_numa_faults = 0;
2083
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2084
	}
2085

2086 2087 2088 2089 2090 2091 2092 2093
	/*
	 * 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);
2094
		if (!priv && !(flags & TNF_NO_GROUP))
2095
			task_numa_group(p, last_cpupid, flags, &priv);
2096 2097
	}

2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108
	/*
	 * 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;

2109
	task_numa_placement(p);
2110

2111 2112 2113 2114 2115
	/*
	 * 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))
2116 2117
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
2118 2119
	if (migrated)
		p->numa_pages_migrated += pages;
2120 2121
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2122

2123 2124
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2125
	p->numa_faults_locality[local] += pages;
2126 2127
}

2128 2129
static void reset_ptenuma_scan(struct task_struct *p)
{
2130 2131 2132 2133 2134 2135 2136 2137
	/*
	 * We only did a read acquisition of the mmap sem, so
	 * p->mm->numa_scan_seq is written to without exclusive access
	 * and the update is not guaranteed to be atomic. That's not
	 * much of an issue though, since this is just used for
	 * statistical sampling. Use READ_ONCE/WRITE_ONCE, which are not
	 * expensive, to avoid any form of compiler optimizations:
	 */
2138
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2139 2140 2141
	p->mm->numa_scan_offset = 0;
}

2142 2143 2144 2145 2146 2147 2148 2149 2150
/*
 * 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;
2151
	struct vm_area_struct *vma;
2152
	unsigned long start, end;
2153
	unsigned long nr_pte_updates = 0;
2154
	long pages;
2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169

	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;

2170
	if (!mm->numa_next_scan) {
2171 2172
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2173 2174
	}

2175 2176 2177 2178 2179 2180 2181
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2182 2183 2184 2185
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
2186

2187
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2188 2189 2190
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2191 2192 2193 2194 2195 2196
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2197 2198 2199 2200 2201
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
	if (!pages)
		return;
2202

2203
	down_read(&mm->mmap_sem);
2204
	vma = find_vma(mm, start);
2205 2206
	if (!vma) {
		reset_ptenuma_scan(p);
2207
		start = 0;
2208 2209
		vma = mm->mmap;
	}
2210
	for (; vma; vma = vma->vm_next) {
2211
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2212
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2213
			continue;
2214
		}
2215

2216 2217 2218 2219 2220 2221 2222 2223 2224 2225
		/*
		 * 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
2226 2227 2228 2229 2230 2231
		/*
		 * 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;
2232

2233 2234 2235 2236
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2237 2238 2239 2240 2241 2242 2243 2244 2245
			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;
2246

2247 2248 2249
			start = end;
			if (pages <= 0)
				goto out;
2250 2251

			cond_resched();
2252
		} while (end != vma->vm_end);
2253
	}
2254

2255
out:
2256
	/*
P
Peter Zijlstra 已提交
2257 2258 2259 2260
	 * 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.
2261 2262
	 */
	if (vma)
2263
		mm->numa_scan_offset = start;
2264 2265 2266
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292
}

/*
 * 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) {
2293
		if (!curr->node_stamp)
2294
			curr->numa_scan_period = task_scan_min(curr);
2295
		curr->node_stamp += period;
2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306

		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)
{
}
2307 2308 2309 2310 2311 2312 2313 2314

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)
{
}
2315 2316
#endif /* CONFIG_NUMA_BALANCING */

2317 2318 2319 2320
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2321
	if (!parent_entity(se))
2322
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2323
#ifdef CONFIG_SMP
2324 2325 2326 2327 2328 2329
	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);
	}
2330
#endif
2331 2332 2333 2334 2335 2336 2337
	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);
2338
	if (!parent_entity(se))
2339
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2340 2341
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2342
		list_del_init(&se->group_node);
2343
	}
2344 2345 2346
	cfs_rq->nr_running--;
}

2347 2348
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2349 2350 2351 2352 2353 2354 2355 2356 2357
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().
	 */
2358
	tg_weight = atomic_long_read(&tg->load_avg);
2359
	tg_weight -= cfs_rq->tg_load_contrib;
2360 2361 2362 2363 2364
	tg_weight += cfs_rq->load.weight;

	return tg_weight;
}

2365
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2366
{
2367
	long tg_weight, load, shares;
2368

2369
	tg_weight = calc_tg_weight(tg, cfs_rq);
2370
	load = cfs_rq->load.weight;
2371 2372

	shares = (tg->shares * load);
2373 2374
	if (tg_weight)
		shares /= tg_weight;
2375 2376 2377 2378 2379 2380 2381 2382 2383

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

	return shares;
}
# else /* CONFIG_SMP */
2384
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2385 2386 2387 2388
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
2389 2390 2391
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2392 2393 2394 2395
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2396
		account_entity_dequeue(cfs_rq, se);
2397
	}
P
Peter Zijlstra 已提交
2398 2399 2400 2401 2402 2403 2404

	update_load_set(&se->load, weight);

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

2405 2406
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2407
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2408 2409 2410
{
	struct task_group *tg;
	struct sched_entity *se;
2411
	long shares;
P
Peter Zijlstra 已提交
2412 2413 2414

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2415
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2416
		return;
2417 2418 2419 2420
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2421
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2422 2423 2424 2425

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2426
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2427 2428 2429 2430
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2431
#ifdef CONFIG_SMP
2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459
/*
 * 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,
};

2460 2461 2462 2463 2464 2465
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477
	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
2478 2479
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
2480 2481 2482 2483 2484 2485
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
2486 2487
	}

2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518
	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];
2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548
}

/*
 * 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}]
 */
2549
static __always_inline int __update_entity_runnable_avg(u64 now, int cpu,
2550
							struct sched_avg *sa,
2551 2552
							int runnable,
							int running)
2553
{
2554 2555
	u64 delta, periods;
	u32 runnable_contrib;
2556
	int delta_w, decayed = 0;
2557
	unsigned long scale_freq = arch_scale_freq_capacity(NULL, cpu);
2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578

	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 */
2579
	delta_w = sa->avg_period % 1024;
2580 2581 2582 2583 2584 2585 2586 2587 2588 2589
	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;
2590 2591
		if (runnable)
			sa->runnable_avg_sum += delta_w;
2592
		if (running)
2593 2594
			sa->running_avg_sum += delta_w * scale_freq
				>> SCHED_CAPACITY_SHIFT;
2595
		sa->avg_period += delta_w;
2596 2597 2598 2599 2600 2601 2602 2603 2604

		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);
2605 2606 2607
		sa->running_avg_sum = decay_load(sa->running_avg_sum,
						  periods + 1);
		sa->avg_period = decay_load(sa->avg_period,
2608 2609 2610 2611 2612 2613
						     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;
2614
		if (running)
2615 2616
			sa->running_avg_sum += runnable_contrib * scale_freq
				>> SCHED_CAPACITY_SHIFT;
2617
		sa->avg_period += runnable_contrib;
2618 2619 2620 2621 2622
	}

	/* Remainder of delta accrued against u_0` */
	if (runnable)
		sa->runnable_avg_sum += delta;
2623
	if (running)
2624 2625
		sa->running_avg_sum += delta * scale_freq
			>> SCHED_CAPACITY_SHIFT;
2626
	sa->avg_period += delta;
2627 2628 2629 2630

	return decayed;
}

2631
/* Synchronize an entity's decay with its parenting cfs_rq.*/
2632
static inline u64 __synchronize_entity_decay(struct sched_entity *se)
2633 2634 2635 2636 2637
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 decays = atomic64_read(&cfs_rq->decay_counter);

	decays -= se->avg.decay_count;
2638
	se->avg.decay_count = 0;
2639
	if (!decays)
2640
		return 0;
2641 2642

	se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
2643 2644
	se->avg.utilization_avg_contrib =
		decay_load(se->avg.utilization_avg_contrib, decays);
2645 2646

	return decays;
2647 2648
}

2649 2650 2651 2652 2653
#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;
2654
	long tg_contrib;
2655 2656 2657 2658

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

2659 2660 2661
	if (!tg_contrib)
		return;

2662 2663
	if (force_update || abs(tg_contrib) > cfs_rq->tg_load_contrib / 8) {
		atomic_long_add(tg_contrib, &tg->load_avg);
2664 2665 2666
		cfs_rq->tg_load_contrib += tg_contrib;
	}
}
2667

2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678
/*
 * 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 */
2679
	contrib = div_u64((u64)sa->runnable_avg_sum << NICE_0_SHIFT,
2680
			  sa->avg_period + 1);
2681 2682 2683 2684 2685 2686 2687 2688
	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;
	}
}

2689 2690 2691 2692
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;
2693 2694
	int runnable_avg;

2695 2696 2697
	u64 contrib;

	contrib = cfs_rq->tg_load_contrib * tg->shares;
2698 2699
	se->avg.load_avg_contrib = div_u64(contrib,
				     atomic_long_read(&tg->load_avg) + 1);
2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728

	/*
	 * 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;
	}
2729
}
2730 2731 2732

static inline void update_rq_runnable_avg(struct rq *rq, int runnable)
{
2733 2734
	__update_entity_runnable_avg(rq_clock_task(rq), cpu_of(rq), &rq->avg,
			runnable, runnable);
2735 2736
	__update_tg_runnable_avg(&rq->avg, &rq->cfs);
}
2737
#else /* CONFIG_FAIR_GROUP_SCHED */
2738 2739
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update) {}
2740 2741
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq) {}
2742
static inline void __update_group_entity_contrib(struct sched_entity *se) {}
2743
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2744
#endif /* CONFIG_FAIR_GROUP_SCHED */
2745

2746 2747 2748 2749 2750 2751
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);
2752
	contrib /= (se->avg.avg_period + 1);
2753 2754 2755
	se->avg.load_avg_contrib = scale_load(contrib);
}

2756 2757 2758 2759 2760
/* 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;

2761 2762 2763
	if (entity_is_task(se)) {
		__update_task_entity_contrib(se);
	} else {
2764
		__update_tg_runnable_avg(&se->avg, group_cfs_rq(se));
2765 2766
		__update_group_entity_contrib(se);
	}
2767 2768 2769 2770

	return se->avg.load_avg_contrib - old_contrib;
}

2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787

static inline void __update_task_entity_utilization(struct sched_entity *se)
{
	u32 contrib;

	/* avoid overflowing a 32-bit type w/ SCHED_LOAD_SCALE */
	contrib = se->avg.running_avg_sum * scale_load_down(SCHED_LOAD_SCALE);
	contrib /= (se->avg.avg_period + 1);
	se->avg.utilization_avg_contrib = scale_load(contrib);
}

static long __update_entity_utilization_avg_contrib(struct sched_entity *se)
{
	long old_contrib = se->avg.utilization_avg_contrib;

	if (entity_is_task(se))
		__update_task_entity_utilization(se);
2788 2789 2790
	else
		se->avg.utilization_avg_contrib =
					group_cfs_rq(se)->utilization_load_avg;
2791 2792 2793 2794

	return se->avg.utilization_avg_contrib - old_contrib;
}

2795 2796 2797 2798 2799 2800 2801 2802 2803
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;
}

2804 2805
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);

2806
/* Update a sched_entity's runnable average */
2807 2808
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq)
2809
{
2810
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
2811
	long contrib_delta, utilization_delta;
2812
	int cpu = cpu_of(rq_of(cfs_rq));
2813
	u64 now;
2814

2815 2816 2817 2818 2819 2820 2821 2822 2823
	/*
	 * 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));

2824
	if (!__update_entity_runnable_avg(now, cpu, &se->avg, se->on_rq,
2825
					cfs_rq->curr == se))
2826 2827 2828
		return;

	contrib_delta = __update_entity_load_avg_contrib(se);
2829
	utilization_delta = __update_entity_utilization_avg_contrib(se);
2830 2831 2832 2833

	if (!update_cfs_rq)
		return;

2834
	if (se->on_rq) {
2835
		cfs_rq->runnable_load_avg += contrib_delta;
2836 2837
		cfs_rq->utilization_load_avg += utilization_delta;
	} else {
2838
		subtract_blocked_load_contrib(cfs_rq, -contrib_delta);
2839
	}
2840 2841 2842 2843 2844 2845
}

/*
 * Decay the load contributed by all blocked children and account this so that
 * their contribution may appropriately discounted when they wake up.
 */
2846
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq, int force_update)
2847
{
2848
	u64 now = cfs_rq_clock_task(cfs_rq) >> 20;
2849 2850 2851
	u64 decays;

	decays = now - cfs_rq->last_decay;
2852
	if (!decays && !force_update)
2853 2854
		return;

2855 2856 2857
	if (atomic_long_read(&cfs_rq->removed_load)) {
		unsigned long removed_load;
		removed_load = atomic_long_xchg(&cfs_rq->removed_load, 0);
2858 2859
		subtract_blocked_load_contrib(cfs_rq, removed_load);
	}
2860

2861 2862 2863 2864 2865 2866
	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;
	}
2867 2868

	__update_cfs_rq_tg_load_contrib(cfs_rq, force_update);
2869
}
2870

2871 2872
/* 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,
2873 2874
						  struct sched_entity *se,
						  int wakeup)
2875
{
2876 2877 2878 2879
	/*
	 * 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.
2880 2881 2882 2883
	 *
	 * 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.
2884 2885
	 */
	if (unlikely(se->avg.decay_count <= 0)) {
2886
		se->avg.last_runnable_update = rq_clock_task(rq_of(cfs_rq));
2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901
		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;
		}
2902 2903
		wakeup = 0;
	} else {
2904
		__synchronize_entity_decay(se);
2905 2906
	}

2907 2908
	/* migrated tasks did not contribute to our blocked load */
	if (wakeup) {
2909
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
2910 2911
		update_entity_load_avg(se, 0);
	}
2912

2913
	cfs_rq->runnable_load_avg += se->avg.load_avg_contrib;
2914
	cfs_rq->utilization_load_avg += se->avg.utilization_avg_contrib;
2915 2916
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !wakeup);
2917 2918
}

2919 2920 2921 2922 2923
/*
 * 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.
 */
2924
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2925 2926
						  struct sched_entity *se,
						  int sleep)
2927
{
2928
	update_entity_load_avg(se, 1);
2929 2930
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !sleep);
2931

2932
	cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib;
2933
	cfs_rq->utilization_load_avg -= se->avg.utilization_avg_contrib;
2934 2935 2936 2937
	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 */
2938
}
2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959

/*
 * 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);
}

2960 2961
static int idle_balance(struct rq *this_rq);

2962 2963
#else /* CONFIG_SMP */

2964 2965
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq) {}
2966
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2967
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2968 2969
					   struct sched_entity *se,
					   int wakeup) {}
2970
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2971 2972
					   struct sched_entity *se,
					   int sleep) {}
2973 2974
static inline void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
					      int force_update) {}
2975 2976 2977 2978 2979 2980

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

2981
#endif /* CONFIG_SMP */
2982

2983
static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
2984 2985
{
#ifdef CONFIG_SCHEDSTATS
2986 2987 2988 2989 2990
	struct task_struct *tsk = NULL;

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

2991
	if (se->statistics.sleep_start) {
2992
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.sleep_start;
2993 2994 2995 2996

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

2997 2998
		if (unlikely(delta > se->statistics.sleep_max))
			se->statistics.sleep_max = delta;
2999

3000
		se->statistics.sleep_start = 0;
3001
		se->statistics.sum_sleep_runtime += delta;
A
Arjan van de Ven 已提交
3002

3003
		if (tsk) {
3004
			account_scheduler_latency(tsk, delta >> 10, 1);
3005 3006
			trace_sched_stat_sleep(tsk, delta);
		}
3007
	}
3008
	if (se->statistics.block_start) {
3009
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.block_start;
3010 3011 3012 3013

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

3014 3015
		if (unlikely(delta > se->statistics.block_max))
			se->statistics.block_max = delta;
3016

3017
		se->statistics.block_start = 0;
3018
		se->statistics.sum_sleep_runtime += delta;
I
Ingo Molnar 已提交
3019

3020
		if (tsk) {
3021
			if (tsk->in_iowait) {
3022 3023
				se->statistics.iowait_sum += delta;
				se->statistics.iowait_count++;
3024
				trace_sched_stat_iowait(tsk, delta);
3025 3026
			}

3027 3028
			trace_sched_stat_blocked(tsk, delta);

3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039
			/*
			 * 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
3040
		}
3041 3042 3043 3044
	}
#endif
}

P
Peter Zijlstra 已提交
3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057
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
}

3058 3059 3060
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3061
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3062

3063 3064 3065 3066 3067 3068
	/*
	 * 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
3069
	if (initial && sched_feat(START_DEBIT))
3070
		vruntime += sched_vslice(cfs_rq, se);
3071

3072
	/* sleeps up to a single latency don't count. */
3073
	if (!initial) {
3074
		unsigned long thresh = sysctl_sched_latency;
3075

3076 3077 3078 3079 3080 3081
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3082

3083
		vruntime -= thresh;
3084 3085
	}

3086
	/* ensure we never gain time by being placed backwards. */
3087
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3088 3089
}

3090 3091
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3092
static void
3093
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3094
{
3095 3096
	/*
	 * Update the normalized vruntime before updating min_vruntime
3097
	 * through calling update_curr().
3098
	 */
3099
	if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
3100 3101
		se->vruntime += cfs_rq->min_vruntime;

3102
	/*
3103
	 * Update run-time statistics of the 'current'.
3104
	 */
3105
	update_curr(cfs_rq);
3106
	enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP);
3107 3108
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
3109

3110
	if (flags & ENQUEUE_WAKEUP) {
3111
		place_entity(cfs_rq, se, 0);
3112
		enqueue_sleeper(cfs_rq, se);
I
Ingo Molnar 已提交
3113
	}
3114

3115
	update_stats_enqueue(cfs_rq, se);
P
Peter Zijlstra 已提交
3116
	check_spread(cfs_rq, se);
3117 3118
	if (se != cfs_rq->curr)
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3119
	se->on_rq = 1;
3120

3121
	if (cfs_rq->nr_running == 1) {
3122
		list_add_leaf_cfs_rq(cfs_rq);
3123 3124
		check_enqueue_throttle(cfs_rq);
	}
3125 3126
}

3127
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3128
{
3129 3130
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3131
		if (cfs_rq->last != se)
3132
			break;
3133 3134

		cfs_rq->last = NULL;
3135 3136
	}
}
P
Peter Zijlstra 已提交
3137

3138 3139 3140 3141
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3142
		if (cfs_rq->next != se)
3143
			break;
3144 3145

		cfs_rq->next = NULL;
3146
	}
P
Peter Zijlstra 已提交
3147 3148
}

3149 3150 3151 3152
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3153
		if (cfs_rq->skip != se)
3154
			break;
3155 3156

		cfs_rq->skip = NULL;
3157 3158 3159
	}
}

P
Peter Zijlstra 已提交
3160 3161
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3162 3163 3164 3165 3166
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
3167 3168 3169

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
3170 3171
}

3172
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3173

3174
static void
3175
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3176
{
3177 3178 3179 3180
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
3181
	dequeue_entity_load_avg(cfs_rq, se, flags & DEQUEUE_SLEEP);
3182

3183
	update_stats_dequeue(cfs_rq, se);
3184
	if (flags & DEQUEUE_SLEEP) {
P
Peter Zijlstra 已提交
3185
#ifdef CONFIG_SCHEDSTATS
3186 3187 3188 3189
		if (entity_is_task(se)) {
			struct task_struct *tsk = task_of(se);

			if (tsk->state & TASK_INTERRUPTIBLE)
3190
				se->statistics.sleep_start = rq_clock(rq_of(cfs_rq));
3191
			if (tsk->state & TASK_UNINTERRUPTIBLE)
3192
				se->statistics.block_start = rq_clock(rq_of(cfs_rq));
3193
		}
3194
#endif
P
Peter Zijlstra 已提交
3195 3196
	}

P
Peter Zijlstra 已提交
3197
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3198

3199
	if (se != cfs_rq->curr)
3200
		__dequeue_entity(cfs_rq, se);
3201
	se->on_rq = 0;
3202
	account_entity_dequeue(cfs_rq, se);
3203 3204 3205 3206 3207 3208

	/*
	 * 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.
	 */
3209
	if (!(flags & DEQUEUE_SLEEP))
3210
		se->vruntime -= cfs_rq->min_vruntime;
3211

3212 3213 3214
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

3215
	update_min_vruntime(cfs_rq);
3216
	update_cfs_shares(cfs_rq);
3217 3218 3219 3220 3221
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
3222
static void
I
Ingo Molnar 已提交
3223
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3224
{
3225
	unsigned long ideal_runtime, delta_exec;
3226 3227
	struct sched_entity *se;
	s64 delta;
3228

P
Peter Zijlstra 已提交
3229
	ideal_runtime = sched_slice(cfs_rq, curr);
3230
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
3231
	if (delta_exec > ideal_runtime) {
3232
		resched_curr(rq_of(cfs_rq));
3233 3234 3235 3236 3237
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248
		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;

3249 3250
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
3251

3252 3253
	if (delta < 0)
		return;
3254

3255
	if (delta > ideal_runtime)
3256
		resched_curr(rq_of(cfs_rq));
3257 3258
}

3259
static void
3260
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
3261
{
3262 3263 3264 3265 3266 3267 3268 3269 3270
	/* '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);
3271
		update_entity_load_avg(se, 1);
3272 3273
	}

3274
	update_stats_curr_start(cfs_rq, se);
3275
	cfs_rq->curr = se;
I
Ingo Molnar 已提交
3276 3277 3278 3279 3280 3281
#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):
	 */
3282
	if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
3283
		se->statistics.slice_max = max(se->statistics.slice_max,
I
Ingo Molnar 已提交
3284 3285 3286
			se->sum_exec_runtime - se->prev_sum_exec_runtime);
	}
#endif
3287
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
3288 3289
}

3290 3291 3292
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

3293 3294 3295 3296 3297 3298 3299
/*
 * 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
 */
3300 3301
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3302
{
3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313
	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 */
3314

3315 3316 3317 3318 3319
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3320 3321 3322 3323 3324 3325 3326 3327 3328 3329
		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;
		}

3330 3331 3332
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3333

3334 3335 3336 3337 3338 3339
	/*
	 * 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;

3340 3341 3342 3343 3344 3345
	/*
	 * 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;

3346
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3347 3348

	return se;
3349 3350
}

3351
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3352

3353
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3354 3355 3356 3357 3358 3359
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3360
		update_curr(cfs_rq);
3361

3362 3363 3364
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

P
Peter Zijlstra 已提交
3365
	check_spread(cfs_rq, prev);
3366
	if (prev->on_rq) {
3367
		update_stats_wait_start(cfs_rq, prev);
3368 3369
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3370
		/* in !on_rq case, update occurred at dequeue */
3371
		update_entity_load_avg(prev, 1);
3372
	}
3373
	cfs_rq->curr = NULL;
3374 3375
}

P
Peter Zijlstra 已提交
3376 3377
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3378 3379
{
	/*
3380
	 * Update run-time statistics of the 'current'.
3381
	 */
3382
	update_curr(cfs_rq);
3383

3384 3385 3386
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3387
	update_entity_load_avg(curr, 1);
3388
	update_cfs_rq_blocked_load(cfs_rq, 1);
3389
	update_cfs_shares(cfs_rq);
3390

P
Peter Zijlstra 已提交
3391 3392 3393 3394 3395
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3396
	if (queued) {
3397
		resched_curr(rq_of(cfs_rq));
3398 3399
		return;
	}
P
Peter Zijlstra 已提交
3400 3401 3402 3403 3404 3405 3406 3407
	/*
	 * 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
3408
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3409
		check_preempt_tick(cfs_rq, curr);
3410 3411
}

3412 3413 3414 3415 3416 3417

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

#ifdef CONFIG_CFS_BANDWIDTH
3418 3419

#ifdef HAVE_JUMP_LABEL
3420
static struct static_key __cfs_bandwidth_used;
3421 3422 3423

static inline bool cfs_bandwidth_used(void)
{
3424
	return static_key_false(&__cfs_bandwidth_used);
3425 3426
}

3427
void cfs_bandwidth_usage_inc(void)
3428
{
3429 3430 3431 3432 3433 3434
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3435 3436 3437 3438 3439 3440 3441
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3442 3443
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3444 3445
#endif /* HAVE_JUMP_LABEL */

3446 3447 3448 3449 3450 3451 3452 3453
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3454 3455 3456 3457 3458 3459

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

P
Paul Turner 已提交
3460 3461 3462 3463 3464 3465 3466
/*
 * 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
 */
3467
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478
{
	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);
}

3479 3480 3481 3482 3483
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3484 3485 3486 3487 3488 3489
/* 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;

3490
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3491 3492
}

3493 3494
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3495 3496 3497
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3498
	u64 amount = 0, min_amount, expires;
3499 3500 3501 3502 3503 3504 3505

	/* 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;
3506
	else {
P
Peter Zijlstra 已提交
3507
		start_cfs_bandwidth(cfs_b);
3508 3509 3510 3511 3512 3513

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3514
	}
P
Paul Turner 已提交
3515
	expires = cfs_b->runtime_expires;
3516 3517 3518
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3519 3520 3521 3522 3523 3524 3525
	/*
	 * 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;
3526 3527

	return cfs_rq->runtime_remaining > 0;
3528 3529
}

P
Paul Turner 已提交
3530 3531 3532 3533 3534
/*
 * 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)
3535
{
P
Paul Turner 已提交
3536 3537 3538
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3539
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3540 3541
		return;

P
Paul Turner 已提交
3542 3543 3544 3545 3546 3547 3548 3549 3550
	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
3551 3552 3553
	 * 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
3554 3555
	 */

3556
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3557 3558 3559 3560 3561 3562 3563 3564
		/* 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;
	}
}

3565
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3566 3567
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3568
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3569 3570 3571
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3572 3573
		return;

3574 3575 3576 3577 3578
	/*
	 * 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))
3579
		resched_curr(rq_of(cfs_rq));
3580 3581
}

3582
static __always_inline
3583
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3584
{
3585
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3586 3587 3588 3589 3590
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3591 3592
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3593
	return cfs_bandwidth_used() && cfs_rq->throttled;
3594 3595
}

3596 3597 3598
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3599
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
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
}

/*
 * 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) {
3628
		/* adjust cfs_rq_clock_task() */
3629
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3630
					     cfs_rq->throttled_clock_task;
3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641
	}
#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)];

3642 3643
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3644
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3645 3646 3647 3648 3649
	cfs_rq->throttle_count++;

	return 0;
}

3650
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3651 3652 3653 3654 3655
{
	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;
P
Peter Zijlstra 已提交
3656
	bool empty;
3657 3658 3659

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

3660
	/* freeze hierarchy runnable averages while throttled */
3661 3662 3663
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680

	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)
3681
		sub_nr_running(rq, task_delta);
3682 3683

	cfs_rq->throttled = 1;
3684
	cfs_rq->throttled_clock = rq_clock(rq);
3685
	raw_spin_lock(&cfs_b->lock);
P
Peter Zijlstra 已提交
3686 3687
	empty = list_empty(&cfs_rq->throttled_list);

3688 3689 3690 3691 3692
	/*
	 * 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);
P
Peter Zijlstra 已提交
3693 3694 3695 3696 3697 3698 3699 3700

	/*
	 * If we're the first throttled task, make sure the bandwidth
	 * timer is running.
	 */
	if (empty)
		start_cfs_bandwidth(cfs_b);

3701 3702 3703
	raw_spin_unlock(&cfs_b->lock);
}

3704
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3705 3706 3707 3708 3709 3710 3711
{
	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;

3712
	se = cfs_rq->tg->se[cpu_of(rq)];
3713 3714

	cfs_rq->throttled = 0;
3715 3716 3717

	update_rq_clock(rq);

3718
	raw_spin_lock(&cfs_b->lock);
3719
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
3720 3721 3722
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

3723 3724 3725
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743
	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)
3744
		add_nr_running(rq, task_delta);
3745 3746 3747

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
3748
		resched_curr(rq);
3749 3750 3751 3752 3753 3754
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
3755 3756
	u64 runtime;
	u64 starting_runtime = remaining;
3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786

	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();

3787
	return starting_runtime - remaining;
3788 3789
}

3790 3791 3792 3793 3794 3795 3796 3797
/*
 * 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)
{
3798
	u64 runtime, runtime_expires;
3799
	int throttled;
3800 3801 3802

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
3803
		goto out_deactivate;
3804

3805
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
3806
	cfs_b->nr_periods += overrun;
3807

3808 3809 3810 3811 3812 3813
	/*
	 * 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
3814 3815 3816

	__refill_cfs_bandwidth_runtime(cfs_b);

3817 3818 3819
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
3820
		return 0;
3821 3822
	}

3823 3824 3825
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

3826 3827 3828
	runtime_expires = cfs_b->runtime_expires;

	/*
3829 3830 3831 3832 3833
	 * 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.
3834
	 */
3835 3836
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
3837 3838 3839 3840 3841 3842 3843
		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);
3844 3845

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3846
	}
3847

3848 3849 3850 3851 3852 3853 3854
	/*
	 * 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;
3855

3856 3857 3858 3859
	return 0;

out_deactivate:
	return 1;
3860
}
3861

3862 3863 3864 3865 3866 3867 3868
/* 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;

3869 3870 3871 3872
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
3873
 * hrtimer base being cleared by hrtimer_start. In the case of
3874 3875
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900
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;

P
Peter Zijlstra 已提交
3901 3902 3903
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932
}

/* 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)
{
3933 3934 3935
	if (!cfs_bandwidth_used())
		return;

3936
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951
		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 */
3952 3953 3954
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
3955
		return;
3956
	}
3957

3958
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
3959
		runtime = cfs_b->runtime;
3960

3961 3962 3963 3964 3965 3966 3967 3968 3969 3970
	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)
3971
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3972 3973 3974
	raw_spin_unlock(&cfs_b->lock);
}

3975 3976 3977 3978 3979 3980 3981
/*
 * 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)
{
3982 3983 3984
	if (!cfs_bandwidth_used())
		return;

3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999
	/* 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() */
4000
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4001
{
4002
	if (!cfs_bandwidth_used())
4003
		return false;
4004

4005
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4006
		return false;
4007 4008 4009 4010 4011 4012

	/*
	 * 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))
4013
		return true;
4014 4015

	throttle_cfs_rq(cfs_rq);
4016
	return true;
4017
}
4018 4019 4020 4021 4022

static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, slack_timer);
P
Peter Zijlstra 已提交
4023

4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035
	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);
	int overrun;
	int idle = 0;

4036
	raw_spin_lock(&cfs_b->lock);
4037
	for (;;) {
P
Peter Zijlstra 已提交
4038
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4039 4040 4041 4042 4043
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4044 4045
	if (idle)
		cfs_b->period_active = 0;
4046
	raw_spin_unlock(&cfs_b->lock);
4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058

	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);
P
Peter Zijlstra 已提交
4059
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070
	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);
}

P
Peter Zijlstra 已提交
4071
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4072
{
P
Peter Zijlstra 已提交
4073
	lockdep_assert_held(&cfs_b->lock);
4074

P
Peter Zijlstra 已提交
4075 4076 4077 4078 4079
	if (!cfs_b->period_active) {
		cfs_b->period_active = 1;
		hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
		hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
	}
4080 4081 4082 4083
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4084 4085 4086 4087
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4088 4089 4090 4091
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104
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);
	}
}

4105
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116
{
	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
		 */
4117
		cfs_rq->runtime_remaining = 1;
4118 4119 4120 4121 4122 4123
		/*
		 * 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;

4124 4125 4126 4127 4128 4129
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
4130 4131
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
4132
	return rq_clock_task(rq_of(cfs_rq));
4133 4134
}

4135
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
4136
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
4137
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
4138
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
4139 4140 4141 4142 4143

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154

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;
}
4155 4156 4157 4158 4159

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) {}
4160 4161
#endif

4162 4163 4164 4165 4166
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) {}
4167
static inline void update_runtime_enabled(struct rq *rq) {}
4168
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
4169 4170 4171

#endif /* CONFIG_CFS_BANDWIDTH */

4172 4173 4174 4175
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
4176 4177 4178 4179 4180 4181 4182 4183
#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);

4184
	if (cfs_rq->nr_running > 1) {
P
Peter Zijlstra 已提交
4185 4186 4187 4188 4189 4190
		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)
4191
				resched_curr(rq);
P
Peter Zijlstra 已提交
4192 4193
			return;
		}
4194
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
4195 4196
	}
}
4197 4198 4199 4200 4201 4202 4203 4204 4205 4206

/*
 * 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;

4207
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
4208 4209 4210 4211 4212
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
4213
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
4214 4215 4216 4217
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
4218 4219 4220 4221

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
4222 4223
#endif

4224 4225 4226 4227 4228
/*
 * 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:
 */
4229
static void
4230
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4231 4232
{
	struct cfs_rq *cfs_rq;
4233
	struct sched_entity *se = &p->se;
4234 4235

	for_each_sched_entity(se) {
4236
		if (se->on_rq)
4237 4238
			break;
		cfs_rq = cfs_rq_of(se);
4239
		enqueue_entity(cfs_rq, se, flags);
4240 4241 4242 4243 4244 4245 4246 4247 4248

		/*
		 * 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;
4249
		cfs_rq->h_nr_running++;
4250

4251
		flags = ENQUEUE_WAKEUP;
4252
	}
P
Peter Zijlstra 已提交
4253

P
Peter Zijlstra 已提交
4254
	for_each_sched_entity(se) {
4255
		cfs_rq = cfs_rq_of(se);
4256
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
4257

4258 4259 4260
		if (cfs_rq_throttled(cfs_rq))
			break;

4261
		update_cfs_shares(cfs_rq);
4262
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
4263 4264
	}

4265 4266
	if (!se) {
		update_rq_runnable_avg(rq, rq->nr_running);
4267
		add_nr_running(rq, 1);
4268
	}
4269
	hrtick_update(rq);
4270 4271
}

4272 4273
static void set_next_buddy(struct sched_entity *se);

4274 4275 4276 4277 4278
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
4279
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4280 4281
{
	struct cfs_rq *cfs_rq;
4282
	struct sched_entity *se = &p->se;
4283
	int task_sleep = flags & DEQUEUE_SLEEP;
4284 4285 4286

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4287
		dequeue_entity(cfs_rq, se, flags);
4288 4289 4290 4291 4292 4293 4294 4295 4296

		/*
		 * 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;
4297
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4298

4299
		/* Don't dequeue parent if it has other entities besides us */
4300 4301 4302 4303 4304 4305 4306
		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));
4307 4308 4309

			/* avoid re-evaluating load for this entity */
			se = parent_entity(se);
4310
			break;
4311
		}
4312
		flags |= DEQUEUE_SLEEP;
4313
	}
P
Peter Zijlstra 已提交
4314

P
Peter Zijlstra 已提交
4315
	for_each_sched_entity(se) {
4316
		cfs_rq = cfs_rq_of(se);
4317
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4318

4319 4320 4321
		if (cfs_rq_throttled(cfs_rq))
			break;

4322
		update_cfs_shares(cfs_rq);
4323
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
4324 4325
	}

4326
	if (!se) {
4327
		sub_nr_running(rq, 1);
4328 4329
		update_rq_runnable_avg(rq, 1);
	}
4330
	hrtick_update(rq);
4331 4332
}

4333
#ifdef CONFIG_SMP
4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461

/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
 * The exact cpuload at various idx values, calculated at every tick would be
 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
 *
 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
 * on nth tick when cpu may be busy, then we have:
 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
 *
 * decay_load_missed() below does efficient calculation of
 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
 *
 * The calculation is approximated on a 128 point scale.
 * degrade_zero_ticks is the number of ticks after which load at any
 * particular idx is approximated to be zero.
 * degrade_factor is a precomputed table, a row for each load idx.
 * Each column corresponds to degradation factor for a power of two ticks,
 * based on 128 point scale.
 * Example:
 * row 2, col 3 (=12) says that the degradation at load idx 2 after
 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
 *
 * With this power of 2 load factors, we can degrade the load n times
 * by looking at 1 bits in n and doing as many mult/shift instead of
 * n mult/shifts needed by the exact degradation.
 */
#define DEGRADE_SHIFT		7
static const unsigned char
		degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
static const unsigned char
		degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
					{0, 0, 0, 0, 0, 0, 0, 0},
					{64, 32, 8, 0, 0, 0, 0, 0},
					{96, 72, 40, 12, 1, 0, 0},
					{112, 98, 75, 43, 15, 1, 0},
					{120, 112, 98, 76, 45, 16, 2} };

/*
 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
 * would be when CPU is idle and so we just decay the old load without
 * adding any new load.
 */
static unsigned long
decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
{
	int j = 0;

	if (!missed_updates)
		return load;

	if (missed_updates >= degrade_zero_ticks[idx])
		return 0;

	if (idx == 1)
		return load >> missed_updates;

	while (missed_updates) {
		if (missed_updates % 2)
			load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;

		missed_updates >>= 1;
		j++;
	}
	return load;
}

/*
 * Update rq->cpu_load[] statistics. This function is usually called every
 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
 * every tick. We fix it up based on jiffies.
 */
static void __update_cpu_load(struct rq *this_rq, unsigned long this_load,
			      unsigned long pending_updates)
{
	int i, scale;

	this_rq->nr_load_updates++;

	/* Update our load: */
	this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
	for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
		unsigned long old_load, new_load;

		/* scale is effectively 1 << i now, and >> i divides by scale */

		old_load = this_rq->cpu_load[i];
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
		new_load = this_load;
		/*
		 * Round up the averaging division if load is increasing. This
		 * prevents us from getting stuck on 9 if the load is 10, for
		 * example.
		 */
		if (new_load > old_load)
			new_load += scale - 1;

		this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
	}

	sched_avg_update(this_rq);
}

#ifdef CONFIG_NO_HZ_COMMON
/*
 * There is no sane way to deal with nohz on smp when using jiffies because the
 * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading
 * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}.
 *
 * Therefore we cannot use the delta approach from the regular tick since that
 * would seriously skew the load calculation. However we'll make do for those
 * updates happening while idle (nohz_idle_balance) or coming out of idle
 * (tick_nohz_idle_exit).
 *
 * This means we might still be one tick off for nohz periods.
 */

/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
static void update_idle_cpu_load(struct rq *this_rq)
{
4462
	unsigned long curr_jiffies = READ_ONCE(jiffies);
4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483
	unsigned long load = this_rq->cfs.runnable_load_avg;
	unsigned long pending_updates;

	/*
	 * bail if there's load or we're actually up-to-date.
	 */
	if (load || curr_jiffies == this_rq->last_load_update_tick)
		return;

	pending_updates = curr_jiffies - this_rq->last_load_update_tick;
	this_rq->last_load_update_tick = curr_jiffies;

	__update_cpu_load(this_rq, load, pending_updates);
}

/*
 * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed.
 */
void update_cpu_load_nohz(void)
{
	struct rq *this_rq = this_rq();
4484
	unsigned long curr_jiffies = READ_ONCE(jiffies);
4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516
	unsigned long pending_updates;

	if (curr_jiffies == this_rq->last_load_update_tick)
		return;

	raw_spin_lock(&this_rq->lock);
	pending_updates = curr_jiffies - this_rq->last_load_update_tick;
	if (pending_updates) {
		this_rq->last_load_update_tick = curr_jiffies;
		/*
		 * We were idle, this means load 0, the current load might be
		 * !0 due to remote wakeups and the sort.
		 */
		__update_cpu_load(this_rq, 0, pending_updates);
	}
	raw_spin_unlock(&this_rq->lock);
}
#endif /* CONFIG_NO_HZ */

/*
 * Called from scheduler_tick()
 */
void update_cpu_load_active(struct rq *this_rq)
{
	unsigned long load = this_rq->cfs.runnable_load_avg;
	/*
	 * See the mess around update_idle_cpu_load() / update_cpu_load_nohz().
	 */
	this_rq->last_load_update_tick = jiffies;
	__update_cpu_load(this_rq, load, 1);
}

4517 4518 4519
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
4520
	return cpu_rq(cpu)->cfs.runnable_load_avg;
4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555
}

/*
 * 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);
}

4556
static unsigned long capacity_of(int cpu)
4557
{
4558
	return cpu_rq(cpu)->cpu_capacity;
4559 4560
}

4561 4562 4563 4564 4565
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

4566 4567 4568
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
4569
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
4570
	unsigned long load_avg = rq->cfs.runnable_load_avg;
4571 4572

	if (nr_running)
4573
		return load_avg / nr_running;
4574 4575 4576 4577

	return 0;
}

4578 4579 4580 4581 4582 4583 4584
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.
	 */
4585
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
4586
		current->wakee_flips >>= 1;
4587 4588 4589 4590 4591 4592 4593 4594
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}
4595

4596
static void task_waking_fair(struct task_struct *p)
4597 4598 4599
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
4600 4601 4602 4603
	u64 min_vruntime;

#ifndef CONFIG_64BIT
	u64 min_vruntime_copy;
4604

4605 4606 4607 4608 4609 4610 4611 4612
	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
4613

4614
	se->vruntime -= min_vruntime;
4615
	record_wakee(p);
4616 4617
}

4618
#ifdef CONFIG_FAIR_GROUP_SCHED
4619 4620 4621 4622 4623 4624
/*
 * 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.
4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667
 *
 * 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.
4668
 */
P
Peter Zijlstra 已提交
4669
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
4670
{
P
Peter Zijlstra 已提交
4671
	struct sched_entity *se = tg->se[cpu];
4672

4673
	if (!tg->parent)	/* the trivial, non-cgroup case */
4674 4675
		return wl;

P
Peter Zijlstra 已提交
4676
	for_each_sched_entity(se) {
4677
		long w, W;
P
Peter Zijlstra 已提交
4678

4679
		tg = se->my_q->tg;
4680

4681 4682 4683 4684
		/*
		 * W = @wg + \Sum rw_j
		 */
		W = wg + calc_tg_weight(tg, se->my_q);
P
Peter Zijlstra 已提交
4685

4686 4687 4688 4689
		/*
		 * w = rw_i + @wl
		 */
		w = se->my_q->load.weight + wl;
4690

4691 4692 4693 4694
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
4695
			wl = (w * (long)tg->shares) / W;
4696 4697
		else
			wl = tg->shares;
4698

4699 4700 4701 4702 4703
		/*
		 * 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().
		 */
4704 4705
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
4706 4707 4708 4709

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
4710
		wl -= se->load.weight;
4711 4712 4713 4714 4715 4716 4717 4718

		/*
		 * 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
4719 4720
		wg = 0;
	}
4721

P
Peter Zijlstra 已提交
4722
	return wl;
4723 4724
}
#else
P
Peter Zijlstra 已提交
4725

4726
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
4727
{
4728
	return wl;
4729
}
P
Peter Zijlstra 已提交
4730

4731 4732
#endif

4733 4734
static int wake_wide(struct task_struct *p)
{
4735
	int factor = this_cpu_read(sd_llc_size);
4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754

	/*
	 * 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;
}

4755
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
4756
{
4757
	s64 this_load, load;
4758
	s64 this_eff_load, prev_eff_load;
4759 4760
	int idx, this_cpu, prev_cpu;
	struct task_group *tg;
4761
	unsigned long weight;
4762
	int balanced;
4763

4764 4765 4766 4767 4768 4769 4770
	/*
	 * If we wake multiple tasks be careful to not bounce
	 * ourselves around too much.
	 */
	if (wake_wide(p))
		return 0;

4771 4772 4773 4774 4775
	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);
4776

4777 4778 4779 4780 4781
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
4782 4783 4784 4785
	if (sync) {
		tg = task_group(current);
		weight = current->se.load.weight;

4786
		this_load += effective_load(tg, this_cpu, -weight, -weight);
4787 4788
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
4789

4790 4791
	tg = task_group(p);
	weight = p->se.load.weight;
4792

4793 4794
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
4795 4796 4797
	 * 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.
4798 4799 4800 4801
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
4802 4803
	this_eff_load = 100;
	this_eff_load *= capacity_of(prev_cpu);
4804

4805 4806
	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
4807

4808
	if (this_load > 0) {
4809 4810 4811 4812
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
4813
	}
4814

4815
	balanced = this_eff_load <= prev_eff_load;
4816

4817
	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
4818

4819 4820
	if (!balanced)
		return 0;
4821

4822 4823 4824 4825
	schedstat_inc(sd, ttwu_move_affine);
	schedstat_inc(p, se.statistics.nr_wakeups_affine);

	return 1;
4826 4827
}

4828 4829 4830 4831 4832
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
4833
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
4834
		  int this_cpu, int sd_flag)
4835
{
4836
	struct sched_group *idlest = NULL, *group = sd->groups;
4837
	unsigned long min_load = ULONG_MAX, this_load = 0;
4838
	int load_idx = sd->forkexec_idx;
4839
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
4840

4841 4842 4843
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

4844 4845 4846 4847
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
4848

4849 4850
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
4851
					tsk_cpus_allowed(p)))
4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869
			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;
		}

4870
		/* Adjust by relative CPU capacity of the group */
4871
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892

		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;
4893 4894 4895 4896
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
4897 4898 4899
	int i;

	/* Traverse only the allowed CPUs */
4900
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922
		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;
			}
4923
		} else if (shallowest_idle_cpu == -1) {
4924 4925 4926 4927 4928
			load = weighted_cpuload(i);
			if (load < min_load || (load == min_load && i == this_cpu)) {
				min_load = load;
				least_loaded_cpu = i;
			}
4929 4930 4931
		}
	}

4932
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
4933
}
4934

4935 4936 4937
/*
 * Try and locate an idle CPU in the sched_domain.
 */
4938
static int select_idle_sibling(struct task_struct *p, int target)
4939
{
4940
	struct sched_domain *sd;
4941
	struct sched_group *sg;
4942
	int i = task_cpu(p);
4943

4944 4945
	if (idle_cpu(target))
		return target;
4946 4947

	/*
4948
	 * If the prevous cpu is cache affine and idle, don't be stupid.
4949
	 */
4950 4951
	if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
		return i;
4952 4953

	/*
4954
	 * Otherwise, iterate the domains and find an elegible idle cpu.
4955
	 */
4956
	sd = rcu_dereference(per_cpu(sd_llc, target));
4957
	for_each_lower_domain(sd) {
4958 4959 4960 4961 4962 4963 4964
		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)) {
4965
				if (i == target || !idle_cpu(i))
4966 4967
					goto next;
			}
4968

4969 4970 4971 4972 4973 4974 4975 4976
			target = cpumask_first_and(sched_group_cpus(sg),
					tsk_cpus_allowed(p));
			goto done;
next:
			sg = sg->next;
		} while (sg != sd->groups);
	}
done:
4977 4978
	return target;
}
4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005
/*
 * get_cpu_usage returns the amount of capacity of a CPU that is used by CFS
 * tasks. The unit of the return value must be the one of capacity so we can
 * compare the usage with the capacity of the CPU that is available for CFS
 * task (ie cpu_capacity).
 * cfs.utilization_load_avg is the sum of running time of runnable tasks on a
 * CPU. It represents the amount of utilization of a CPU in the range
 * [0..SCHED_LOAD_SCALE].  The usage of a CPU can't be higher than the full
 * capacity of the CPU because it's about the running time on this CPU.
 * Nevertheless, cfs.utilization_load_avg can be higher than SCHED_LOAD_SCALE
 * because of unfortunate rounding in avg_period and running_load_avg or just
 * after migrating tasks until the average stabilizes with the new running
 * time. So we need to check that the usage stays into the range
 * [0..cpu_capacity_orig] and cap if necessary.
 * Without capping the usage, a group could be seen as overloaded (CPU0 usage
 * at 121% + CPU1 usage at 80%) whereas CPU1 has 20% of available capacity
 */
static int get_cpu_usage(int cpu)
{
	unsigned long usage = cpu_rq(cpu)->cfs.utilization_load_avg;
	unsigned long capacity = capacity_orig_of(cpu);

	if (usage >= SCHED_LOAD_SCALE)
		return capacity;

	return (usage * capacity) >> SCHED_LOAD_SHIFT;
}
5006

5007
/*
5008 5009 5010
 * 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.
5011
 *
5012 5013
 * 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.
5014
 *
5015
 * Returns the target cpu number.
5016 5017 5018
 *
 * preempt must be disabled.
 */
5019
static int
5020
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
5021
{
5022
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
5023 5024
	int cpu = smp_processor_id();
	int new_cpu = cpu;
5025
	int want_affine = 0;
5026
	int sync = wake_flags & WF_SYNC;
5027

5028 5029
	if (sd_flag & SD_BALANCE_WAKE)
		want_affine = cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
5030

5031
	rcu_read_lock();
5032
	for_each_domain(cpu, tmp) {
5033 5034 5035
		if (!(tmp->flags & SD_LOAD_BALANCE))
			continue;

5036
		/*
5037 5038
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
5039
		 */
5040 5041 5042
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
5043
			break;
5044
		}
5045

5046
		if (tmp->flags & sd_flag)
5047 5048 5049
			sd = tmp;
	}

5050 5051
	if (affine_sd && cpu != prev_cpu && wake_affine(affine_sd, p, sync))
		prev_cpu = cpu;
5052

5053
	if (sd_flag & SD_BALANCE_WAKE) {
5054 5055
		new_cpu = select_idle_sibling(p, prev_cpu);
		goto unlock;
5056
	}
5057

5058 5059
	while (sd) {
		struct sched_group *group;
5060
		int weight;
5061

5062
		if (!(sd->flags & sd_flag)) {
5063 5064 5065
			sd = sd->child;
			continue;
		}
5066

5067
		group = find_idlest_group(sd, p, cpu, sd_flag);
5068 5069 5070 5071
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
5072

5073
		new_cpu = find_idlest_cpu(group, p, cpu);
5074 5075 5076 5077
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
5078
		}
5079 5080 5081

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
5082
		weight = sd->span_weight;
5083 5084
		sd = NULL;
		for_each_domain(cpu, tmp) {
5085
			if (weight <= tmp->span_weight)
5086
				break;
5087
			if (tmp->flags & sd_flag)
5088 5089 5090
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
5091
	}
5092 5093
unlock:
	rcu_read_unlock();
5094

5095
	return new_cpu;
5096
}
5097 5098 5099 5100 5101 5102 5103 5104 5105 5106

/*
 * 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)
{
5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117
	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);
5118 5119
		atomic_long_add(se->avg.load_avg_contrib,
						&cfs_rq->removed_load);
5120
	}
5121 5122 5123

	/* We have migrated, no longer consider this task hot */
	se->exec_start = 0;
5124
}
5125 5126
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
5127 5128
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
5129 5130 5131 5132
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
5133 5134
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
5135 5136 5137 5138 5139 5140 5141 5142 5143
	 *
	 * 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.
5144
	 */
5145
	return calc_delta_fair(gran, se);
5146 5147
}

5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169
/*
 * 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
5170
	gran = wakeup_gran(curr, se);
5171 5172 5173 5174 5175 5176
	if (vdiff > gran)
		return 1;

	return 0;
}

5177 5178
static void set_last_buddy(struct sched_entity *se)
{
5179 5180 5181 5182 5183
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
5184 5185 5186 5187
}

static void set_next_buddy(struct sched_entity *se)
{
5188 5189 5190 5191 5192
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
5193 5194
}

5195 5196
static void set_skip_buddy(struct sched_entity *se)
{
5197 5198
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
5199 5200
}

5201 5202 5203
/*
 * Preempt the current task with a newly woken task if needed:
 */
5204
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
5205 5206
{
	struct task_struct *curr = rq->curr;
5207
	struct sched_entity *se = &curr->se, *pse = &p->se;
5208
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
5209
	int scale = cfs_rq->nr_running >= sched_nr_latency;
5210
	int next_buddy_marked = 0;
5211

I
Ingo Molnar 已提交
5212 5213 5214
	if (unlikely(se == pse))
		return;

5215
	/*
5216
	 * This is possible from callers such as attach_tasks(), in which we
5217 5218 5219 5220 5221 5222 5223
	 * 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;

5224
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
5225
		set_next_buddy(pse);
5226 5227
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
5228

5229 5230 5231
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
5232 5233 5234 5235 5236 5237
	 *
	 * 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.
5238 5239 5240 5241
	 */
	if (test_tsk_need_resched(curr))
		return;

5242 5243 5244 5245 5246
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

5247
	/*
5248 5249
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
5250
	 */
5251
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
5252
		return;
5253

5254
	find_matching_se(&se, &pse);
5255
	update_curr(cfs_rq_of(se));
5256
	BUG_ON(!pse);
5257 5258 5259 5260 5261 5262 5263
	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);
5264
		goto preempt;
5265
	}
5266

5267
	return;
5268

5269
preempt:
5270
	resched_curr(rq);
5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284
	/*
	 * 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);
5285 5286
}

5287 5288
static struct task_struct *
pick_next_task_fair(struct rq *rq, struct task_struct *prev)
5289 5290 5291
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
5292
	struct task_struct *p;
5293
	int new_tasks;
5294

5295
again:
5296 5297
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
5298
		goto idle;
5299

5300
	if (prev->sched_class != &fair_sched_class)
5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319
		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.
		 */
5320 5321 5322 5323 5324
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
5325

5326 5327 5328 5329 5330 5331 5332 5333 5334
			/*
			 * 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;
		}
5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374

		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
5375

5376
	if (!cfs_rq->nr_running)
5377
		goto idle;
5378

5379
	put_prev_task(rq, prev);
5380

5381
	do {
5382
		se = pick_next_entity(cfs_rq, NULL);
5383
		set_next_entity(cfs_rq, se);
5384 5385 5386
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
5387
	p = task_of(se);
5388

5389 5390
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
5391 5392

	return p;
5393 5394

idle:
5395 5396 5397 5398 5399 5400 5401
	/*
	 * This is OK, because current is on_cpu, which avoids it being picked
	 * for load-balance and preemption/IRQs are still disabled avoiding
	 * further scheduler activity on it and we're being very careful to
	 * re-start the picking loop.
	 */
	lockdep_unpin_lock(&rq->lock);
5402
	new_tasks = idle_balance(rq);
5403
	lockdep_pin_lock(&rq->lock);
5404 5405 5406 5407 5408
	/*
	 * 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.
	 */
5409
	if (new_tasks < 0)
5410 5411
		return RETRY_TASK;

5412
	if (new_tasks > 0)
5413 5414 5415
		goto again;

	return NULL;
5416 5417 5418 5419 5420
}

/*
 * Account for a descheduled task:
 */
5421
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
5422 5423 5424 5425 5426 5427
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5428
		put_prev_entity(cfs_rq, se);
5429 5430 5431
	}
}

5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456
/*
 * 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);
5457 5458 5459 5460 5461
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
5462
		rq_clock_skip_update(rq, true);
5463 5464 5465 5466 5467
	}

	set_skip_buddy(se);
}

5468 5469 5470 5471
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

5472 5473
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
5474 5475 5476 5477 5478 5479 5480 5481 5482 5483
		return false;

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

	yield_task_fair(rq);

	return true;
}

5484
#ifdef CONFIG_SMP
5485
/**************************************************
P
Peter Zijlstra 已提交
5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508
 * 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)
 *
5509
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
5510 5511 5512 5513 5514 5515
 * 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):
 *
5516
 *   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
5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601
 *
 * 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.]
 */ 
5602

5603 5604
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

5605 5606
enum fbq_type { regular, remote, all };

5607
#define LBF_ALL_PINNED	0x01
5608
#define LBF_NEED_BREAK	0x02
5609 5610
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
5611 5612 5613 5614 5615

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
5616
	int			src_cpu;
5617 5618 5619 5620

	int			dst_cpu;
	struct rq		*dst_rq;

5621 5622
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
5623
	enum cpu_idle_type	idle;
5624
	long			imbalance;
5625 5626 5627
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

5628
	unsigned int		flags;
5629 5630 5631 5632

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
5633 5634

	enum fbq_type		fbq_type;
5635
	struct list_head	tasks;
5636 5637
};

5638 5639 5640
/*
 * Is this task likely cache-hot:
 */
5641
static int task_hot(struct task_struct *p, struct lb_env *env)
5642 5643 5644
{
	s64 delta;

5645 5646
	lockdep_assert_held(&env->src_rq->lock);

5647 5648 5649 5650 5651 5652 5653 5654 5655
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
5656
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
5657 5658 5659 5660 5661 5662 5663 5664 5665
			(&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;

5666
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
5667 5668 5669 5670

	return delta < (s64)sysctl_sched_migration_cost;
}

5671
#ifdef CONFIG_NUMA_BALANCING
5672 5673 5674 5675 5676
/*
 * Returns true if the destination node is the preferred node.
 * Needs to match fbq_classify_rq(): if there is a runnable task
 * that is not on its preferred node, we should identify it.
 */
5677 5678
static bool migrate_improves_locality(struct task_struct *p, struct lb_env *env)
{
5679
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5680
	unsigned long src_faults, dst_faults;
5681 5682
	int src_nid, dst_nid;

5683
	if (!sched_feat(NUMA_FAVOUR_HIGHER) || !p->numa_faults ||
5684 5685 5686 5687 5688 5689 5690
	    !(env->sd->flags & SD_NUMA)) {
		return false;
	}

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

5691
	if (src_nid == dst_nid)
5692 5693
		return false;

5694 5695
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
5696 5697
		return true;

5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710
	/* Migrating away from the preferred node is bad. */
	if (src_nid == p->numa_preferred_nid)
		return false;

	if (numa_group) {
		src_faults = group_faults(p, src_nid);
		dst_faults = group_faults(p, dst_nid);
	} else {
		src_faults = task_faults(p, src_nid);
		dst_faults = task_faults(p, dst_nid);
	}

	return dst_faults > src_faults;
5711
}
5712 5713 5714 5715


static bool migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
{
5716
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5717
	unsigned long src_faults, dst_faults;
5718 5719 5720 5721 5722
	int src_nid, dst_nid;

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

5723
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
5724 5725 5726 5727 5728
		return false;

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

5729
	if (src_nid == dst_nid)
5730 5731
		return false;

5732 5733 5734
	/* Migrating away from the preferred node is bad. */
	if (src_nid == p->numa_preferred_nid)
		return true;
5735

5736 5737 5738
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
		return false;
5739

5740 5741 5742 5743 5744 5745
	if (numa_group) {
		src_faults = group_faults(p, src_nid);
		dst_faults = group_faults(p, dst_nid);
	} else {
		src_faults = task_faults(p, src_nid);
		dst_faults = task_faults(p, dst_nid);
5746 5747
	}

5748
	return dst_faults < src_faults;
5749 5750
}

5751 5752 5753 5754 5755 5756
#else
static inline bool migrate_improves_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5757 5758 5759 5760 5761 5762

static inline bool migrate_degrades_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
5763 5764
#endif

5765 5766 5767 5768
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
5769
int can_migrate_task(struct task_struct *p, struct lb_env *env)
5770 5771
{
	int tsk_cache_hot = 0;
5772 5773 5774

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

5775 5776
	/*
	 * We do not migrate tasks that are:
5777
	 * 1) throttled_lb_pair, or
5778
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
5779 5780
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
5781
	 */
5782 5783 5784
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

5785
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
5786
		int cpu;
5787

5788
		schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
5789

5790 5791
		env->flags |= LBF_SOME_PINNED;

5792 5793 5794 5795 5796 5797 5798 5799
		/*
		 * 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.
		 */
5800
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
5801 5802
			return 0;

5803 5804 5805
		/* 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))) {
5806
				env->flags |= LBF_DST_PINNED;
5807 5808 5809
				env->new_dst_cpu = cpu;
				break;
			}
5810
		}
5811

5812 5813
		return 0;
	}
5814 5815

	/* Record that we found atleast one task that could run on dst_cpu */
5816
	env->flags &= ~LBF_ALL_PINNED;
5817

5818
	if (task_running(env->src_rq, p)) {
5819
		schedstat_inc(p, se.statistics.nr_failed_migrations_running);
5820 5821 5822 5823 5824
		return 0;
	}

	/*
	 * Aggressive migration if:
5825 5826 5827
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
5828
	 */
5829
	tsk_cache_hot = task_hot(p, env);
5830 5831
	if (!tsk_cache_hot)
		tsk_cache_hot = migrate_degrades_locality(p, env);
5832

5833 5834
	if (migrate_improves_locality(p, env) || !tsk_cache_hot ||
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
5835 5836 5837 5838
		if (tsk_cache_hot) {
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
			schedstat_inc(p, se.statistics.nr_forced_migrations);
		}
5839 5840 5841
		return 1;
	}

Z
Zhang Hang 已提交
5842 5843
	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
	return 0;
5844 5845
}

5846
/*
5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857
 * 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);
}

5858
/*
5859
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
5860 5861
 * part of active balancing operations within "domain".
 *
5862
 * Returns a task if successful and NULL otherwise.
5863
 */
5864
static struct task_struct *detach_one_task(struct lb_env *env)
5865 5866 5867
{
	struct task_struct *p, *n;

5868 5869
	lockdep_assert_held(&env->src_rq->lock);

5870 5871 5872
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
5873

5874
		detach_task(p, env);
5875

5876
		/*
5877
		 * Right now, this is only the second place where
5878
		 * lb_gained[env->idle] is updated (other is detach_tasks)
5879
		 * so we can safely collect stats here rather than
5880
		 * inside detach_tasks().
5881 5882
		 */
		schedstat_inc(env->sd, lb_gained[env->idle]);
5883
		return p;
5884
	}
5885
	return NULL;
5886 5887
}

5888 5889
static const unsigned int sched_nr_migrate_break = 32;

5890
/*
5891 5892
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
5893
 *
5894
 * Returns number of detached tasks if successful and 0 otherwise.
5895
 */
5896
static int detach_tasks(struct lb_env *env)
5897
{
5898 5899
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
5900
	unsigned long load;
5901 5902 5903
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
5904

5905
	if (env->imbalance <= 0)
5906
		return 0;
5907

5908 5909
	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
5910

5911 5912
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
5913
		if (env->loop > env->loop_max)
5914
			break;
5915 5916

		/* take a breather every nr_migrate tasks */
5917
		if (env->loop > env->loop_break) {
5918
			env->loop_break += sched_nr_migrate_break;
5919
			env->flags |= LBF_NEED_BREAK;
5920
			break;
5921
		}
5922

5923
		if (!can_migrate_task(p, env))
5924 5925 5926
			goto next;

		load = task_h_load(p);
5927

5928
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
5929 5930
			goto next;

5931
		if ((load / 2) > env->imbalance)
5932
			goto next;
5933

5934 5935 5936 5937
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
5938
		env->imbalance -= load;
5939 5940

#ifdef CONFIG_PREEMPT
5941 5942
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
5943
		 * kernels will stop after the first task is detached to minimize
5944 5945
		 * the critical section.
		 */
5946
		if (env->idle == CPU_NEWLY_IDLE)
5947
			break;
5948 5949
#endif

5950 5951 5952 5953
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
5954
		if (env->imbalance <= 0)
5955
			break;
5956 5957 5958

		continue;
next:
5959
		list_move_tail(&p->se.group_node, tasks);
5960
	}
5961

5962
	/*
5963 5964 5965
	 * 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().
5966
	 */
5967
	schedstat_add(env->sd, lb_gained[env->idle], detached);
5968

5969 5970 5971 5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009
	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);
6010

6011 6012 6013 6014
		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
6015 6016
}

P
Peter Zijlstra 已提交
6017
#ifdef CONFIG_FAIR_GROUP_SCHED
6018 6019 6020
/*
 * update tg->load_weight by folding this cpu's load_avg
 */
6021
static void __update_blocked_averages_cpu(struct task_group *tg, int cpu)
6022
{
6023 6024
	struct sched_entity *se = tg->se[cpu];
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
6025

6026 6027 6028
	/* throttled entities do not contribute to load */
	if (throttled_hierarchy(cfs_rq))
		return;
6029

6030
	update_cfs_rq_blocked_load(cfs_rq, 1);
6031

6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045
	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 {
6046
		struct rq *rq = rq_of(cfs_rq);
6047 6048
		update_rq_runnable_avg(rq, rq->nr_running);
	}
6049 6050
}

6051
static void update_blocked_averages(int cpu)
6052 6053
{
	struct rq *rq = cpu_rq(cpu);
6054 6055
	struct cfs_rq *cfs_rq;
	unsigned long flags;
6056

6057 6058
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
6059 6060 6061 6062
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
6063
	for_each_leaf_cfs_rq(rq, cfs_rq) {
6064 6065 6066 6067 6068 6069
		/*
		 * 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);
6070
	}
6071 6072

	raw_spin_unlock_irqrestore(&rq->lock, flags);
6073 6074
}

6075
/*
6076
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
6077 6078 6079
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
6080
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
6081
{
6082 6083
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
6084
	unsigned long now = jiffies;
6085
	unsigned long load;
6086

6087
	if (cfs_rq->last_h_load_update == now)
6088 6089
		return;

6090 6091 6092 6093 6094 6095 6096
	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;
	}
6097

6098
	if (!se) {
6099
		cfs_rq->h_load = cfs_rq->runnable_load_avg;
6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110
		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;
	}
6111 6112
}

6113
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
6114
{
6115
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
6116

6117
	update_cfs_rq_h_load(cfs_rq);
6118 6119
	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
6120 6121
}
#else
6122
static inline void update_blocked_averages(int cpu)
6123 6124 6125
{
}

6126
static unsigned long task_h_load(struct task_struct *p)
6127
{
6128
	return p->se.avg.load_avg_contrib;
6129
}
P
Peter Zijlstra 已提交
6130
#endif
6131 6132

/********** Helpers for find_busiest_group ************************/
6133 6134 6135 6136 6137 6138 6139

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

6140 6141 6142 6143 6144 6145 6146
/*
 * 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
6147
	unsigned long load_per_task;
6148
	unsigned long group_capacity;
6149
	unsigned long group_usage; /* Total usage of the group */
6150 6151 6152
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
6153
	enum group_type group_type;
6154
	int group_no_capacity;
6155 6156 6157 6158
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
6159 6160
};

J
Joonsoo Kim 已提交
6161 6162 6163 6164 6165 6166 6167 6168
/*
 * 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 */
6169
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
6170 6171 6172
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
6173
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
6174 6175
};

6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187
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,
6188
		.total_capacity = 0UL,
6189 6190
		.busiest_stat = {
			.avg_load = 0UL,
6191 6192
			.sum_nr_running = 0,
			.group_type = group_other,
6193 6194 6195 6196
		},
	};
}

6197 6198 6199
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
6200
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
6201 6202
 *
 * Return: The load index.
6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224
 */
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;
}

6225
static unsigned long default_scale_cpu_capacity(struct sched_domain *sd, int cpu)
6226
{
6227 6228
	if ((sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
		return sd->smt_gain / sd->span_weight;
6229

6230
	return SCHED_CAPACITY_SCALE;
6231 6232
}

6233
unsigned long __weak arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
6234
{
6235
	return default_scale_cpu_capacity(sd, cpu);
6236 6237
}

6238
static unsigned long scale_rt_capacity(int cpu)
6239 6240
{
	struct rq *rq = cpu_rq(cpu);
6241
	u64 total, used, age_stamp, avg;
6242
	s64 delta;
6243

6244 6245 6246 6247
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
6248 6249
	age_stamp = READ_ONCE(rq->age_stamp);
	avg = READ_ONCE(rq->rt_avg);
6250
	delta = __rq_clock_broken(rq) - age_stamp;
6251

6252 6253 6254 6255
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
6256

6257
	used = div_u64(avg, total);
6258

6259 6260
	if (likely(used < SCHED_CAPACITY_SCALE))
		return SCHED_CAPACITY_SCALE - used;
6261

6262
	return 1;
6263 6264
}

6265
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
6266
{
6267
	unsigned long capacity = SCHED_CAPACITY_SCALE;
6268 6269
	struct sched_group *sdg = sd->groups;

6270 6271 6272 6273
	if (sched_feat(ARCH_CAPACITY))
		capacity *= arch_scale_cpu_capacity(sd, cpu);
	else
		capacity *= default_scale_cpu_capacity(sd, cpu);
6274

6275
	capacity >>= SCHED_CAPACITY_SHIFT;
6276

6277
	cpu_rq(cpu)->cpu_capacity_orig = capacity;
6278

6279
	capacity *= scale_rt_capacity(cpu);
6280
	capacity >>= SCHED_CAPACITY_SHIFT;
6281

6282 6283
	if (!capacity)
		capacity = 1;
6284

6285 6286
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
6287 6288
}

6289
void update_group_capacity(struct sched_domain *sd, int cpu)
6290 6291 6292
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
6293
	unsigned long capacity;
6294 6295 6296 6297
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
6298
	sdg->sgc->next_update = jiffies + interval;
6299 6300

	if (!child) {
6301
		update_cpu_capacity(sd, cpu);
6302 6303 6304
		return;
	}

6305
	capacity = 0;
6306

P
Peter Zijlstra 已提交
6307 6308 6309 6310 6311 6312
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

6313
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
6314
			struct sched_group_capacity *sgc;
6315
			struct rq *rq = cpu_rq(cpu);
6316

6317
			/*
6318
			 * build_sched_domains() -> init_sched_groups_capacity()
6319 6320 6321
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
6322 6323
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
6324
			 *
6325
			 * This avoids capacity from being 0 and
6326 6327 6328
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
6329
				capacity += capacity_of(cpu);
6330 6331
				continue;
			}
6332

6333 6334
			sgc = rq->sd->groups->sgc;
			capacity += sgc->capacity;
6335
		}
P
Peter Zijlstra 已提交
6336 6337 6338 6339 6340 6341 6342 6343
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
		 */ 

		group = child->groups;
		do {
6344
			capacity += group->sgc->capacity;
P
Peter Zijlstra 已提交
6345 6346 6347
			group = group->next;
		} while (group != child->groups);
	}
6348

6349
	sdg->sgc->capacity = capacity;
6350 6351
}

6352
/*
6353 6354 6355
 * Check whether the capacity of the rq has been noticeably reduced by side
 * activity. The imbalance_pct is used for the threshold.
 * Return true is the capacity is reduced
6356 6357
 */
static inline int
6358
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
6359
{
6360 6361
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
6362 6363
}

6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379
/*
 * 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
6380 6381
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
6382 6383
 *
 * When this is so detected; this group becomes a candidate for busiest; see
6384
 * update_sd_pick_busiest(). And calculate_imbalance() and
6385
 * find_busiest_group() avoid some of the usual balance conditions to allow it
6386 6387 6388 6389 6390 6391 6392
 * 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.
 */

6393
static inline int sg_imbalanced(struct sched_group *group)
6394
{
6395
	return group->sgc->imbalance;
6396 6397
}

6398
/*
6399 6400 6401 6402 6403 6404 6405 6406 6407 6408
 * group_has_capacity returns true if the group has spare capacity that could
 * be used by some tasks.
 * We consider that a group has spare capacity if the  * number of task is
 * smaller than the number of CPUs or if the usage is lower than the available
 * capacity for CFS tasks.
 * For the latter, we use a threshold to stabilize the state, to take into
 * account the variance of the tasks' load and to return true if the available
 * capacity in meaningful for the load balancer.
 * As an example, an available capacity of 1% can appear but it doesn't make
 * any benefit for the load balance.
6409
 */
6410 6411
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
6412
{
6413 6414
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
6415

6416 6417 6418
	if ((sgs->group_capacity * 100) >
			(sgs->group_usage * env->sd->imbalance_pct))
		return true;
6419

6420 6421 6422 6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435
	return false;
}

/*
 *  group_is_overloaded returns true if the group has more tasks than it can
 *  handle.
 *  group_is_overloaded is not equals to !group_has_capacity because a group
 *  with the exact right number of tasks, has no more spare capacity but is not
 *  overloaded so both group_has_capacity and group_is_overloaded return
 *  false.
 */
static inline bool
group_is_overloaded(struct lb_env *env, struct sg_lb_stats *sgs)
{
	if (sgs->sum_nr_running <= sgs->group_weight)
		return false;
6436

6437 6438 6439
	if ((sgs->group_capacity * 100) <
			(sgs->group_usage * env->sd->imbalance_pct))
		return true;
6440

6441
	return false;
6442 6443
}

6444 6445 6446
static enum group_type group_classify(struct lb_env *env,
		struct sched_group *group,
		struct sg_lb_stats *sgs)
6447
{
6448
	if (sgs->group_no_capacity)
6449 6450 6451 6452 6453 6454 6455 6456
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

6457 6458
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
6459
 * @env: The load balancing environment.
6460 6461 6462 6463
 * @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.
6464
 * @overload: Indicate more than one runnable task for any CPU.
6465
 */
6466 6467
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
6468 6469
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
6470
{
6471
	unsigned long load;
6472
	int i;
6473

6474 6475
	memset(sgs, 0, sizeof(*sgs));

6476
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6477 6478 6479
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
6480
		if (local_group)
6481
			load = target_load(i, load_idx);
6482
		else
6483 6484 6485
			load = source_load(i, load_idx);

		sgs->group_load += load;
6486
		sgs->group_usage += get_cpu_usage(i);
6487
		sgs->sum_nr_running += rq->cfs.h_nr_running;
6488 6489 6490 6491

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

6492 6493 6494 6495
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
6496
		sgs->sum_weighted_load += weighted_cpuload(i);
6497 6498
		if (idle_cpu(i))
			sgs->idle_cpus++;
6499 6500
	}

6501 6502
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
6503
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
6504

6505
	if (sgs->sum_nr_running)
6506
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
6507

6508
	sgs->group_weight = group->group_weight;
6509

6510 6511
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
	sgs->group_type = group_classify(env, group, sgs);
6512 6513
}

6514 6515
/**
 * update_sd_pick_busiest - return 1 on busiest group
6516
 * @env: The load balancing environment.
6517 6518
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
6519
 * @sgs: sched_group statistics
6520 6521 6522
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
6523 6524 6525
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
6526
 */
6527
static bool update_sd_pick_busiest(struct lb_env *env,
6528 6529
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
6530
				   struct sg_lb_stats *sgs)
6531
{
6532
	struct sg_lb_stats *busiest = &sds->busiest_stat;
6533

6534
	if (sgs->group_type > busiest->group_type)
6535 6536
		return true;

6537 6538 6539 6540 6541 6542 6543 6544
	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))
6545 6546 6547 6548 6549 6550 6551
		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.
	 */
6552
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
6553 6554 6555 6556 6557 6558 6559 6560 6561 6562
		if (!sds->busiest)
			return true;

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

	return false;
}

6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592
#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 */

6593
/**
6594
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
6595
 * @env: The load balancing environment.
6596 6597
 * @sds: variable to hold the statistics for this sched_domain.
 */
6598
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
6599
{
6600 6601
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
6602
	struct sg_lb_stats tmp_sgs;
6603
	int load_idx, prefer_sibling = 0;
6604
	bool overload = false;
6605 6606 6607 6608

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

6609
	load_idx = get_sd_load_idx(env->sd, env->idle);
6610 6611

	do {
J
Joonsoo Kim 已提交
6612
		struct sg_lb_stats *sgs = &tmp_sgs;
6613 6614
		int local_group;

6615
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
6616 6617 6618
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
6619 6620

			if (env->idle != CPU_NEWLY_IDLE ||
6621 6622
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
6623
		}
6624

6625 6626
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
6627

6628 6629 6630
		if (local_group)
			goto next_group;

6631 6632
		/*
		 * In case the child domain prefers tasks go to siblings
6633
		 * first, lower the sg capacity so that we'll try
6634 6635
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
6636 6637 6638 6639
		 * these excess tasks. The 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).
6640
		 */
6641
		if (prefer_sibling && sds->local &&
6642 6643 6644 6645
		    group_has_capacity(env, &sds->local_stat) &&
		    (sgs->sum_nr_running > 1)) {
			sgs->group_no_capacity = 1;
			sgs->group_type = group_overloaded;
6646
		}
6647

6648
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
6649
			sds->busiest = sg;
J
Joonsoo Kim 已提交
6650
			sds->busiest_stat = *sgs;
6651 6652
		}

6653 6654 6655
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
6656
		sds->total_capacity += sgs->group_capacity;
6657

6658
		sg = sg->next;
6659
	} while (sg != env->sd->groups);
6660 6661 6662

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
6663 6664 6665 6666 6667 6668 6669

	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;
	}

6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688
}

/**
 * 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.
 *
6689
 * Return: 1 when packing is required and a task should be moved to
6690 6691
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
6692
 * @env: The load balancing environment.
6693 6694
 * @sds: Statistics of the sched_domain which is to be packed
 */
6695
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
6696 6697 6698
{
	int busiest_cpu;

6699
	if (!(env->sd->flags & SD_ASYM_PACKING))
6700 6701 6702 6703 6704 6705
		return 0;

	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
6706
	if (env->dst_cpu > busiest_cpu)
6707 6708
		return 0;

6709
	env->imbalance = DIV_ROUND_CLOSEST(
6710
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
6711
		SCHED_CAPACITY_SCALE);
6712

6713
	return 1;
6714 6715 6716 6717 6718 6719
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
6720
 * @env: The load balancing environment.
6721 6722
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
6723 6724
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6725
{
6726
	unsigned long tmp, capa_now = 0, capa_move = 0;
6727
	unsigned int imbn = 2;
6728
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
6729
	struct sg_lb_stats *local, *busiest;
6730

J
Joonsoo Kim 已提交
6731 6732
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6733

J
Joonsoo Kim 已提交
6734 6735 6736 6737
	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;
6738

J
Joonsoo Kim 已提交
6739
	scaled_busy_load_per_task =
6740
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6741
		busiest->group_capacity;
J
Joonsoo Kim 已提交
6742

6743 6744
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
6745
		env->imbalance = busiest->load_per_task;
6746 6747 6748 6749 6750
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
6751
	 * however we may be able to increase total CPU capacity used by
6752 6753 6754
	 * moving them.
	 */

6755
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
6756
			min(busiest->load_per_task, busiest->avg_load);
6757
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
6758
			min(local->load_per_task, local->avg_load);
6759
	capa_now /= SCHED_CAPACITY_SCALE;
6760 6761

	/* Amount of load we'd subtract */
6762
	if (busiest->avg_load > scaled_busy_load_per_task) {
6763
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
6764
			    min(busiest->load_per_task,
6765
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
6766
	}
6767 6768

	/* Amount of load we'd add */
6769
	if (busiest->avg_load * busiest->group_capacity <
6770
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
6771 6772
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
6773
	} else {
6774
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6775
		      local->group_capacity;
J
Joonsoo Kim 已提交
6776
	}
6777
	capa_move += local->group_capacity *
6778
		    min(local->load_per_task, local->avg_load + tmp);
6779
	capa_move /= SCHED_CAPACITY_SCALE;
6780 6781

	/* Move if we gain throughput */
6782
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
6783
		env->imbalance = busiest->load_per_task;
6784 6785 6786 6787 6788
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
6789
 * @env: load balance environment
6790 6791
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
6792
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6793
{
6794
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
6795 6796 6797 6798
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6799

6800
	if (busiest->group_type == group_imbalanced) {
6801 6802 6803 6804
		/*
		 * 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
6805 6806
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
6807 6808
	}

6809 6810 6811
	/*
	 * 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
6812
	 * its cpu_capacity, while calculating max_load..)
6813
	 */
6814 6815
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
6816 6817
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
6818 6819
	}

6820 6821 6822 6823 6824
	/*
	 * If there aren't any idle cpus, avoid creating some.
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
6825 6826 6827 6828 6829 6830
		load_above_capacity = busiest->sum_nr_running *
					SCHED_LOAD_SCALE;
		if (load_above_capacity > busiest->group_capacity)
			load_above_capacity -= busiest->group_capacity;
		else
			load_above_capacity = ~0UL;
6831 6832 6833 6834 6835 6836 6837 6838 6839 6840
	}

	/*
	 * 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.
	 */
6841
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
6842 6843

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
6844
	env->imbalance = min(
6845 6846
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
6847
	) / SCHED_CAPACITY_SCALE;
6848 6849 6850

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
6851
	 * there is no guarantee that any tasks will be moved so we'll have
6852 6853 6854
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
6855
	if (env->imbalance < busiest->load_per_task)
6856
		return fix_small_imbalance(env, sds);
6857
}
6858

6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870
/******* 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.
 *
6871
 * @env: The load balancing environment.
6872
 *
6873
 * Return:	- The busiest group if imbalance exists.
6874 6875 6876 6877
 *		- 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
6878
static struct sched_group *find_busiest_group(struct lb_env *env)
6879
{
J
Joonsoo Kim 已提交
6880
	struct sg_lb_stats *local, *busiest;
6881 6882
	struct sd_lb_stats sds;

6883
	init_sd_lb_stats(&sds);
6884 6885 6886 6887 6888

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
6889
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
6890 6891
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
6892

6893
	/* ASYM feature bypasses nice load balance check */
6894 6895
	if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
	    check_asym_packing(env, &sds))
6896 6897
		return sds.busiest;

6898
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
6899
	if (!sds.busiest || busiest->sum_nr_running == 0)
6900 6901
		goto out_balanced;

6902 6903
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
6904

P
Peter Zijlstra 已提交
6905 6906
	/*
	 * If the busiest group is imbalanced the below checks don't
6907
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
6908 6909
	 * isn't true due to cpus_allowed constraints and the like.
	 */
6910
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
6911 6912
		goto force_balance;

6913
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
6914 6915
	if (env->idle == CPU_NEWLY_IDLE && group_has_capacity(env, local) &&
	    busiest->group_no_capacity)
6916 6917
		goto force_balance;

6918
	/*
6919
	 * If the local group is busier than the selected busiest group
6920 6921
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
6922
	if (local->avg_load >= busiest->avg_load)
6923 6924
		goto out_balanced;

6925 6926 6927 6928
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
6929
	if (local->avg_load >= sds.avg_load)
6930 6931
		goto out_balanced;

6932
	if (env->idle == CPU_IDLE) {
6933
		/*
6934 6935 6936 6937 6938
		 * 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
6939
		 */
6940 6941
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
6942
			goto out_balanced;
6943 6944 6945 6946 6947
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
6948 6949
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
6950
			goto out_balanced;
6951
	}
6952

6953
force_balance:
6954
	/* Looks like there is an imbalance. Compute it */
6955
	calculate_imbalance(env, &sds);
6956 6957 6958
	return sds.busiest;

out_balanced:
6959
	env->imbalance = 0;
6960 6961 6962 6963 6964 6965
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
6966
static struct rq *find_busiest_queue(struct lb_env *env,
6967
				     struct sched_group *group)
6968 6969
{
	struct rq *busiest = NULL, *rq;
6970
	unsigned long busiest_load = 0, busiest_capacity = 1;
6971 6972
	int i;

6973
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6974
		unsigned long capacity, wl;
6975 6976 6977 6978
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
6979

6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001
		/*
		 * 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;

7002
		capacity = capacity_of(i);
7003

7004
		wl = weighted_cpuload(i);
7005

7006 7007
		/*
		 * When comparing with imbalance, use weighted_cpuload()
7008
		 * which is not scaled with the cpu capacity.
7009
		 */
7010 7011 7012

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
7013 7014
			continue;

7015 7016
		/*
		 * For the load comparisons with the other cpu's, consider
7017 7018 7019
		 * 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.
7020
		 *
7021
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
7022
		 * multiplication to rid ourselves of the division works out
7023 7024
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
7025
		 */
7026
		if (wl * busiest_capacity > busiest_load * capacity) {
7027
			busiest_load = wl;
7028
			busiest_capacity = capacity;
7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041 7042
			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. */
7043
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
7044

7045
static int need_active_balance(struct lb_env *env)
7046
{
7047 7048 7049
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
7050 7051 7052 7053 7054 7055

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
7056
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
7057
			return 1;
7058 7059
	}

7060 7061 7062 7063 7064 7065 7066 7067 7068 7069 7070 7071 7072
	/*
	 * The dst_cpu is idle and the src_cpu CPU has only 1 CFS task.
	 * It's worth migrating the task if the src_cpu's capacity is reduced
	 * because of other sched_class or IRQs if more capacity stays
	 * available on dst_cpu.
	 */
	if ((env->idle != CPU_NOT_IDLE) &&
	    (env->src_rq->cfs.h_nr_running == 1)) {
		if ((check_cpu_capacity(env->src_rq, sd)) &&
		    (capacity_of(env->src_cpu)*sd->imbalance_pct < capacity_of(env->dst_cpu)*100))
			return 1;
	}

7073 7074 7075
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

7076 7077
static int active_load_balance_cpu_stop(void *data);

7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095 7096 7097 7098 7099 7100 7101 7102 7103 7104 7105 7106 7107 7108
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.
	 */
7109
	return balance_cpu == env->dst_cpu;
7110 7111
}

7112 7113 7114 7115 7116 7117
/*
 * 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,
7118
			int *continue_balancing)
7119
{
7120
	int ld_moved, cur_ld_moved, active_balance = 0;
7121
	struct sched_domain *sd_parent = sd->parent;
7122 7123 7124
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
7125
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
7126

7127 7128
	struct lb_env env = {
		.sd		= sd,
7129 7130
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
7131
		.dst_grpmask    = sched_group_cpus(sd->groups),
7132
		.idle		= idle,
7133
		.loop_break	= sched_nr_migrate_break,
7134
		.cpus		= cpus,
7135
		.fbq_type	= all,
7136
		.tasks		= LIST_HEAD_INIT(env.tasks),
7137 7138
	};

7139 7140 7141 7142
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
7143
	if (idle == CPU_NEWLY_IDLE)
7144 7145
		env.dst_grpmask = NULL;

7146 7147 7148 7149 7150
	cpumask_copy(cpus, cpu_active_mask);

	schedstat_inc(sd, lb_count[idle]);

redo:
7151 7152
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
7153
		goto out_balanced;
7154
	}
7155

7156
	group = find_busiest_group(&env);
7157 7158 7159 7160 7161
	if (!group) {
		schedstat_inc(sd, lb_nobusyg[idle]);
		goto out_balanced;
	}

7162
	busiest = find_busiest_queue(&env, group);
7163 7164 7165 7166 7167
	if (!busiest) {
		schedstat_inc(sd, lb_nobusyq[idle]);
		goto out_balanced;
	}

7168
	BUG_ON(busiest == env.dst_rq);
7169

7170
	schedstat_add(sd, lb_imbalance[idle], env.imbalance);
7171

7172 7173 7174
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

7175 7176 7177 7178 7179 7180 7181 7182
	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.
		 */
7183
		env.flags |= LBF_ALL_PINNED;
7184
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
7185

7186
more_balance:
7187
		raw_spin_lock_irqsave(&busiest->lock, flags);
7188 7189 7190 7191 7192

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
7193
		cur_ld_moved = detach_tasks(&env);
7194 7195

		/*
7196 7197 7198 7199 7200
		 * 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.
7201
		 */
7202 7203 7204 7205 7206 7207 7208 7209

		raw_spin_unlock(&busiest->lock);

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

7210
		local_irq_restore(flags);
7211

7212 7213 7214 7215 7216
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

7217 7218 7219 7220 7221 7222 7223 7224 7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235
		/*
		 * 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.
		 */
7236
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
7237

7238 7239 7240
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

7241
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
7242
			env.dst_cpu	 = env.new_dst_cpu;
7243
			env.flags	&= ~LBF_DST_PINNED;
7244 7245
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
7246

7247 7248 7249 7250 7251 7252
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
7253

7254 7255 7256 7257
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
7258
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
7259

7260
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
7261 7262 7263
				*group_imbalance = 1;
		}

7264
		/* All tasks on this runqueue were pinned by CPU affinity */
7265
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
7266
			cpumask_clear_cpu(cpu_of(busiest), cpus);
7267 7268 7269
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
7270
				goto redo;
7271
			}
7272
			goto out_all_pinned;
7273 7274 7275 7276 7277
		}
	}

	if (!ld_moved) {
		schedstat_inc(sd, lb_failed[idle]);
7278 7279 7280 7281 7282 7283 7284 7285
		/*
		 * 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++;
7286

7287
		if (need_active_balance(&env)) {
7288 7289
			raw_spin_lock_irqsave(&busiest->lock, flags);

7290 7291 7292
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
7293 7294
			 */
			if (!cpumask_test_cpu(this_cpu,
7295
					tsk_cpus_allowed(busiest->curr))) {
7296 7297
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
7298
				env.flags |= LBF_ALL_PINNED;
7299 7300 7301
				goto out_one_pinned;
			}

7302 7303 7304 7305 7306
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
7307 7308 7309 7310 7311 7312
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
7313

7314
			if (active_balance) {
7315 7316 7317
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
7318
			}
7319 7320 7321 7322 7323 7324 7325 7326 7327 7328 7329 7330 7331 7332 7333 7334 7335 7336

			/*
			 * 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
7337
		 * detach_tasks).
7338 7339 7340 7341 7342 7343 7344 7345
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362
	/*
	 * 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.
	 */
7363 7364 7365 7366 7367 7368
	schedstat_inc(sd, lb_balanced[idle]);

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
7369
	if (((env.flags & LBF_ALL_PINNED) &&
7370
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
7371 7372 7373
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

7374
	ld_moved = 0;
7375 7376 7377 7378
out:
	return ld_moved;
}

7379 7380 7381 7382 7383 7384 7385 7386 7387 7388 7389 7390 7391 7392 7393 7394 7395 7396 7397 7398 7399 7400 7401 7402 7403 7404 7405
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;
}

7406 7407 7408 7409
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
7410
static int idle_balance(struct rq *this_rq)
7411
{
7412 7413
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
7414 7415
	struct sched_domain *sd;
	int pulled_task = 0;
7416
	u64 curr_cost = 0;
7417

7418
	idle_enter_fair(this_rq);
7419

7420 7421 7422 7423 7424 7425
	/*
	 * 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);

7426 7427
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
7428 7429 7430 7431 7432 7433
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, 0, &next_balance);
		rcu_read_unlock();

7434
		goto out;
7435
	}
7436

7437 7438
	raw_spin_unlock(&this_rq->lock);

7439
	update_blocked_averages(this_cpu);
7440
	rcu_read_lock();
7441
	for_each_domain(this_cpu, sd) {
7442
		int continue_balancing = 1;
7443
		u64 t0, domain_cost;
7444 7445 7446 7447

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

7448 7449
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, 0, &next_balance);
7450
			break;
7451
		}
7452

7453
		if (sd->flags & SD_BALANCE_NEWIDLE) {
7454 7455
			t0 = sched_clock_cpu(this_cpu);

7456
			pulled_task = load_balance(this_cpu, this_rq,
7457 7458
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
7459 7460 7461 7462 7463 7464

			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;
7465
		}
7466

7467
		update_next_balance(sd, 0, &next_balance);
7468 7469 7470 7471 7472 7473

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
7474 7475
			break;
	}
7476
	rcu_read_unlock();
7477 7478 7479

	raw_spin_lock(&this_rq->lock);

7480 7481 7482
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

7483
	/*
7484 7485 7486
	 * 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.
7487
	 */
7488
	if (this_rq->cfs.h_nr_running && !pulled_task)
7489
		pulled_task = 1;
7490

7491 7492 7493
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
7494
		this_rq->next_balance = next_balance;
7495

7496
	/* Is there a task of a high priority class? */
7497
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
7498 7499 7500 7501
		pulled_task = -1;

	if (pulled_task) {
		idle_exit_fair(this_rq);
7502
		this_rq->idle_stamp = 0;
7503
	}
7504

7505
	return pulled_task;
7506 7507 7508
}

/*
7509 7510 7511 7512
 * 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.
7513
 */
7514
static int active_load_balance_cpu_stop(void *data)
7515
{
7516 7517
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
7518
	int target_cpu = busiest_rq->push_cpu;
7519
	struct rq *target_rq = cpu_rq(target_cpu);
7520
	struct sched_domain *sd;
7521
	struct task_struct *p = NULL;
7522 7523 7524 7525 7526 7527 7528

	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;
7529 7530 7531

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
7532
		goto out_unlock;
7533 7534 7535 7536 7537 7538 7539 7540 7541

	/*
	 * 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. */
7542
	rcu_read_lock();
7543 7544 7545 7546 7547 7548 7549
	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)) {
7550 7551
		struct lb_env env = {
			.sd		= sd,
7552 7553 7554 7555
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
7556 7557 7558
			.idle		= CPU_IDLE,
		};

7559 7560
		schedstat_inc(sd, alb_count);

7561 7562
		p = detach_one_task(&env);
		if (p)
7563 7564 7565 7566
			schedstat_inc(sd, alb_pushed);
		else
			schedstat_inc(sd, alb_failed);
	}
7567
	rcu_read_unlock();
7568 7569
out_unlock:
	busiest_rq->active_balance = 0;
7570 7571 7572 7573 7574 7575 7576
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

7577
	return 0;
7578 7579
}

7580 7581 7582 7583 7584
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

7585
#ifdef CONFIG_NO_HZ_COMMON
7586 7587 7588 7589 7590 7591
/*
 * 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.
 */
7592
static struct {
7593
	cpumask_var_t idle_cpus_mask;
7594
	atomic_t nr_cpus;
7595 7596
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
7597

7598
static inline int find_new_ilb(void)
7599
{
7600
	int ilb = cpumask_first(nohz.idle_cpus_mask);
7601

7602 7603 7604 7605
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
7606 7607
}

7608 7609 7610 7611 7612
/*
 * 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).
 */
7613
static void nohz_balancer_kick(void)
7614 7615 7616 7617 7618
{
	int ilb_cpu;

	nohz.next_balance++;

7619
	ilb_cpu = find_new_ilb();
7620

7621 7622
	if (ilb_cpu >= nr_cpu_ids)
		return;
7623

7624
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
7625 7626 7627 7628 7629 7630 7631 7632
		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);
7633 7634 7635
	return;
}

7636
static inline void nohz_balance_exit_idle(int cpu)
7637 7638
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
7639 7640 7641 7642 7643 7644 7645
		/*
		 * 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);
		}
7646 7647 7648 7649
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

7650 7651 7652
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
7653
	int cpu = smp_processor_id();
7654 7655

	rcu_read_lock();
7656
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7657 7658 7659 7660 7661

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

7662
	atomic_inc(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7663
unlock:
7664 7665 7666 7667 7668 7669
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
7670
	int cpu = smp_processor_id();
7671 7672

	rcu_read_lock();
7673
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7674 7675 7676 7677 7678

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

7679
	atomic_dec(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7680
unlock:
7681 7682 7683
	rcu_read_unlock();
}

7684
/*
7685
 * This routine will record that the cpu is going idle with tick stopped.
7686
 * This info will be used in performing idle load balancing in the future.
7687
 */
7688
void nohz_balance_enter_idle(int cpu)
7689
{
7690 7691 7692 7693 7694 7695
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

7696 7697
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
7698

7699 7700 7701 7702 7703 7704
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

7705 7706 7707
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
7708
}
7709

7710
static int sched_ilb_notifier(struct notifier_block *nfb,
7711 7712 7713 7714
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DYING:
7715
		nohz_balance_exit_idle(smp_processor_id());
7716 7717 7718 7719 7720
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}
7721 7722 7723 7724
#endif

static DEFINE_SPINLOCK(balancing);

7725 7726 7727 7728
/*
 * 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.
 */
7729
void update_max_interval(void)
7730 7731 7732 7733
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

7734 7735 7736 7737
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
7738
 * Balancing parameters are set up in init_sched_domains.
7739
 */
7740
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
7741
{
7742
	int continue_balancing = 1;
7743
	int cpu = rq->cpu;
7744
	unsigned long interval;
7745
	struct sched_domain *sd;
7746 7747 7748
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
7749 7750
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
7751

7752
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
7753

7754
	rcu_read_lock();
7755
	for_each_domain(cpu, sd) {
7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767
		/*
		 * 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;

7768 7769 7770
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781
		/*
		 * 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;
		}

7782
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7783 7784 7785 7786 7787 7788 7789 7790

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

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
7791
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
7792
				/*
7793
				 * The LBF_DST_PINNED logic could have changed
7794 7795
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
7796
				 */
7797
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
7798 7799
			}
			sd->last_balance = jiffies;
7800
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7801 7802 7803 7804 7805 7806 7807 7808
		}
		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;
		}
7809 7810
	}
	if (need_decay) {
7811
		/*
7812 7813
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
7814
		 */
7815 7816
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
7817
	}
7818
	rcu_read_unlock();
7819 7820 7821 7822 7823 7824 7825 7826 7827 7828

	/*
	 * 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;
}

7829
#ifdef CONFIG_NO_HZ_COMMON
7830
/*
7831
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
7832 7833
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
7834
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
7835
{
7836
	int this_cpu = this_rq->cpu;
7837 7838 7839
	struct rq *rq;
	int balance_cpu;

7840 7841 7842
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
7843 7844

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
7845
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
7846 7847 7848 7849 7850 7851 7852
			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.
		 */
7853
		if (need_resched())
7854 7855
			break;

V
Vincent Guittot 已提交
7856 7857
		rq = cpu_rq(balance_cpu);

7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868
		/*
		 * 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);
		}
7869 7870 7871 7872 7873

		if (time_after(this_rq->next_balance, rq->next_balance))
			this_rq->next_balance = rq->next_balance;
	}
	nohz.next_balance = this_rq->next_balance;
7874 7875
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
7876 7877 7878
}

/*
7879
 * Current heuristic for kicking the idle load balancer in the presence
7880
 * of an idle cpu in the system.
7881
 *   - This rq has more than one task.
7882 7883 7884 7885
 *   - This rq has at least one CFS task and the capacity of the CPU is
 *     significantly reduced because of RT tasks or IRQs.
 *   - At parent of LLC scheduler domain level, this cpu's scheduler group has
 *     multiple busy cpu.
7886 7887
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
7888
 */
7889
static inline bool nohz_kick_needed(struct rq *rq)
7890 7891
{
	unsigned long now = jiffies;
7892
	struct sched_domain *sd;
7893
	struct sched_group_capacity *sgc;
7894
	int nr_busy, cpu = rq->cpu;
7895
	bool kick = false;
7896

7897
	if (unlikely(rq->idle_balance))
7898
		return false;
7899

7900 7901 7902 7903
       /*
	* 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.
	*/
7904
	set_cpu_sd_state_busy();
7905
	nohz_balance_exit_idle(cpu);
7906 7907 7908 7909 7910 7911

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
7912
		return false;
7913 7914

	if (time_before(now, nohz.next_balance))
7915
		return false;
7916

7917
	if (rq->nr_running >= 2)
7918
		return true;
7919

7920
	rcu_read_lock();
7921 7922
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
	if (sd) {
7923 7924
		sgc = sd->groups->sgc;
		nr_busy = atomic_read(&sgc->nr_busy_cpus);
7925

7926 7927 7928 7929 7930
		if (nr_busy > 1) {
			kick = true;
			goto unlock;
		}

7931
	}
7932

7933 7934 7935 7936 7937 7938 7939 7940
	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
			kick = true;
			goto unlock;
		}
	}
7941

7942
	sd = rcu_dereference(per_cpu(sd_asym, cpu));
7943
	if (sd && (cpumask_first_and(nohz.idle_cpus_mask,
7944 7945 7946 7947
				  sched_domain_span(sd)) < cpu)) {
		kick = true;
		goto unlock;
	}
7948

7949
unlock:
7950
	rcu_read_unlock();
7951
	return kick;
7952 7953
}
#else
7954
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
7955 7956 7957 7958 7959 7960
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
7961 7962
static void run_rebalance_domains(struct softirq_action *h)
{
7963
	struct rq *this_rq = this_rq();
7964
	enum cpu_idle_type idle = this_rq->idle_balance ?
7965 7966 7967
						CPU_IDLE : CPU_NOT_IDLE;

	/*
7968
	 * If this cpu has a pending nohz_balance_kick, then do the
7969
	 * balancing on behalf of the other idle cpus whose ticks are
7970 7971 7972 7973
	 * stopped. Do nohz_idle_balance *before* rebalance_domains to
	 * give the idle cpus a chance to load balance. Else we may
	 * load balance only within the local sched_domain hierarchy
	 * and abort nohz_idle_balance altogether if we pull some load.
7974
	 */
7975
	nohz_idle_balance(this_rq, idle);
7976
	rebalance_domains(this_rq, idle);
7977 7978 7979 7980 7981
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
7982
void trigger_load_balance(struct rq *rq)
7983 7984
{
	/* Don't need to rebalance while attached to NULL domain */
7985 7986 7987 7988
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
7989
		raise_softirq(SCHED_SOFTIRQ);
7990
#ifdef CONFIG_NO_HZ_COMMON
7991
	if (nohz_kick_needed(rq))
7992
		nohz_balancer_kick();
7993
#endif
7994 7995
}

7996 7997 7998
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
7999 8000

	update_runtime_enabled(rq);
8001 8002 8003 8004 8005
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
8006 8007 8008

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
8009 8010
}

8011
#endif /* CONFIG_SMP */
8012

8013 8014 8015
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
8016
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
8017 8018 8019 8020 8021 8022
{
	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
8023
		entity_tick(cfs_rq, se, queued);
8024
	}
8025

8026
	if (numabalancing_enabled)
8027
		task_tick_numa(rq, curr);
8028

8029
	update_rq_runnable_avg(rq, 1);
8030 8031 8032
}

/*
P
Peter Zijlstra 已提交
8033 8034 8035
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
8036
 */
P
Peter Zijlstra 已提交
8037
static void task_fork_fair(struct task_struct *p)
8038
{
8039 8040
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
8041
	int this_cpu = smp_processor_id();
P
Peter Zijlstra 已提交
8042 8043 8044
	struct rq *rq = this_rq();
	unsigned long flags;

8045
	raw_spin_lock_irqsave(&rq->lock, flags);
8046

8047 8048
	update_rq_clock(rq);

8049 8050 8051
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;

8052 8053 8054 8055 8056 8057 8058 8059 8060
	/*
	 * 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();
8061

8062
	update_curr(cfs_rq);
P
Peter Zijlstra 已提交
8063

8064 8065
	if (curr)
		se->vruntime = curr->vruntime;
8066
	place_entity(cfs_rq, se, 1);
8067

P
Peter Zijlstra 已提交
8068
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
8069
		/*
8070 8071 8072
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
8073
		swap(curr->vruntime, se->vruntime);
8074
		resched_curr(rq);
8075
	}
8076

8077 8078
	se->vruntime -= cfs_rq->min_vruntime;

8079
	raw_spin_unlock_irqrestore(&rq->lock, flags);
8080 8081
}

8082 8083 8084 8085
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
8086 8087
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
8088
{
8089
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
8090 8091
		return;

8092 8093 8094 8095 8096
	/*
	 * 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
8097
	if (rq->curr == p) {
8098
		if (p->prio > oldprio)
8099
			resched_curr(rq);
8100
	} else
8101
		check_preempt_curr(rq, p, 0);
8102 8103
}

P
Peter Zijlstra 已提交
8104 8105 8106 8107 8108 8109
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);

	/*
8110
	 * Ensure the task's vruntime is normalized, so that when it's
P
Peter Zijlstra 已提交
8111 8112 8113
	 * switched back to the fair class the enqueue_entity(.flags=0) will
	 * do the right thing.
	 *
8114 8115
	 * 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
8116 8117
	 * the task is sleeping will it still have non-normalized vruntime.
	 */
8118
	if (!task_on_rq_queued(p) && p->state != TASK_RUNNING) {
P
Peter Zijlstra 已提交
8119 8120 8121 8122 8123 8124 8125
		/*
		 * 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;
	}
8126

8127
#ifdef CONFIG_SMP
8128 8129 8130 8131 8132
	/*
	* 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.
	*/
8133 8134 8135
	if (se->avg.decay_count) {
		__synchronize_entity_decay(se);
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
8136 8137
	}
#endif
P
Peter Zijlstra 已提交
8138 8139
}

8140 8141 8142
/*
 * We switched to the sched_fair class.
 */
P
Peter Zijlstra 已提交
8143
static void switched_to_fair(struct rq *rq, struct task_struct *p)
8144
{
8145
#ifdef CONFIG_FAIR_GROUP_SCHED
8146
	struct sched_entity *se = &p->se;
8147 8148 8149 8150 8151 8152
	/*
	 * 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
8153
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
8154 8155
		return;

8156 8157 8158 8159 8160
	/*
	 * 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
8161
	if (rq->curr == p)
8162
		resched_curr(rq);
8163
	else
8164
		check_preempt_curr(rq, p, 0);
8165 8166
}

8167 8168 8169 8170 8171 8172 8173 8174 8175
/* 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;

8176 8177 8178 8179 8180 8181 8182
	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);
	}
8183 8184
}

8185 8186 8187 8188 8189 8190 8191
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
8192
#ifdef CONFIG_SMP
8193
	atomic64_set(&cfs_rq->decay_counter, 1);
8194
	atomic_long_set(&cfs_rq->removed_load, 0);
8195
#endif
8196 8197
}

P
Peter Zijlstra 已提交
8198
#ifdef CONFIG_FAIR_GROUP_SCHED
8199
static void task_move_group_fair(struct task_struct *p, int queued)
P
Peter Zijlstra 已提交
8200
{
P
Peter Zijlstra 已提交
8201
	struct sched_entity *se = &p->se;
8202
	struct cfs_rq *cfs_rq;
P
Peter Zijlstra 已提交
8203

8204 8205 8206 8207 8208 8209 8210 8211 8212 8213 8214 8215 8216
	/*
	 * 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.
	 */
8217
	/*
8218
	 * When !queued, vruntime of the task has usually NOT been normalized.
8219 8220 8221 8222
	 * 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().
8223 8224
	 * - Moving a task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
8225 8226 8227 8228
	 *
	 * To prevent boost or penalty in the new cfs_rq caused by delta
	 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
	 */
8229 8230
	if (!queued && (!se->sum_exec_runtime || p->state == TASK_WAKING))
		queued = 1;
8231

8232
	if (!queued)
P
Peter Zijlstra 已提交
8233
		se->vruntime -= cfs_rq_of(se)->min_vruntime;
8234
	set_task_rq(p, task_cpu(p));
P
Peter Zijlstra 已提交
8235
	se->depth = se->parent ? se->parent->depth + 1 : 0;
8236
	if (!queued) {
P
Peter Zijlstra 已提交
8237 8238
		cfs_rq = cfs_rq_of(se);
		se->vruntime += cfs_rq->min_vruntime;
8239 8240 8241 8242 8243 8244
#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
8245 8246
		se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
		cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
8247 8248
#endif
	}
P
Peter Zijlstra 已提交
8249
}
8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272 8273 8274 8275 8276 8277 8278 8279 8280 8281 8282 8283 8284 8285 8286 8287 8288 8289 8290 8291 8292 8293 8294 8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315 8316 8317 8318 8319 8320 8321 8322 8323 8324 8325 8326 8327 8328 8329 8330 8331 8332 8333 8334 8335 8336 8337 8338 8339 8340 8341

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
8342
	if (!parent) {
8343
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
8344 8345
		se->depth = 0;
	} else {
8346
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
8347 8348
		se->depth = parent->depth + 1;
	}
8349 8350

	se->my_q = cfs_rq;
8351 8352
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378 8379 8380 8381 8382
	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);
8383 8384 8385

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
8386
		for_each_sched_entity(se)
8387 8388 8389 8390 8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402 8403 8404 8405 8406 8407
			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
8408

8409
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
8410 8411 8412 8413 8414 8415 8416 8417 8418
{
	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)
8419
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
8420 8421 8422 8423

	return rr_interval;
}

8424 8425 8426
/*
 * All the scheduling class methods:
 */
8427
const struct sched_class fair_sched_class = {
8428
	.next			= &idle_sched_class,
8429 8430 8431
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
8432
	.yield_to_task		= yield_to_task_fair,
8433

I
Ingo Molnar 已提交
8434
	.check_preempt_curr	= check_preempt_wakeup,
8435 8436 8437 8438

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

8439
#ifdef CONFIG_SMP
L
Li Zefan 已提交
8440
	.select_task_rq		= select_task_rq_fair,
8441
	.migrate_task_rq	= migrate_task_rq_fair,
8442

8443 8444
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
8445 8446

	.task_waking		= task_waking_fair,
8447
#endif
8448

8449
	.set_curr_task          = set_curr_task_fair,
8450
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
8451
	.task_fork		= task_fork_fair,
8452 8453

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
8454
	.switched_from		= switched_from_fair,
8455
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
8456

8457 8458
	.get_rr_interval	= get_rr_interval_fair,

8459 8460
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
8461
#ifdef CONFIG_FAIR_GROUP_SCHED
8462
	.task_move_group	= task_move_group_fair,
P
Peter Zijlstra 已提交
8463
#endif
8464 8465 8466
};

#ifdef CONFIG_SCHED_DEBUG
8467
void print_cfs_stats(struct seq_file *m, int cpu)
8468 8469 8470
{
	struct cfs_rq *cfs_rq;

8471
	rcu_read_lock();
8472
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
8473
		print_cfs_rq(m, cpu, cfs_rq);
8474
	rcu_read_unlock();
8475
}
8476 8477 8478 8479 8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492 8493 8494 8495 8496

#ifdef CONFIG_NUMA_BALANCING
void show_numa_stats(struct task_struct *p, struct seq_file *m)
{
	int node;
	unsigned long tsf = 0, tpf = 0, gsf = 0, gpf = 0;

	for_each_online_node(node) {
		if (p->numa_faults) {
			tsf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 0)];
			tpf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		if (p->numa_group) {
			gsf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 0)],
			gpf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		print_numa_stats(m, node, tsf, tpf, gsf, gpf);
	}
}
#endif /* CONFIG_NUMA_BALANCING */
#endif /* CONFIG_SCHED_DEBUG */
8497 8498 8499 8500 8501 8502

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

8503
#ifdef CONFIG_NO_HZ_COMMON
8504
	nohz.next_balance = jiffies;
8505
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
8506
	cpu_notifier(sched_ilb_notifier, 0);
8507 8508 8509 8510
#endif
#endif /* SMP */

}