fair.c 222.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
/*
 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
 *
 *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 *
 *  Interactivity improvements by Mike Galbraith
 *  (C) 2007 Mike Galbraith <efault@gmx.de>
 *
 *  Various enhancements by Dmitry Adamushko.
 *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
 *
 *  Group scheduling enhancements by Srivatsa Vaddagiri
 *  Copyright IBM Corporation, 2007
 *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
 *
 *  Scaled math optimizations by Thomas Gleixner
 *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
18 19
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
20
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
21 22
 */

23
#include <linux/sched.h>
24
#include <linux/latencytop.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;
}

286 287 288
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
289 290 291 292 293 294 295 296 297 298 299 300
		/*
		 * 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,
301
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
302
		}
303 304 305 306 307 308 309 310 311 312 313 314 315

		cfs_rq->on_list = 1;
	}
}

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
316 317 318 319 320
/* 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
321
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
322 323 324
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
325
		return se->cfs_rq;
P
Peter Zijlstra 已提交
326

P
Peter Zijlstra 已提交
327
	return NULL;
P
Peter Zijlstra 已提交
328 329 330 331 332 333 334
}

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

335 336 337 338 339 340 341 342 343 344 345 346 347
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
348 349
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366

	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);
	}
}

367 368 369 370 371 372
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
373

374 375 376
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
377 378 379 380
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
381 382
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
383

P
Peter Zijlstra 已提交
384
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
385
{
P
Peter Zijlstra 已提交
386
	return &task_rq(p)->cfs;
387 388
}

P
Peter Zijlstra 已提交
389 390 391 392 393 394 395 396 397 398 399 400 401 402
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;
}

403 404 405 406 407 408 409 410
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
411 412 413 414 415 416 417 418
#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;
}

419 420 421 422 423
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
424 425
#endif	/* CONFIG_FAIR_GROUP_SCHED */

426
static __always_inline
427
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
428 429 430 431 432

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

433
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
434
{
435
	s64 delta = (s64)(vruntime - max_vruntime);
436
	if (delta > 0)
437
		max_vruntime = vruntime;
438

439
	return max_vruntime;
440 441
}

442
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
443 444 445 446 447 448 449 450
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

451 452 453 454 455 456
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

457 458 459 460 461 462 463 464 465 466 467 468
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
469
		if (!cfs_rq->curr)
470 471 472 473 474
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

475
	/* ensure we never gain time by being placed backwards. */
476
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
477 478 479 480
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
481 482
}

483 484 485
/*
 * Enqueue an entity into the rb-tree:
 */
486
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502
{
	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.
		 */
503
		if (entity_before(se, entry)) {
504 505 506 507 508 509 510 511 512 513 514
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
			leftmost = 0;
		}
	}

	/*
	 * Maintain a cache of leftmost tree entries (it is frequently
	 * used):
	 */
515
	if (leftmost)
I
Ingo Molnar 已提交
516
		cfs_rq->rb_leftmost = &se->run_node;
517 518 519 520 521

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

522
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
523
{
P
Peter Zijlstra 已提交
524 525 526 527 528 529
	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
530

531 532 533
	rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
}

534
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
535
{
536 537 538 539 540 541
	struct rb_node *left = cfs_rq->rb_leftmost;

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
542 543
}

544 545 546 547 548 549 550 551 552 553 554
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
555
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
556
{
557
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
558

559 560
	if (!last)
		return NULL;
561 562

	return rb_entry(last, struct sched_entity, run_node);
563 564
}

565 566 567 568
/**************************************************************
 * Scheduling class statistics methods:
 */

569
int sched_proc_update_handler(struct ctl_table *table, int write,
570
		void __user *buffer, size_t *lenp,
571 572
		loff_t *ppos)
{
573
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
574
	unsigned int factor = get_update_sysctl_factor();
575 576 577 578 579 580 581

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

582 583 584 585 586 587 588
#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

589 590 591
	return 0;
}
#endif
592

593
/*
594
 * delta /= w
595
 */
596
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
597
{
598
	if (unlikely(se->load.weight != NICE_0_LOAD))
599
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
600 601 602 603

	return delta;
}

604 605 606
/*
 * The idea is to set a period in which each task runs once.
 *
607
 * When there are too many tasks (sched_nr_latency) we have to stretch
608 609 610 611
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
612 613
static u64 __sched_period(unsigned long nr_running)
{
614 615 616 617
	if (unlikely(nr_running > sched_nr_latency))
		return nr_running * sysctl_sched_min_granularity;
	else
		return sysctl_sched_latency;
618 619
}

620 621 622 623
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
624
 * s = p*P[w/rw]
625
 */
P
Peter Zijlstra 已提交
626
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
627
{
M
Mike Galbraith 已提交
628
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
629

M
Mike Galbraith 已提交
630
	for_each_sched_entity(se) {
L
Lin Ming 已提交
631
		struct load_weight *load;
632
		struct load_weight lw;
L
Lin Ming 已提交
633 634 635

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
636

M
Mike Galbraith 已提交
637
		if (unlikely(!se->on_rq)) {
638
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
639 640 641 642

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
643
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
644 645
	}
	return slice;
646 647
}

648
/*
A
Andrei Epure 已提交
649
 * We calculate the vruntime slice of a to-be-inserted task.
650
 *
651
 * vs = s/w
652
 */
653
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
654
{
655
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
656 657
}

658
#ifdef CONFIG_SMP
659
static int select_idle_sibling(struct task_struct *p, int cpu);
660 661
static unsigned long task_h_load(struct task_struct *p);

662 663
/*
 * We choose a half-life close to 1 scheduling period.
664 665
 * Note: The tables runnable_avg_yN_inv and runnable_avg_yN_sum are
 * dependent on this value.
666 667 668
 */
#define LOAD_AVG_PERIOD 32
#define LOAD_AVG_MAX 47742 /* maximum possible load avg */
669
#define LOAD_AVG_MAX_N 345 /* number of full periods to produce LOAD_AVG_MAX */
670

671 672
/* Give new sched_entity start runnable values to heavy its load in infant time */
void init_entity_runnable_average(struct sched_entity *se)
673
{
674
	struct sched_avg *sa = &se->avg;
675

676 677 678 679 680 681 682
	sa->last_update_time = 0;
	/*
	 * sched_avg's period_contrib should be strictly less then 1024, so
	 * we give it 1023 to make sure it is almost a period (1024us), and
	 * will definitely be update (after enqueue).
	 */
	sa->period_contrib = 1023;
683
	sa->load_avg = scale_load_down(se->load.weight);
684 685
	sa->load_sum = sa->load_avg * LOAD_AVG_MAX;
	sa->util_avg = scale_load_down(SCHED_LOAD_SCALE);
686
	sa->util_sum = sa->util_avg * LOAD_AVG_MAX;
687
	/* when this task enqueue'ed, it will contribute to its cfs_rq's load_avg */
688
}
689 690 691

static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq);
static inline unsigned long cfs_rq_load_avg(struct cfs_rq *cfs_rq);
692
#else
693
void init_entity_runnable_average(struct sched_entity *se)
694 695 696 697
{
}
#endif

698
/*
699
 * Update the current task's runtime statistics.
700
 */
701
static void update_curr(struct cfs_rq *cfs_rq)
702
{
703
	struct sched_entity *curr = cfs_rq->curr;
704
	u64 now = rq_clock_task(rq_of(cfs_rq));
705
	u64 delta_exec;
706 707 708 709

	if (unlikely(!curr))
		return;

710 711
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
712
		return;
713

I
Ingo Molnar 已提交
714
	curr->exec_start = now;
715

716 717 718 719 720 721 722 723 724
	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);

725 726 727
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

728
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
729
		cpuacct_charge(curtask, delta_exec);
730
		account_group_exec_runtime(curtask, delta_exec);
731
	}
732 733

	account_cfs_rq_runtime(cfs_rq, delta_exec);
734 735
}

736 737 738 739 740
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

741
#ifdef CONFIG_SCHEDSTATS
742
static inline void
743
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
744
{
745 746 747 748 749 750 751
	u64 wait_start = rq_clock(rq_of(cfs_rq));

	if (entity_is_task(se) && task_on_rq_migrating(task_of(se)) &&
	    likely(wait_start > se->statistics.wait_start))
		wait_start -= se->statistics.wait_start;

	se->statistics.wait_start = wait_start;
752 753
}

754 755 756 757
static void
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *p;
758 759 760
	u64 delta;

	delta = rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start;
761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781

	if (entity_is_task(se)) {
		p = task_of(se);
		if (task_on_rq_migrating(p)) {
			/*
			 * Preserve migrating task's wait time so wait_start
			 * time stamp can be adjusted to accumulate wait time
			 * prior to migration.
			 */
			se->statistics.wait_start = delta;
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

	se->statistics.wait_max = max(se->statistics.wait_max, delta);
	se->statistics.wait_count++;
	se->statistics.wait_sum += delta;
	se->statistics.wait_start = 0;
}

782 783 784
/*
 * Task is being enqueued - update stats:
 */
785 786
static inline void
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
787 788 789 790 791
{
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
792
	if (se != cfs_rq->curr)
793
		update_stats_wait_start(cfs_rq, se);
794 795 796
}

static inline void
797
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
798 799 800 801 802
{
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
803
	if (se != cfs_rq->curr)
804
		update_stats_wait_end(cfs_rq, se);
805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836

	if (flags & DEQUEUE_SLEEP) {
		if (entity_is_task(se)) {
			struct task_struct *tsk = task_of(se);

			if (tsk->state & TASK_INTERRUPTIBLE)
				se->statistics.sleep_start = rq_clock(rq_of(cfs_rq));
			if (tsk->state & TASK_UNINTERRUPTIBLE)
				se->statistics.block_start = rq_clock(rq_of(cfs_rq));
		}
	}

}
#else
static inline void
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
}

static inline void
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
}

static inline void
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
}

static inline void
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
{
837
}
838
#endif
839 840 841 842 843

/*
 * We are picking a new current task - update its stats:
 */
static inline void
844
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
845 846 847 848
{
	/*
	 * We are starting a new run period:
	 */
849
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
850 851 852 853 854 855
}

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

856 857
#ifdef CONFIG_NUMA_BALANCING
/*
858 859 860
 * 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.
861
 */
862 863
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
864 865 866

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
867

868 869 870
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894
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)
{
895
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
896 897 898
	unsigned int scan, floor;
	unsigned int windows = 1;

899 900
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916
	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);
}

917 918 919 920 921 922 923 924 925 926 927 928
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));
}

929 930 931 932 933
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
934
	pid_t gid;
935
	int active_nodes;
936 937

	struct rcu_head rcu;
938
	unsigned long total_faults;
939
	unsigned long max_faults_cpu;
940 941 942 943 944
	/*
	 * 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.
	 */
945
	unsigned long *faults_cpu;
946
	unsigned long faults[0];
947 948
};

949 950 951 952 953 954 955 956 957
/* 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)

958 959 960 961 962
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

963 964 965 966 967 968 969
/*
 * 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)
970
{
971
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
972 973 974 975
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
976
	if (!p->numa_faults)
977 978
		return 0;

979 980
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
981 982
}

983 984 985 986 987
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

988 989
	return p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 1)];
990 991
}

992 993
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
994 995
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
996 997
}

998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
/*
 * A node triggering more than 1/3 as many NUMA faults as the maximum is
 * considered part of a numa group's pseudo-interleaving set. Migrations
 * between these nodes are slowed down, to allow things to settle down.
 */
#define ACTIVE_NODE_FRACTION 3

static bool numa_is_active_node(int nid, struct numa_group *ng)
{
	return group_faults_cpu(ng, nid) * ACTIVE_NODE_FRACTION > ng->max_faults_cpu;
}

1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
/* 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;
}

1075 1076 1077 1078 1079 1080
/*
 * 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.
 */
1081 1082
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1083
{
1084
	unsigned long faults, total_faults;
1085

1086
	if (!p->numa_faults)
1087 1088 1089 1090 1091 1092 1093
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1094
	faults = task_faults(p, nid);
1095 1096
	faults += score_nearby_nodes(p, nid, dist, true);

1097
	return 1000 * faults / total_faults;
1098 1099
}

1100 1101
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1102
{
1103 1104 1105 1106 1107 1108 1109 1110
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1111 1112
		return 0;

1113
	faults = group_faults(p, nid);
1114 1115
	faults += score_nearby_nodes(p, nid, dist, false);

1116
	return 1000 * faults / total_faults;
1117 1118
}

1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158
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;

	/*
1159 1160
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1161
	 */
1162 1163
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1164 1165 1166
		return true;

	/*
1167 1168 1169 1170 1171 1172
	 * Distribute memory according to CPU & memory use on each node,
	 * with 3/4 hysteresis to avoid unnecessary memory migrations:
	 *
	 * faults_cpu(dst)   3   faults_cpu(src)
	 * --------------- * - > ---------------
	 * faults_mem(dst)   4   faults_mem(src)
1173
	 */
1174 1175
	return group_faults_cpu(ng, dst_nid) * group_faults(p, src_nid) * 3 >
	       group_faults_cpu(ng, src_nid) * group_faults(p, dst_nid) * 4;
1176 1177
}

1178
static unsigned long weighted_cpuload(const int cpu);
1179 1180
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1181
static unsigned long capacity_of(int cpu);
1182 1183
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1184
/* Cached statistics for all CPUs within a node */
1185
struct numa_stats {
1186
	unsigned long nr_running;
1187
	unsigned long load;
1188 1189

	/* Total compute capacity of CPUs on a node */
1190
	unsigned long compute_capacity;
1191 1192

	/* Approximate capacity in terms of runnable tasks on a node */
1193
	unsigned long task_capacity;
1194
	int has_free_capacity;
1195
};
1196

1197 1198 1199 1200 1201
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1202 1203
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1204 1205 1206 1207 1208 1209 1210

	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);
1211
		ns->compute_capacity += capacity_of(cpu);
1212 1213

		cpus++;
1214 1215
	}

1216 1217 1218 1219 1220
	/*
	 * 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.
	 *
1221 1222
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1223 1224 1225 1226
	 */
	if (!cpus)
		return;

1227 1228 1229 1230 1231 1232
	/* 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));
1233
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1234 1235
}

1236 1237
struct task_numa_env {
	struct task_struct *p;
1238

1239 1240
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1241

1242
	struct numa_stats src_stats, dst_stats;
1243

1244
	int imbalance_pct;
1245
	int dist;
1246 1247 1248

	struct task_struct *best_task;
	long best_imp;
1249 1250 1251
	int best_cpu;
};

1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
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);

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

1263
static bool load_too_imbalanced(long src_load, long dst_load,
1264 1265
				struct task_numa_env *env)
{
1266 1267
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278
	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;
1279 1280

	/* We care about the slope of the imbalance, not the direction. */
1281 1282
	if (dst_load < src_load)
		swap(dst_load, src_load);
1283 1284

	/* Is the difference below the threshold? */
1285 1286
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1287 1288 1289 1290 1291
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
1292
	 * Compare it with the old imbalance.
1293
	 */
1294
	orig_src_load = env->src_stats.load;
1295
	orig_dst_load = env->dst_stats.load;
1296

1297 1298
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);
1299

1300 1301 1302 1303 1304
	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);
1305 1306
}

1307 1308 1309 1310 1311 1312
/*
 * 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
 */
1313 1314
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1315 1316 1317 1318
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1319
	long src_load, dst_load;
1320
	long load;
1321
	long imp = env->p->numa_group ? groupimp : taskimp;
1322
	long moveimp = imp;
1323
	int dist = env->dist;
1324
	bool assigned = false;
1325 1326

	rcu_read_lock();
1327 1328 1329 1330

	raw_spin_lock_irq(&dst_rq->lock);
	cur = dst_rq->curr;
	/*
1331
	 * No need to move the exiting task or idle task.
1332 1333
	 */
	if ((cur->flags & PF_EXITING) || is_idle_task(cur))
1334
		cur = NULL;
1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347
	else {
		/*
		 * The task_struct must be protected here to protect the
		 * p->numa_faults access in the task_weight since the
		 * numa_faults could already be freed in the following path:
		 * finish_task_switch()
		 *     --> put_task_struct()
		 *         --> __put_task_struct()
		 *             --> task_numa_free()
		 */
		get_task_struct(cur);
	}

1348
	raw_spin_unlock_irq(&dst_rq->lock);
1349

1350 1351 1352 1353 1354 1355 1356
	/*
	 * 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;

1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368
	/*
	 * "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;

1369 1370
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1371
		 * in any group then look only at task weights.
1372
		 */
1373
		if (cur->numa_group == env->p->numa_group) {
1374 1375
			imp = taskimp + task_weight(cur, env->src_nid, dist) -
			      task_weight(cur, env->dst_nid, dist);
1376 1377 1378 1379 1380 1381
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1382
		} else {
1383 1384 1385 1386 1387 1388
			/*
			 * 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)
1389 1390
				imp += group_weight(cur, env->src_nid, dist) -
				       group_weight(cur, env->dst_nid, dist);
1391
			else
1392 1393
				imp += task_weight(cur, env->src_nid, dist) -
				       task_weight(cur, env->dst_nid, dist);
1394
		}
1395 1396
	}

1397
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1398 1399 1400 1401
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1402
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1403
		    !env->dst_stats.has_free_capacity)
1404 1405 1406 1407 1408 1409
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1410 1411
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1412 1413 1414 1415 1416 1417
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1418 1419 1420
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1421

1422 1423 1424 1425 1426 1427 1428 1429 1430
	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;
1431
			put_task_struct(cur);
1432 1433 1434 1435 1436 1437 1438 1439
			cur = NULL;
			goto assign;
		}
	}

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

1440
	if (cur) {
1441 1442 1443
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1444 1445
	}

1446
	if (load_too_imbalanced(src_load, dst_load, env))
1447 1448
		goto unlock;

1449 1450 1451 1452 1453 1454 1455
	/*
	 * 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);

1456
assign:
1457
	assigned = true;
1458 1459 1460
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
1461 1462 1463 1464 1465 1466
	/*
	 * The dst_rq->curr isn't assigned. The protection for task_struct is
	 * finished.
	 */
	if (cur && !assigned)
		put_task_struct(cur);
1467 1468
}

1469 1470
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1471 1472 1473 1474 1475 1476 1477 1478 1479
{
	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;
1480
		task_numa_compare(env, taskimp, groupimp);
1481 1482 1483
	}
}

1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500
/* 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
	 */
1501 1502 1503
	if (src->load * dst->compute_capacity * env->imbalance_pct >

	    dst->load * src->compute_capacity * 100)
1504 1505 1506 1507 1508
		return true;

	return false;
}

1509 1510 1511 1512
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1513

1514
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1515
		.src_nid = task_node(p),
1516 1517 1518 1519 1520

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1521
		.best_cpu = -1,
1522 1523
	};
	struct sched_domain *sd;
1524
	unsigned long taskweight, groupweight;
1525
	int nid, ret, dist;
1526
	long taskimp, groupimp;
1527

1528
	/*
1529 1530 1531 1532 1533 1534
	 * 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.
1535 1536
	 */
	rcu_read_lock();
1537
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1538 1539
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1540 1541
	rcu_read_unlock();

1542 1543 1544 1545 1546 1547 1548
	/*
	 * 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)) {
1549
		p->numa_preferred_nid = task_node(p);
1550 1551 1552
		return -EINVAL;
	}

1553
	env.dst_nid = p->numa_preferred_nid;
1554 1555 1556 1557 1558 1559
	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;
1560
	update_numa_stats(&env.dst_stats, env.dst_nid);
1561

1562
	/* Try to find a spot on the preferred nid. */
1563 1564
	if (numa_has_capacity(&env))
		task_numa_find_cpu(&env, taskimp, groupimp);
1565

1566 1567 1568 1569 1570 1571 1572
	/*
	 * 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.
	 */
1573
	if (env.best_cpu == -1 || (p->numa_group && p->numa_group->active_nodes > 1)) {
1574 1575 1576
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1577

1578
			dist = node_distance(env.src_nid, env.dst_nid);
1579 1580 1581 1582 1583
			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);
			}
1584

1585
			/* Only consider nodes where both task and groups benefit */
1586 1587
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1588
			if (taskimp < 0 && groupimp < 0)
1589 1590
				continue;

1591
			env.dist = dist;
1592 1593
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1594 1595
			if (numa_has_capacity(&env))
				task_numa_find_cpu(&env, taskimp, groupimp);
1596 1597 1598
		}
	}

1599 1600 1601 1602 1603 1604 1605 1606
	/*
	 * 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.
	 */
1607
	if (p->numa_group) {
1608 1609
		struct numa_group *ng = p->numa_group;

1610 1611 1612 1613 1614
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
			nid = env.dst_nid;

1615
		if (ng->active_nodes > 1 && numa_is_active_node(env.dst_nid, ng))
1616 1617 1618 1619 1620 1621
			sched_setnuma(p, env.dst_nid);
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1622

1623 1624 1625 1626 1627 1628
	/*
	 * 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);

1629
	if (env.best_task == NULL) {
1630 1631 1632
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1633 1634 1635 1636
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1637 1638
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1639 1640
	put_task_struct(env.best_task);
	return ret;
1641 1642
}

1643 1644 1645
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1646 1647
	unsigned long interval = HZ;

1648
	/* This task has no NUMA fault statistics yet */
1649
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1650 1651
		return;

1652
	/* Periodically retry migrating the task to the preferred node */
1653 1654
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1655 1656

	/* Success if task is already running on preferred CPU */
1657
	if (task_node(p) == p->numa_preferred_nid)
1658 1659 1660
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1661
	task_numa_migrate(p);
1662 1663
}

1664
/*
1665
 * Find out how many nodes on the workload is actively running on. Do this by
1666 1667 1668 1669
 * 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.
 */
1670
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1671 1672
{
	unsigned long faults, max_faults = 0;
1673
	int nid, active_nodes = 0;
1674 1675 1676 1677 1678 1679 1680 1681 1682

	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);
1683 1684
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1685
	}
1686 1687 1688

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1689 1690
}

1691 1692 1693
/*
 * 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
1694 1695 1696
 * 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.
1697 1698
 */
#define NUMA_PERIOD_SLOTS 10
1699
#define NUMA_PERIOD_THRESHOLD 7
1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719

/*
 * 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
1720 1721 1722
	 * 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
1723
	 */
1724
	if (local + shared == 0 || p->numa_faults_locality[2]) {
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
		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
		 */
1758
		ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared + 1));
1759 1760 1761 1762 1763 1764 1765 1766
		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));
}

1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784
/*
 * 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 {
1785 1786
		delta = p->se.avg.load_sum / p->se.load.weight;
		*period = LOAD_AVG_MAX;
1787 1788 1789 1790 1791 1792 1793 1794
	}

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

	return delta;
}

1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841
/*
 * 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;
1842
		nodemask_t max_group = NODE_MASK_NONE;
1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875
		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. */
1876 1877
		if (!max_faults)
			break;
1878 1879 1880 1881 1882
		nodes = max_group;
	}
	return nid;
}

1883 1884
static void task_numa_placement(struct task_struct *p)
{
1885 1886
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
1887
	unsigned long fault_types[2] = { 0, 0 };
1888 1889
	unsigned long total_faults;
	u64 runtime, period;
1890
	spinlock_t *group_lock = NULL;
1891

1892 1893 1894 1895 1896
	/*
	 * 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:
	 */
1897
	seq = READ_ONCE(p->mm->numa_scan_seq);
1898 1899 1900
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
1901
	p->numa_scan_period_max = task_scan_max(p);
1902

1903 1904 1905 1906
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

1907 1908 1909
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
1910
		spin_lock_irq(group_lock);
1911 1912
	}

1913 1914
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
1915 1916
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
1917
		unsigned long faults = 0, group_faults = 0;
1918
		int priv;
1919

1920
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
1921
			long diff, f_diff, f_weight;
1922

1923 1924 1925 1926
			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);
1927

1928
			/* Decay existing window, copy faults since last scan */
1929 1930 1931
			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;
1932

1933 1934 1935 1936 1937 1938 1939 1940
			/*
			 * 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);
1941
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
1942
				   (total_faults + 1);
1943 1944
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
1945

1946 1947 1948
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
1949
			p->total_numa_faults += diff;
1950
			if (p->numa_group) {
1951 1952 1953 1954 1955 1956 1957 1958 1959
				/*
				 * 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;
1960
				p->numa_group->total_faults += diff;
1961
				group_faults += p->numa_group->faults[mem_idx];
1962
			}
1963 1964
		}

1965 1966 1967 1968
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
1969 1970 1971 1972 1973 1974 1975

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

1976 1977
	update_task_scan_period(p, fault_types[0], fault_types[1]);

1978
	if (p->numa_group) {
1979
		numa_group_count_active_nodes(p->numa_group);
1980
		spin_unlock_irq(group_lock);
1981
		max_nid = preferred_group_nid(p, max_group_nid);
1982 1983
	}

1984 1985 1986 1987 1988 1989 1990
	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);
1991
	}
1992 1993
}

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
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);
}

2005 2006
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2007 2008 2009 2010 2011 2012 2013 2014 2015
{
	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) +
2016
				    4*nr_node_ids*sizeof(unsigned long);
2017 2018 2019 2020 2021 2022

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

		atomic_set(&grp->refcount, 1);
2023 2024
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2025
		spin_lock_init(&grp->lock);
2026
		grp->gid = p->pid;
2027
		/* Second half of the array tracks nids where faults happen */
2028 2029
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2030

2031
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2032
			grp->faults[i] = p->numa_faults[i];
2033

2034
		grp->total_faults = p->total_numa_faults;
2035

2036 2037 2038 2039 2040
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2041
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2042 2043

	if (!cpupid_match_pid(tsk, cpupid))
2044
		goto no_join;
2045 2046 2047

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2048
		goto no_join;
2049 2050 2051

	my_grp = p->numa_group;
	if (grp == my_grp)
2052
		goto no_join;
2053 2054 2055 2056 2057 2058

	/*
	 * 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)
2059
		goto no_join;
2060 2061 2062 2063 2064

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2065
		goto no_join;
2066

2067 2068 2069 2070 2071 2072 2073
	/* 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;
2074

2075 2076 2077
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2078
	if (join && !get_numa_group(grp))
2079
		goto no_join;
2080 2081 2082 2083 2084 2085

	rcu_read_unlock();

	if (!join)
		return;

2086 2087
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2088

2089
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2090 2091
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2092
	}
2093 2094
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2095 2096 2097 2098 2099

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

	spin_unlock(&my_grp->lock);
2100
	spin_unlock_irq(&grp->lock);
2101 2102 2103 2104

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2105 2106 2107 2108 2109
	return;

no_join:
	rcu_read_unlock();
	return;
2110 2111 2112 2113 2114
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2115
	void *numa_faults = p->numa_faults;
2116 2117
	unsigned long flags;
	int i;
2118 2119

	if (grp) {
2120
		spin_lock_irqsave(&grp->lock, flags);
2121
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2122
			grp->faults[i] -= p->numa_faults[i];
2123
		grp->total_faults -= p->total_numa_faults;
2124

2125
		grp->nr_tasks--;
2126
		spin_unlock_irqrestore(&grp->lock, flags);
2127
		RCU_INIT_POINTER(p->numa_group, NULL);
2128 2129 2130
		put_numa_group(grp);
	}

2131
	p->numa_faults = NULL;
2132
	kfree(numa_faults);
2133 2134
}

2135 2136 2137
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2138
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2139 2140
{
	struct task_struct *p = current;
2141
	bool migrated = flags & TNF_MIGRATED;
2142
	int cpu_node = task_node(current);
2143
	int local = !!(flags & TNF_FAULT_LOCAL);
2144
	struct numa_group *ng;
2145
	int priv;
2146

2147
	if (!static_branch_likely(&sched_numa_balancing))
2148 2149
		return;

2150 2151 2152 2153
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2154
	/* Allocate buffer to track faults on a per-node basis */
2155 2156
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2157
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2158

2159 2160
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2161
			return;
2162

2163
		p->total_numa_faults = 0;
2164
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2165
	}
2166

2167 2168 2169 2170 2171 2172 2173 2174
	/*
	 * 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);
2175
		if (!priv && !(flags & TNF_NO_GROUP))
2176
			task_numa_group(p, last_cpupid, flags, &priv);
2177 2178
	}

2179 2180 2181 2182 2183 2184
	/*
	 * 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.
	 */
2185 2186 2187 2188
	ng = p->numa_group;
	if (!priv && !local && ng && ng->active_nodes > 1 &&
				numa_is_active_node(cpu_node, ng) &&
				numa_is_active_node(mem_node, ng))
2189 2190
		local = 1;

2191
	task_numa_placement(p);
2192

2193 2194 2195 2196 2197
	/*
	 * 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))
2198 2199
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
2200 2201
	if (migrated)
		p->numa_pages_migrated += pages;
2202 2203
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2204

2205 2206
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2207
	p->numa_faults_locality[local] += pages;
2208 2209
}

2210 2211
static void reset_ptenuma_scan(struct task_struct *p)
{
2212 2213 2214 2215 2216 2217 2218 2219
	/*
	 * 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:
	 */
2220
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2221 2222 2223
	p->mm->numa_scan_offset = 0;
}

2224 2225 2226 2227 2228 2229 2230 2231 2232
/*
 * 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;
2233
	u64 runtime = p->se.sum_exec_runtime;
2234
	struct vm_area_struct *vma;
2235
	unsigned long start, end;
2236
	unsigned long nr_pte_updates = 0;
2237
	long pages, virtpages;
2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252

	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;

2253
	if (!mm->numa_next_scan) {
2254 2255
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2256 2257
	}

2258 2259 2260 2261 2262 2263 2264
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2265 2266 2267 2268
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
2269

2270
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2271 2272 2273
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2274 2275 2276 2277 2278 2279
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2280 2281 2282
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2283
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2284 2285
	if (!pages)
		return;
2286

2287

2288
	down_read(&mm->mmap_sem);
2289
	vma = find_vma(mm, start);
2290 2291
	if (!vma) {
		reset_ptenuma_scan(p);
2292
		start = 0;
2293 2294
		vma = mm->mmap;
	}
2295
	for (; vma; vma = vma->vm_next) {
2296
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2297
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2298
			continue;
2299
		}
2300

2301 2302 2303 2304 2305 2306 2307 2308 2309 2310
		/*
		 * 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
2311 2312 2313 2314 2315 2316
		/*
		 * 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;
2317

2318 2319 2320 2321
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2322
			nr_pte_updates = change_prot_numa(vma, start, end);
2323 2324

			/*
2325 2326 2327 2328 2329 2330
			 * Try to scan sysctl_numa_balancing_size worth of
			 * hpages that have at least one present PTE that
			 * is not already pte-numa. If the VMA contains
			 * areas that are unused or already full of prot_numa
			 * PTEs, scan up to virtpages, to skip through those
			 * areas faster.
2331 2332 2333
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2334
			virtpages -= (end - start) >> PAGE_SHIFT;
2335

2336
			start = end;
2337
			if (pages <= 0 || virtpages <= 0)
2338
				goto out;
2339 2340

			cond_resched();
2341
		} while (end != vma->vm_end);
2342
	}
2343

2344
out:
2345
	/*
P
Peter Zijlstra 已提交
2346 2347 2348 2349
	 * 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.
2350 2351
	 */
	if (vma)
2352
		mm->numa_scan_offset = start;
2353 2354 2355
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366

	/*
	 * Make sure tasks use at least 32x as much time to run other code
	 * than they used here, to limit NUMA PTE scanning overhead to 3% max.
	 * Usually update_task_scan_period slows down scanning enough; on an
	 * overloaded system we need to limit overhead on a per task basis.
	 */
	if (unlikely(p->se.sum_exec_runtime != runtime)) {
		u64 diff = p->se.sum_exec_runtime - runtime;
		p->node_stamp += 32 * diff;
	}
2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391
}

/*
 * 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;

2392
	if (now > curr->node_stamp + period) {
2393
		if (!curr->node_stamp)
2394
			curr->numa_scan_period = task_scan_min(curr);
2395
		curr->node_stamp += period;
2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406

		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)
{
}
2407 2408 2409 2410 2411 2412 2413 2414

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)
{
}
2415 2416
#endif /* CONFIG_NUMA_BALANCING */

2417 2418 2419 2420
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2421
	if (!parent_entity(se))
2422
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2423
#ifdef CONFIG_SMP
2424 2425 2426 2427 2428 2429
	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);
	}
2430
#endif
2431 2432 2433 2434 2435 2436 2437
	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);
2438
	if (!parent_entity(se))
2439
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2440 2441
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2442
		list_del_init(&se->group_node);
2443
	}
2444 2445 2446
	cfs_rq->nr_running--;
}

2447 2448
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2449 2450 2451 2452 2453
static inline long calc_tg_weight(struct task_group *tg, struct cfs_rq *cfs_rq)
{
	long tg_weight;

	/*
2454 2455 2456
	 * Use this CPU's real-time load instead of the last load contribution
	 * as the updating of the contribution is delayed, and we will use the
	 * the real-time load to calc the share. See update_tg_load_avg().
2457
	 */
2458
	tg_weight = atomic_long_read(&tg->load_avg);
2459
	tg_weight -= cfs_rq->tg_load_avg_contrib;
2460
	tg_weight += cfs_rq->load.weight;
2461 2462 2463 2464

	return tg_weight;
}

2465
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2466
{
2467
	long tg_weight, load, shares;
2468

2469
	tg_weight = calc_tg_weight(tg, cfs_rq);
2470
	load = cfs_rq->load.weight;
2471 2472

	shares = (tg->shares * load);
2473 2474
	if (tg_weight)
		shares /= tg_weight;
2475 2476 2477 2478 2479 2480 2481 2482 2483

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

	return shares;
}
# else /* CONFIG_SMP */
2484
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2485 2486 2487 2488
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
2489 2490 2491
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2492 2493 2494 2495
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2496
		account_entity_dequeue(cfs_rq, se);
2497
	}
P
Peter Zijlstra 已提交
2498 2499 2500 2501 2502 2503 2504

	update_load_set(&se->load, weight);

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

2505 2506
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2507
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2508 2509 2510
{
	struct task_group *tg;
	struct sched_entity *se;
2511
	long shares;
P
Peter Zijlstra 已提交
2512 2513 2514

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2515
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2516
		return;
2517 2518 2519 2520
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2521
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2522 2523 2524 2525

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2526
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2527 2528 2529 2530
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2531
#ifdef CONFIG_SMP
2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551
/* 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,
};

2552 2553 2554 2555 2556 2557
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569
	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
2570 2571
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
2572 2573 2574 2575 2576 2577
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
2578 2579
	}

2580 2581
	val = mul_u64_u32_shr(val, runnable_avg_yN_inv[local_n], 32);
	return val;
2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609
}

/*
 * 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];
2610 2611
}

2612 2613 2614 2615
#if (SCHED_LOAD_SHIFT - SCHED_LOAD_RESOLUTION) != 10 || SCHED_CAPACITY_SHIFT != 10
#error "load tracking assumes 2^10 as unit"
#endif

2616
#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
2617

2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645
/*
 * 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}]
 */
2646 2647
static __always_inline int
__update_load_avg(u64 now, int cpu, struct sched_avg *sa,
2648
		  unsigned long weight, int running, struct cfs_rq *cfs_rq)
2649
{
2650
	u64 delta, scaled_delta, periods;
2651
	u32 contrib;
2652
	unsigned int delta_w, scaled_delta_w, decayed = 0;
2653
	unsigned long scale_freq, scale_cpu;
2654

2655
	delta = now - sa->last_update_time;
2656 2657 2658 2659 2660
	/*
	 * 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) {
2661
		sa->last_update_time = now;
2662 2663 2664 2665 2666 2667 2668 2669 2670 2671
		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;
2672
	sa->last_update_time = now;
2673

2674 2675 2676
	scale_freq = arch_scale_freq_capacity(NULL, cpu);
	scale_cpu = arch_scale_cpu_capacity(NULL, cpu);

2677
	/* delta_w is the amount already accumulated against our next period */
2678
	delta_w = sa->period_contrib;
2679 2680 2681
	if (delta + delta_w >= 1024) {
		decayed = 1;

2682 2683 2684
		/* how much left for next period will start over, we don't know yet */
		sa->period_contrib = 0;

2685 2686 2687 2688 2689 2690
		/*
		 * 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;
2691
		scaled_delta_w = cap_scale(delta_w, scale_freq);
2692
		if (weight) {
2693 2694 2695 2696 2697
			sa->load_sum += weight * scaled_delta_w;
			if (cfs_rq) {
				cfs_rq->runnable_load_sum +=
						weight * scaled_delta_w;
			}
2698
		}
2699
		if (running)
2700
			sa->util_sum += scaled_delta_w * scale_cpu;
2701 2702 2703 2704 2705 2706 2707

		delta -= delta_w;

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

2708
		sa->load_sum = decay_load(sa->load_sum, periods + 1);
2709 2710 2711 2712
		if (cfs_rq) {
			cfs_rq->runnable_load_sum =
				decay_load(cfs_rq->runnable_load_sum, periods + 1);
		}
2713
		sa->util_sum = decay_load((u64)(sa->util_sum), periods + 1);
2714 2715

		/* Efficiently calculate \sum (1..n_period) 1024*y^i */
2716
		contrib = __compute_runnable_contrib(periods);
2717
		contrib = cap_scale(contrib, scale_freq);
2718
		if (weight) {
2719
			sa->load_sum += weight * contrib;
2720 2721 2722
			if (cfs_rq)
				cfs_rq->runnable_load_sum += weight * contrib;
		}
2723
		if (running)
2724
			sa->util_sum += contrib * scale_cpu;
2725 2726 2727
	}

	/* Remainder of delta accrued against u_0` */
2728
	scaled_delta = cap_scale(delta, scale_freq);
2729
	if (weight) {
2730
		sa->load_sum += weight * scaled_delta;
2731
		if (cfs_rq)
2732
			cfs_rq->runnable_load_sum += weight * scaled_delta;
2733
	}
2734
	if (running)
2735
		sa->util_sum += scaled_delta * scale_cpu;
2736

2737
	sa->period_contrib += delta;
2738

2739 2740
	if (decayed) {
		sa->load_avg = div_u64(sa->load_sum, LOAD_AVG_MAX);
2741 2742 2743 2744
		if (cfs_rq) {
			cfs_rq->runnable_load_avg =
				div_u64(cfs_rq->runnable_load_sum, LOAD_AVG_MAX);
		}
2745
		sa->util_avg = sa->util_sum / LOAD_AVG_MAX;
2746
	}
2747

2748
	return decayed;
2749 2750
}

2751
#ifdef CONFIG_FAIR_GROUP_SCHED
2752
/*
2753 2754
 * Updating tg's load_avg is necessary before update_cfs_share (which is done)
 * and effective_load (which is not done because it is too costly).
2755
 */
2756
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
2757
{
2758
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
2759

2760 2761 2762 2763 2764 2765
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

2766 2767 2768
	if (force || abs(delta) > cfs_rq->tg_load_avg_contrib / 64) {
		atomic_long_add(delta, &cfs_rq->tg->load_avg);
		cfs_rq->tg_load_avg_contrib = cfs_rq->avg.load_avg;
2769
	}
2770
}
2771

2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817
/*
 * Called within set_task_rq() right before setting a task's cpu. The
 * caller only guarantees p->pi_lock is held; no other assumptions,
 * including the state of rq->lock, should be made.
 */
void set_task_rq_fair(struct sched_entity *se,
		      struct cfs_rq *prev, struct cfs_rq *next)
{
	if (!sched_feat(ATTACH_AGE_LOAD))
		return;

	/*
	 * We are supposed to update the task to "current" time, then its up to
	 * date and ready to go to new CPU/cfs_rq. But we have difficulty in
	 * getting what current time is, so simply throw away the out-of-date
	 * time. This will result in the wakee task is less decayed, but giving
	 * the wakee more load sounds not bad.
	 */
	if (se->avg.last_update_time && prev) {
		u64 p_last_update_time;
		u64 n_last_update_time;

#ifndef CONFIG_64BIT
		u64 p_last_update_time_copy;
		u64 n_last_update_time_copy;

		do {
			p_last_update_time_copy = prev->load_last_update_time_copy;
			n_last_update_time_copy = next->load_last_update_time_copy;

			smp_rmb();

			p_last_update_time = prev->avg.last_update_time;
			n_last_update_time = next->avg.last_update_time;

		} while (p_last_update_time != p_last_update_time_copy ||
			 n_last_update_time != n_last_update_time_copy);
#else
		p_last_update_time = prev->avg.last_update_time;
		n_last_update_time = next->avg.last_update_time;
#endif
		__update_load_avg(p_last_update_time, cpu_of(rq_of(prev)),
				  &se->avg, 0, 0, NULL);
		se->avg.last_update_time = n_last_update_time;
	}
}
2818
#else /* CONFIG_FAIR_GROUP_SCHED */
2819
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
2820
#endif /* CONFIG_FAIR_GROUP_SCHED */
2821

2822
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
2823

2824 2825
/* Group cfs_rq's load_avg is used for task_h_load and update_cfs_share */
static inline int update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
2826
{
2827
	struct sched_avg *sa = &cfs_rq->avg;
2828
	int decayed, removed = 0;
2829

2830
	if (atomic_long_read(&cfs_rq->removed_load_avg)) {
2831
		s64 r = atomic_long_xchg(&cfs_rq->removed_load_avg, 0);
2832 2833
		sa->load_avg = max_t(long, sa->load_avg - r, 0);
		sa->load_sum = max_t(s64, sa->load_sum - r * LOAD_AVG_MAX, 0);
2834
		removed = 1;
2835
	}
2836

2837 2838 2839
	if (atomic_long_read(&cfs_rq->removed_util_avg)) {
		long r = atomic_long_xchg(&cfs_rq->removed_util_avg, 0);
		sa->util_avg = max_t(long, sa->util_avg - r, 0);
2840
		sa->util_sum = max_t(s32, sa->util_sum - r * LOAD_AVG_MAX, 0);
2841
	}
2842

2843
	decayed = __update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
2844
		scale_load_down(cfs_rq->load.weight), cfs_rq->curr != NULL, cfs_rq);
2845

2846 2847 2848 2849
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
2850

2851
	return decayed || removed;
2852 2853
}

2854 2855
/* Update task and its cfs_rq load average */
static inline void update_load_avg(struct sched_entity *se, int update_tg)
2856
{
2857
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
2858
	u64 now = cfs_rq_clock_task(cfs_rq);
2859
	int cpu = cpu_of(rq_of(cfs_rq));
2860

2861
	/*
2862 2863
	 * Track task load average for carrying it to new CPU after migrated, and
	 * track group sched_entity load average for task_h_load calc in migration
2864
	 */
2865
	__update_load_avg(now, cpu, &se->avg,
2866 2867
			  se->on_rq * scale_load_down(se->load.weight),
			  cfs_rq->curr == se, NULL);
2868

2869 2870
	if (update_cfs_rq_load_avg(now, cfs_rq) && update_tg)
		update_tg_load_avg(cfs_rq, 0);
2871 2872
}

2873 2874
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2875 2876 2877
	if (!sched_feat(ATTACH_AGE_LOAD))
		goto skip_aging;

2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891
	/*
	 * If we got migrated (either between CPUs or between cgroups) we'll
	 * have aged the average right before clearing @last_update_time.
	 */
	if (se->avg.last_update_time) {
		__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
				  &se->avg, 0, 0, NULL);

		/*
		 * XXX: we could have just aged the entire load away if we've been
		 * absent from the fair class for too long.
		 */
	}

2892
skip_aging:
2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911
	se->avg.last_update_time = cfs_rq->avg.last_update_time;
	cfs_rq->avg.load_avg += se->avg.load_avg;
	cfs_rq->avg.load_sum += se->avg.load_sum;
	cfs_rq->avg.util_avg += se->avg.util_avg;
	cfs_rq->avg.util_sum += se->avg.util_sum;
}

static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
			  &se->avg, se->on_rq * scale_load_down(se->load.weight),
			  cfs_rq->curr == se, NULL);

	cfs_rq->avg.load_avg = max_t(long, cfs_rq->avg.load_avg - se->avg.load_avg, 0);
	cfs_rq->avg.load_sum = max_t(s64,  cfs_rq->avg.load_sum - se->avg.load_sum, 0);
	cfs_rq->avg.util_avg = max_t(long, cfs_rq->avg.util_avg - se->avg.util_avg, 0);
	cfs_rq->avg.util_sum = max_t(s32,  cfs_rq->avg.util_sum - se->avg.util_sum, 0);
}

2912 2913 2914
/* Add the load generated by se into cfs_rq's load average */
static inline void
enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
2915
{
2916 2917
	struct sched_avg *sa = &se->avg;
	u64 now = cfs_rq_clock_task(cfs_rq);
2918
	int migrated, decayed;
2919

2920 2921
	migrated = !sa->last_update_time;
	if (!migrated) {
2922
		__update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
2923 2924
			se->on_rq * scale_load_down(se->load.weight),
			cfs_rq->curr == se, NULL);
2925
	}
2926

2927
	decayed = update_cfs_rq_load_avg(now, cfs_rq);
2928

2929 2930 2931
	cfs_rq->runnable_load_avg += sa->load_avg;
	cfs_rq->runnable_load_sum += sa->load_sum;

2932 2933
	if (migrated)
		attach_entity_load_avg(cfs_rq, se);
2934

2935 2936
	if (decayed || migrated)
		update_tg_load_avg(cfs_rq, 0);
2937 2938
}

2939 2940 2941 2942 2943 2944 2945 2946 2947
/* Remove the runnable load generated by se from cfs_rq's runnable load average */
static inline void
dequeue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_avg(se, 1);

	cfs_rq->runnable_load_avg =
		max_t(long, cfs_rq->runnable_load_avg - se->avg.load_avg, 0);
	cfs_rq->runnable_load_sum =
2948
		max_t(s64,  cfs_rq->runnable_load_sum - se->avg.load_sum, 0);
2949 2950
}

2951
#ifndef CONFIG_64BIT
2952 2953
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
2954
	u64 last_update_time_copy;
2955
	u64 last_update_time;
2956

2957 2958 2959 2960 2961
	do {
		last_update_time_copy = cfs_rq->load_last_update_time_copy;
		smp_rmb();
		last_update_time = cfs_rq->avg.last_update_time;
	} while (last_update_time != last_update_time_copy);
2962 2963 2964

	return last_update_time;
}
2965
#else
2966 2967 2968 2969
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
2970 2971
#endif

2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989
/*
 * Task first catches up with cfs_rq, and then subtract
 * itself from the cfs_rq (task must be off the queue now).
 */
void remove_entity_load_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 last_update_time;

	/*
	 * Newly created task or never used group entity should not be removed
	 * from its (source) cfs_rq
	 */
	if (se->avg.last_update_time == 0)
		return;

	last_update_time = cfs_rq_last_update_time(cfs_rq);

2990
	__update_load_avg(last_update_time, cpu_of(rq_of(cfs_rq)), &se->avg, 0, 0, NULL);
2991 2992
	atomic_long_add(se->avg.load_avg, &cfs_rq->removed_load_avg);
	atomic_long_add(se->avg.util_avg, &cfs_rq->removed_util_avg);
2993
}
2994

2995 2996 2997 2998 2999 3000 3001 3002 3003 3004
static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq)
{
	return cfs_rq->runnable_load_avg;
}

static inline unsigned long cfs_rq_load_avg(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.load_avg;
}

3005 3006
static int idle_balance(struct rq *this_rq);

3007 3008
#else /* CONFIG_SMP */

3009 3010 3011
static inline void update_load_avg(struct sched_entity *se, int update_tg) {}
static inline void
enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
3012 3013
static inline void
dequeue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
3014
static inline void remove_entity_load_avg(struct sched_entity *se) {}
3015

3016 3017 3018 3019 3020
static inline void
attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
static inline void
detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}

3021 3022 3023 3024 3025
static inline int idle_balance(struct rq *rq)
{
	return 0;
}

3026
#endif /* CONFIG_SMP */
3027

3028
static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
3029 3030
{
#ifdef CONFIG_SCHEDSTATS
3031 3032 3033 3034 3035
	struct task_struct *tsk = NULL;

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

3036
	if (se->statistics.sleep_start) {
3037
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.sleep_start;
3038 3039 3040 3041

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

3042 3043
		if (unlikely(delta > se->statistics.sleep_max))
			se->statistics.sleep_max = delta;
3044

3045
		se->statistics.sleep_start = 0;
3046
		se->statistics.sum_sleep_runtime += delta;
A
Arjan van de Ven 已提交
3047

3048
		if (tsk) {
3049
			account_scheduler_latency(tsk, delta >> 10, 1);
3050 3051
			trace_sched_stat_sleep(tsk, delta);
		}
3052
	}
3053
	if (se->statistics.block_start) {
3054
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.block_start;
3055 3056 3057 3058

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

3059 3060
		if (unlikely(delta > se->statistics.block_max))
			se->statistics.block_max = delta;
3061

3062
		se->statistics.block_start = 0;
3063
		se->statistics.sum_sleep_runtime += delta;
I
Ingo Molnar 已提交
3064

3065
		if (tsk) {
3066
			if (tsk->in_iowait) {
3067 3068
				se->statistics.iowait_sum += delta;
				se->statistics.iowait_count++;
3069
				trace_sched_stat_iowait(tsk, delta);
3070 3071
			}

3072 3073
			trace_sched_stat_blocked(tsk, delta);

3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084
			/*
			 * 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
3085
		}
3086 3087 3088 3089
	}
#endif
}

P
Peter Zijlstra 已提交
3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102
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
}

3103 3104 3105
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3106
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3107

3108 3109 3110 3111 3112 3113
	/*
	 * 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
3114
	if (initial && sched_feat(START_DEBIT))
3115
		vruntime += sched_vslice(cfs_rq, se);
3116

3117
	/* sleeps up to a single latency don't count. */
3118
	if (!initial) {
3119
		unsigned long thresh = sysctl_sched_latency;
3120

3121 3122 3123 3124 3125 3126
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3127

3128
		vruntime -= thresh;
3129 3130
	}

3131
	/* ensure we never gain time by being placed backwards. */
3132
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3133 3134
}

3135 3136
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156
static inline void check_schedstat_required(void)
{
#ifdef CONFIG_SCHEDSTATS
	if (schedstat_enabled())
		return;

	/* Force schedstat enabled if a dependent tracepoint is active */
	if (trace_sched_stat_wait_enabled()    ||
			trace_sched_stat_sleep_enabled()   ||
			trace_sched_stat_iowait_enabled()  ||
			trace_sched_stat_blocked_enabled() ||
			trace_sched_stat_runtime_enabled())  {
		pr_warn_once("Scheduler tracepoints stat_sleep, stat_iowait, "
			     "stat_blocked and stat_runtime require the "
			     "kernel parameter schedstats=enabled or "
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

3157
static void
3158
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3159
{
3160 3161
	/*
	 * Update the normalized vruntime before updating min_vruntime
3162
	 * through calling update_curr().
3163
	 */
3164
	if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
3165 3166
		se->vruntime += cfs_rq->min_vruntime;

3167
	/*
3168
	 * Update run-time statistics of the 'current'.
3169
	 */
3170
	update_curr(cfs_rq);
3171
	enqueue_entity_load_avg(cfs_rq, se);
3172 3173
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
3174

3175
	if (flags & ENQUEUE_WAKEUP) {
3176
		place_entity(cfs_rq, se, 0);
3177 3178
		if (schedstat_enabled())
			enqueue_sleeper(cfs_rq, se);
I
Ingo Molnar 已提交
3179
	}
3180

3181 3182 3183 3184 3185
	check_schedstat_required();
	if (schedstat_enabled()) {
		update_stats_enqueue(cfs_rq, se);
		check_spread(cfs_rq, se);
	}
3186 3187
	if (se != cfs_rq->curr)
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3188
	se->on_rq = 1;
3189

3190
	if (cfs_rq->nr_running == 1) {
3191
		list_add_leaf_cfs_rq(cfs_rq);
3192 3193
		check_enqueue_throttle(cfs_rq);
	}
3194 3195
}

3196
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3197
{
3198 3199
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3200
		if (cfs_rq->last != se)
3201
			break;
3202 3203

		cfs_rq->last = NULL;
3204 3205
	}
}
P
Peter Zijlstra 已提交
3206

3207 3208 3209 3210
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3211
		if (cfs_rq->next != se)
3212
			break;
3213 3214

		cfs_rq->next = NULL;
3215
	}
P
Peter Zijlstra 已提交
3216 3217
}

3218 3219 3220 3221
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3222
		if (cfs_rq->skip != se)
3223
			break;
3224 3225

		cfs_rq->skip = NULL;
3226 3227 3228
	}
}

P
Peter Zijlstra 已提交
3229 3230
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3231 3232 3233 3234 3235
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
3236 3237 3238

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
3239 3240
}

3241
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3242

3243
static void
3244
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3245
{
3246 3247 3248 3249
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
3250
	dequeue_entity_load_avg(cfs_rq, se);
3251

3252 3253
	if (schedstat_enabled())
		update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
3254

P
Peter Zijlstra 已提交
3255
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3256

3257
	if (se != cfs_rq->curr)
3258
		__dequeue_entity(cfs_rq, se);
3259
	se->on_rq = 0;
3260
	account_entity_dequeue(cfs_rq, se);
3261 3262 3263 3264 3265 3266

	/*
	 * 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.
	 */
3267
	if (!(flags & DEQUEUE_SLEEP))
3268
		se->vruntime -= cfs_rq->min_vruntime;
3269

3270 3271 3272
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

3273
	update_min_vruntime(cfs_rq);
3274
	update_cfs_shares(cfs_rq);
3275 3276 3277 3278 3279
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
3280
static void
I
Ingo Molnar 已提交
3281
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3282
{
3283
	unsigned long ideal_runtime, delta_exec;
3284 3285
	struct sched_entity *se;
	s64 delta;
3286

P
Peter Zijlstra 已提交
3287
	ideal_runtime = sched_slice(cfs_rq, curr);
3288
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
3289
	if (delta_exec > ideal_runtime) {
3290
		resched_curr(rq_of(cfs_rq));
3291 3292 3293 3294 3295
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306
		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;

3307 3308
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
3309

3310 3311
	if (delta < 0)
		return;
3312

3313
	if (delta > ideal_runtime)
3314
		resched_curr(rq_of(cfs_rq));
3315 3316
}

3317
static void
3318
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
3319
{
3320 3321 3322 3323 3324 3325 3326
	/* '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.
		 */
3327 3328
		if (schedstat_enabled())
			update_stats_wait_end(cfs_rq, se);
3329
		__dequeue_entity(cfs_rq, se);
3330
		update_load_avg(se, 1);
3331 3332
	}

3333
	update_stats_curr_start(cfs_rq, se);
3334
	cfs_rq->curr = se;
I
Ingo Molnar 已提交
3335 3336 3337 3338 3339 3340
#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):
	 */
3341
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
3342
		se->statistics.slice_max = max(se->statistics.slice_max,
I
Ingo Molnar 已提交
3343 3344 3345
			se->sum_exec_runtime - se->prev_sum_exec_runtime);
	}
#endif
3346
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
3347 3348
}

3349 3350 3351
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

3352 3353 3354 3355 3356 3357 3358
/*
 * 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
 */
3359 3360
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3361
{
3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372
	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 */
3373

3374 3375 3376 3377 3378
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3379 3380 3381 3382 3383 3384 3385 3386 3387 3388
		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;
		}

3389 3390 3391
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3392

3393 3394 3395 3396 3397 3398
	/*
	 * 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;

3399 3400 3401 3402 3403 3404
	/*
	 * 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;

3405
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3406 3407

	return se;
3408 3409
}

3410
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3411

3412
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3413 3414 3415 3416 3417 3418
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3419
		update_curr(cfs_rq);
3420

3421 3422 3423
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

3424 3425 3426 3427 3428 3429
	if (schedstat_enabled()) {
		check_spread(cfs_rq, prev);
		if (prev->on_rq)
			update_stats_wait_start(cfs_rq, prev);
	}

3430 3431 3432
	if (prev->on_rq) {
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3433
		/* in !on_rq case, update occurred at dequeue */
3434
		update_load_avg(prev, 0);
3435
	}
3436
	cfs_rq->curr = NULL;
3437 3438
}

P
Peter Zijlstra 已提交
3439 3440
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3441 3442
{
	/*
3443
	 * Update run-time statistics of the 'current'.
3444
	 */
3445
	update_curr(cfs_rq);
3446

3447 3448 3449
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3450
	update_load_avg(curr, 1);
3451
	update_cfs_shares(cfs_rq);
3452

P
Peter Zijlstra 已提交
3453 3454 3455 3456 3457
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3458
	if (queued) {
3459
		resched_curr(rq_of(cfs_rq));
3460 3461
		return;
	}
P
Peter Zijlstra 已提交
3462 3463 3464 3465 3466 3467 3468 3469
	/*
	 * 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
3470
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3471
		check_preempt_tick(cfs_rq, curr);
3472 3473
}

3474 3475 3476 3477 3478 3479

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

#ifdef CONFIG_CFS_BANDWIDTH
3480 3481

#ifdef HAVE_JUMP_LABEL
3482
static struct static_key __cfs_bandwidth_used;
3483 3484 3485

static inline bool cfs_bandwidth_used(void)
{
3486
	return static_key_false(&__cfs_bandwidth_used);
3487 3488
}

3489
void cfs_bandwidth_usage_inc(void)
3490
{
3491 3492 3493 3494 3495 3496
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3497 3498 3499 3500 3501 3502 3503
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3504 3505
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3506 3507
#endif /* HAVE_JUMP_LABEL */

3508 3509 3510 3511 3512 3513 3514 3515
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3516 3517 3518 3519 3520 3521

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

P
Paul Turner 已提交
3522 3523 3524 3525 3526 3527 3528
/*
 * 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
 */
3529
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540
{
	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);
}

3541 3542 3543 3544 3545
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3546 3547 3548 3549 3550 3551
/* 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;

3552
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3553 3554
}

3555 3556
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3557 3558 3559
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3560
	u64 amount = 0, min_amount, expires;
3561 3562 3563 3564 3565 3566 3567

	/* 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;
3568
	else {
P
Peter Zijlstra 已提交
3569
		start_cfs_bandwidth(cfs_b);
3570 3571 3572 3573 3574 3575

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3576
	}
P
Paul Turner 已提交
3577
	expires = cfs_b->runtime_expires;
3578 3579 3580
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3581 3582 3583 3584 3585 3586 3587
	/*
	 * 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;
3588 3589

	return cfs_rq->runtime_remaining > 0;
3590 3591
}

P
Paul Turner 已提交
3592 3593 3594 3595 3596
/*
 * 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)
3597
{
P
Paul Turner 已提交
3598 3599 3600
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3601
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3602 3603
		return;

P
Paul Turner 已提交
3604 3605 3606 3607 3608 3609 3610 3611 3612
	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
3613 3614 3615
	 * 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
3616 3617
	 */

3618
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3619 3620 3621 3622 3623 3624 3625 3626
		/* 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;
	}
}

3627
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3628 3629
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3630
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3631 3632 3633
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3634 3635
		return;

3636 3637 3638 3639 3640
	/*
	 * 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))
3641
		resched_curr(rq_of(cfs_rq));
3642 3643
}

3644
static __always_inline
3645
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3646
{
3647
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3648 3649 3650 3651 3652
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3653 3654
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3655
	return cfs_bandwidth_used() && cfs_rq->throttled;
3656 3657
}

3658 3659 3660
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3661
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689
}

/*
 * 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) {
3690
		/* adjust cfs_rq_clock_task() */
3691
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3692
					     cfs_rq->throttled_clock_task;
3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703
	}
#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)];

3704 3705
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3706
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3707 3708 3709 3710 3711
	cfs_rq->throttle_count++;

	return 0;
}

3712
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3713 3714 3715 3716 3717
{
	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
3718
	bool empty;
3719 3720 3721

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

3722
	/* freeze hierarchy runnable averages while throttled */
3723 3724 3725
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742

	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)
3743
		sub_nr_running(rq, task_delta);
3744 3745

	cfs_rq->throttled = 1;
3746
	cfs_rq->throttled_clock = rq_clock(rq);
3747
	raw_spin_lock(&cfs_b->lock);
3748
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
3749

3750 3751 3752 3753 3754
	/*
	 * 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
3755 3756 3757 3758 3759 3760 3761 3762

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

3763 3764 3765
	raw_spin_unlock(&cfs_b->lock);
}

3766
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3767 3768 3769 3770 3771 3772 3773
{
	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;

3774
	se = cfs_rq->tg->se[cpu_of(rq)];
3775 3776

	cfs_rq->throttled = 0;
3777 3778 3779

	update_rq_clock(rq);

3780
	raw_spin_lock(&cfs_b->lock);
3781
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
3782 3783 3784
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

3785 3786 3787
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805
	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)
3806
		add_nr_running(rq, task_delta);
3807 3808 3809

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
3810
		resched_curr(rq);
3811 3812 3813 3814 3815 3816
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
3817 3818
	u64 runtime;
	u64 starting_runtime = remaining;
3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848

	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();

3849
	return starting_runtime - remaining;
3850 3851
}

3852 3853 3854 3855 3856 3857 3858 3859
/*
 * 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)
{
3860
	u64 runtime, runtime_expires;
3861
	int throttled;
3862 3863 3864

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
3865
		goto out_deactivate;
3866

3867
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
3868
	cfs_b->nr_periods += overrun;
3869

3870 3871 3872 3873 3874 3875
	/*
	 * 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
3876 3877 3878

	__refill_cfs_bandwidth_runtime(cfs_b);

3879 3880 3881
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
3882
		return 0;
3883 3884
	}

3885 3886 3887
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

3888 3889 3890
	runtime_expires = cfs_b->runtime_expires;

	/*
3891 3892 3893 3894 3895
	 * 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.
3896
	 */
3897 3898
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
3899 3900 3901 3902 3903 3904 3905
		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);
3906 3907

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
3908
	}
3909

3910 3911 3912 3913 3914 3915 3916
	/*
	 * 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;
3917

3918 3919 3920 3921
	return 0;

out_deactivate:
	return 1;
3922
}
3923

3924 3925 3926 3927 3928 3929 3930
/* 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;

3931 3932 3933 3934
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
3935
 * hrtimer base being cleared by hrtimer_start. In the case of
3936 3937
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962
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
3963 3964 3965
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994
}

/* 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)
{
3995 3996 3997
	if (!cfs_bandwidth_used())
		return;

3998
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013
		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 */
4014 4015 4016
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
4017
		return;
4018
	}
4019

4020
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
4021
		runtime = cfs_b->runtime;
4022

4023 4024 4025 4026 4027 4028 4029 4030 4031 4032
	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)
4033
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4034 4035 4036
	raw_spin_unlock(&cfs_b->lock);
}

4037 4038 4039 4040 4041 4042 4043
/*
 * 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)
{
4044 4045 4046
	if (!cfs_bandwidth_used())
		return;

4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061
	/* 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() */
4062
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4063
{
4064
	if (!cfs_bandwidth_used())
4065
		return false;
4066

4067
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4068
		return false;
4069 4070 4071 4072 4073 4074

	/*
	 * 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))
4075
		return true;
4076 4077

	throttle_cfs_rq(cfs_rq);
4078
	return true;
4079
}
4080 4081 4082 4083 4084

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
4085

4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097
	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;

4098
	raw_spin_lock(&cfs_b->lock);
4099
	for (;;) {
P
Peter Zijlstra 已提交
4100
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4101 4102 4103 4104 4105
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4106 4107
	if (idle)
		cfs_b->period_active = 0;
4108
	raw_spin_unlock(&cfs_b->lock);
4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120

	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
4121
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132
	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
4133
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4134
{
P
Peter Zijlstra 已提交
4135
	lockdep_assert_held(&cfs_b->lock);
4136

P
Peter Zijlstra 已提交
4137 4138 4139 4140 4141
	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);
	}
4142 4143 4144 4145
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4146 4147 4148 4149
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4150 4151 4152 4153
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166
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);
	}
}

4167
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178
{
	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
		 */
4179
		cfs_rq->runtime_remaining = 1;
4180 4181 4182 4183 4184 4185
		/*
		 * 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;

4186 4187 4188 4189 4190 4191
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
4192 4193
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
4194
	return rq_clock_task(rq_of(cfs_rq));
4195 4196
}

4197
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
4198
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
4199
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
4200
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
4201 4202 4203 4204 4205

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216

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;
}
4217 4218 4219 4220 4221

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) {}
4222 4223
#endif

4224 4225 4226 4227 4228
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) {}
4229
static inline void update_runtime_enabled(struct rq *rq) {}
4230
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
4231 4232 4233

#endif /* CONFIG_CFS_BANDWIDTH */

4234 4235 4236 4237
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
4238 4239 4240 4241 4242 4243 4244 4245
#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);

4246
	if (cfs_rq->nr_running > 1) {
P
Peter Zijlstra 已提交
4247 4248 4249 4250 4251 4252
		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)
4253
				resched_curr(rq);
P
Peter Zijlstra 已提交
4254 4255
			return;
		}
4256
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
4257 4258
	}
}
4259 4260 4261 4262 4263 4264 4265 4266 4267 4268

/*
 * 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;

4269
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
4270 4271 4272 4273 4274
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
4275
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
4276 4277 4278 4279
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
4280 4281 4282 4283

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
4284 4285
#endif

4286 4287 4288 4289 4290
/*
 * 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:
 */
4291
static void
4292
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4293 4294
{
	struct cfs_rq *cfs_rq;
4295
	struct sched_entity *se = &p->se;
4296 4297

	for_each_sched_entity(se) {
4298
		if (se->on_rq)
4299 4300
			break;
		cfs_rq = cfs_rq_of(se);
4301
		enqueue_entity(cfs_rq, se, flags);
4302 4303 4304 4305 4306 4307 4308 4309 4310

		/*
		 * 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;
4311
		cfs_rq->h_nr_running++;
4312

4313
		flags = ENQUEUE_WAKEUP;
4314
	}
P
Peter Zijlstra 已提交
4315

P
Peter Zijlstra 已提交
4316
	for_each_sched_entity(se) {
4317
		cfs_rq = cfs_rq_of(se);
4318
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
4319

4320 4321 4322
		if (cfs_rq_throttled(cfs_rq))
			break;

4323
		update_load_avg(se, 1);
4324
		update_cfs_shares(cfs_rq);
P
Peter Zijlstra 已提交
4325 4326
	}

Y
Yuyang Du 已提交
4327
	if (!se)
4328
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
4329

4330
	hrtick_update(rq);
4331 4332
}

4333 4334
static void set_next_buddy(struct sched_entity *se);

4335 4336 4337 4338 4339
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
4340
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4341 4342
{
	struct cfs_rq *cfs_rq;
4343
	struct sched_entity *se = &p->se;
4344
	int task_sleep = flags & DEQUEUE_SLEEP;
4345 4346 4347

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4348
		dequeue_entity(cfs_rq, se, flags);
4349 4350 4351 4352 4353 4354 4355 4356 4357

		/*
		 * 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;
4358
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4359

4360
		/* Don't dequeue parent if it has other entities besides us */
4361 4362 4363 4364 4365 4366 4367
		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));
4368 4369 4370

			/* avoid re-evaluating load for this entity */
			se = parent_entity(se);
4371
			break;
4372
		}
4373
		flags |= DEQUEUE_SLEEP;
4374
	}
P
Peter Zijlstra 已提交
4375

P
Peter Zijlstra 已提交
4376
	for_each_sched_entity(se) {
4377
		cfs_rq = cfs_rq_of(se);
4378
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4379

4380 4381 4382
		if (cfs_rq_throttled(cfs_rq))
			break;

4383
		update_load_avg(se, 1);
4384
		update_cfs_shares(cfs_rq);
P
Peter Zijlstra 已提交
4385 4386
	}

Y
Yuyang Du 已提交
4387
	if (!se)
4388
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
4389

4390
	hrtick_update(rq);
4391 4392
}

4393
#ifdef CONFIG_SMP
4394 4395 4396 4397 4398 4399

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

/*
4400
 * The exact cpuload calculated at every tick would be:
4401
 *
4402 4403 4404 4405 4406 4407 4408
 *   load' = (1 - 1/2^i) * load + (1/2^i) * cur_load
 *
 * If a cpu misses updates for n ticks (as it was idle) and update gets
 * called on the n+1-th tick when cpu may be busy, then we have:
 *
 *   load_n   = (1 - 1/2^i)^n * load_0
 *   load_n+1 = (1 - 1/2^i)   * load_n + (1/2^i) * cur_load
4409 4410 4411
 *
 * decay_load_missed() below does efficient calculation of
 *
4412 4413 4414 4415 4416 4417
 *   load' = (1 - 1/2^i)^n * load
 *
 * Because x^(n+m) := x^n * x^m we can decompose any x^n in power-of-2 factors.
 * This allows us to precompute the above in said factors, thereby allowing the
 * reduction of an arbitrary n in O(log_2 n) steps. (See also
 * fixed_power_int())
4418
 *
4419
 * The calculation is approximated on a 128 point scale.
4420 4421
 */
#define DEGRADE_SHIFT		7
4422 4423 4424 4425 4426 4427 4428 4429 4430

static const u8 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
static const u8 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, 0 },
	{ 112,  98, 75, 43, 15,  1, 0, 0 },
	{ 120, 112, 98, 76, 45, 16, 2, 0 }
};
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

/*
 * 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;
}

4461 4462 4463 4464 4465 4466 4467
/**
 * __update_cpu_load - update the rq->cpu_load[] statistics
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 * @active: !0 for NOHZ_FULL
 *
4468
 * Update rq->cpu_load[] statistics. This function is usually called every
4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495
 * scheduler tick (TICK_NSEC).
 *
 * This function computes a decaying average:
 *
 *   load[i]' = (1 - 1/2^i) * load[i] + (1/2^i) * load
 *
 * Because of NOHZ it might not get called on every tick which gives need for
 * the @pending_updates argument.
 *
 *   load[i]_n = (1 - 1/2^i) * load[i]_n-1 + (1/2^i) * load_n-1
 *             = A * load[i]_n-1 + B ; A := (1 - 1/2^i), B := (1/2^i) * load
 *             = A * (A * load[i]_n-2 + B) + B
 *             = A * (A * (A * load[i]_n-3 + B) + B) + B
 *             = A^3 * load[i]_n-3 + (A^2 + A + 1) * B
 *             = A^n * load[i]_0 + (A^(n-1) + A^(n-2) + ... + 1) * B
 *             = A^n * load[i]_0 + ((1 - A^n) / (1 - A)) * B
 *             = (1 - 1/2^i)^n * (load[i]_0 - load) + load
 *
 * In the above we've assumed load_n := load, which is true for NOHZ_FULL as
 * any change in load would have resulted in the tick being turned back on.
 *
 * For regular NOHZ, this reduces to:
 *
 *   load[i]_n = (1 - 1/2^i)^n * load[i]_0
 *
 * see decay_load_misses(). For NOHZ_FULL we get to subtract and add the extra
 * term. See the @active paramter.
4496 4497
 */
static void __update_cpu_load(struct rq *this_rq, unsigned long this_load,
4498
			      unsigned long pending_updates, int active)
4499
{
4500
	unsigned long tickless_load = active ? this_rq->cpu_load[0] : 0;
4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511
	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 */

4512
		old_load = this_rq->cpu_load[i] - tickless_load;
4513
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
4514
		old_load += tickless_load;
4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529
		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);
}

4530 4531 4532 4533 4534 4535
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
	return cfs_rq_runnable_load_avg(&cpu_rq(cpu)->cfs);
}

4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555
#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)
{
4556
	unsigned long curr_jiffies = READ_ONCE(jiffies);
4557
	unsigned long load = weighted_cpuload(cpu_of(this_rq));
4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568
	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;

4569
	__update_cpu_load(this_rq, load, pending_updates, 0);
4570 4571 4572 4573 4574
}

/*
 * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed.
 */
4575
void update_cpu_load_nohz(int active)
4576 4577
{
	struct rq *this_rq = this_rq();
4578
	unsigned long curr_jiffies = READ_ONCE(jiffies);
4579
	unsigned long load = active ? weighted_cpuload(cpu_of(this_rq)) : 0;
4580 4581 4582 4583 4584 4585 4586 4587 4588 4589
	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;
		/*
4590 4591 4592
		 * In the regular NOHZ case, we were idle, this means load 0.
		 * In the NOHZ_FULL case, we were non-idle, we should consider
		 * its weighted load.
4593
		 */
4594
		__update_cpu_load(this_rq, load, pending_updates, active);
4595 4596 4597 4598 4599 4600 4601 4602 4603 4604
	}
	raw_spin_unlock(&this_rq->lock);
}
#endif /* CONFIG_NO_HZ */

/*
 * Called from scheduler_tick()
 */
void update_cpu_load_active(struct rq *this_rq)
{
4605
	unsigned long load = weighted_cpuload(cpu_of(this_rq));
4606 4607 4608 4609
	/*
	 * See the mess around update_idle_cpu_load() / update_cpu_load_nohz().
	 */
	this_rq->last_load_update_tick = jiffies;
4610
	__update_cpu_load(this_rq, load, 1, 1);
4611 4612
}

4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645
/*
 * 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);
}

4646
static unsigned long capacity_of(int cpu)
4647
{
4648
	return cpu_rq(cpu)->cpu_capacity;
4649 4650
}

4651 4652 4653 4654 4655
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

4656 4657 4658
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
4659
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
4660
	unsigned long load_avg = weighted_cpuload(cpu);
4661 4662

	if (nr_running)
4663
		return load_avg / nr_running;
4664 4665 4666 4667

	return 0;
}

4668 4669 4670 4671 4672 4673 4674
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.
	 */
4675
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
4676
		current->wakee_flips >>= 1;
4677 4678 4679 4680 4681 4682 4683 4684
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}
4685

4686
static void task_waking_fair(struct task_struct *p)
4687 4688 4689
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
4690 4691 4692 4693
	u64 min_vruntime;

#ifndef CONFIG_64BIT
	u64 min_vruntime_copy;
4694

4695 4696 4697 4698 4699 4700 4701 4702
	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
4703

4704
	se->vruntime -= min_vruntime;
4705
	record_wakee(p);
4706 4707
}

4708
#ifdef CONFIG_FAIR_GROUP_SCHED
4709 4710 4711 4712 4713 4714
/*
 * 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.
4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757
 *
 * 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.
4758
 */
P
Peter Zijlstra 已提交
4759
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
4760
{
P
Peter Zijlstra 已提交
4761
	struct sched_entity *se = tg->se[cpu];
4762

4763
	if (!tg->parent)	/* the trivial, non-cgroup case */
4764 4765
		return wl;

P
Peter Zijlstra 已提交
4766
	for_each_sched_entity(se) {
4767
		long w, W;
P
Peter Zijlstra 已提交
4768

4769
		tg = se->my_q->tg;
4770

4771 4772 4773 4774
		/*
		 * W = @wg + \Sum rw_j
		 */
		W = wg + calc_tg_weight(tg, se->my_q);
P
Peter Zijlstra 已提交
4775

4776 4777 4778
		/*
		 * w = rw_i + @wl
		 */
4779
		w = cfs_rq_load_avg(se->my_q) + wl;
4780

4781 4782 4783 4784
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
4785
			wl = (w * (long)tg->shares) / W;
4786 4787
		else
			wl = tg->shares;
4788

4789 4790 4791 4792 4793
		/*
		 * 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().
		 */
4794 4795
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
4796 4797 4798 4799

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
4800
		wl -= se->avg.load_avg;
4801 4802 4803 4804 4805 4806 4807 4808

		/*
		 * 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
4809 4810
		wg = 0;
	}
4811

P
Peter Zijlstra 已提交
4812
	return wl;
4813 4814
}
#else
P
Peter Zijlstra 已提交
4815

4816
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
4817
{
4818
	return wl;
4819
}
P
Peter Zijlstra 已提交
4820

4821 4822
#endif

M
Mike Galbraith 已提交
4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
 * A waker of many should wake a different task than the one last awakened
 * at a frequency roughly N times higher than one of its wakees.  In order
 * to determine whether we should let the load spread vs consolodating to
 * shared cache, we look for a minimum 'flip' frequency of llc_size in one
 * partner, and a factor of lls_size higher frequency in the other.  With
 * both conditions met, we can be relatively sure that the relationship is
 * non-monogamous, with partner count exceeding socket size.  Waker/wakee
 * being client/server, worker/dispatcher, interrupt source or whatever is
 * irrelevant, spread criteria is apparent partner count exceeds socket size.
 */
4835 4836
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
4837 4838
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
4839
	int factor = this_cpu_read(sd_llc_size);
4840

M
Mike Galbraith 已提交
4841 4842 4843 4844 4845
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
4846 4847
}

4848
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
4849
{
4850
	s64 this_load, load;
4851
	s64 this_eff_load, prev_eff_load;
4852 4853
	int idx, this_cpu, prev_cpu;
	struct task_group *tg;
4854
	unsigned long weight;
4855
	int balanced;
4856

4857 4858 4859 4860 4861
	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);
4862

4863 4864 4865 4866 4867
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
4868 4869
	if (sync) {
		tg = task_group(current);
4870
		weight = current->se.avg.load_avg;
4871

4872
		this_load += effective_load(tg, this_cpu, -weight, -weight);
4873 4874
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
4875

4876
	tg = task_group(p);
4877
	weight = p->se.avg.load_avg;
4878

4879 4880
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
4881 4882 4883
	 * 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.
4884 4885 4886 4887
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
4888 4889
	this_eff_load = 100;
	this_eff_load *= capacity_of(prev_cpu);
4890

4891 4892
	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
4893

4894
	if (this_load > 0) {
4895 4896 4897 4898
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
4899
	}
4900

4901
	balanced = this_eff_load <= prev_eff_load;
4902

4903
	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
4904

4905 4906
	if (!balanced)
		return 0;
4907

4908 4909 4910 4911
	schedstat_inc(sd, ttwu_move_affine);
	schedstat_inc(p, se.statistics.nr_wakeups_affine);

	return 1;
4912 4913
}

4914 4915 4916 4917 4918
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
4919
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
4920
		  int this_cpu, int sd_flag)
4921
{
4922
	struct sched_group *idlest = NULL, *group = sd->groups;
4923
	unsigned long min_load = ULONG_MAX, this_load = 0;
4924
	int load_idx = sd->forkexec_idx;
4925
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
4926

4927 4928 4929
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

4930 4931 4932 4933
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
4934

4935 4936
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
4937
					tsk_cpus_allowed(p)))
4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955
			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;
		}

4956
		/* Adjust by relative CPU capacity of the group */
4957
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978

		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;
4979 4980 4981 4982
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
4983 4984 4985
	int i;

	/* Traverse only the allowed CPUs */
4986
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008
		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;
			}
5009
		} else if (shallowest_idle_cpu == -1) {
5010 5011 5012 5013 5014
			load = weighted_cpuload(i);
			if (load < min_load || (load == min_load && i == this_cpu)) {
				min_load = load;
				least_loaded_cpu = i;
			}
5015 5016 5017
		}
	}

5018
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
5019
}
5020

5021 5022 5023
/*
 * Try and locate an idle CPU in the sched_domain.
 */
5024
static int select_idle_sibling(struct task_struct *p, int target)
5025
{
5026
	struct sched_domain *sd;
5027
	struct sched_group *sg;
5028
	int i = task_cpu(p);
5029

5030 5031
	if (idle_cpu(target))
		return target;
5032 5033

	/*
5034
	 * If the prevous cpu is cache affine and idle, don't be stupid.
5035
	 */
5036 5037
	if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
		return i;
5038 5039

	/*
5040
	 * Otherwise, iterate the domains and find an elegible idle cpu.
5041
	 */
5042
	sd = rcu_dereference(per_cpu(sd_llc, target));
5043
	for_each_lower_domain(sd) {
5044 5045 5046 5047 5048 5049 5050
		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)) {
5051
				if (i == target || !idle_cpu(i))
5052 5053
					goto next;
			}
5054

5055 5056 5057 5058 5059 5060 5061 5062
			target = cpumask_first_and(sched_group_cpus(sg),
					tsk_cpus_allowed(p));
			goto done;
next:
			sg = sg->next;
		} while (sg != sd->groups);
	}
done:
5063 5064
	return target;
}
5065

5066
/*
5067
 * cpu_util returns the amount of capacity of a CPU that is used by CFS
5068
 * tasks. The unit of the return value must be the one of capacity so we can
5069 5070
 * compare the utilization with the capacity of the CPU that is available for
 * CFS task (ie cpu_capacity).
5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090
 *
 * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the
 * recent utilization of currently non-runnable tasks on a CPU. It represents
 * the amount of utilization of a CPU in the range [0..capacity_orig] where
 * capacity_orig is the cpu_capacity available at the highest frequency
 * (arch_scale_freq_capacity()).
 * The utilization of a CPU converges towards a sum equal to or less than the
 * current capacity (capacity_curr <= capacity_orig) of the CPU because it is
 * the running time on this CPU scaled by capacity_curr.
 *
 * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even
 * higher than capacity_orig because of unfortunate rounding in
 * cfs.avg.util_avg or just after migrating tasks and new task wakeups until
 * the average stabilizes with the new running time. We need to check that the
 * utilization stays within the range of [0..capacity_orig] and cap it if
 * necessary. Without utilization capping, a group could be seen as overloaded
 * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of
 * available capacity. We allow utilization to overshoot capacity_curr (but not
 * capacity_orig) as it useful for predicting the capacity required after task
 * migrations (scheduler-driven DVFS).
5091
 */
5092
static int cpu_util(int cpu)
5093
{
5094
	unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
5095 5096
	unsigned long capacity = capacity_orig_of(cpu);

5097
	return (util >= capacity) ? capacity : util;
5098
}
5099

5100
/*
5101 5102 5103
 * 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.
5104
 *
5105 5106
 * 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.
5107
 *
5108
 * Returns the target cpu number.
5109 5110 5111
 *
 * preempt must be disabled.
 */
5112
static int
5113
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
5114
{
5115
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
5116
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
5117
	int new_cpu = prev_cpu;
5118
	int want_affine = 0;
5119
	int sync = wake_flags & WF_SYNC;
5120

5121
	if (sd_flag & SD_BALANCE_WAKE)
M
Mike Galbraith 已提交
5122
		want_affine = !wake_wide(p) && cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
5123

5124
	rcu_read_lock();
5125
	for_each_domain(cpu, tmp) {
5126
		if (!(tmp->flags & SD_LOAD_BALANCE))
M
Mike Galbraith 已提交
5127
			break;
5128

5129
		/*
5130 5131
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
5132
		 */
5133 5134 5135
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
5136
			break;
5137
		}
5138

5139
		if (tmp->flags & sd_flag)
5140
			sd = tmp;
M
Mike Galbraith 已提交
5141 5142
		else if (!want_affine)
			break;
5143 5144
	}

M
Mike Galbraith 已提交
5145 5146 5147 5148
	if (affine_sd) {
		sd = NULL; /* Prefer wake_affine over balance flags */
		if (cpu != prev_cpu && wake_affine(affine_sd, p, sync))
			new_cpu = cpu;
5149
	}
5150

M
Mike Galbraith 已提交
5151 5152 5153 5154 5155
	if (!sd) {
		if (sd_flag & SD_BALANCE_WAKE) /* XXX always ? */
			new_cpu = select_idle_sibling(p, new_cpu);

	} else while (sd) {
5156
		struct sched_group *group;
5157
		int weight;
5158

5159
		if (!(sd->flags & sd_flag)) {
5160 5161 5162
			sd = sd->child;
			continue;
		}
5163

5164
		group = find_idlest_group(sd, p, cpu, sd_flag);
5165 5166 5167 5168
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
5169

5170
		new_cpu = find_idlest_cpu(group, p, cpu);
5171 5172 5173 5174
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
5175
		}
5176 5177 5178

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
5179
		weight = sd->span_weight;
5180 5181
		sd = NULL;
		for_each_domain(cpu, tmp) {
5182
			if (weight <= tmp->span_weight)
5183
				break;
5184
			if (tmp->flags & sd_flag)
5185 5186 5187
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
5188
	}
5189
	rcu_read_unlock();
5190

5191
	return new_cpu;
5192
}
5193 5194 5195 5196

/*
 * 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
5197
 * previous cpu. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
5198
 */
5199
static void migrate_task_rq_fair(struct task_struct *p)
5200
{
5201
	/*
5202 5203 5204 5205 5206
	 * We are supposed to update the task to "current" time, then its up to date
	 * and ready to go to new CPU/cfs_rq. But we have difficulty in getting
	 * what current time is, so simply throw away the out-of-date time. This
	 * will result in the wakee task is less decayed, but giving the wakee more
	 * load sounds not bad.
5207
	 */
5208 5209 5210 5211
	remove_entity_load_avg(&p->se);

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
5212 5213

	/* We have migrated, no longer consider this task hot */
5214
	p->se.exec_start = 0;
5215
}
5216 5217 5218 5219 5220

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
5221 5222
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
5223 5224
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
5225 5226 5227 5228
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
5229 5230
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
5231 5232 5233 5234 5235 5236 5237 5238 5239
	 *
	 * 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.
5240
	 */
5241
	return calc_delta_fair(gran, se);
5242 5243
}

5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265
/*
 * 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
5266
	gran = wakeup_gran(curr, se);
5267 5268 5269 5270 5271 5272
	if (vdiff > gran)
		return 1;

	return 0;
}

5273 5274
static void set_last_buddy(struct sched_entity *se)
{
5275 5276 5277 5278 5279
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
5280 5281 5282 5283
}

static void set_next_buddy(struct sched_entity *se)
{
5284 5285 5286 5287 5288
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
5289 5290
}

5291 5292
static void set_skip_buddy(struct sched_entity *se)
{
5293 5294
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
5295 5296
}

5297 5298 5299
/*
 * Preempt the current task with a newly woken task if needed:
 */
5300
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
5301 5302
{
	struct task_struct *curr = rq->curr;
5303
	struct sched_entity *se = &curr->se, *pse = &p->se;
5304
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
5305
	int scale = cfs_rq->nr_running >= sched_nr_latency;
5306
	int next_buddy_marked = 0;
5307

I
Ingo Molnar 已提交
5308 5309 5310
	if (unlikely(se == pse))
		return;

5311
	/*
5312
	 * This is possible from callers such as attach_tasks(), in which we
5313 5314 5315 5316 5317 5318 5319
	 * 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;

5320
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
5321
		set_next_buddy(pse);
5322 5323
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
5324

5325 5326 5327
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
5328 5329 5330 5331 5332 5333
	 *
	 * 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.
5334 5335 5336 5337
	 */
	if (test_tsk_need_resched(curr))
		return;

5338 5339 5340 5341 5342
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

5343
	/*
5344 5345
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
5346
	 */
5347
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
5348
		return;
5349

5350
	find_matching_se(&se, &pse);
5351
	update_curr(cfs_rq_of(se));
5352
	BUG_ON(!pse);
5353 5354 5355 5356 5357 5358 5359
	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);
5360
		goto preempt;
5361
	}
5362

5363
	return;
5364

5365
preempt:
5366
	resched_curr(rq);
5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380
	/*
	 * 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);
5381 5382
}

5383 5384
static struct task_struct *
pick_next_task_fair(struct rq *rq, struct task_struct *prev)
5385 5386 5387
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
5388
	struct task_struct *p;
5389
	int new_tasks;
5390

5391
again:
5392 5393
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
5394
		goto idle;
5395

5396
	if (prev->sched_class != &fair_sched_class)
5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415
		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.
		 */
5416 5417 5418 5419 5420
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
5421

5422 5423 5424 5425 5426 5427 5428 5429 5430
			/*
			 * 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;
		}
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 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470

		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
5471

5472
	if (!cfs_rq->nr_running)
5473
		goto idle;
5474

5475
	put_prev_task(rq, prev);
5476

5477
	do {
5478
		se = pick_next_entity(cfs_rq, NULL);
5479
		set_next_entity(cfs_rq, se);
5480 5481 5482
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
5483
	p = task_of(se);
5484

5485 5486
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
5487 5488

	return p;
5489 5490

idle:
5491 5492 5493 5494 5495 5496 5497
	/*
	 * 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);
5498
	new_tasks = idle_balance(rq);
5499
	lockdep_pin_lock(&rq->lock);
5500 5501 5502 5503 5504
	/*
	 * 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.
	 */
5505
	if (new_tasks < 0)
5506 5507
		return RETRY_TASK;

5508
	if (new_tasks > 0)
5509 5510 5511
		goto again;

	return NULL;
5512 5513 5514 5515 5516
}

/*
 * Account for a descheduled task:
 */
5517
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
5518 5519 5520 5521 5522 5523
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5524
		put_prev_entity(cfs_rq, se);
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
/*
 * 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);
5553 5554 5555 5556 5557
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
5558
		rq_clock_skip_update(rq, true);
5559 5560 5561 5562 5563
	}

	set_skip_buddy(se);
}

5564 5565 5566 5567
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

5568 5569
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
5570 5571 5572 5573 5574 5575 5576 5577 5578 5579
		return false;

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

	yield_task_fair(rq);

	return true;
}

5580
#ifdef CONFIG_SMP
5581
/**************************************************
P
Peter Zijlstra 已提交
5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604
 * 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)
 *
5605
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
5606 5607 5608 5609 5610 5611
 * 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):
 *
5612
 *   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
5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697
 *
 * 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.]
 */ 
5698

5699 5700
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

5701 5702
enum fbq_type { regular, remote, all };

5703
#define LBF_ALL_PINNED	0x01
5704
#define LBF_NEED_BREAK	0x02
5705 5706
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
5707 5708 5709 5710 5711

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
5712
	int			src_cpu;
5713 5714 5715 5716

	int			dst_cpu;
	struct rq		*dst_rq;

5717 5718
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
5719
	enum cpu_idle_type	idle;
5720
	long			imbalance;
5721 5722 5723
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

5724
	unsigned int		flags;
5725 5726 5727 5728

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
5729 5730

	enum fbq_type		fbq_type;
5731
	struct list_head	tasks;
5732 5733
};

5734 5735 5736
/*
 * Is this task likely cache-hot:
 */
5737
static int task_hot(struct task_struct *p, struct lb_env *env)
5738 5739 5740
{
	s64 delta;

5741 5742
	lockdep_assert_held(&env->src_rq->lock);

5743 5744 5745 5746 5747 5748 5749 5750 5751
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
5752
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
5753 5754 5755 5756 5757 5758 5759 5760 5761
			(&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;

5762
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
5763 5764 5765 5766

	return delta < (s64)sysctl_sched_migration_cost;
}

5767
#ifdef CONFIG_NUMA_BALANCING
5768
/*
5769 5770 5771
 * Returns 1, if task migration degrades locality
 * Returns 0, if task migration improves locality i.e migration preferred.
 * Returns -1, if task migration is not affected by locality.
5772
 */
5773
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
5774
{
5775
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
5776
	unsigned long src_faults, dst_faults;
5777 5778
	int src_nid, dst_nid;

5779
	if (!static_branch_likely(&sched_numa_balancing))
5780 5781
		return -1;

5782
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
5783
		return -1;
5784 5785 5786 5787

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

5788
	if (src_nid == dst_nid)
5789
		return -1;
5790

5791 5792 5793 5794 5795 5796 5797
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid) {
		if (env->src_rq->nr_running > env->src_rq->nr_preferred_running)
			return 1;
		else
			return -1;
	}
5798

5799 5800
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
5801
		return 0;
5802

5803 5804 5805 5806 5807 5808
	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);
5809 5810
	}

5811
	return dst_faults < src_faults;
5812 5813
}

5814
#else
5815
static inline int migrate_degrades_locality(struct task_struct *p,
5816 5817
					     struct lb_env *env)
{
5818
	return -1;
5819
}
5820 5821
#endif

5822 5823 5824 5825
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
5826
int can_migrate_task(struct task_struct *p, struct lb_env *env)
5827
{
5828
	int tsk_cache_hot;
5829 5830 5831

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

5832 5833
	/*
	 * We do not migrate tasks that are:
5834
	 * 1) throttled_lb_pair, or
5835
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
5836 5837
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
5838
	 */
5839 5840 5841
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

5842
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
5843
		int cpu;
5844

5845
		schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
5846

5847 5848
		env->flags |= LBF_SOME_PINNED;

5849 5850 5851 5852 5853 5854 5855 5856
		/*
		 * 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.
		 */
5857
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
5858 5859
			return 0;

5860 5861 5862
		/* 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))) {
5863
				env->flags |= LBF_DST_PINNED;
5864 5865 5866
				env->new_dst_cpu = cpu;
				break;
			}
5867
		}
5868

5869 5870
		return 0;
	}
5871 5872

	/* Record that we found atleast one task that could run on dst_cpu */
5873
	env->flags &= ~LBF_ALL_PINNED;
5874

5875
	if (task_running(env->src_rq, p)) {
5876
		schedstat_inc(p, se.statistics.nr_failed_migrations_running);
5877 5878 5879 5880 5881
		return 0;
	}

	/*
	 * Aggressive migration if:
5882 5883 5884
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
5885
	 */
5886 5887 5888
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
5889

5890
	if (tsk_cache_hot <= 0 ||
5891
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
5892
		if (tsk_cache_hot == 1) {
5893 5894 5895
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
			schedstat_inc(p, se.statistics.nr_forced_migrations);
		}
5896 5897 5898
		return 1;
	}

Z
Zhang Hang 已提交
5899 5900
	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
	return 0;
5901 5902
}

5903
/*
5904 5905 5906 5907 5908 5909 5910
 * 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);

	p->on_rq = TASK_ON_RQ_MIGRATING;
5911
	deactivate_task(env->src_rq, p, 0);
5912 5913 5914
	set_task_cpu(p, env->dst_cpu);
}

5915
/*
5916
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
5917 5918
 * part of active balancing operations within "domain".
 *
5919
 * Returns a task if successful and NULL otherwise.
5920
 */
5921
static struct task_struct *detach_one_task(struct lb_env *env)
5922 5923 5924
{
	struct task_struct *p, *n;

5925 5926
	lockdep_assert_held(&env->src_rq->lock);

5927 5928 5929
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
5930

5931
		detach_task(p, env);
5932

5933
		/*
5934
		 * Right now, this is only the second place where
5935
		 * lb_gained[env->idle] is updated (other is detach_tasks)
5936
		 * so we can safely collect stats here rather than
5937
		 * inside detach_tasks().
5938 5939
		 */
		schedstat_inc(env->sd, lb_gained[env->idle]);
5940
		return p;
5941
	}
5942
	return NULL;
5943 5944
}

5945 5946
static const unsigned int sched_nr_migrate_break = 32;

5947
/*
5948 5949
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
5950
 *
5951
 * Returns number of detached tasks if successful and 0 otherwise.
5952
 */
5953
static int detach_tasks(struct lb_env *env)
5954
{
5955 5956
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
5957
	unsigned long load;
5958 5959 5960
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
5961

5962
	if (env->imbalance <= 0)
5963
		return 0;
5964

5965
	while (!list_empty(tasks)) {
5966 5967 5968 5969 5970 5971 5972
		/*
		 * We don't want to steal all, otherwise we may be treated likewise,
		 * which could at worst lead to a livelock crash.
		 */
		if (env->idle != CPU_NOT_IDLE && env->src_rq->nr_running <= 1)
			break;

5973
		p = list_first_entry(tasks, struct task_struct, se.group_node);
5974

5975 5976
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
5977
		if (env->loop > env->loop_max)
5978
			break;
5979 5980

		/* take a breather every nr_migrate tasks */
5981
		if (env->loop > env->loop_break) {
5982
			env->loop_break += sched_nr_migrate_break;
5983
			env->flags |= LBF_NEED_BREAK;
5984
			break;
5985
		}
5986

5987
		if (!can_migrate_task(p, env))
5988 5989 5990
			goto next;

		load = task_h_load(p);
5991

5992
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
5993 5994
			goto next;

5995
		if ((load / 2) > env->imbalance)
5996
			goto next;
5997

5998 5999 6000 6001
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
6002
		env->imbalance -= load;
6003 6004

#ifdef CONFIG_PREEMPT
6005 6006
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
6007
		 * kernels will stop after the first task is detached to minimize
6008 6009
		 * the critical section.
		 */
6010
		if (env->idle == CPU_NEWLY_IDLE)
6011
			break;
6012 6013
#endif

6014 6015 6016 6017
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
6018
		if (env->imbalance <= 0)
6019
			break;
6020 6021 6022

		continue;
next:
6023
		list_move_tail(&p->se.group_node, tasks);
6024
	}
6025

6026
	/*
6027 6028 6029
	 * 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().
6030
	 */
6031
	schedstat_add(env->sd, lb_gained[env->idle], detached);
6032

6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044
	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);
	activate_task(rq, p, 0);
6045
	p->on_rq = TASK_ON_RQ_QUEUED;
6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073
	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);
6074

6075 6076 6077 6078
		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
6079 6080
}

P
Peter Zijlstra 已提交
6081
#ifdef CONFIG_FAIR_GROUP_SCHED
6082
static void update_blocked_averages(int cpu)
6083 6084
{
	struct rq *rq = cpu_rq(cpu);
6085 6086
	struct cfs_rq *cfs_rq;
	unsigned long flags;
6087

6088 6089
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
6090

6091 6092 6093 6094
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
6095
	for_each_leaf_cfs_rq(rq, cfs_rq) {
6096 6097 6098
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
6099

6100 6101 6102
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq))
			update_tg_load_avg(cfs_rq, 0);
	}
6103
	raw_spin_unlock_irqrestore(&rq->lock, flags);
6104 6105
}

6106
/*
6107
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
6108 6109 6110
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
6111
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
6112
{
6113 6114
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
6115
	unsigned long now = jiffies;
6116
	unsigned long load;
6117

6118
	if (cfs_rq->last_h_load_update == now)
6119 6120
		return;

6121 6122 6123 6124 6125 6126 6127
	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;
	}
6128

6129
	if (!se) {
6130
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
6131 6132 6133 6134 6135
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
6136 6137
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
6138 6139 6140 6141
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
6142 6143
}

6144
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
6145
{
6146
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
6147

6148
	update_cfs_rq_h_load(cfs_rq);
6149
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
6150
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
6151 6152
}
#else
6153
static inline void update_blocked_averages(int cpu)
6154
{
6155 6156 6157 6158 6159 6160 6161 6162
	struct rq *rq = cpu_rq(cpu);
	struct cfs_rq *cfs_rq = &rq->cfs;
	unsigned long flags;

	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq);
	raw_spin_unlock_irqrestore(&rq->lock, flags);
6163 6164
}

6165
static unsigned long task_h_load(struct task_struct *p)
6166
{
6167
	return p->se.avg.load_avg;
6168
}
P
Peter Zijlstra 已提交
6169
#endif
6170 6171

/********** Helpers for find_busiest_group ************************/
6172 6173 6174 6175 6176 6177 6178

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

6179 6180 6181 6182 6183 6184 6185
/*
 * 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
6186
	unsigned long load_per_task;
6187
	unsigned long group_capacity;
6188
	unsigned long group_util; /* Total utilization of the group */
6189 6190 6191
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
6192
	enum group_type group_type;
6193
	int group_no_capacity;
6194 6195 6196 6197
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
6198 6199
};

J
Joonsoo Kim 已提交
6200 6201 6202 6203 6204 6205 6206 6207
/*
 * 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 */
6208
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
6209 6210 6211
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
6212
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
6213 6214
};

6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226
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,
6227
		.total_capacity = 0UL,
6228 6229
		.busiest_stat = {
			.avg_load = 0UL,
6230 6231
			.sum_nr_running = 0,
			.group_type = group_other,
6232 6233 6234 6235
		},
	};
}

6236 6237 6238
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
6239
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
6240 6241
 *
 * Return: The load index.
6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263
 */
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;
}

6264
static unsigned long scale_rt_capacity(int cpu)
6265 6266
{
	struct rq *rq = cpu_rq(cpu);
6267
	u64 total, used, age_stamp, avg;
6268
	s64 delta;
6269

6270 6271 6272 6273
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
6274 6275
	age_stamp = READ_ONCE(rq->age_stamp);
	avg = READ_ONCE(rq->rt_avg);
6276
	delta = __rq_clock_broken(rq) - age_stamp;
6277

6278 6279 6280 6281
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
6282

6283
	used = div_u64(avg, total);
6284

6285 6286
	if (likely(used < SCHED_CAPACITY_SCALE))
		return SCHED_CAPACITY_SCALE - used;
6287

6288
	return 1;
6289 6290
}

6291
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
6292
{
6293
	unsigned long capacity = arch_scale_cpu_capacity(sd, cpu);
6294 6295
	struct sched_group *sdg = sd->groups;

6296
	cpu_rq(cpu)->cpu_capacity_orig = capacity;
6297

6298
	capacity *= scale_rt_capacity(cpu);
6299
	capacity >>= SCHED_CAPACITY_SHIFT;
6300

6301 6302
	if (!capacity)
		capacity = 1;
6303

6304 6305
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
6306 6307
}

6308
void update_group_capacity(struct sched_domain *sd, int cpu)
6309 6310 6311
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
6312
	unsigned long capacity;
6313 6314 6315 6316
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
6317
	sdg->sgc->next_update = jiffies + interval;
6318 6319

	if (!child) {
6320
		update_cpu_capacity(sd, cpu);
6321 6322 6323
		return;
	}

6324
	capacity = 0;
6325

P
Peter Zijlstra 已提交
6326 6327 6328 6329 6330 6331
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

6332
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
6333
			struct sched_group_capacity *sgc;
6334
			struct rq *rq = cpu_rq(cpu);
6335

6336
			/*
6337
			 * build_sched_domains() -> init_sched_groups_capacity()
6338 6339 6340
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
6341 6342
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
6343
			 *
6344
			 * This avoids capacity from being 0 and
6345 6346 6347
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
6348
				capacity += capacity_of(cpu);
6349 6350
				continue;
			}
6351

6352 6353
			sgc = rq->sd->groups->sgc;
			capacity += sgc->capacity;
6354
		}
P
Peter Zijlstra 已提交
6355 6356 6357 6358 6359 6360 6361 6362
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
		 */ 

		group = child->groups;
		do {
6363
			capacity += group->sgc->capacity;
P
Peter Zijlstra 已提交
6364 6365 6366
			group = group->next;
		} while (group != child->groups);
	}
6367

6368
	sdg->sgc->capacity = capacity;
6369 6370
}

6371
/*
6372 6373 6374
 * 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
6375 6376
 */
static inline int
6377
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
6378
{
6379 6380
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
6381 6382
}

6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398
/*
 * 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
6399 6400
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
6401 6402
 *
 * When this is so detected; this group becomes a candidate for busiest; see
6403
 * update_sd_pick_busiest(). And calculate_imbalance() and
6404
 * find_busiest_group() avoid some of the usual balance conditions to allow it
6405 6406 6407 6408 6409 6410 6411
 * 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.
 */

6412
static inline int sg_imbalanced(struct sched_group *group)
6413
{
6414
	return group->sgc->imbalance;
6415 6416
}

6417
/*
6418 6419 6420
 * 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
6421 6422
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
6423 6424 6425 6426 6427
 * 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.
6428
 */
6429 6430
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
6431
{
6432 6433
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
6434

6435
	if ((sgs->group_capacity * 100) >
6436
			(sgs->group_util * env->sd->imbalance_pct))
6437
		return true;
6438

6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452 6453 6454
	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;
6455

6456
	if ((sgs->group_capacity * 100) <
6457
			(sgs->group_util * env->sd->imbalance_pct))
6458
		return true;
6459

6460
	return false;
6461 6462
}

6463 6464 6465
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
6466
{
6467
	if (sgs->group_no_capacity)
6468 6469 6470 6471 6472 6473 6474 6475
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

6476 6477
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
6478
 * @env: The load balancing environment.
6479 6480 6481 6482
 * @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.
6483
 * @overload: Indicate more than one runnable task for any CPU.
6484
 */
6485 6486
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
6487 6488
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
6489
{
6490
	unsigned long load;
6491
	int i, nr_running;
6492

6493 6494
	memset(sgs, 0, sizeof(*sgs));

6495
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6496 6497 6498
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
6499
		if (local_group)
6500
			load = target_load(i, load_idx);
6501
		else
6502 6503 6504
			load = source_load(i, load_idx);

		sgs->group_load += load;
6505
		sgs->group_util += cpu_util(i);
6506
		sgs->sum_nr_running += rq->cfs.h_nr_running;
6507

6508 6509
		nr_running = rq->nr_running;
		if (nr_running > 1)
6510 6511
			*overload = true;

6512 6513 6514 6515
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
6516
		sgs->sum_weighted_load += weighted_cpuload(i);
6517 6518 6519 6520
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
6521
			sgs->idle_cpus++;
6522 6523
	}

6524 6525
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
6526
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
6527

6528
	if (sgs->sum_nr_running)
6529
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
6530

6531
	sgs->group_weight = group->group_weight;
6532

6533
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
6534
	sgs->group_type = group_classify(group, sgs);
6535 6536
}

6537 6538
/**
 * update_sd_pick_busiest - return 1 on busiest group
6539
 * @env: The load balancing environment.
6540 6541
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
6542
 * @sgs: sched_group statistics
6543 6544 6545
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
6546 6547 6548
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
6549
 */
6550
static bool update_sd_pick_busiest(struct lb_env *env,
6551 6552
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
6553
				   struct sg_lb_stats *sgs)
6554
{
6555
	struct sg_lb_stats *busiest = &sds->busiest_stat;
6556

6557
	if (sgs->group_type > busiest->group_type)
6558 6559
		return true;

6560 6561 6562 6563 6564 6565 6566 6567
	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))
6568 6569 6570 6571 6572 6573 6574
		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.
	 */
6575
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
6576 6577 6578 6579 6580 6581 6582 6583 6584 6585
		if (!sds->busiest)
			return true;

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

	return false;
}

6586 6587 6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615
#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 */

6616
/**
6617
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
6618
 * @env: The load balancing environment.
6619 6620
 * @sds: variable to hold the statistics for this sched_domain.
 */
6621
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
6622
{
6623 6624
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
6625
	struct sg_lb_stats tmp_sgs;
6626
	int load_idx, prefer_sibling = 0;
6627
	bool overload = false;
6628 6629 6630 6631

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

6632
	load_idx = get_sd_load_idx(env->sd, env->idle);
6633 6634

	do {
J
Joonsoo Kim 已提交
6635
		struct sg_lb_stats *sgs = &tmp_sgs;
6636 6637
		int local_group;

6638
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
6639 6640 6641
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
6642 6643

			if (env->idle != CPU_NEWLY_IDLE ||
6644 6645
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
6646
		}
6647

6648 6649
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
6650

6651 6652 6653
		if (local_group)
			goto next_group;

6654 6655
		/*
		 * In case the child domain prefers tasks go to siblings
6656
		 * first, lower the sg capacity so that we'll try
6657 6658
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
6659 6660 6661 6662
		 * 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).
6663
		 */
6664
		if (prefer_sibling && sds->local &&
6665 6666 6667
		    group_has_capacity(env, &sds->local_stat) &&
		    (sgs->sum_nr_running > 1)) {
			sgs->group_no_capacity = 1;
6668
			sgs->group_type = group_classify(sg, sgs);
6669
		}
6670

6671
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
6672
			sds->busiest = sg;
J
Joonsoo Kim 已提交
6673
			sds->busiest_stat = *sgs;
6674 6675
		}

6676 6677 6678
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
6679
		sds->total_capacity += sgs->group_capacity;
6680

6681
		sg = sg->next;
6682
	} while (sg != env->sd->groups);
6683 6684 6685

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
6686 6687 6688 6689 6690 6691 6692

	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;
	}

6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711
}

/**
 * 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.
 *
6712
 * Return: 1 when packing is required and a task should be moved to
6713 6714
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
6715
 * @env: The load balancing environment.
6716 6717
 * @sds: Statistics of the sched_domain which is to be packed
 */
6718
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
6719 6720 6721
{
	int busiest_cpu;

6722
	if (!(env->sd->flags & SD_ASYM_PACKING))
6723 6724 6725 6726 6727 6728
		return 0;

	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
6729
	if (env->dst_cpu > busiest_cpu)
6730 6731
		return 0;

6732
	env->imbalance = DIV_ROUND_CLOSEST(
6733
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
6734
		SCHED_CAPACITY_SCALE);
6735

6736
	return 1;
6737 6738 6739 6740 6741 6742
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
6743
 * @env: The load balancing environment.
6744 6745
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
6746 6747
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6748
{
6749
	unsigned long tmp, capa_now = 0, capa_move = 0;
6750
	unsigned int imbn = 2;
6751
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
6752
	struct sg_lb_stats *local, *busiest;
6753

J
Joonsoo Kim 已提交
6754 6755
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6756

J
Joonsoo Kim 已提交
6757 6758 6759 6760
	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;
6761

J
Joonsoo Kim 已提交
6762
	scaled_busy_load_per_task =
6763
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6764
		busiest->group_capacity;
J
Joonsoo Kim 已提交
6765

6766 6767
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
6768
		env->imbalance = busiest->load_per_task;
6769 6770 6771 6772 6773
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
6774
	 * however we may be able to increase total CPU capacity used by
6775 6776 6777
	 * moving them.
	 */

6778
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
6779
			min(busiest->load_per_task, busiest->avg_load);
6780
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
6781
			min(local->load_per_task, local->avg_load);
6782
	capa_now /= SCHED_CAPACITY_SCALE;
6783 6784

	/* Amount of load we'd subtract */
6785
	if (busiest->avg_load > scaled_busy_load_per_task) {
6786
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
6787
			    min(busiest->load_per_task,
6788
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
6789
	}
6790 6791

	/* Amount of load we'd add */
6792
	if (busiest->avg_load * busiest->group_capacity <
6793
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
6794 6795
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
6796
	} else {
6797
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
6798
		      local->group_capacity;
J
Joonsoo Kim 已提交
6799
	}
6800
	capa_move += local->group_capacity *
6801
		    min(local->load_per_task, local->avg_load + tmp);
6802
	capa_move /= SCHED_CAPACITY_SCALE;
6803 6804

	/* Move if we gain throughput */
6805
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
6806
		env->imbalance = busiest->load_per_task;
6807 6808 6809 6810 6811
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
6812
 * @env: load balance environment
6813 6814
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
6815
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
6816
{
6817
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
6818 6819 6820 6821
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
6822

6823
	if (busiest->group_type == group_imbalanced) {
6824 6825 6826 6827
		/*
		 * 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
6828 6829
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
6830 6831
	}

6832 6833 6834
	/*
	 * 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
6835
	 * its cpu_capacity, while calculating max_load..)
6836
	 */
6837 6838
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
6839 6840
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
6841 6842
	}

6843 6844 6845 6846 6847
	/*
	 * If there aren't any idle cpus, avoid creating some.
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
6848 6849 6850 6851 6852 6853
		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;
6854 6855 6856 6857 6858 6859 6860 6861 6862 6863
	}

	/*
	 * 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.
	 */
6864
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
6865 6866

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
6867
	env->imbalance = min(
6868 6869
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
6870
	) / SCHED_CAPACITY_SCALE;
6871 6872 6873

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
6874
	 * there is no guarantee that any tasks will be moved so we'll have
6875 6876 6877
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
6878
	if (env->imbalance < busiest->load_per_task)
6879
		return fix_small_imbalance(env, sds);
6880
}
6881

6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893
/******* 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.
 *
6894
 * @env: The load balancing environment.
6895
 *
6896
 * Return:	- The busiest group if imbalance exists.
6897 6898 6899 6900
 *		- 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
6901
static struct sched_group *find_busiest_group(struct lb_env *env)
6902
{
J
Joonsoo Kim 已提交
6903
	struct sg_lb_stats *local, *busiest;
6904 6905
	struct sd_lb_stats sds;

6906
	init_sd_lb_stats(&sds);
6907 6908 6909 6910 6911

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
6912
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
6913 6914
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
6915

6916
	/* ASYM feature bypasses nice load balance check */
6917 6918
	if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
	    check_asym_packing(env, &sds))
6919 6920
		return sds.busiest;

6921
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
6922
	if (!sds.busiest || busiest->sum_nr_running == 0)
6923 6924
		goto out_balanced;

6925 6926
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
6927

P
Peter Zijlstra 已提交
6928 6929
	/*
	 * If the busiest group is imbalanced the below checks don't
6930
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
6931 6932
	 * isn't true due to cpus_allowed constraints and the like.
	 */
6933
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
6934 6935
		goto force_balance;

6936
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
6937 6938
	if (env->idle == CPU_NEWLY_IDLE && group_has_capacity(env, local) &&
	    busiest->group_no_capacity)
6939 6940
		goto force_balance;

6941
	/*
6942
	 * If the local group is busier than the selected busiest group
6943 6944
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
6945
	if (local->avg_load >= busiest->avg_load)
6946 6947
		goto out_balanced;

6948 6949 6950 6951
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
6952
	if (local->avg_load >= sds.avg_load)
6953 6954
		goto out_balanced;

6955
	if (env->idle == CPU_IDLE) {
6956
		/*
6957 6958 6959 6960 6961
		 * 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
6962
		 */
6963 6964
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
6965
			goto out_balanced;
6966 6967 6968 6969 6970
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
6971 6972
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
6973
			goto out_balanced;
6974
	}
6975

6976
force_balance:
6977
	/* Looks like there is an imbalance. Compute it */
6978
	calculate_imbalance(env, &sds);
6979 6980 6981
	return sds.busiest;

out_balanced:
6982
	env->imbalance = 0;
6983 6984 6985 6986 6987 6988
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
6989
static struct rq *find_busiest_queue(struct lb_env *env,
6990
				     struct sched_group *group)
6991 6992
{
	struct rq *busiest = NULL, *rq;
6993
	unsigned long busiest_load = 0, busiest_capacity = 1;
6994 6995
	int i;

6996
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
6997
		unsigned long capacity, wl;
6998 6999 7000 7001
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
7002

7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024
		/*
		 * 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;

7025
		capacity = capacity_of(i);
7026

7027
		wl = weighted_cpuload(i);
7028

7029 7030
		/*
		 * When comparing with imbalance, use weighted_cpuload()
7031
		 * which is not scaled with the cpu capacity.
7032
		 */
7033 7034 7035

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
7036 7037
			continue;

7038 7039
		/*
		 * For the load comparisons with the other cpu's, consider
7040 7041 7042
		 * 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.
7043
		 *
7044
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
7045
		 * multiplication to rid ourselves of the division works out
7046 7047
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
7048
		 */
7049
		if (wl * busiest_capacity > busiest_load * capacity) {
7050
			busiest_load = wl;
7051
			busiest_capacity = capacity;
7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064 7065
			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. */
7066
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
7067

7068
static int need_active_balance(struct lb_env *env)
7069
{
7070 7071 7072
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
7073 7074 7075 7076 7077 7078

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
7079
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
7080
			return 1;
7081 7082
	}

7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095
	/*
	 * 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;
	}

7096 7097 7098
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

7099 7100
static int active_load_balance_cpu_stop(void *data);

7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119 7120 7121 7122 7123 7124 7125 7126 7127 7128 7129 7130 7131
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.
	 */
7132
	return balance_cpu == env->dst_cpu;
7133 7134
}

7135 7136 7137 7138 7139 7140
/*
 * 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,
7141
			int *continue_balancing)
7142
{
7143
	int ld_moved, cur_ld_moved, active_balance = 0;
7144
	struct sched_domain *sd_parent = sd->parent;
7145 7146 7147
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
7148
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
7149

7150 7151
	struct lb_env env = {
		.sd		= sd,
7152 7153
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
7154
		.dst_grpmask    = sched_group_cpus(sd->groups),
7155
		.idle		= idle,
7156
		.loop_break	= sched_nr_migrate_break,
7157
		.cpus		= cpus,
7158
		.fbq_type	= all,
7159
		.tasks		= LIST_HEAD_INIT(env.tasks),
7160 7161
	};

7162 7163 7164 7165
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
7166
	if (idle == CPU_NEWLY_IDLE)
7167 7168
		env.dst_grpmask = NULL;

7169 7170 7171 7172 7173
	cpumask_copy(cpus, cpu_active_mask);

	schedstat_inc(sd, lb_count[idle]);

redo:
7174 7175
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
7176
		goto out_balanced;
7177
	}
7178

7179
	group = find_busiest_group(&env);
7180 7181 7182 7183 7184
	if (!group) {
		schedstat_inc(sd, lb_nobusyg[idle]);
		goto out_balanced;
	}

7185
	busiest = find_busiest_queue(&env, group);
7186 7187 7188 7189 7190
	if (!busiest) {
		schedstat_inc(sd, lb_nobusyq[idle]);
		goto out_balanced;
	}

7191
	BUG_ON(busiest == env.dst_rq);
7192

7193
	schedstat_add(sd, lb_imbalance[idle], env.imbalance);
7194

7195 7196 7197
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

7198 7199 7200 7201 7202 7203 7204 7205
	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.
		 */
7206
		env.flags |= LBF_ALL_PINNED;
7207
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
7208

7209
more_balance:
7210
		raw_spin_lock_irqsave(&busiest->lock, flags);
7211 7212 7213 7214 7215

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
7216
		cur_ld_moved = detach_tasks(&env);
7217 7218

		/*
7219 7220 7221 7222 7223
		 * 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.
7224
		 */
7225 7226 7227 7228 7229 7230 7231 7232

		raw_spin_unlock(&busiest->lock);

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

7233
		local_irq_restore(flags);
7234

7235 7236 7237 7238 7239
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253 7254 7255 7256 7257 7258
		/*
		 * 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.
		 */
7259
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
7260

7261 7262 7263
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

7264
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
7265
			env.dst_cpu	 = env.new_dst_cpu;
7266
			env.flags	&= ~LBF_DST_PINNED;
7267 7268
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
7269

7270 7271 7272 7273 7274 7275
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
7276

7277 7278 7279 7280
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
7281
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
7282

7283
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
7284 7285 7286
				*group_imbalance = 1;
		}

7287
		/* All tasks on this runqueue were pinned by CPU affinity */
7288
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
7289
			cpumask_clear_cpu(cpu_of(busiest), cpus);
7290 7291 7292
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
7293
				goto redo;
7294
			}
7295
			goto out_all_pinned;
7296 7297 7298 7299 7300
		}
	}

	if (!ld_moved) {
		schedstat_inc(sd, lb_failed[idle]);
7301 7302 7303 7304 7305 7306 7307 7308
		/*
		 * 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++;
7309

7310
		if (need_active_balance(&env)) {
7311 7312
			raw_spin_lock_irqsave(&busiest->lock, flags);

7313 7314 7315
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
7316 7317
			 */
			if (!cpumask_test_cpu(this_cpu,
7318
					tsk_cpus_allowed(busiest->curr))) {
7319 7320
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
7321
				env.flags |= LBF_ALL_PINNED;
7322 7323 7324
				goto out_one_pinned;
			}

7325 7326 7327 7328 7329
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
7330 7331 7332 7333 7334 7335
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
7336

7337
			if (active_balance) {
7338 7339 7340
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
7341
			}
7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357 7358 7359

			/*
			 * 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
7360
		 * detach_tasks).
7361 7362 7363 7364 7365 7366 7367 7368
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384 7385
	/*
	 * 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.
	 */
7386 7387 7388 7389 7390 7391
	schedstat_inc(sd, lb_balanced[idle]);

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
7392
	if (((env.flags & LBF_ALL_PINNED) &&
7393
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
7394 7395 7396
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

7397
	ld_moved = 0;
7398 7399 7400 7401
out:
	return ld_moved;
}

7402 7403 7404 7405 7406 7407 7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423 7424 7425 7426 7427 7428
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;
}

7429 7430 7431 7432
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
7433
static int idle_balance(struct rq *this_rq)
7434
{
7435 7436
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
7437 7438
	struct sched_domain *sd;
	int pulled_task = 0;
7439
	u64 curr_cost = 0;
7440

7441 7442 7443 7444 7445 7446
	/*
	 * 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);

7447 7448
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
7449 7450 7451 7452 7453 7454
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, 0, &next_balance);
		rcu_read_unlock();

7455
		goto out;
7456
	}
7457

7458 7459
	raw_spin_unlock(&this_rq->lock);

7460
	update_blocked_averages(this_cpu);
7461
	rcu_read_lock();
7462
	for_each_domain(this_cpu, sd) {
7463
		int continue_balancing = 1;
7464
		u64 t0, domain_cost;
7465 7466 7467 7468

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

7469 7470
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, 0, &next_balance);
7471
			break;
7472
		}
7473

7474
		if (sd->flags & SD_BALANCE_NEWIDLE) {
7475 7476
			t0 = sched_clock_cpu(this_cpu);

7477
			pulled_task = load_balance(this_cpu, this_rq,
7478 7479
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
7480 7481 7482 7483 7484 7485

			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;
7486
		}
7487

7488
		update_next_balance(sd, 0, &next_balance);
7489 7490 7491 7492 7493 7494

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
7495 7496
			break;
	}
7497
	rcu_read_unlock();
7498 7499 7500

	raw_spin_lock(&this_rq->lock);

7501 7502 7503
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

7504
	/*
7505 7506 7507
	 * 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.
7508
	 */
7509
	if (this_rq->cfs.h_nr_running && !pulled_task)
7510
		pulled_task = 1;
7511

7512 7513 7514
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
7515
		this_rq->next_balance = next_balance;
7516

7517
	/* Is there a task of a high priority class? */
7518
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
7519 7520
		pulled_task = -1;

7521
	if (pulled_task)
7522 7523
		this_rq->idle_stamp = 0;

7524
	return pulled_task;
7525 7526 7527
}

/*
7528 7529 7530 7531
 * 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.
7532
 */
7533
static int active_load_balance_cpu_stop(void *data)
7534
{
7535 7536
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
7537
	int target_cpu = busiest_rq->push_cpu;
7538
	struct rq *target_rq = cpu_rq(target_cpu);
7539
	struct sched_domain *sd;
7540
	struct task_struct *p = NULL;
7541 7542 7543 7544 7545 7546 7547

	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;
7548 7549 7550

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
7551
		goto out_unlock;
7552 7553 7554 7555 7556 7557 7558 7559 7560

	/*
	 * 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. */
7561
	rcu_read_lock();
7562 7563 7564 7565 7566 7567 7568
	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)) {
7569 7570
		struct lb_env env = {
			.sd		= sd,
7571 7572 7573 7574
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
7575 7576 7577
			.idle		= CPU_IDLE,
		};

7578 7579
		schedstat_inc(sd, alb_count);

7580 7581
		p = detach_one_task(&env);
		if (p)
7582 7583 7584 7585
			schedstat_inc(sd, alb_pushed);
		else
			schedstat_inc(sd, alb_failed);
	}
7586
	rcu_read_unlock();
7587 7588
out_unlock:
	busiest_rq->active_balance = 0;
7589 7590 7591 7592 7593 7594 7595
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

7596
	return 0;
7597 7598
}

7599 7600 7601 7602 7603
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

7604
#ifdef CONFIG_NO_HZ_COMMON
7605 7606 7607 7608 7609 7610
/*
 * 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.
 */
7611
static struct {
7612
	cpumask_var_t idle_cpus_mask;
7613
	atomic_t nr_cpus;
7614 7615
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
7616

7617
static inline int find_new_ilb(void)
7618
{
7619
	int ilb = cpumask_first(nohz.idle_cpus_mask);
7620

7621 7622 7623 7624
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
7625 7626
}

7627 7628 7629 7630 7631
/*
 * 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).
 */
7632
static void nohz_balancer_kick(void)
7633 7634 7635 7636 7637
{
	int ilb_cpu;

	nohz.next_balance++;

7638
	ilb_cpu = find_new_ilb();
7639

7640 7641
	if (ilb_cpu >= nr_cpu_ids)
		return;
7642

7643
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
7644 7645 7646 7647 7648 7649 7650 7651
		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);
7652 7653 7654
	return;
}

7655
static inline void nohz_balance_exit_idle(int cpu)
7656 7657
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
7658 7659 7660 7661 7662 7663 7664
		/*
		 * 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);
		}
7665 7666 7667 7668
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

7669 7670 7671
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
7672
	int cpu = smp_processor_id();
7673 7674

	rcu_read_lock();
7675
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7676 7677 7678 7679 7680

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

7681
	atomic_inc(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7682
unlock:
7683 7684 7685 7686 7687 7688
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
7689
	int cpu = smp_processor_id();
7690 7691

	rcu_read_lock();
7692
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
V
Vincent Guittot 已提交
7693 7694 7695 7696 7697

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

7698
	atomic_dec(&sd->groups->sgc->nr_busy_cpus);
V
Vincent Guittot 已提交
7699
unlock:
7700 7701 7702
	rcu_read_unlock();
}

7703
/*
7704
 * This routine will record that the cpu is going idle with tick stopped.
7705
 * This info will be used in performing idle load balancing in the future.
7706
 */
7707
void nohz_balance_enter_idle(int cpu)
7708
{
7709 7710 7711 7712 7713 7714
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

7715 7716
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
7717

7718 7719 7720 7721 7722 7723
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

7724 7725 7726
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
7727
}
7728

7729
static int sched_ilb_notifier(struct notifier_block *nfb,
7730 7731 7732 7733
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DYING:
7734
		nohz_balance_exit_idle(smp_processor_id());
7735 7736 7737 7738 7739
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}
7740 7741 7742 7743
#endif

static DEFINE_SPINLOCK(balancing);

7744 7745 7746 7747
/*
 * 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.
 */
7748
void update_max_interval(void)
7749 7750 7751 7752
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

7753 7754 7755 7756
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
7757
 * Balancing parameters are set up in init_sched_domains.
7758
 */
7759
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
7760
{
7761
	int continue_balancing = 1;
7762
	int cpu = rq->cpu;
7763
	unsigned long interval;
7764
	struct sched_domain *sd;
7765 7766 7767
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
7768 7769
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
7770

7771
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
7772

7773
	rcu_read_lock();
7774
	for_each_domain(cpu, sd) {
7775 7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786
		/*
		 * 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;

7787 7788 7789
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800
		/*
		 * 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;
		}

7801
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7802 7803 7804 7805 7806 7807 7808 7809

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

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
7810
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
7811
				/*
7812
				 * The LBF_DST_PINNED logic could have changed
7813 7814
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
7815
				 */
7816
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
7817 7818
			}
			sd->last_balance = jiffies;
7819
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
7820 7821 7822 7823 7824 7825 7826 7827
		}
		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;
		}
7828 7829
	}
	if (need_decay) {
7830
		/*
7831 7832
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
7833
		 */
7834 7835
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
7836
	}
7837
	rcu_read_unlock();
7838 7839 7840 7841 7842 7843

	/*
	 * 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.
	 */
7844
	if (likely(update_next_balance)) {
7845
		rq->next_balance = next_balance;
7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859

#ifdef CONFIG_NO_HZ_COMMON
		/*
		 * If this CPU has been elected to perform the nohz idle
		 * balance. Other idle CPUs have already rebalanced with
		 * nohz_idle_balance() and nohz.next_balance has been
		 * updated accordingly. This CPU is now running the idle load
		 * balance for itself and we need to update the
		 * nohz.next_balance accordingly.
		 */
		if ((idle == CPU_IDLE) && time_after(nohz.next_balance, rq->next_balance))
			nohz.next_balance = rq->next_balance;
#endif
	}
7860 7861
}

7862
#ifdef CONFIG_NO_HZ_COMMON
7863
/*
7864
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
7865 7866
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
7867
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
7868
{
7869
	int this_cpu = this_rq->cpu;
7870 7871
	struct rq *rq;
	int balance_cpu;
7872 7873 7874
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
7875

7876 7877 7878
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
7879 7880

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
7881
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
7882 7883 7884 7885 7886 7887 7888
			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.
		 */
7889
		if (need_resched())
7890 7891
			break;

V
Vincent Guittot 已提交
7892 7893
		rq = cpu_rq(balance_cpu);

7894 7895 7896 7897 7898 7899 7900 7901 7902 7903 7904
		/*
		 * 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);
		}
7905

7906 7907 7908 7909
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
7910
	}
7911 7912 7913 7914 7915 7916 7917 7918

	/*
	 * 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))
		nohz.next_balance = next_balance;
7919 7920
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
7921 7922 7923
}

/*
7924
 * Current heuristic for kicking the idle load balancer in the presence
7925
 * of an idle cpu in the system.
7926
 *   - This rq has more than one task.
7927 7928 7929 7930
 *   - 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.
7931 7932
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
7933
 */
7934
static inline bool nohz_kick_needed(struct rq *rq)
7935 7936
{
	unsigned long now = jiffies;
7937
	struct sched_domain *sd;
7938
	struct sched_group_capacity *sgc;
7939
	int nr_busy, cpu = rq->cpu;
7940
	bool kick = false;
7941

7942
	if (unlikely(rq->idle_balance))
7943
		return false;
7944

7945 7946 7947 7948
       /*
	* 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.
	*/
7949
	set_cpu_sd_state_busy();
7950
	nohz_balance_exit_idle(cpu);
7951 7952 7953 7954 7955 7956

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
7957
		return false;
7958 7959

	if (time_before(now, nohz.next_balance))
7960
		return false;
7961

7962
	if (rq->nr_running >= 2)
7963
		return true;
7964

7965
	rcu_read_lock();
7966 7967
	sd = rcu_dereference(per_cpu(sd_busy, cpu));
	if (sd) {
7968 7969
		sgc = sd->groups->sgc;
		nr_busy = atomic_read(&sgc->nr_busy_cpus);
7970

7971 7972 7973 7974 7975
		if (nr_busy > 1) {
			kick = true;
			goto unlock;
		}

7976
	}
7977

7978 7979 7980 7981 7982 7983 7984 7985
	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
			kick = true;
			goto unlock;
		}
	}
7986

7987
	sd = rcu_dereference(per_cpu(sd_asym, cpu));
7988
	if (sd && (cpumask_first_and(nohz.idle_cpus_mask,
7989 7990 7991 7992
				  sched_domain_span(sd)) < cpu)) {
		kick = true;
		goto unlock;
	}
7993

7994
unlock:
7995
	rcu_read_unlock();
7996
	return kick;
7997 7998
}
#else
7999
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
8000 8001 8002 8003 8004 8005
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
8006 8007
static void run_rebalance_domains(struct softirq_action *h)
{
8008
	struct rq *this_rq = this_rq();
8009
	enum cpu_idle_type idle = this_rq->idle_balance ?
8010 8011 8012
						CPU_IDLE : CPU_NOT_IDLE;

	/*
8013
	 * If this cpu has a pending nohz_balance_kick, then do the
8014
	 * balancing on behalf of the other idle cpus whose ticks are
8015 8016 8017 8018
	 * 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.
8019
	 */
8020
	nohz_idle_balance(this_rq, idle);
8021
	rebalance_domains(this_rq, idle);
8022 8023 8024 8025 8026
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
8027
void trigger_load_balance(struct rq *rq)
8028 8029
{
	/* Don't need to rebalance while attached to NULL domain */
8030 8031 8032 8033
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
8034
		raise_softirq(SCHED_SOFTIRQ);
8035
#ifdef CONFIG_NO_HZ_COMMON
8036
	if (nohz_kick_needed(rq))
8037
		nohz_balancer_kick();
8038
#endif
8039 8040
}

8041 8042 8043
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
8044 8045

	update_runtime_enabled(rq);
8046 8047 8048 8049 8050
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
8051 8052 8053

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
8054 8055
}

8056
#endif /* CONFIG_SMP */
8057

8058 8059 8060
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
8061
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
8062 8063 8064 8065 8066 8067
{
	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
8068
		entity_tick(cfs_rq, se, queued);
8069
	}
8070

8071
	if (static_branch_unlikely(&sched_numa_balancing))
8072
		task_tick_numa(rq, curr);
8073 8074 8075
}

/*
P
Peter Zijlstra 已提交
8076 8077 8078
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
8079
 */
P
Peter Zijlstra 已提交
8080
static void task_fork_fair(struct task_struct *p)
8081
{
8082 8083
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
8084
	int this_cpu = smp_processor_id();
P
Peter Zijlstra 已提交
8085 8086 8087
	struct rq *rq = this_rq();
	unsigned long flags;

8088
	raw_spin_lock_irqsave(&rq->lock, flags);
8089

8090 8091
	update_rq_clock(rq);

8092 8093 8094
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;

8095 8096 8097 8098 8099 8100 8101 8102 8103
	/*
	 * 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();
8104

8105
	update_curr(cfs_rq);
P
Peter Zijlstra 已提交
8106

8107 8108
	if (curr)
		se->vruntime = curr->vruntime;
8109
	place_entity(cfs_rq, se, 1);
8110

P
Peter Zijlstra 已提交
8111
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
8112
		/*
8113 8114 8115
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
8116
		swap(curr->vruntime, se->vruntime);
8117
		resched_curr(rq);
8118
	}
8119

8120 8121
	se->vruntime -= cfs_rq->min_vruntime;

8122
	raw_spin_unlock_irqrestore(&rq->lock, flags);
8123 8124
}

8125 8126 8127 8128
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
8129 8130
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
8131
{
8132
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
8133 8134
		return;

8135 8136 8137 8138 8139
	/*
	 * 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
8140
	if (rq->curr == p) {
8141
		if (p->prio > oldprio)
8142
			resched_curr(rq);
8143
	} else
8144
		check_preempt_curr(rq, p, 0);
8145 8146
}

8147
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
8148 8149 8150 8151
{
	struct sched_entity *se = &p->se;

	/*
8152 8153 8154 8155 8156 8157 8158 8159 8160 8161
	 * In both the TASK_ON_RQ_QUEUED and TASK_ON_RQ_MIGRATING cases,
	 * the dequeue_entity(.flags=0) will already have normalized the
	 * vruntime.
	 */
	if (p->on_rq)
		return true;

	/*
	 * When !on_rq, vruntime of the task has usually NOT been normalized.
	 * But there are some cases where it has already been normalized:
P
Peter Zijlstra 已提交
8162
	 *
8163 8164 8165 8166
	 * - A forked child which is waiting for being woken up by
	 *   wake_up_new_task().
	 * - A task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
P
Peter Zijlstra 已提交
8167
	 */
8168 8169 8170 8171 8172 8173 8174 8175 8176 8177 8178 8179
	if (!se->sum_exec_runtime || p->state == TASK_WAKING)
		return true;

	return false;
}

static void detach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	if (!vruntime_normalized(p)) {
P
Peter Zijlstra 已提交
8180 8181 8182 8183 8184 8185 8186
		/*
		 * 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;
	}
8187

8188
	/* Catch up with the cfs_rq and remove our load when we leave */
8189
	detach_entity_load_avg(cfs_rq, se);
P
Peter Zijlstra 已提交
8190 8191
}

8192
static void attach_task_cfs_rq(struct task_struct *p)
8193
{
8194
	struct sched_entity *se = &p->se;
8195
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
8196 8197

#ifdef CONFIG_FAIR_GROUP_SCHED
8198 8199 8200 8201 8202 8203
	/*
	 * 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
8204

8205
	/* Synchronize task with its cfs_rq */
8206 8207 8208 8209 8210
	attach_entity_load_avg(cfs_rq, se);

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
8211

8212 8213 8214 8215 8216 8217 8218 8219
static void switched_from_fair(struct rq *rq, struct task_struct *p)
{
	detach_task_cfs_rq(p);
}

static void switched_to_fair(struct rq *rq, struct task_struct *p)
{
	attach_task_cfs_rq(p);
8220

8221
	if (task_on_rq_queued(p)) {
8222
		/*
8223 8224 8225
		 * 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.
8226
		 */
8227 8228 8229 8230
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
8231
	}
8232 8233
}

8234 8235 8236 8237 8238 8239 8240 8241 8242
/* 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;

8243 8244 8245 8246 8247 8248 8249
	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);
	}
8250 8251
}

8252 8253 8254 8255 8256 8257 8258
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
8259
#ifdef CONFIG_SMP
8260 8261
	atomic_long_set(&cfs_rq->removed_load_avg, 0);
	atomic_long_set(&cfs_rq->removed_util_avg, 0);
8262
#endif
8263 8264
}

P
Peter Zijlstra 已提交
8265
#ifdef CONFIG_FAIR_GROUP_SCHED
8266
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
8267
{
8268
	detach_task_cfs_rq(p);
8269
	set_task_rq(p, task_cpu(p));
8270 8271 8272 8273 8274

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
8275
	attach_task_cfs_rq(p);
P
Peter Zijlstra 已提交
8276
}
8277 8278 8279 8280 8281 8282 8283 8284 8285 8286

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]);
8287
		if (tg->se)
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
			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]);
8325
		init_entity_runnable_average(se);
8326 8327 8328 8329 8330 8331 8332 8333 8334 8335
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

8336
void unregister_fair_sched_group(struct task_group *tg)
8337 8338
{
	unsigned long flags;
8339 8340
	struct rq *rq;
	int cpu;
8341

8342 8343 8344
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
8345

8346 8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358
		/*
		 * 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)
			continue;

		rq = cpu_rq(cpu);

		raw_spin_lock_irqsave(&rq->lock, flags);
		list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
		raw_spin_unlock_irqrestore(&rq->lock, flags);
	}
8359 8360 8361 8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377
}

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
8378
	if (!parent) {
8379
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
8380 8381
		se->depth = 0;
	} else {
8382
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
8383 8384
		se->depth = parent->depth + 1;
	}
8385 8386

	se->my_q = cfs_rq;
8387 8388
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
8389 8390 8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402 8403 8404 8405 8406 8407 8408 8409 8410 8411 8412 8413 8414 8415 8416 8417 8418
	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);
8419 8420 8421

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
8422
		for_each_sched_entity(se)
8423 8424 8425 8426 8427 8428 8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439
			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;
}

8440
void unregister_fair_sched_group(struct task_group *tg) { }
8441 8442 8443

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
8444

8445
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
8446 8447 8448 8449 8450 8451 8452 8453 8454
{
	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)
8455
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
8456 8457 8458 8459

	return rr_interval;
}

8460 8461 8462
/*
 * All the scheduling class methods:
 */
8463
const struct sched_class fair_sched_class = {
8464
	.next			= &idle_sched_class,
8465 8466 8467
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
8468
	.yield_to_task		= yield_to_task_fair,
8469

I
Ingo Molnar 已提交
8470
	.check_preempt_curr	= check_preempt_wakeup,
8471 8472 8473 8474

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

8475
#ifdef CONFIG_SMP
L
Li Zefan 已提交
8476
	.select_task_rq		= select_task_rq_fair,
8477
	.migrate_task_rq	= migrate_task_rq_fair,
8478

8479 8480
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
8481 8482

	.task_waking		= task_waking_fair,
8483
	.task_dead		= task_dead_fair,
8484
	.set_cpus_allowed	= set_cpus_allowed_common,
8485
#endif
8486

8487
	.set_curr_task          = set_curr_task_fair,
8488
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
8489
	.task_fork		= task_fork_fair,
8490 8491

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
8492
	.switched_from		= switched_from_fair,
8493
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
8494

8495 8496
	.get_rr_interval	= get_rr_interval_fair,

8497 8498
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
8499
#ifdef CONFIG_FAIR_GROUP_SCHED
8500
	.task_move_group	= task_move_group_fair,
P
Peter Zijlstra 已提交
8501
#endif
8502 8503 8504
};

#ifdef CONFIG_SCHED_DEBUG
8505
void print_cfs_stats(struct seq_file *m, int cpu)
8506 8507 8508
{
	struct cfs_rq *cfs_rq;

8509
	rcu_read_lock();
8510
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
8511
		print_cfs_rq(m, cpu, cfs_rq);
8512
	rcu_read_unlock();
8513
}
8514 8515 8516 8517 8518 8519 8520 8521 8522 8523 8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534

#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 */
8535 8536 8537 8538 8539 8540

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

8541
#ifdef CONFIG_NO_HZ_COMMON
8542
	nohz.next_balance = jiffies;
8543
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
8544
	cpu_notifier(sched_ilb_notifier, 0);
8545 8546 8547 8548
#endif
#endif /* SMP */

}