fair.c 240.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
/*
 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
 *
 *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 *
 *  Interactivity improvements by Mike Galbraith
 *  (C) 2007 Mike Galbraith <efault@gmx.de>
 *
 *  Various enhancements by Dmitry Adamushko.
 *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
 *
 *  Group scheduling enhancements by Srivatsa Vaddagiri
 *  Copyright IBM Corporation, 2007
 *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
 *
 *  Scaled math optimizations by Thomas Gleixner
 *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
18 19
 *
 *  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 99 100 101 102 103 104 105 106 107 108 109
#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

110 111 112 113 114 115
/*
 * The margin used when comparing utilization with CPU capacity:
 * util * 1024 < capacity * margin
 */
unsigned int capacity_margin = 1280; /* ~20% */

116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133
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;
}

134 135 136 137 138 139 140 141 142
/*
 * 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:
 */
143
static unsigned int get_update_sysctl_factor(void)
144
{
145
	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
	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();
}

181
#define WMULT_CONST	(~0U)
182 183
#define WMULT_SHIFT	32

184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199
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;
}
200 201

/*
202 203 204 205
 * delta_exec * weight / lw.weight
 *   OR
 * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT
 *
206
 * Either weight := NICE_0_LOAD and lw \e sched_prio_to_wmult[], in which case
207 208 209 210 211
 * 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.
212
 */
213
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
214
{
215 216
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
217

218
	__update_inv_weight(lw);
219

220 221 222 223 224
	if (unlikely(fact >> 32)) {
		while (fact >> 32) {
			fact >>= 1;
			shift--;
		}
225 226
	}

227 228
	/* hint to use a 32x32->64 mul */
	fact = (u64)(u32)fact * lw->inv_weight;
229

230 231 232 233
	while (fact >> 32) {
		fact >>= 1;
		shift--;
	}
234

235
	return mul_u64_u32_shr(delta_exec, fact, shift);
236 237 238 239
}


const struct sched_class fair_sched_class;
240

241 242 243 244
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

245
#ifdef CONFIG_FAIR_GROUP_SCHED
246

247
/* cpu runqueue to which this cfs_rq is attached */
248 249
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
250
	return cfs_rq->rq;
251 252
}

253 254
/* An entity is a task if it doesn't "own" a runqueue */
#define entity_is_task(se)	(!se->my_q)
255

256 257
static inline struct task_struct *task_of(struct sched_entity *se)
{
258
	SCHED_WARN_ON(!entity_is_task(se));
259 260 261
	return container_of(se, struct task_struct, se);
}

P
Peter Zijlstra 已提交
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
/* 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;
}

283 284 285
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
286 287 288 289 290 291 292 293 294 295 296 297
		/*
		 * 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,
298
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
299
		}
300 301 302 303 304 305 306 307 308 309 310 311 312

		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
313 314 315 316 317
/* 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
318
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
319 320 321
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
322
		return se->cfs_rq;
P
Peter Zijlstra 已提交
323

P
Peter Zijlstra 已提交
324
	return NULL;
P
Peter Zijlstra 已提交
325 326 327 328 329 330 331
}

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

332 333 334 335 336 337 338 339 340 341 342 343 344
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
345 346
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363

	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);
	}
}

364 365 366 367 368 369
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
370

371 372 373
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
374 375 376 377
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
378 379
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
380

P
Peter Zijlstra 已提交
381
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
382
{
P
Peter Zijlstra 已提交
383
	return &task_rq(p)->cfs;
384 385
}

P
Peter Zijlstra 已提交
386 387 388 389 390 391 392 393 394 395 396 397 398 399
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;
}

400 401 402 403 404 405 406 407
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
408 409 410 411 412 413 414 415
#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;
}

416 417 418 419 420
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
421 422
#endif	/* CONFIG_FAIR_GROUP_SCHED */

423
static __always_inline
424
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
425 426 427 428 429

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

430
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
431
{
432
	s64 delta = (s64)(vruntime - max_vruntime);
433
	if (delta > 0)
434
		max_vruntime = vruntime;
435

436
	return max_vruntime;
437 438
}

439
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
440 441 442 443 444 445 446 447
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

448 449 450 451 452 453
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

454 455
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
456 457
	struct sched_entity *curr = cfs_rq->curr;

458 459
	u64 vruntime = cfs_rq->min_vruntime;

460 461 462 463 464 465
	if (curr) {
		if (curr->on_rq)
			vruntime = curr->vruntime;
		else
			curr = NULL;
	}
466 467 468 469 470 471

	if (cfs_rq->rb_leftmost) {
		struct sched_entity *se = rb_entry(cfs_rq->rb_leftmost,
						   struct sched_entity,
						   run_node);

472
		if (!curr)
473 474 475 476 477
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

478
	/* ensure we never gain time by being placed backwards. */
479
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
480 481 482 483
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
484 485
}

486 487 488
/*
 * Enqueue an entity into the rb-tree:
 */
489
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505
{
	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.
		 */
506
		if (entity_before(se, entry)) {
507 508 509 510 511 512 513 514 515 516 517
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
			leftmost = 0;
		}
	}

	/*
	 * Maintain a cache of leftmost tree entries (it is frequently
	 * used):
	 */
518
	if (leftmost)
I
Ingo Molnar 已提交
519
		cfs_rq->rb_leftmost = &se->run_node;
520 521 522 523 524

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

525
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
526
{
P
Peter Zijlstra 已提交
527 528 529 530 531 532
	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
533

534 535 536
	rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
}

537
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
538
{
539 540 541 542 543 544
	struct rb_node *left = cfs_rq->rb_leftmost;

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
545 546
}

547 548 549 550 551 552 553 554 555 556 557
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
558
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
559
{
560
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
561

562 563
	if (!last)
		return NULL;
564 565

	return rb_entry(last, struct sched_entity, run_node);
566 567
}

568 569 570 571
/**************************************************************
 * Scheduling class statistics methods:
 */

572
int sched_proc_update_handler(struct ctl_table *table, int write,
573
		void __user *buffer, size_t *lenp,
574 575
		loff_t *ppos)
{
576
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
577
	unsigned int factor = get_update_sysctl_factor();
578 579 580 581 582 583 584

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

585 586 587 588 589 590 591
#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

592 593 594
	return 0;
}
#endif
595

596
/*
597
 * delta /= w
598
 */
599
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
600
{
601
	if (unlikely(se->load.weight != NICE_0_LOAD))
602
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
603 604 605 606

	return delta;
}

607 608 609
/*
 * The idea is to set a period in which each task runs once.
 *
610
 * When there are too many tasks (sched_nr_latency) we have to stretch
611 612 613 614
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
615 616
static u64 __sched_period(unsigned long nr_running)
{
617 618 619 620
	if (unlikely(nr_running > sched_nr_latency))
		return nr_running * sysctl_sched_min_granularity;
	else
		return sysctl_sched_latency;
621 622
}

623 624 625 626
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
627
 * s = p*P[w/rw]
628
 */
P
Peter Zijlstra 已提交
629
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
630
{
M
Mike Galbraith 已提交
631
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
632

M
Mike Galbraith 已提交
633
	for_each_sched_entity(se) {
L
Lin Ming 已提交
634
		struct load_weight *load;
635
		struct load_weight lw;
L
Lin Ming 已提交
636 637 638

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
639

M
Mike Galbraith 已提交
640
		if (unlikely(!se->on_rq)) {
641
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
642 643 644 645

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
646
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
647 648
	}
	return slice;
649 650
}

651
/*
A
Andrei Epure 已提交
652
 * We calculate the vruntime slice of a to-be-inserted task.
653
 *
654
 * vs = s/w
655
 */
656
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
657
{
658
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
659 660
}

661
#ifdef CONFIG_SMP
662
static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu);
663 664
static unsigned long task_h_load(struct task_struct *p);

665 666
/*
 * We choose a half-life close to 1 scheduling period.
667 668
 * Note: The tables runnable_avg_yN_inv and runnable_avg_yN_sum are
 * dependent on this value.
669 670 671
 */
#define LOAD_AVG_PERIOD 32
#define LOAD_AVG_MAX 47742 /* maximum possible load avg */
672
#define LOAD_AVG_MAX_N 345 /* number of full periods to produce LOAD_AVG_MAX */
673

674 675
/* Give new sched_entity start runnable values to heavy its load in infant time */
void init_entity_runnable_average(struct sched_entity *se)
676
{
677
	struct sched_avg *sa = &se->avg;
678

679 680 681 682 683 684 685
	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;
686 687 688 689 690 691 692 693
	/*
	 * Tasks are intialized with full load to be seen as heavy tasks until
	 * they get a chance to stabilize to their real load level.
	 * Group entities are intialized with zero load to reflect the fact that
	 * nothing has been attached to the task group yet.
	 */
	if (entity_is_task(se))
		sa->load_avg = scale_load_down(se->load.weight);
694
	sa->load_sum = sa->load_avg * LOAD_AVG_MAX;
695 696 697 698 699
	/*
	 * At this point, util_avg won't be used in select_task_rq_fair anyway
	 */
	sa->util_avg = 0;
	sa->util_sum = 0;
700
	/* when this task enqueue'ed, it will contribute to its cfs_rq's load_avg */
701
}
702

703 704
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
static int update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq);
705
static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force);
706 707
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se);

708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736
/*
 * With new tasks being created, their initial util_avgs are extrapolated
 * based on the cfs_rq's current util_avg:
 *
 *   util_avg = cfs_rq->util_avg / (cfs_rq->load_avg + 1) * se.load.weight
 *
 * However, in many cases, the above util_avg does not give a desired
 * value. Moreover, the sum of the util_avgs may be divergent, such
 * as when the series is a harmonic series.
 *
 * To solve this problem, we also cap the util_avg of successive tasks to
 * only 1/2 of the left utilization budget:
 *
 *   util_avg_cap = (1024 - cfs_rq->avg.util_avg) / 2^n
 *
 * where n denotes the nth task.
 *
 * For example, a simplest series from the beginning would be like:
 *
 *  task  util_avg: 512, 256, 128,  64,  32,   16,    8, ...
 * cfs_rq util_avg: 512, 768, 896, 960, 992, 1008, 1016, ...
 *
 * Finally, that extrapolated util_avg is clamped to the cap (util_avg_cap)
 * if util_avg > util_avg_cap.
 */
void post_init_entity_util_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	struct sched_avg *sa = &se->avg;
737
	long cap = (long)(SCHED_CAPACITY_SCALE - cfs_rq->avg.util_avg) / 2;
738
	u64 now = cfs_rq_clock_task(cfs_rq);
739 740 741 742 743 744 745 746 747 748 749 750 751

	if (cap > 0) {
		if (cfs_rq->avg.util_avg != 0) {
			sa->util_avg  = cfs_rq->avg.util_avg * se->load.weight;
			sa->util_avg /= (cfs_rq->avg.load_avg + 1);

			if (sa->util_avg > cap)
				sa->util_avg = cap;
		} else {
			sa->util_avg = cap;
		}
		sa->util_sum = sa->util_avg * LOAD_AVG_MAX;
	}
752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770

	if (entity_is_task(se)) {
		struct task_struct *p = task_of(se);
		if (p->sched_class != &fair_sched_class) {
			/*
			 * For !fair tasks do:
			 *
			update_cfs_rq_load_avg(now, cfs_rq, false);
			attach_entity_load_avg(cfs_rq, se);
			switched_from_fair(rq, p);
			 *
			 * such that the next switched_to_fair() has the
			 * expected state.
			 */
			se->avg.last_update_time = now;
			return;
		}
	}

771
	update_cfs_rq_load_avg(now, cfs_rq, false);
772
	attach_entity_load_avg(cfs_rq, se);
773
	update_tg_load_avg(cfs_rq, false);
774 775
}

776
#else /* !CONFIG_SMP */
777
void init_entity_runnable_average(struct sched_entity *se)
778 779
{
}
780 781 782
void post_init_entity_util_avg(struct sched_entity *se)
{
}
783 784 785
static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
{
}
786
#endif /* CONFIG_SMP */
787

788
/*
789
 * Update the current task's runtime statistics.
790
 */
791
static void update_curr(struct cfs_rq *cfs_rq)
792
{
793
	struct sched_entity *curr = cfs_rq->curr;
794
	u64 now = rq_clock_task(rq_of(cfs_rq));
795
	u64 delta_exec;
796 797 798 799

	if (unlikely(!curr))
		return;

800 801
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
802
		return;
803

I
Ingo Molnar 已提交
804
	curr->exec_start = now;
805

806 807 808 809
	schedstat_set(curr->statistics.exec_max,
		      max(delta_exec, curr->statistics.exec_max));

	curr->sum_exec_runtime += delta_exec;
810
	schedstat_add(cfs_rq->exec_clock, delta_exec);
811 812 813 814

	curr->vruntime += calc_delta_fair(delta_exec, curr);
	update_min_vruntime(cfs_rq);

815 816 817
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

818
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
819
		cpuacct_charge(curtask, delta_exec);
820
		account_group_exec_runtime(curtask, delta_exec);
821
	}
822 823

	account_cfs_rq_runtime(cfs_rq, delta_exec);
824 825
}

826 827 828 829 830
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

831
static inline void
832
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
833
{
834 835 836 837 838 839 840
	u64 wait_start, prev_wait_start;

	if (!schedstat_enabled())
		return;

	wait_start = rq_clock(rq_of(cfs_rq));
	prev_wait_start = schedstat_val(se->statistics.wait_start);
841 842

	if (entity_is_task(se) && task_on_rq_migrating(task_of(se)) &&
843 844
	    likely(wait_start > prev_wait_start))
		wait_start -= prev_wait_start;
845

846
	schedstat_set(se->statistics.wait_start, wait_start);
847 848
}

849
static inline void
850 851 852
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *p;
853 854
	u64 delta;

855 856 857 858
	if (!schedstat_enabled())
		return;

	delta = rq_clock(rq_of(cfs_rq)) - schedstat_val(se->statistics.wait_start);
859 860 861 862 863 864 865 866 867

	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.
			 */
868
			schedstat_set(se->statistics.wait_start, delta);
869 870 871 872 873
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

874 875 876 877 878
	schedstat_set(se->statistics.wait_max,
		      max(schedstat_val(se->statistics.wait_max), delta));
	schedstat_inc(se->statistics.wait_count);
	schedstat_add(se->statistics.wait_sum, delta);
	schedstat_set(se->statistics.wait_start, 0);
879 880
}

881
static inline void
882 883 884
update_stats_enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *tsk = NULL;
885 886 887 888 889 890 891
	u64 sleep_start, block_start;

	if (!schedstat_enabled())
		return;

	sleep_start = schedstat_val(se->statistics.sleep_start);
	block_start = schedstat_val(se->statistics.block_start);
892 893 894 895

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

896 897
	if (sleep_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - sleep_start;
898 899 900 901

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

902 903
		if (unlikely(delta > schedstat_val(se->statistics.sleep_max)))
			schedstat_set(se->statistics.sleep_max, delta);
904

905 906
		schedstat_set(se->statistics.sleep_start, 0);
		schedstat_add(se->statistics.sum_sleep_runtime, delta);
907 908 909 910 911 912

		if (tsk) {
			account_scheduler_latency(tsk, delta >> 10, 1);
			trace_sched_stat_sleep(tsk, delta);
		}
	}
913 914
	if (block_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - block_start;
915 916 917 918

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

919 920
		if (unlikely(delta > schedstat_val(se->statistics.block_max)))
			schedstat_set(se->statistics.block_max, delta);
921

922 923
		schedstat_set(se->statistics.block_start, 0);
		schedstat_add(se->statistics.sum_sleep_runtime, delta);
924 925 926

		if (tsk) {
			if (tsk->in_iowait) {
927 928
				schedstat_add(se->statistics.iowait_sum, delta);
				schedstat_inc(se->statistics.iowait_count);
929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946
				trace_sched_stat_iowait(tsk, delta);
			}

			trace_sched_stat_blocked(tsk, delta);

			/*
			 * 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);
		}
	}
947 948
}

949 950 951
/*
 * Task is being enqueued - update stats:
 */
952
static inline void
953
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
954
{
955 956 957
	if (!schedstat_enabled())
		return;

958 959 960 961
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
962
	if (se != cfs_rq->curr)
963
		update_stats_wait_start(cfs_rq, se);
964 965 966

	if (flags & ENQUEUE_WAKEUP)
		update_stats_enqueue_sleeper(cfs_rq, se);
967 968 969
}

static inline void
970
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
971
{
972 973 974 975

	if (!schedstat_enabled())
		return;

976 977 978 979
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
980
	if (se != cfs_rq->curr)
981
		update_stats_wait_end(cfs_rq, se);
982

983 984
	if ((flags & DEQUEUE_SLEEP) && entity_is_task(se)) {
		struct task_struct *tsk = task_of(se);
985

986 987 988 989 990 991
		if (tsk->state & TASK_INTERRUPTIBLE)
			schedstat_set(se->statistics.sleep_start,
				      rq_clock(rq_of(cfs_rq)));
		if (tsk->state & TASK_UNINTERRUPTIBLE)
			schedstat_set(se->statistics.block_start,
				      rq_clock(rq_of(cfs_rq)));
992 993 994
	}
}

995 996 997 998
/*
 * We are picking a new current task - update its stats:
 */
static inline void
999
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
1000 1001 1002 1003
{
	/*
	 * We are starting a new run period:
	 */
1004
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
1005 1006 1007 1008 1009 1010
}

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

1011 1012
#ifdef CONFIG_NUMA_BALANCING
/*
1013 1014 1015
 * 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.
1016
 */
1017 1018
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
1019 1020 1021

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
1022

1023 1024 1025
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049
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)
{
1050
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
1051 1052 1053
	unsigned int scan, floor;
	unsigned int windows = 1;

1054 1055
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071
	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);
}

1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
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));
}

1084 1085 1086 1087 1088
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
1089
	pid_t gid;
1090
	int active_nodes;
1091 1092

	struct rcu_head rcu;
1093
	unsigned long total_faults;
1094
	unsigned long max_faults_cpu;
1095 1096 1097 1098 1099
	/*
	 * 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.
	 */
1100
	unsigned long *faults_cpu;
1101
	unsigned long faults[0];
1102 1103
};

1104 1105 1106 1107 1108 1109 1110 1111 1112
/* 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)

1113 1114 1115 1116 1117
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

1118 1119 1120 1121 1122 1123 1124
/*
 * 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)
1125
{
1126
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
1127 1128 1129 1130
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
1131
	if (!p->numa_faults)
1132 1133
		return 0;

1134 1135
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
1136 1137
}

1138 1139 1140 1141 1142
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

1143 1144
	return p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 1)];
1145 1146
}

1147 1148
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
1149 1150
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
1151 1152
}

1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164
/*
 * 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;
}

1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229
/* 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;
}

1230 1231 1232 1233 1234 1235
/*
 * 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.
 */
1236 1237
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1238
{
1239
	unsigned long faults, total_faults;
1240

1241
	if (!p->numa_faults)
1242 1243 1244 1245 1246 1247 1248
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1249
	faults = task_faults(p, nid);
1250 1251
	faults += score_nearby_nodes(p, nid, dist, true);

1252
	return 1000 * faults / total_faults;
1253 1254
}

1255 1256
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1257
{
1258 1259 1260 1261 1262 1263 1264 1265
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1266 1267
		return 0;

1268
	faults = group_faults(p, nid);
1269 1270
	faults += score_nearby_nodes(p, nid, dist, false);

1271
	return 1000 * faults / total_faults;
1272 1273
}

1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313
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;

	/*
1314 1315
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1316
	 */
1317 1318
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1319 1320 1321
		return true;

	/*
1322 1323 1324 1325 1326 1327
	 * 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)
1328
	 */
1329 1330
	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;
1331 1332
}

1333
static unsigned long weighted_cpuload(const int cpu);
1334 1335
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1336
static unsigned long capacity_of(int cpu);
1337 1338
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1339
/* Cached statistics for all CPUs within a node */
1340
struct numa_stats {
1341
	unsigned long nr_running;
1342
	unsigned long load;
1343 1344

	/* Total compute capacity of CPUs on a node */
1345
	unsigned long compute_capacity;
1346 1347

	/* Approximate capacity in terms of runnable tasks on a node */
1348
	unsigned long task_capacity;
1349
	int has_free_capacity;
1350
};
1351

1352 1353 1354 1355 1356
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1357 1358
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1359 1360 1361 1362 1363 1364 1365

	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);
1366
		ns->compute_capacity += capacity_of(cpu);
1367 1368

		cpus++;
1369 1370
	}

1371 1372 1373 1374 1375
	/*
	 * 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.
	 *
1376 1377
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1378 1379 1380 1381
	 */
	if (!cpus)
		return;

1382 1383 1384 1385 1386 1387
	/* 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));
1388
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1389 1390
}

1391 1392
struct task_numa_env {
	struct task_struct *p;
1393

1394 1395
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1396

1397
	struct numa_stats src_stats, dst_stats;
1398

1399
	int imbalance_pct;
1400
	int dist;
1401 1402 1403

	struct task_struct *best_task;
	long best_imp;
1404 1405 1406
	int best_cpu;
};

1407 1408 1409 1410 1411
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);
1412 1413
	if (p)
		get_task_struct(p);
1414 1415 1416 1417 1418 1419

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

1420
static bool load_too_imbalanced(long src_load, long dst_load,
1421 1422
				struct task_numa_env *env)
{
1423 1424
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435
	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;
1436 1437

	/* We care about the slope of the imbalance, not the direction. */
1438 1439
	if (dst_load < src_load)
		swap(dst_load, src_load);
1440 1441

	/* Is the difference below the threshold? */
1442 1443
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1444 1445 1446 1447 1448
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
1449
	 * Compare it with the old imbalance.
1450
	 */
1451
	orig_src_load = env->src_stats.load;
1452
	orig_dst_load = env->dst_stats.load;
1453

1454 1455
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);
1456

1457 1458 1459 1460 1461
	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);
1462 1463
}

1464 1465 1466 1467 1468 1469
/*
 * 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
 */
1470 1471
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1472 1473 1474 1475
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1476
	long src_load, dst_load;
1477
	long load;
1478
	long imp = env->p->numa_group ? groupimp : taskimp;
1479
	long moveimp = imp;
1480
	int dist = env->dist;
1481 1482

	rcu_read_lock();
1483 1484
	cur = task_rcu_dereference(&dst_rq->curr);
	if (cur && ((cur->flags & PF_EXITING) || is_idle_task(cur)))
1485 1486
		cur = NULL;

1487 1488 1489 1490 1491 1492 1493
	/*
	 * 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;

1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505
	/*
	 * "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;

1506 1507
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1508
		 * in any group then look only at task weights.
1509
		 */
1510
		if (cur->numa_group == env->p->numa_group) {
1511 1512
			imp = taskimp + task_weight(cur, env->src_nid, dist) -
			      task_weight(cur, env->dst_nid, dist);
1513 1514 1515 1516 1517 1518
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1519
		} else {
1520 1521 1522 1523 1524 1525
			/*
			 * 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)
1526 1527
				imp += group_weight(cur, env->src_nid, dist) -
				       group_weight(cur, env->dst_nid, dist);
1528
			else
1529 1530
				imp += task_weight(cur, env->src_nid, dist) -
				       task_weight(cur, env->dst_nid, dist);
1531
		}
1532 1533
	}

1534
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1535 1536 1537 1538
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1539
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1540
		    !env->dst_stats.has_free_capacity)
1541 1542 1543 1544 1545 1546
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1547 1548
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1549 1550 1551 1552 1553 1554
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1555 1556 1557
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1558

1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575
	if (moveimp > imp && moveimp > env->best_imp) {
		/*
		 * If the improvement from just moving env->p direction is
		 * better than swapping tasks around, check if a move is
		 * possible. Store a slightly smaller score than moveimp,
		 * so an actually idle CPU will win.
		 */
		if (!load_too_imbalanced(src_load, dst_load, env)) {
			imp = moveimp - 1;
			cur = NULL;
			goto assign;
		}
	}

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

1576
	if (cur) {
1577 1578 1579
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1580 1581
	}

1582
	if (load_too_imbalanced(src_load, dst_load, env))
1583 1584
		goto unlock;

1585 1586 1587 1588
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
1589 1590 1591 1592 1593 1594
	if (!cur) {
		/*
		 * select_idle_siblings() uses an per-cpu cpumask that
		 * can be used from IRQ context.
		 */
		local_irq_disable();
1595 1596
		env->dst_cpu = select_idle_sibling(env->p, env->src_cpu,
						   env->dst_cpu);
1597 1598
		local_irq_enable();
	}
1599

1600 1601 1602 1603 1604 1605
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1606 1607
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1608 1609 1610 1611 1612 1613 1614 1615 1616
{
	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;
1617
		task_numa_compare(env, taskimp, groupimp);
1618 1619 1620
	}
}

1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637
/* 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
	 */
1638 1639 1640
	if (src->load * dst->compute_capacity * env->imbalance_pct >

	    dst->load * src->compute_capacity * 100)
1641 1642 1643 1644 1645
		return true;

	return false;
}

1646 1647 1648 1649
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1650

1651
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1652
		.src_nid = task_node(p),
1653 1654 1655 1656 1657

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1658
		.best_cpu = -1,
1659 1660
	};
	struct sched_domain *sd;
1661
	unsigned long taskweight, groupweight;
1662
	int nid, ret, dist;
1663
	long taskimp, groupimp;
1664

1665
	/*
1666 1667 1668 1669 1670 1671
	 * 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.
1672 1673
	 */
	rcu_read_lock();
1674
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1675 1676
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1677 1678
	rcu_read_unlock();

1679 1680 1681 1682 1683 1684 1685
	/*
	 * 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)) {
1686
		p->numa_preferred_nid = task_node(p);
1687 1688 1689
		return -EINVAL;
	}

1690
	env.dst_nid = p->numa_preferred_nid;
1691 1692 1693 1694 1695 1696
	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;
1697
	update_numa_stats(&env.dst_stats, env.dst_nid);
1698

1699
	/* Try to find a spot on the preferred nid. */
1700 1701
	if (numa_has_capacity(&env))
		task_numa_find_cpu(&env, taskimp, groupimp);
1702

1703 1704 1705 1706 1707 1708 1709
	/*
	 * 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.
	 */
1710
	if (env.best_cpu == -1 || (p->numa_group && p->numa_group->active_nodes > 1)) {
1711 1712 1713
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1714

1715
			dist = node_distance(env.src_nid, env.dst_nid);
1716 1717 1718 1719 1720
			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);
			}
1721

1722
			/* Only consider nodes where both task and groups benefit */
1723 1724
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1725
			if (taskimp < 0 && groupimp < 0)
1726 1727
				continue;

1728
			env.dist = dist;
1729 1730
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1731 1732
			if (numa_has_capacity(&env))
				task_numa_find_cpu(&env, taskimp, groupimp);
1733 1734 1735
		}
	}

1736 1737 1738 1739 1740 1741 1742 1743
	/*
	 * 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.
	 */
1744
	if (p->numa_group) {
1745 1746
		struct numa_group *ng = p->numa_group;

1747 1748 1749 1750 1751
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
			nid = env.dst_nid;

1752
		if (ng->active_nodes > 1 && numa_is_active_node(env.dst_nid, ng))
1753 1754 1755 1756 1757 1758
			sched_setnuma(p, env.dst_nid);
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1759

1760 1761 1762 1763 1764 1765
	/*
	 * 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);

1766
	if (env.best_task == NULL) {
1767 1768 1769
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1770 1771 1772 1773
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1774 1775
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1776 1777
	put_task_struct(env.best_task);
	return ret;
1778 1779
}

1780 1781 1782
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1783 1784
	unsigned long interval = HZ;

1785
	/* This task has no NUMA fault statistics yet */
1786
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1787 1788
		return;

1789
	/* Periodically retry migrating the task to the preferred node */
1790 1791
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1792 1793

	/* Success if task is already running on preferred CPU */
1794
	if (task_node(p) == p->numa_preferred_nid)
1795 1796 1797
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1798
	task_numa_migrate(p);
1799 1800
}

1801
/*
1802
 * Find out how many nodes on the workload is actively running on. Do this by
1803 1804 1805 1806
 * 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.
 */
1807
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1808 1809
{
	unsigned long faults, max_faults = 0;
1810
	int nid, active_nodes = 0;
1811 1812 1813 1814 1815 1816 1817 1818 1819

	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);
1820 1821
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1822
	}
1823 1824 1825

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1826 1827
}

1828 1829 1830
/*
 * 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
1831 1832 1833
 * 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.
1834 1835
 */
#define NUMA_PERIOD_SLOTS 10
1836
#define NUMA_PERIOD_THRESHOLD 7
1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856

/*
 * 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
1857 1858 1859
	 * 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
1860
	 */
1861
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894
		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
		 */
1895
		ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared + 1));
1896 1897 1898 1899 1900 1901 1902 1903
		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));
}

1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921
/*
 * 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 {
1922 1923
		delta = p->se.avg.load_sum / p->se.load.weight;
		*period = LOAD_AVG_MAX;
1924 1925 1926 1927 1928 1929 1930 1931
	}

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

	return delta;
}

1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978
/*
 * 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;
1979
		nodemask_t max_group = NODE_MASK_NONE;
1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
		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. */
2013 2014
		if (!max_faults)
			break;
2015 2016 2017 2018 2019
		nodes = max_group;
	}
	return nid;
}

2020 2021
static void task_numa_placement(struct task_struct *p)
{
2022 2023
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
2024
	unsigned long fault_types[2] = { 0, 0 };
2025 2026
	unsigned long total_faults;
	u64 runtime, period;
2027
	spinlock_t *group_lock = NULL;
2028

2029 2030 2031 2032 2033
	/*
	 * 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:
	 */
2034
	seq = READ_ONCE(p->mm->numa_scan_seq);
2035 2036 2037
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
2038
	p->numa_scan_period_max = task_scan_max(p);
2039

2040 2041 2042 2043
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

2044 2045 2046
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
2047
		spin_lock_irq(group_lock);
2048 2049
	}

2050 2051
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
2052 2053
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
2054
		unsigned long faults = 0, group_faults = 0;
2055
		int priv;
2056

2057
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
2058
			long diff, f_diff, f_weight;
2059

2060 2061 2062 2063
			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);
2064

2065
			/* Decay existing window, copy faults since last scan */
2066 2067 2068
			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;
2069

2070 2071 2072 2073 2074 2075 2076 2077
			/*
			 * 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);
2078
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
2079
				   (total_faults + 1);
2080 2081
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
2082

2083 2084 2085
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
2086
			p->total_numa_faults += diff;
2087
			if (p->numa_group) {
2088 2089 2090 2091 2092 2093 2094 2095 2096
				/*
				 * 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;
2097
				p->numa_group->total_faults += diff;
2098
				group_faults += p->numa_group->faults[mem_idx];
2099
			}
2100 2101
		}

2102 2103 2104 2105
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
2106 2107 2108 2109 2110 2111 2112

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

2113 2114
	update_task_scan_period(p, fault_types[0], fault_types[1]);

2115
	if (p->numa_group) {
2116
		numa_group_count_active_nodes(p->numa_group);
2117
		spin_unlock_irq(group_lock);
2118
		max_nid = preferred_group_nid(p, max_group_nid);
2119 2120
	}

2121 2122 2123 2124 2125 2126 2127
	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);
2128
	}
2129 2130
}

2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141
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);
}

2142 2143
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2144 2145 2146 2147 2148 2149 2150 2151 2152
{
	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) +
2153
				    4*nr_node_ids*sizeof(unsigned long);
2154 2155 2156 2157 2158 2159

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

		atomic_set(&grp->refcount, 1);
2160 2161
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2162
		spin_lock_init(&grp->lock);
2163
		grp->gid = p->pid;
2164
		/* Second half of the array tracks nids where faults happen */
2165 2166
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2167

2168
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2169
			grp->faults[i] = p->numa_faults[i];
2170

2171
		grp->total_faults = p->total_numa_faults;
2172

2173 2174 2175 2176 2177
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2178
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2179 2180

	if (!cpupid_match_pid(tsk, cpupid))
2181
		goto no_join;
2182 2183 2184

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2185
		goto no_join;
2186 2187 2188

	my_grp = p->numa_group;
	if (grp == my_grp)
2189
		goto no_join;
2190 2191 2192 2193 2194 2195

	/*
	 * 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)
2196
		goto no_join;
2197 2198 2199 2200 2201

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2202
		goto no_join;
2203

2204 2205 2206 2207 2208 2209 2210
	/* 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;
2211

2212 2213 2214
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2215
	if (join && !get_numa_group(grp))
2216
		goto no_join;
2217 2218 2219 2220 2221 2222

	rcu_read_unlock();

	if (!join)
		return;

2223 2224
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2225

2226
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2227 2228
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2229
	}
2230 2231
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2232 2233 2234 2235 2236

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

	spin_unlock(&my_grp->lock);
2237
	spin_unlock_irq(&grp->lock);
2238 2239 2240 2241

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2242 2243 2244 2245 2246
	return;

no_join:
	rcu_read_unlock();
	return;
2247 2248 2249 2250 2251
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2252
	void *numa_faults = p->numa_faults;
2253 2254
	unsigned long flags;
	int i;
2255 2256

	if (grp) {
2257
		spin_lock_irqsave(&grp->lock, flags);
2258
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2259
			grp->faults[i] -= p->numa_faults[i];
2260
		grp->total_faults -= p->total_numa_faults;
2261

2262
		grp->nr_tasks--;
2263
		spin_unlock_irqrestore(&grp->lock, flags);
2264
		RCU_INIT_POINTER(p->numa_group, NULL);
2265 2266 2267
		put_numa_group(grp);
	}

2268
	p->numa_faults = NULL;
2269
	kfree(numa_faults);
2270 2271
}

2272 2273 2274
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2275
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2276 2277
{
	struct task_struct *p = current;
2278
	bool migrated = flags & TNF_MIGRATED;
2279
	int cpu_node = task_node(current);
2280
	int local = !!(flags & TNF_FAULT_LOCAL);
2281
	struct numa_group *ng;
2282
	int priv;
2283

2284
	if (!static_branch_likely(&sched_numa_balancing))
2285 2286
		return;

2287 2288 2289 2290
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2291
	/* Allocate buffer to track faults on a per-node basis */
2292 2293
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2294
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2295

2296 2297
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2298
			return;
2299

2300
		p->total_numa_faults = 0;
2301
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2302
	}
2303

2304 2305 2306 2307 2308 2309 2310 2311
	/*
	 * 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);
2312
		if (!priv && !(flags & TNF_NO_GROUP))
2313
			task_numa_group(p, last_cpupid, flags, &priv);
2314 2315
	}

2316 2317 2318 2319 2320 2321
	/*
	 * 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.
	 */
2322 2323 2324 2325
	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))
2326 2327
		local = 1;

2328
	task_numa_placement(p);
2329

2330 2331 2332 2333 2334
	/*
	 * 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))
2335 2336
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
2337 2338
	if (migrated)
		p->numa_pages_migrated += pages;
2339 2340
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2341

2342 2343
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2344
	p->numa_faults_locality[local] += pages;
2345 2346
}

2347 2348
static void reset_ptenuma_scan(struct task_struct *p)
{
2349 2350 2351 2352 2353 2354 2355 2356
	/*
	 * 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:
	 */
2357
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2358 2359 2360
	p->mm->numa_scan_offset = 0;
}

2361 2362 2363 2364 2365 2366 2367 2368 2369
/*
 * 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;
2370
	u64 runtime = p->se.sum_exec_runtime;
2371
	struct vm_area_struct *vma;
2372
	unsigned long start, end;
2373
	unsigned long nr_pte_updates = 0;
2374
	long pages, virtpages;
2375

2376
	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389

	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;

2390
	if (!mm->numa_next_scan) {
2391 2392
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2393 2394
	}

2395 2396 2397 2398 2399 2400 2401
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2402 2403 2404 2405
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
2406

2407
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2408 2409 2410
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2411 2412 2413 2414 2415 2416
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2417 2418 2419
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2420
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2421 2422
	if (!pages)
		return;
2423

2424

2425
	down_read(&mm->mmap_sem);
2426
	vma = find_vma(mm, start);
2427 2428
	if (!vma) {
		reset_ptenuma_scan(p);
2429
		start = 0;
2430 2431
		vma = mm->mmap;
	}
2432
	for (; vma; vma = vma->vm_next) {
2433
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2434
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2435
			continue;
2436
		}
2437

2438 2439 2440 2441 2442 2443 2444 2445 2446 2447
		/*
		 * 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
2448 2449 2450 2451 2452 2453
		/*
		 * 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;
2454

2455 2456 2457 2458
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2459
			nr_pte_updates = change_prot_numa(vma, start, end);
2460 2461

			/*
2462 2463 2464 2465 2466 2467
			 * 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.
2468 2469 2470
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2471
			virtpages -= (end - start) >> PAGE_SHIFT;
2472

2473
			start = end;
2474
			if (pages <= 0 || virtpages <= 0)
2475
				goto out;
2476 2477

			cond_resched();
2478
		} while (end != vma->vm_end);
2479
	}
2480

2481
out:
2482
	/*
P
Peter Zijlstra 已提交
2483 2484 2485 2486
	 * 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.
2487 2488
	 */
	if (vma)
2489
		mm->numa_scan_offset = start;
2490 2491 2492
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503

	/*
	 * 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;
	}
2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528
}

/*
 * 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;

2529
	if (now > curr->node_stamp + period) {
2530
		if (!curr->node_stamp)
2531
			curr->numa_scan_period = task_scan_min(curr);
2532
		curr->node_stamp += period;
2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543

		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)
{
}
2544 2545 2546 2547 2548 2549 2550 2551

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)
{
}
2552 2553
#endif /* CONFIG_NUMA_BALANCING */

2554 2555 2556 2557
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2558
	if (!parent_entity(se))
2559
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2560
#ifdef CONFIG_SMP
2561 2562 2563 2564 2565 2566
	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);
	}
2567
#endif
2568 2569 2570 2571 2572 2573 2574
	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);
2575
	if (!parent_entity(se))
2576
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2577
#ifdef CONFIG_SMP
2578 2579
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2580
		list_del_init(&se->group_node);
2581
	}
2582
#endif
2583 2584 2585
	cfs_rq->nr_running--;
}

2586 2587
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2588
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2589
{
2590
	long tg_weight, load, shares;
2591 2592

	/*
2593 2594 2595
	 * This really should be: cfs_rq->avg.load_avg, but instead we use
	 * cfs_rq->load.weight, which is its upper bound. This helps ramp up
	 * the shares for small weight interactive tasks.
2596
	 */
2597
	load = scale_load_down(cfs_rq->load.weight);
2598

2599
	tg_weight = atomic_long_read(&tg->load_avg);
2600

2601 2602 2603
	/* Ensure tg_weight >= load */
	tg_weight -= cfs_rq->tg_load_avg_contrib;
	tg_weight += load;
2604 2605

	shares = (tg->shares * load);
2606 2607
	if (tg_weight)
		shares /= tg_weight;
2608 2609 2610 2611 2612 2613 2614 2615 2616

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

	return shares;
}
# else /* CONFIG_SMP */
2617
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2618 2619 2620 2621
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
2622

P
Peter Zijlstra 已提交
2623 2624 2625
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2626 2627 2628 2629
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2630
		account_entity_dequeue(cfs_rq, se);
2631
	}
P
Peter Zijlstra 已提交
2632 2633 2634 2635 2636 2637 2638

	update_load_set(&se->load, weight);

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

2639 2640
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2641
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2642 2643 2644
{
	struct task_group *tg;
	struct sched_entity *se;
2645
	long shares;
P
Peter Zijlstra 已提交
2646 2647 2648

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2649
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2650
		return;
2651 2652 2653 2654
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2655
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2656 2657 2658 2659

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2660
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2661 2662 2663 2664
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2665
#ifdef CONFIG_SMP
2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685
/* 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,
};

2686 2687 2688 2689 2690 2691 2692 2693 2694 2695
/*
 * Precomputed \Sum y^k { 1<=k<=n, where n%32=0). Values are rolled down to
 * lower integers. See Documentation/scheduler/sched-avg.txt how these
 * were generated:
 */
static const u32 __accumulated_sum_N32[] = {
	    0, 23371, 35056, 40899, 43820, 45281,
	46011, 46376, 46559, 46650, 46696, 46719,
};

2696 2697 2698 2699 2700 2701
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713
	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
2714 2715
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
2716 2717 2718 2719 2720 2721
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
2722 2723
	}

2724 2725
	val = mul_u64_u32_shr(val, runnable_avg_yN_inv[local_n], 32);
	return val;
2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743
}

/*
 * 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;

2744 2745 2746
	/* Since n < LOAD_AVG_MAX_N, n/LOAD_AVG_PERIOD < 11 */
	contrib = __accumulated_sum_N32[n/LOAD_AVG_PERIOD];
	n %= LOAD_AVG_PERIOD;
2747 2748
	contrib = decay_load(contrib, n);
	return contrib + runnable_avg_yN_sum[n];
2749 2750
}

2751
#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
2752

2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780
/*
 * 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}]
 */
2781 2782
static __always_inline int
__update_load_avg(u64 now, int cpu, struct sched_avg *sa,
2783
		  unsigned long weight, int running, struct cfs_rq *cfs_rq)
2784
{
2785
	u64 delta, scaled_delta, periods;
2786
	u32 contrib;
2787
	unsigned int delta_w, scaled_delta_w, decayed = 0;
2788
	unsigned long scale_freq, scale_cpu;
2789

2790
	delta = now - sa->last_update_time;
2791 2792 2793 2794 2795
	/*
	 * 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) {
2796
		sa->last_update_time = now;
2797 2798 2799 2800 2801 2802 2803 2804 2805 2806
		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;
2807
	sa->last_update_time = now;
2808

2809 2810 2811
	scale_freq = arch_scale_freq_capacity(NULL, cpu);
	scale_cpu = arch_scale_cpu_capacity(NULL, cpu);

2812
	/* delta_w is the amount already accumulated against our next period */
2813
	delta_w = sa->period_contrib;
2814 2815 2816
	if (delta + delta_w >= 1024) {
		decayed = 1;

2817 2818 2819
		/* how much left for next period will start over, we don't know yet */
		sa->period_contrib = 0;

2820 2821 2822 2823 2824 2825
		/*
		 * 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;
2826
		scaled_delta_w = cap_scale(delta_w, scale_freq);
2827
		if (weight) {
2828 2829 2830 2831 2832
			sa->load_sum += weight * scaled_delta_w;
			if (cfs_rq) {
				cfs_rq->runnable_load_sum +=
						weight * scaled_delta_w;
			}
2833
		}
2834
		if (running)
2835
			sa->util_sum += scaled_delta_w * scale_cpu;
2836 2837 2838 2839 2840 2841 2842

		delta -= delta_w;

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

2843
		sa->load_sum = decay_load(sa->load_sum, periods + 1);
2844 2845 2846 2847
		if (cfs_rq) {
			cfs_rq->runnable_load_sum =
				decay_load(cfs_rq->runnable_load_sum, periods + 1);
		}
2848
		sa->util_sum = decay_load((u64)(sa->util_sum), periods + 1);
2849 2850

		/* Efficiently calculate \sum (1..n_period) 1024*y^i */
2851
		contrib = __compute_runnable_contrib(periods);
2852
		contrib = cap_scale(contrib, scale_freq);
2853
		if (weight) {
2854
			sa->load_sum += weight * contrib;
2855 2856 2857
			if (cfs_rq)
				cfs_rq->runnable_load_sum += weight * contrib;
		}
2858
		if (running)
2859
			sa->util_sum += contrib * scale_cpu;
2860 2861 2862
	}

	/* Remainder of delta accrued against u_0` */
2863
	scaled_delta = cap_scale(delta, scale_freq);
2864
	if (weight) {
2865
		sa->load_sum += weight * scaled_delta;
2866
		if (cfs_rq)
2867
			cfs_rq->runnable_load_sum += weight * scaled_delta;
2868
	}
2869
	if (running)
2870
		sa->util_sum += scaled_delta * scale_cpu;
2871

2872
	sa->period_contrib += delta;
2873

2874 2875
	if (decayed) {
		sa->load_avg = div_u64(sa->load_sum, LOAD_AVG_MAX);
2876 2877 2878 2879
		if (cfs_rq) {
			cfs_rq->runnable_load_avg =
				div_u64(cfs_rq->runnable_load_sum, LOAD_AVG_MAX);
		}
2880
		sa->util_avg = sa->util_sum / LOAD_AVG_MAX;
2881
	}
2882

2883
	return decayed;
2884 2885
}

2886
#ifdef CONFIG_FAIR_GROUP_SCHED
2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901
/**
 * update_tg_load_avg - update the tg's load avg
 * @cfs_rq: the cfs_rq whose avg changed
 * @force: update regardless of how small the difference
 *
 * This function 'ensures': tg->load_avg := \Sum tg->cfs_rq[]->avg.load.
 * However, because tg->load_avg is a global value there are performance
 * considerations.
 *
 * In order to avoid having to look at the other cfs_rq's, we use a
 * differential update where we store the last value we propagated. This in
 * turn allows skipping updates if the differential is 'small'.
 *
 * 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).
2902
 */
2903
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
2904
{
2905
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
2906

2907 2908 2909 2910 2911 2912
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

2913 2914 2915
	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;
2916
	}
2917
}
2918

2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964
/*
 * 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;
	}
}
2965
#else /* CONFIG_FAIR_GROUP_SCHED */
2966
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
2967
#endif /* CONFIG_FAIR_GROUP_SCHED */
2968

2969 2970
static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq)
{
2971
	if (&this_rq()->cfs == cfs_rq) {
2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987
		/*
		 * There are a few boundary cases this might miss but it should
		 * get called often enough that that should (hopefully) not be
		 * a real problem -- added to that it only calls on the local
		 * CPU, so if we enqueue remotely we'll miss an update, but
		 * the next tick/schedule should update.
		 *
		 * It will not get called when we go idle, because the idle
		 * thread is a different class (!fair), nor will the utilization
		 * number include things like RT tasks.
		 *
		 * As is, the util number is not freq-invariant (we'd have to
		 * implement arch_scale_freq_capacity() for that).
		 *
		 * See cpu_util().
		 */
2988
		cpufreq_update_util(rq_of(cfs_rq), 0);
2989 2990 2991
	}
}

2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008
/*
 * Unsigned subtract and clamp on underflow.
 *
 * Explicitly do a load-store to ensure the intermediate value never hits
 * memory. This allows lockless observations without ever seeing the negative
 * values.
 */
#define sub_positive(_ptr, _val) do {				\
	typeof(_ptr) ptr = (_ptr);				\
	typeof(*ptr) val = (_val);				\
	typeof(*ptr) res, var = READ_ONCE(*ptr);		\
	res = var - val;					\
	if (res > var)						\
		res = 0;					\
	WRITE_ONCE(*ptr, res);					\
} while (0)

3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020
/**
 * update_cfs_rq_load_avg - update the cfs_rq's load/util averages
 * @now: current time, as per cfs_rq_clock_task()
 * @cfs_rq: cfs_rq to update
 * @update_freq: should we call cfs_rq_util_change() or will the call do so
 *
 * The cfs_rq avg is the direct sum of all its entities (blocked and runnable)
 * avg. The immediate corollary is that all (fair) tasks must be attached, see
 * post_init_entity_util_avg().
 *
 * cfs_rq->avg is used for task_h_load() and update_cfs_share() for example.
 *
3021 3022 3023 3024
 * Returns true if the load decayed or we removed load.
 *
 * Since both these conditions indicate a changed cfs_rq->avg.load we should
 * call update_tg_load_avg() when this function returns true.
3025
 */
3026 3027
static inline int
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
3028
{
3029
	struct sched_avg *sa = &cfs_rq->avg;
3030
	int decayed, removed_load = 0, removed_util = 0;
3031

3032
	if (atomic_long_read(&cfs_rq->removed_load_avg)) {
3033
		s64 r = atomic_long_xchg(&cfs_rq->removed_load_avg, 0);
3034 3035
		sub_positive(&sa->load_avg, r);
		sub_positive(&sa->load_sum, r * LOAD_AVG_MAX);
3036
		removed_load = 1;
3037
	}
3038

3039 3040
	if (atomic_long_read(&cfs_rq->removed_util_avg)) {
		long r = atomic_long_xchg(&cfs_rq->removed_util_avg, 0);
3041 3042
		sub_positive(&sa->util_avg, r);
		sub_positive(&sa->util_sum, r * LOAD_AVG_MAX);
3043
		removed_util = 1;
3044
	}
3045

3046
	decayed = __update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
3047
		scale_load_down(cfs_rq->load.weight), cfs_rq->curr != NULL, cfs_rq);
3048

3049 3050 3051 3052
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
3053

3054 3055
	if (update_freq && (decayed || removed_util))
		cfs_rq_util_change(cfs_rq);
3056

3057
	return decayed || removed_load;
3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075
}

/* Update task and its cfs_rq load average */
static inline void update_load_avg(struct sched_entity *se, int update_tg)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 now = cfs_rq_clock_task(cfs_rq);
	struct rq *rq = rq_of(cfs_rq);
	int cpu = cpu_of(rq);

	/*
	 * 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
	 */
	__update_load_avg(now, cpu, &se->avg,
			  se->on_rq * scale_load_down(se->load.weight),
			  cfs_rq->curr == se, NULL);

3076
	if (update_cfs_rq_load_avg(now, cfs_rq, true) && update_tg)
3077
		update_tg_load_avg(cfs_rq, 0);
3078 3079
}

3080 3081 3082 3083 3084 3085 3086 3087
/**
 * attach_entity_load_avg - attach this entity to its cfs_rq load avg
 * @cfs_rq: cfs_rq to attach to
 * @se: sched_entity to attach
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3088 3089
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3090 3091 3092
	if (!sched_feat(ATTACH_AGE_LOAD))
		goto skip_aging;

3093 3094 3095
	/*
	 * If we got migrated (either between CPUs or between cgroups) we'll
	 * have aged the average right before clearing @last_update_time.
3096 3097
	 *
	 * Or we're fresh through post_init_entity_util_avg().
3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108
	 */
	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.
		 */
	}

3109
skip_aging:
3110 3111 3112 3113 3114
	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;
3115 3116

	cfs_rq_util_change(cfs_rq);
3117 3118
}

3119 3120 3121 3122 3123 3124 3125 3126
/**
 * detach_entity_load_avg - detach this entity from its cfs_rq load avg
 * @cfs_rq: cfs_rq to detach from
 * @se: sched_entity to detach
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3127 3128 3129 3130 3131 3132
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);

3133 3134 3135 3136
	sub_positive(&cfs_rq->avg.load_avg, se->avg.load_avg);
	sub_positive(&cfs_rq->avg.load_sum, se->avg.load_sum);
	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
3137 3138

	cfs_rq_util_change(cfs_rq);
3139 3140
}

3141 3142 3143
/* 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)
3144
{
3145 3146
	struct sched_avg *sa = &se->avg;
	u64 now = cfs_rq_clock_task(cfs_rq);
3147
	int migrated, decayed;
3148

3149 3150
	migrated = !sa->last_update_time;
	if (!migrated) {
3151
		__update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
3152 3153
			se->on_rq * scale_load_down(se->load.weight),
			cfs_rq->curr == se, NULL);
3154
	}
3155

3156
	decayed = update_cfs_rq_load_avg(now, cfs_rq, !migrated);
3157

3158 3159 3160
	cfs_rq->runnable_load_avg += sa->load_avg;
	cfs_rq->runnable_load_sum += sa->load_sum;

3161 3162
	if (migrated)
		attach_entity_load_avg(cfs_rq, se);
3163

3164 3165
	if (decayed || migrated)
		update_tg_load_avg(cfs_rq, 0);
3166 3167
}

3168 3169 3170 3171 3172 3173 3174 3175 3176
/* 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 =
3177
		max_t(s64,  cfs_rq->runnable_load_sum - se->avg.load_sum, 0);
3178 3179
}

3180
#ifndef CONFIG_64BIT
3181 3182
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
3183
	u64 last_update_time_copy;
3184
	u64 last_update_time;
3185

3186 3187 3188 3189 3190
	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);
3191 3192 3193

	return last_update_time;
}
3194
#else
3195 3196 3197 3198
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
3199 3200
#endif

3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213
/*
 * Synchronize entity load avg of dequeued entity without locking
 * the previous rq.
 */
void sync_entity_load_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 last_update_time;

	last_update_time = cfs_rq_last_update_time(cfs_rq);
	__update_load_avg(last_update_time, cpu_of(rq_of(cfs_rq)), &se->avg, 0, 0, NULL);
}

3214 3215 3216 3217 3218 3219 3220 3221 3222
/*
 * 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);

	/*
3223 3224 3225 3226 3227 3228 3229
	 * tasks cannot exit without having gone through wake_up_new_task() ->
	 * post_init_entity_util_avg() which will have added things to the
	 * cfs_rq, so we can remove unconditionally.
	 *
	 * Similarly for groups, they will have passed through
	 * post_init_entity_util_avg() before unregister_sched_fair_group()
	 * calls this.
3230 3231
	 */

3232
	sync_entity_load_avg(se);
3233 3234
	atomic_long_add(se->avg.load_avg, &cfs_rq->removed_load_avg);
	atomic_long_add(se->avg.util_avg, &cfs_rq->removed_util_avg);
3235
}
3236

3237 3238 3239 3240 3241 3242 3243 3244 3245 3246
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;
}

3247 3248
static int idle_balance(struct rq *this_rq);

3249 3250
#else /* CONFIG_SMP */

3251 3252 3253 3254 3255 3256
static inline int
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
{
	return 0;
}

3257 3258
static inline void update_load_avg(struct sched_entity *se, int not_used)
{
3259
	cpufreq_update_util(rq_of(cfs_rq_of(se)), 0);
3260 3261
}

3262 3263
static inline void
enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
3264 3265
static inline void
dequeue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
3266
static inline void remove_entity_load_avg(struct sched_entity *se) {}
3267

3268 3269 3270 3271 3272
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) {}

3273 3274 3275 3276 3277
static inline int idle_balance(struct rq *rq)
{
	return 0;
}

3278
#endif /* CONFIG_SMP */
3279

P
Peter Zijlstra 已提交
3280 3281 3282 3283 3284 3285 3286 3287 3288
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)
3289
		schedstat_inc(cfs_rq->nr_spread_over);
P
Peter Zijlstra 已提交
3290 3291 3292
#endif
}

3293 3294 3295
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3296
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3297

3298 3299 3300 3301 3302 3303
	/*
	 * 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
3304
	if (initial && sched_feat(START_DEBIT))
3305
		vruntime += sched_vslice(cfs_rq, se);
3306

3307
	/* sleeps up to a single latency don't count. */
3308
	if (!initial) {
3309
		unsigned long thresh = sysctl_sched_latency;
3310

3311 3312 3313 3314 3315 3316
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3317

3318
		vruntime -= thresh;
3319 3320
	}

3321
	/* ensure we never gain time by being placed backwards. */
3322
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3323 3324
}

3325 3326
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338
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())  {
3339
		printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
3340 3341 3342 3343 3344 3345 3346
			     "stat_blocked and stat_runtime require the "
			     "kernel parameter schedstats=enabled or "
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365

/*
 * MIGRATION
 *
 *	dequeue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime -= min_vruntime
 *
 *	enqueue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime += min_vruntime
 *
 * this way the vruntime transition between RQs is done when both
 * min_vruntime are up-to-date.
 *
 * WAKEUP (remote)
 *
3366
 *	->migrate_task_rq_fair() (p->state == TASK_WAKING)
3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377
 *	  vruntime -= min_vruntime
 *
 *	enqueue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime += min_vruntime
 *
 * this way we don't have the most up-to-date min_vruntime on the originating
 * CPU and an up-to-date min_vruntime on the destination CPU.
 */

3378
static void
3379
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3380
{
3381 3382 3383
	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
	bool curr = cfs_rq->curr == se;

3384
	/*
3385 3386
	 * If we're the current task, we must renormalise before calling
	 * update_curr().
3387
	 */
3388
	if (renorm && curr)
3389 3390
		se->vruntime += cfs_rq->min_vruntime;

3391 3392
	update_curr(cfs_rq);

3393
	/*
3394 3395 3396 3397
	 * Otherwise, renormalise after, such that we're placed at the current
	 * moment in time, instead of some random moment in the past. Being
	 * placed in the past could significantly boost this task to the
	 * fairness detriment of existing tasks.
3398
	 */
3399 3400 3401
	if (renorm && !curr)
		se->vruntime += cfs_rq->min_vruntime;

3402
	enqueue_entity_load_avg(cfs_rq, se);
3403 3404
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
3405

3406
	if (flags & ENQUEUE_WAKEUP)
3407
		place_entity(cfs_rq, se, 0);
3408

3409
	check_schedstat_required();
3410 3411
	update_stats_enqueue(cfs_rq, se, flags);
	check_spread(cfs_rq, se);
3412
	if (!curr)
3413
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3414
	se->on_rq = 1;
3415

3416
	if (cfs_rq->nr_running == 1) {
3417
		list_add_leaf_cfs_rq(cfs_rq);
3418 3419
		check_enqueue_throttle(cfs_rq);
	}
3420 3421
}

3422
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3423
{
3424 3425
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3426
		if (cfs_rq->last != se)
3427
			break;
3428 3429

		cfs_rq->last = NULL;
3430 3431
	}
}
P
Peter Zijlstra 已提交
3432

3433 3434 3435 3436
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3437
		if (cfs_rq->next != se)
3438
			break;
3439 3440

		cfs_rq->next = NULL;
3441
	}
P
Peter Zijlstra 已提交
3442 3443
}

3444 3445 3446 3447
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3448
		if (cfs_rq->skip != se)
3449
			break;
3450 3451

		cfs_rq->skip = NULL;
3452 3453 3454
	}
}

P
Peter Zijlstra 已提交
3455 3456
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3457 3458 3459 3460 3461
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
3462 3463 3464

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
3465 3466
}

3467
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3468

3469
static void
3470
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3471
{
3472 3473 3474 3475
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
3476
	dequeue_entity_load_avg(cfs_rq, se);
3477

3478
	update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
3479

P
Peter Zijlstra 已提交
3480
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3481

3482
	if (se != cfs_rq->curr)
3483
		__dequeue_entity(cfs_rq, se);
3484
	se->on_rq = 0;
3485
	account_entity_dequeue(cfs_rq, se);
3486 3487

	/*
3488 3489 3490 3491
	 * Normalize after update_curr(); which will also have moved
	 * min_vruntime if @se is the one holding it back. But before doing
	 * update_min_vruntime() again, which will discount @se's position and
	 * can move min_vruntime forward still more.
3492
	 */
3493
	if (!(flags & DEQUEUE_SLEEP))
3494
		se->vruntime -= cfs_rq->min_vruntime;
3495

3496 3497 3498
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

3499
	update_cfs_shares(cfs_rq);
3500 3501 3502 3503 3504 3505 3506 3507 3508

	/*
	 * Now advance min_vruntime if @se was the entity holding it back,
	 * except when: DEQUEUE_SAVE && !DEQUEUE_MOVE, in this case we'll be
	 * put back on, and if we advance min_vruntime, we'll be placed back
	 * further than we started -- ie. we'll be penalized.
	 */
	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
		update_min_vruntime(cfs_rq);
3509 3510 3511 3512 3513
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
3514
static void
I
Ingo Molnar 已提交
3515
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3516
{
3517
	unsigned long ideal_runtime, delta_exec;
3518 3519
	struct sched_entity *se;
	s64 delta;
3520

P
Peter Zijlstra 已提交
3521
	ideal_runtime = sched_slice(cfs_rq, curr);
3522
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
3523
	if (delta_exec > ideal_runtime) {
3524
		resched_curr(rq_of(cfs_rq));
3525 3526 3527 3528 3529
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540
		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;

3541 3542
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
3543

3544 3545
	if (delta < 0)
		return;
3546

3547
	if (delta > ideal_runtime)
3548
		resched_curr(rq_of(cfs_rq));
3549 3550
}

3551
static void
3552
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
3553
{
3554 3555 3556 3557 3558 3559 3560
	/* '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.
		 */
3561
		update_stats_wait_end(cfs_rq, se);
3562
		__dequeue_entity(cfs_rq, se);
3563
		update_load_avg(se, 1);
3564 3565
	}

3566
	update_stats_curr_start(cfs_rq, se);
3567
	cfs_rq->curr = se;
3568

I
Ingo Molnar 已提交
3569 3570 3571 3572 3573
	/*
	 * 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):
	 */
3574
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
3575 3576 3577
		schedstat_set(se->statistics.slice_max,
			max((u64)schedstat_val(se->statistics.slice_max),
			    se->sum_exec_runtime - se->prev_sum_exec_runtime));
I
Ingo Molnar 已提交
3578
	}
3579

3580
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
3581 3582
}

3583 3584 3585
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

3586 3587 3588 3589 3590 3591 3592
/*
 * 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
 */
3593 3594
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3595
{
3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606
	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 */
3607

3608 3609 3610 3611 3612
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3613 3614 3615 3616 3617 3618 3619 3620 3621 3622
		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;
		}

3623 3624 3625
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3626

3627 3628 3629 3630 3631 3632
	/*
	 * 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;

3633 3634 3635 3636 3637 3638
	/*
	 * 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;

3639
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3640 3641

	return se;
3642 3643
}

3644
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3645

3646
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3647 3648 3649 3650 3651 3652
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3653
		update_curr(cfs_rq);
3654

3655 3656 3657
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

3658
	check_spread(cfs_rq, prev);
3659

3660
	if (prev->on_rq) {
3661
		update_stats_wait_start(cfs_rq, prev);
3662 3663
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3664
		/* in !on_rq case, update occurred at dequeue */
3665
		update_load_avg(prev, 0);
3666
	}
3667
	cfs_rq->curr = NULL;
3668 3669
}

P
Peter Zijlstra 已提交
3670 3671
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3672 3673
{
	/*
3674
	 * Update run-time statistics of the 'current'.
3675
	 */
3676
	update_curr(cfs_rq);
3677

3678 3679 3680
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3681
	update_load_avg(curr, 1);
3682
	update_cfs_shares(cfs_rq);
3683

P
Peter Zijlstra 已提交
3684 3685 3686 3687 3688
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3689
	if (queued) {
3690
		resched_curr(rq_of(cfs_rq));
3691 3692
		return;
	}
P
Peter Zijlstra 已提交
3693 3694 3695 3696 3697 3698 3699 3700
	/*
	 * 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
3701
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3702
		check_preempt_tick(cfs_rq, curr);
3703 3704
}

3705 3706 3707 3708 3709 3710

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

#ifdef CONFIG_CFS_BANDWIDTH
3711 3712

#ifdef HAVE_JUMP_LABEL
3713
static struct static_key __cfs_bandwidth_used;
3714 3715 3716

static inline bool cfs_bandwidth_used(void)
{
3717
	return static_key_false(&__cfs_bandwidth_used);
3718 3719
}

3720
void cfs_bandwidth_usage_inc(void)
3721
{
3722 3723 3724 3725 3726 3727
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3728 3729 3730 3731 3732 3733 3734
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3735 3736
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3737 3738
#endif /* HAVE_JUMP_LABEL */

3739 3740 3741 3742 3743 3744 3745 3746
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3747 3748 3749 3750 3751 3752

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

P
Paul Turner 已提交
3753 3754 3755 3756 3757 3758 3759
/*
 * 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
 */
3760
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771
{
	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);
}

3772 3773 3774 3775 3776
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3777 3778 3779 3780
/* 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))
3781
		return cfs_rq->throttled_clock_task - cfs_rq->throttled_clock_task_time;
3782

3783
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3784 3785
}

3786 3787
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3788 3789 3790
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3791
	u64 amount = 0, min_amount, expires;
3792 3793 3794 3795 3796 3797 3798

	/* 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;
3799
	else {
P
Peter Zijlstra 已提交
3800
		start_cfs_bandwidth(cfs_b);
3801 3802 3803 3804 3805 3806

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3807
	}
P
Paul Turner 已提交
3808
	expires = cfs_b->runtime_expires;
3809 3810 3811
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3812 3813 3814 3815 3816 3817 3818
	/*
	 * 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;
3819 3820

	return cfs_rq->runtime_remaining > 0;
3821 3822
}

P
Paul Turner 已提交
3823 3824 3825 3826 3827
/*
 * 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)
3828
{
P
Paul Turner 已提交
3829 3830 3831
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3832
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3833 3834
		return;

P
Paul Turner 已提交
3835 3836 3837 3838 3839 3840 3841 3842 3843
	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
3844 3845 3846
	 * 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
3847 3848
	 */

3849
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3850 3851 3852 3853 3854 3855 3856 3857
		/* 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;
	}
}

3858
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3859 3860
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3861
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3862 3863 3864
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3865 3866
		return;

3867 3868 3869 3870 3871
	/*
	 * 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))
3872
		resched_curr(rq_of(cfs_rq));
3873 3874
}

3875
static __always_inline
3876
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3877
{
3878
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3879 3880 3881 3882 3883
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3884 3885
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3886
	return cfs_bandwidth_used() && cfs_rq->throttled;
3887 3888
}

3889 3890 3891
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3892
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919
}

/*
 * 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--;
	if (!cfs_rq->throttle_count) {
3920
		/* adjust cfs_rq_clock_task() */
3921
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3922
					     cfs_rq->throttled_clock_task;
3923 3924 3925 3926 3927 3928 3929 3930 3931 3932
	}

	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)];

3933 3934
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3935
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3936 3937 3938 3939 3940
	cfs_rq->throttle_count++;

	return 0;
}

3941
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3942 3943 3944 3945 3946
{
	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
3947
	bool empty;
3948 3949 3950

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

3951
	/* freeze hierarchy runnable averages while throttled */
3952 3953 3954
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971

	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)
3972
		sub_nr_running(rq, task_delta);
3973 3974

	cfs_rq->throttled = 1;
3975
	cfs_rq->throttled_clock = rq_clock(rq);
3976
	raw_spin_lock(&cfs_b->lock);
3977
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
3978

3979 3980 3981 3982 3983
	/*
	 * 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
3984 3985 3986 3987 3988 3989 3990 3991

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

3992 3993 3994
	raw_spin_unlock(&cfs_b->lock);
}

3995
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3996 3997 3998 3999 4000 4001 4002
{
	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;

4003
	se = cfs_rq->tg->se[cpu_of(rq)];
4004 4005

	cfs_rq->throttled = 0;
4006 4007 4008

	update_rq_clock(rq);

4009
	raw_spin_lock(&cfs_b->lock);
4010
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
4011 4012 4013
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

4014 4015 4016
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034
	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)
4035
		add_nr_running(rq, task_delta);
4036 4037 4038

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
4039
		resched_curr(rq);
4040 4041 4042 4043 4044 4045
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
4046 4047
	u64 runtime;
	u64 starting_runtime = remaining;
4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077

	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();

4078
	return starting_runtime - remaining;
4079 4080
}

4081 4082 4083 4084 4085 4086 4087 4088
/*
 * 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)
{
4089
	u64 runtime, runtime_expires;
4090
	int throttled;
4091 4092 4093

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
4094
		goto out_deactivate;
4095

4096
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4097
	cfs_b->nr_periods += overrun;
4098

4099 4100 4101 4102 4103 4104
	/*
	 * 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
4105 4106 4107

	__refill_cfs_bandwidth_runtime(cfs_b);

4108 4109 4110
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
4111
		return 0;
4112 4113
	}

4114 4115 4116
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

4117 4118 4119
	runtime_expires = cfs_b->runtime_expires;

	/*
4120 4121 4122 4123 4124
	 * 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.
4125
	 */
4126 4127
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
4128 4129 4130 4131 4132 4133 4134
		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);
4135 4136

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4137
	}
4138

4139 4140 4141 4142 4143 4144 4145
	/*
	 * 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;
4146

4147 4148 4149 4150
	return 0;

out_deactivate:
	return 1;
4151
}
4152

4153 4154 4155 4156 4157 4158 4159
/* 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;

4160 4161 4162 4163
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
4164
 * hrtimer base being cleared by hrtimer_start. In the case of
4165 4166
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191
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
4192 4193 4194
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223
}

/* 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)
{
4224 4225 4226
	if (!cfs_bandwidth_used())
		return;

4227
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242
		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 */
4243 4244 4245
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
4246
		return;
4247
	}
4248

4249
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
4250
		runtime = cfs_b->runtime;
4251

4252 4253 4254 4255 4256 4257 4258 4259 4260 4261
	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)
4262
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4263 4264 4265
	raw_spin_unlock(&cfs_b->lock);
}

4266 4267 4268 4269 4270 4271 4272
/*
 * 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)
{
4273 4274 4275
	if (!cfs_bandwidth_used())
		return;

4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289
	/* 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);
}

4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303
static void sync_throttle(struct task_group *tg, int cpu)
{
	struct cfs_rq *pcfs_rq, *cfs_rq;

	if (!cfs_bandwidth_used())
		return;

	if (!tg->parent)
		return;

	cfs_rq = tg->cfs_rq[cpu];
	pcfs_rq = tg->parent->cfs_rq[cpu];

	cfs_rq->throttle_count = pcfs_rq->throttle_count;
4304
	cfs_rq->throttled_clock_task = rq_clock_task(cpu_rq(cpu));
4305 4306
}

4307
/* conditionally throttle active cfs_rq's from put_prev_entity() */
4308
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4309
{
4310
	if (!cfs_bandwidth_used())
4311
		return false;
4312

4313
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4314
		return false;
4315 4316 4317 4318 4319 4320

	/*
	 * 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))
4321
		return true;
4322 4323

	throttle_cfs_rq(cfs_rq);
4324
	return true;
4325
}
4326 4327 4328 4329 4330

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
4331

4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343
	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;

4344
	raw_spin_lock(&cfs_b->lock);
4345
	for (;;) {
P
Peter Zijlstra 已提交
4346
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4347 4348 4349 4350 4351
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4352 4353
	if (idle)
		cfs_b->period_active = 0;
4354
	raw_spin_unlock(&cfs_b->lock);
4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366

	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
4367
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378
	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
4379
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4380
{
P
Peter Zijlstra 已提交
4381
	lockdep_assert_held(&cfs_b->lock);
4382

P
Peter Zijlstra 已提交
4383 4384 4385 4386 4387
	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);
	}
4388 4389 4390 4391
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4392 4393 4394 4395
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4396 4397 4398 4399
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412
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);
	}
}

4413
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424
{
	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
		 */
4425
		cfs_rq->runtime_remaining = 1;
4426 4427 4428 4429 4430 4431
		/*
		 * 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;

4432 4433 4434 4435 4436 4437
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
4438 4439
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
4440
	return rq_clock_task(rq_of(cfs_rq));
4441 4442
}

4443
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
4444
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
4445
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
4446
static inline void sync_throttle(struct task_group *tg, int cpu) {}
4447
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
4448 4449 4450 4451 4452

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463

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;
}
4464 4465 4466 4467 4468

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) {}
4469 4470
#endif

4471 4472 4473 4474 4475
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) {}
4476
static inline void update_runtime_enabled(struct rq *rq) {}
4477
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
4478 4479 4480

#endif /* CONFIG_CFS_BANDWIDTH */

4481 4482 4483 4484
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
4485 4486 4487 4488 4489 4490
#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);

4491
	SCHED_WARN_ON(task_rq(p) != rq);
P
Peter Zijlstra 已提交
4492

4493
	if (rq->cfs.h_nr_running > 1) {
P
Peter Zijlstra 已提交
4494 4495 4496 4497 4498 4499
		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)
4500
				resched_curr(rq);
P
Peter Zijlstra 已提交
4501 4502
			return;
		}
4503
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
4504 4505
	}
}
4506 4507 4508 4509 4510 4511 4512 4513 4514 4515

/*
 * 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;

4516
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
4517 4518 4519 4520 4521
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
4522
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
4523 4524 4525 4526
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
4527 4528 4529 4530

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
4531 4532
#endif

4533 4534 4535 4536 4537
/*
 * 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:
 */
4538
static void
4539
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4540 4541
{
	struct cfs_rq *cfs_rq;
4542
	struct sched_entity *se = &p->se;
4543

4544 4545 4546 4547 4548 4549 4550 4551
	/*
	 * If in_iowait is set, the code below may not trigger any cpufreq
	 * utilization updates, so do it here explicitly with the IOWAIT flag
	 * passed.
	 */
	if (p->in_iowait)
		cpufreq_update_this_cpu(rq, SCHED_CPUFREQ_IOWAIT);

4552
	for_each_sched_entity(se) {
4553
		if (se->on_rq)
4554 4555
			break;
		cfs_rq = cfs_rq_of(se);
4556
		enqueue_entity(cfs_rq, se, flags);
4557 4558 4559 4560 4561 4562

		/*
		 * 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.
4563
		 */
4564 4565
		if (cfs_rq_throttled(cfs_rq))
			break;
4566
		cfs_rq->h_nr_running++;
4567

4568
		flags = ENQUEUE_WAKEUP;
4569
	}
P
Peter Zijlstra 已提交
4570

P
Peter Zijlstra 已提交
4571
	for_each_sched_entity(se) {
4572
		cfs_rq = cfs_rq_of(se);
4573
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
4574

4575 4576 4577
		if (cfs_rq_throttled(cfs_rq))
			break;

4578
		update_load_avg(se, 1);
4579
		update_cfs_shares(cfs_rq);
P
Peter Zijlstra 已提交
4580 4581
	}

Y
Yuyang Du 已提交
4582
	if (!se)
4583
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
4584

4585
	hrtick_update(rq);
4586 4587
}

4588 4589
static void set_next_buddy(struct sched_entity *se);

4590 4591 4592 4593 4594
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
4595
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4596 4597
{
	struct cfs_rq *cfs_rq;
4598
	struct sched_entity *se = &p->se;
4599
	int task_sleep = flags & DEQUEUE_SLEEP;
4600 4601 4602

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4603
		dequeue_entity(cfs_rq, se, flags);
4604 4605 4606 4607 4608 4609 4610 4611 4612

		/*
		 * 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;
4613
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4614

4615
		/* Don't dequeue parent if it has other entities besides us */
4616
		if (cfs_rq->load.weight) {
4617 4618
			/* Avoid re-evaluating load for this entity: */
			se = parent_entity(se);
4619 4620 4621 4622
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
4623 4624
			if (task_sleep && se && !throttled_hierarchy(cfs_rq))
				set_next_buddy(se);
4625
			break;
4626
		}
4627
		flags |= DEQUEUE_SLEEP;
4628
	}
P
Peter Zijlstra 已提交
4629

P
Peter Zijlstra 已提交
4630
	for_each_sched_entity(se) {
4631
		cfs_rq = cfs_rq_of(se);
4632
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4633

4634 4635 4636
		if (cfs_rq_throttled(cfs_rq))
			break;

4637
		update_load_avg(se, 1);
4638
		update_cfs_shares(cfs_rq);
P
Peter Zijlstra 已提交
4639 4640
	}

Y
Yuyang Du 已提交
4641
	if (!se)
4642
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
4643

4644
	hrtick_update(rq);
4645 4646
}

4647
#ifdef CONFIG_SMP
4648 4649 4650 4651 4652

/* Working cpumask for: load_balance, load_balance_newidle. */
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);

4653
#ifdef CONFIG_NO_HZ_COMMON
4654 4655 4656 4657 4658
/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
4659
 * The exact cpuload calculated at every tick would be:
4660
 *
4661 4662 4663 4664 4665 4666 4667
 *   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
4668 4669 4670
 *
 * decay_load_missed() below does efficient calculation of
 *
4671 4672 4673 4674 4675 4676
 *   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())
4677
 *
4678
 * The calculation is approximated on a 128 point scale.
4679 4680
 */
#define DEGRADE_SHIFT		7
4681 4682 4683 4684 4685 4686 4687 4688 4689

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 }
};
4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718

/*
 * 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;
}
4719
#endif /* CONFIG_NO_HZ_COMMON */
4720

4721
/**
4722
 * __cpu_load_update - update the rq->cpu_load[] statistics
4723 4724 4725 4726
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 *
4727
 * Update rq->cpu_load[] statistics. This function is usually called every
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
 * 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
4754
 * term.
4755
 */
4756 4757
static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
			    unsigned long pending_updates)
4758
{
4759
	unsigned long __maybe_unused tickless_load = this_rq->cpu_load[0];
4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770
	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 */

4771
		old_load = this_rq->cpu_load[i];
4772
#ifdef CONFIG_NO_HZ_COMMON
4773
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
4774 4775 4776 4777 4778 4779 4780 4781 4782
		if (tickless_load) {
			old_load -= decay_load_missed(tickless_load, pending_updates - 1, i);
			/*
			 * old_load can never be a negative value because a
			 * decayed tickless_load cannot be greater than the
			 * original tickless_load.
			 */
			old_load += tickless_load;
		}
4783
#endif
4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798
		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);
}

4799 4800 4801 4802 4803 4804
/* 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);
}

4805
#ifdef CONFIG_NO_HZ_COMMON
4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822
/*
 * 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 need to avoid the delta approach from the regular tick when
 * possible since that would seriously skew the load calculation. This is why we
 * use cpu_load_update_periodic() for CPUs out of nohz. However we'll rely on
 * jiffies deltas for updates happening while in nohz mode (idle ticks, idle
 * loop exit, nohz_idle_balance, nohz full exit...)
 *
 * This means we might still be one tick off for nohz periods.
 */

static void cpu_load_update_nohz(struct rq *this_rq,
				 unsigned long curr_jiffies,
				 unsigned long load)
4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833
{
	unsigned long pending_updates;

	pending_updates = curr_jiffies - this_rq->last_load_update_tick;
	if (pending_updates) {
		this_rq->last_load_update_tick = curr_jiffies;
		/*
		 * 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.
		 */
4834
		cpu_load_update(this_rq, load, pending_updates);
4835 4836 4837
	}
}

4838 4839 4840 4841
/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
4842
static void cpu_load_update_idle(struct rq *this_rq)
4843 4844 4845 4846
{
	/*
	 * bail if there's load or we're actually up-to-date.
	 */
4847
	if (weighted_cpuload(cpu_of(this_rq)))
4848 4849
		return;

4850
	cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), 0);
4851 4852 4853
}

/*
4854 4855 4856 4857
 * Record CPU load on nohz entry so we know the tickless load to account
 * on nohz exit. cpu_load[0] happens then to be updated more frequently
 * than other cpu_load[idx] but it should be fine as cpu_load readers
 * shouldn't rely into synchronized cpu_load[*] updates.
4858
 */
4859
void cpu_load_update_nohz_start(void)
4860 4861
{
	struct rq *this_rq = this_rq();
4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875

	/*
	 * This is all lockless but should be fine. If weighted_cpuload changes
	 * concurrently we'll exit nohz. And cpu_load write can race with
	 * cpu_load_update_idle() but both updater would be writing the same.
	 */
	this_rq->cpu_load[0] = weighted_cpuload(cpu_of(this_rq));
}

/*
 * Account the tickless load in the end of a nohz frame.
 */
void cpu_load_update_nohz_stop(void)
{
4876
	unsigned long curr_jiffies = READ_ONCE(jiffies);
4877 4878
	struct rq *this_rq = this_rq();
	unsigned long load;
4879 4880 4881 4882

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

4883
	load = weighted_cpuload(cpu_of(this_rq));
4884
	raw_spin_lock(&this_rq->lock);
4885
	update_rq_clock(this_rq);
4886
	cpu_load_update_nohz(this_rq, curr_jiffies, load);
4887 4888
	raw_spin_unlock(&this_rq->lock);
}
4889 4890 4891 4892 4893 4894 4895 4896
#else /* !CONFIG_NO_HZ_COMMON */
static inline void cpu_load_update_nohz(struct rq *this_rq,
					unsigned long curr_jiffies,
					unsigned long load) { }
#endif /* CONFIG_NO_HZ_COMMON */

static void cpu_load_update_periodic(struct rq *this_rq, unsigned long load)
{
4897
#ifdef CONFIG_NO_HZ_COMMON
4898 4899
	/* See the mess around cpu_load_update_nohz(). */
	this_rq->last_load_update_tick = READ_ONCE(jiffies);
4900
#endif
4901 4902
	cpu_load_update(this_rq, load, 1);
}
4903 4904 4905 4906

/*
 * Called from scheduler_tick()
 */
4907
void cpu_load_update_active(struct rq *this_rq)
4908
{
4909
	unsigned long load = weighted_cpuload(cpu_of(this_rq));
4910 4911 4912 4913 4914

	if (tick_nohz_tick_stopped())
		cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), load);
	else
		cpu_load_update_periodic(this_rq, load);
4915 4916
}

4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949
/*
 * 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);
}

4950
static unsigned long capacity_of(int cpu)
4951
{
4952
	return cpu_rq(cpu)->cpu_capacity;
4953 4954
}

4955 4956 4957 4958 4959
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

4960 4961 4962
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
4963
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
4964
	unsigned long load_avg = weighted_cpuload(cpu);
4965 4966

	if (nr_running)
4967
		return load_avg / nr_running;
4968 4969 4970 4971

	return 0;
}

4972
#ifdef CONFIG_FAIR_GROUP_SCHED
4973 4974 4975 4976 4977 4978
/*
 * 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.
4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021
 *
 * 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.
5022
 */
P
Peter Zijlstra 已提交
5023
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
5024
{
P
Peter Zijlstra 已提交
5025
	struct sched_entity *se = tg->se[cpu];
5026

5027
	if (!tg->parent)	/* the trivial, non-cgroup case */
5028 5029
		return wl;

P
Peter Zijlstra 已提交
5030
	for_each_sched_entity(se) {
5031 5032
		struct cfs_rq *cfs_rq = se->my_q;
		long W, w = cfs_rq_load_avg(cfs_rq);
P
Peter Zijlstra 已提交
5033

5034
		tg = cfs_rq->tg;
5035

5036 5037 5038
		/*
		 * W = @wg + \Sum rw_j
		 */
5039 5040 5041 5042 5043
		W = wg + atomic_long_read(&tg->load_avg);

		/* Ensure \Sum rw_j >= rw_i */
		W -= cfs_rq->tg_load_avg_contrib;
		W += w;
P
Peter Zijlstra 已提交
5044

5045 5046 5047
		/*
		 * w = rw_i + @wl
		 */
5048
		w += wl;
5049

5050 5051 5052 5053
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
5054
			wl = (w * (long)scale_load_down(tg->shares)) / W;
5055
		else
5056
			wl = scale_load_down(tg->shares);
5057

5058 5059 5060 5061 5062
		/*
		 * 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().
		 */
5063 5064
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
5065 5066 5067 5068

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
5069
		wl -= se->avg.load_avg;
5070 5071 5072 5073 5074 5075 5076 5077

		/*
		 * 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
5078 5079
		wg = 0;
	}
5080

P
Peter Zijlstra 已提交
5081
	return wl;
5082 5083
}
#else
P
Peter Zijlstra 已提交
5084

5085
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
5086
{
5087
	return wl;
5088
}
P
Peter Zijlstra 已提交
5089

5090 5091
#endif

P
Peter Zijlstra 已提交
5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108
static void record_wakee(struct task_struct *p)
{
	/*
	 * Only decay a single time; tasks that have less then 1 wakeup per
	 * jiffy will not have built up many flips.
	 */
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
		current->wakee_flips >>= 1;
		current->wakee_flip_decay_ts = jiffies;
	}

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

M
Mike Galbraith 已提交
5109 5110
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
P
Peter Zijlstra 已提交
5111
 *
M
Mike Galbraith 已提交
5112
 * A waker of many should wake a different task than the one last awakened
P
Peter Zijlstra 已提交
5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124
 * at a frequency roughly N times higher than one of its wakees.
 *
 * In order to determine whether we should let the load spread vs consolidating
 * 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.
M
Mike Galbraith 已提交
5125
 */
5126 5127
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
5128 5129
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
5130
	int factor = this_cpu_read(sd_llc_size);
5131

M
Mike Galbraith 已提交
5132 5133 5134 5135 5136
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
5137 5138
}

5139 5140
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
		       int prev_cpu, int sync)
5141
{
5142
	s64 this_load, load;
5143
	s64 this_eff_load, prev_eff_load;
5144
	int idx, this_cpu;
5145
	struct task_group *tg;
5146
	unsigned long weight;
5147
	int balanced;
5148

5149 5150 5151 5152
	idx	  = sd->wake_idx;
	this_cpu  = smp_processor_id();
	load	  = source_load(prev_cpu, idx);
	this_load = target_load(this_cpu, idx);
5153

5154 5155 5156 5157 5158
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
5159 5160
	if (sync) {
		tg = task_group(current);
5161
		weight = current->se.avg.load_avg;
5162

5163
		this_load += effective_load(tg, this_cpu, -weight, -weight);
5164 5165
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
5166

5167
	tg = task_group(p);
5168
	weight = p->se.avg.load_avg;
5169

5170 5171
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
5172 5173 5174
	 * 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.
5175 5176 5177 5178
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
5179 5180
	this_eff_load = 100;
	this_eff_load *= capacity_of(prev_cpu);
5181

5182 5183
	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
5184

5185
	if (this_load > 0) {
5186 5187 5188 5189
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
5190
	}
5191

5192
	balanced = this_eff_load <= prev_eff_load;
5193

5194
	schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts);
5195

5196 5197
	if (!balanced)
		return 0;
5198

5199 5200
	schedstat_inc(sd->ttwu_move_affine);
	schedstat_inc(p->se.statistics.nr_wakeups_affine);
5201 5202

	return 1;
5203 5204
}

5205 5206 5207 5208 5209 5210 5211 5212
static inline int task_util(struct task_struct *p);
static int cpu_util_wake(int cpu, struct task_struct *p);

static unsigned long capacity_spare_wake(int cpu, struct task_struct *p)
{
	return capacity_orig_of(cpu) - cpu_util_wake(cpu, p);
}

5213 5214 5215 5216 5217
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
5218
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
5219
		  int this_cpu, int sd_flag)
5220
{
5221
	struct sched_group *idlest = NULL, *group = sd->groups;
5222
	struct sched_group *most_spare_sg = NULL;
5223
	unsigned long min_load = ULONG_MAX, this_load = 0;
5224
	unsigned long most_spare = 0, this_spare = 0;
5225
	int load_idx = sd->forkexec_idx;
5226
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
5227

5228 5229 5230
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

5231
	do {
5232
		unsigned long load, avg_load, spare_cap, max_spare_cap;
5233 5234
		int local_group;
		int i;
5235

5236 5237
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
5238
					tsk_cpus_allowed(p)))
5239 5240 5241 5242 5243
			continue;

		local_group = cpumask_test_cpu(this_cpu,
					       sched_group_cpus(group));

5244 5245 5246 5247
		/*
		 * Tally up the load of all CPUs in the group and find
		 * the group containing the CPU with most spare capacity.
		 */
5248
		avg_load = 0;
5249
		max_spare_cap = 0;
5250 5251 5252 5253 5254 5255 5256 5257 5258

		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;
5259 5260 5261 5262 5263

			spare_cap = capacity_spare_wake(i, p);

			if (spare_cap > max_spare_cap)
				max_spare_cap = spare_cap;
5264 5265
		}

5266
		/* Adjust by relative CPU capacity of the group */
5267
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
5268 5269 5270

		if (local_group) {
			this_load = avg_load;
5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281
			this_spare = max_spare_cap;
		} else {
			if (avg_load < min_load) {
				min_load = avg_load;
				idlest = group;
			}

			if (most_spare < max_spare_cap) {
				most_spare = max_spare_cap;
				most_spare_sg = group;
			}
5282 5283 5284
		}
	} while (group = group->next, group != sd->groups);

5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297
	/*
	 * The cross-over point between using spare capacity or least load
	 * is too conservative for high utilization tasks on partially
	 * utilized systems if we require spare_capacity > task_util(p),
	 * so we allow for some task stuffing by using
	 * spare_capacity > task_util(p)/2.
	 */
	if (this_spare > task_util(p) / 2 &&
	    imbalance*this_spare > 100*most_spare)
		return NULL;
	else if (most_spare > task_util(p) / 2)
		return most_spare_sg;

5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309
	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;
5310 5311 5312 5313
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
5314 5315
	int i;

5316 5317 5318 5319
	/* Check if we have any choice: */
	if (group->group_weight == 1)
		return cpumask_first(sched_group_cpus(group));

5320
	/* Traverse only the allowed CPUs */
5321
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343
		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;
			}
5344
		} else if (shallowest_idle_cpu == -1) {
5345 5346 5347 5348 5349
			load = weighted_cpuload(i);
			if (load < min_load || (load == min_load && i == this_cpu)) {
				min_load = load;
				least_loaded_cpu = i;
			}
5350 5351 5352
		}
	}

5353
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
5354
}
5355

5356
/*
5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421
 * Implement a for_each_cpu() variant that starts the scan at a given cpu
 * (@start), and wraps around.
 *
 * This is used to scan for idle CPUs; such that not all CPUs looking for an
 * idle CPU find the same CPU. The down-side is that tasks tend to cycle
 * through the LLC domain.
 *
 * Especially tbench is found sensitive to this.
 */

static int cpumask_next_wrap(int n, const struct cpumask *mask, int start, int *wrapped)
{
	int next;

again:
	next = find_next_bit(cpumask_bits(mask), nr_cpumask_bits, n+1);

	if (*wrapped) {
		if (next >= start)
			return nr_cpumask_bits;
	} else {
		if (next >= nr_cpumask_bits) {
			*wrapped = 1;
			n = -1;
			goto again;
		}
	}

	return next;
}

#define for_each_cpu_wrap(cpu, mask, start, wrap)				\
	for ((wrap) = 0, (cpu) = (start)-1;					\
		(cpu) = cpumask_next_wrap((cpu), (mask), (start), &(wrap)),	\
		(cpu) < nr_cpumask_bits; )

#ifdef CONFIG_SCHED_SMT

static inline void set_idle_cores(int cpu, int val)
{
	struct sched_domain_shared *sds;

	sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
	if (sds)
		WRITE_ONCE(sds->has_idle_cores, val);
}

static inline bool test_idle_cores(int cpu, bool def)
{
	struct sched_domain_shared *sds;

	sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
	if (sds)
		return READ_ONCE(sds->has_idle_cores);

	return def;
}

/*
 * Scans the local SMT mask to see if the entire core is idle, and records this
 * information in sd_llc_shared->has_idle_cores.
 *
 * Since SMT siblings share all cache levels, inspecting this limited remote
 * state should be fairly cheap.
 */
P
Peter Zijlstra 已提交
5422
void __update_idle_core(struct rq *rq)
5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453
{
	int core = cpu_of(rq);
	int cpu;

	rcu_read_lock();
	if (test_idle_cores(core, true))
		goto unlock;

	for_each_cpu(cpu, cpu_smt_mask(core)) {
		if (cpu == core)
			continue;

		if (!idle_cpu(cpu))
			goto unlock;
	}

	set_idle_cores(core, 1);
unlock:
	rcu_read_unlock();
}

/*
 * Scan the entire LLC domain for idle cores; this dynamically switches off if
 * there are no idle cores left in the system; tracked through
 * sd_llc->shared->has_idle_cores and enabled through update_idle_core() above.
 */
static int select_idle_core(struct task_struct *p, struct sched_domain *sd, int target)
{
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(select_idle_mask);
	int core, cpu, wrap;

P
Peter Zijlstra 已提交
5454 5455 5456
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489
	if (!test_idle_cores(target, false))
		return -1;

	cpumask_and(cpus, sched_domain_span(sd), tsk_cpus_allowed(p));

	for_each_cpu_wrap(core, cpus, target, wrap) {
		bool idle = true;

		for_each_cpu(cpu, cpu_smt_mask(core)) {
			cpumask_clear_cpu(cpu, cpus);
			if (!idle_cpu(cpu))
				idle = false;
		}

		if (idle)
			return core;
	}

	/*
	 * Failed to find an idle core; stop looking for one.
	 */
	set_idle_cores(target, 0);

	return -1;
}

/*
 * Scan the local SMT mask for idle CPUs.
 */
static int select_idle_smt(struct task_struct *p, struct sched_domain *sd, int target)
{
	int cpu;

P
Peter Zijlstra 已提交
5490 5491 5492
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520
	for_each_cpu(cpu, cpu_smt_mask(target)) {
		if (!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
			continue;
		if (idle_cpu(cpu))
			return cpu;
	}

	return -1;
}

#else /* CONFIG_SCHED_SMT */

static inline int select_idle_core(struct task_struct *p, struct sched_domain *sd, int target)
{
	return -1;
}

static inline int select_idle_smt(struct task_struct *p, struct sched_domain *sd, int target)
{
	return -1;
}

#endif /* CONFIG_SCHED_SMT */

/*
 * Scan the LLC domain for idle CPUs; this is dynamically regulated by
 * comparing the average scan cost (tracked in sd->avg_scan_cost) against the
 * average idle time for this rq (as found in rq->avg_idle).
5521
 */
5522 5523
static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int target)
{
5524 5525
	struct sched_domain *this_sd;
	u64 avg_cost, avg_idle = this_rq()->avg_idle;
5526 5527 5528 5529
	u64 time, cost;
	s64 delta;
	int cpu, wrap;

5530 5531 5532 5533 5534 5535
	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
	if (!this_sd)
		return -1;

	avg_cost = this_sd->avg_scan_cost;

5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561
	/*
	 * Due to large variance we need a large fuzz factor; hackbench in
	 * particularly is sensitive here.
	 */
	if ((avg_idle / 512) < avg_cost)
		return -1;

	time = local_clock();

	for_each_cpu_wrap(cpu, sched_domain_span(sd), target, wrap) {
		if (!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
			continue;
		if (idle_cpu(cpu))
			break;
	}

	time = local_clock() - time;
	cost = this_sd->avg_scan_cost;
	delta = (s64)(time - cost) / 8;
	this_sd->avg_scan_cost += delta;

	return cpu;
}

/*
 * Try and locate an idle core/thread in the LLC cache domain.
5562
 */
5563
static int select_idle_sibling(struct task_struct *p, int prev, int target)
5564
{
5565
	struct sched_domain *sd;
5566
	int i;
5567

5568 5569
	if (idle_cpu(target))
		return target;
5570 5571

	/*
5572
	 * If the previous cpu is cache affine and idle, don't be stupid.
5573
	 */
5574 5575
	if (prev != target && cpus_share_cache(prev, target) && idle_cpu(prev))
		return prev;
5576

5577
	sd = rcu_dereference(per_cpu(sd_llc, target));
5578 5579
	if (!sd)
		return target;
5580

5581 5582 5583
	i = select_idle_core(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
5584

5585 5586 5587 5588 5589 5590 5591
	i = select_idle_cpu(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;

	i = select_idle_smt(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
5592

5593 5594
	return target;
}
5595

5596
/*
5597
 * cpu_util returns the amount of capacity of a CPU that is used by CFS
5598
 * tasks. The unit of the return value must be the one of capacity so we can
5599 5600
 * compare the utilization with the capacity of the CPU that is available for
 * CFS task (ie cpu_capacity).
5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620
 *
 * 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).
5621
 */
5622
static int cpu_util(int cpu)
5623
{
5624
	unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
5625 5626
	unsigned long capacity = capacity_orig_of(cpu);

5627
	return (util >= capacity) ? capacity : util;
5628
}
5629

5630 5631 5632 5633 5634
static inline int task_util(struct task_struct *p)
{
	return p->se.avg.util_avg;
}

5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652
/*
 * cpu_util_wake: Compute cpu utilization with any contributions from
 * the waking task p removed.
 */
static int cpu_util_wake(int cpu, struct task_struct *p)
{
	unsigned long util, capacity;

	/* Task has no contribution or is new */
	if (cpu != task_cpu(p) || !p->se.avg.last_update_time)
		return cpu_util(cpu);

	capacity = capacity_orig_of(cpu);
	util = max_t(long, cpu_rq(cpu)->cfs.avg.util_avg - task_util(p), 0);

	return (util >= capacity) ? capacity : util;
}

5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670
/*
 * Disable WAKE_AFFINE in the case where task @p doesn't fit in the
 * capacity of either the waking CPU @cpu or the previous CPU @prev_cpu.
 *
 * In that case WAKE_AFFINE doesn't make sense and we'll let
 * BALANCE_WAKE sort things out.
 */
static int wake_cap(struct task_struct *p, int cpu, int prev_cpu)
{
	long min_cap, max_cap;

	min_cap = min(capacity_orig_of(prev_cpu), capacity_orig_of(cpu));
	max_cap = cpu_rq(cpu)->rd->max_cpu_capacity;

	/* Minimum capacity is close to max, no need to abort wake_affine */
	if (max_cap - min_cap < max_cap >> 3)
		return 0;

5671 5672 5673
	/* Bring task utilization in sync with prev_cpu */
	sync_entity_load_avg(&p->se);

5674 5675 5676
	return min_cap * 1024 < task_util(p) * capacity_margin;
}

5677
/*
5678 5679 5680
 * 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.
5681
 *
5682 5683
 * 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.
5684
 *
5685
 * Returns the target cpu number.
5686 5687 5688
 *
 * preempt must be disabled.
 */
5689
static int
5690
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
5691
{
5692
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
5693
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
5694
	int new_cpu = prev_cpu;
5695
	int want_affine = 0;
5696
	int sync = wake_flags & WF_SYNC;
5697

P
Peter Zijlstra 已提交
5698 5699
	if (sd_flag & SD_BALANCE_WAKE) {
		record_wakee(p);
5700 5701
		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu)
			      && cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
P
Peter Zijlstra 已提交
5702
	}
5703

5704
	rcu_read_lock();
5705
	for_each_domain(cpu, tmp) {
5706
		if (!(tmp->flags & SD_LOAD_BALANCE))
M
Mike Galbraith 已提交
5707
			break;
5708

5709
		/*
5710 5711
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
5712
		 */
5713 5714 5715
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
5716
			break;
5717
		}
5718

5719
		if (tmp->flags & sd_flag)
5720
			sd = tmp;
M
Mike Galbraith 已提交
5721 5722
		else if (!want_affine)
			break;
5723 5724
	}

M
Mike Galbraith 已提交
5725 5726
	if (affine_sd) {
		sd = NULL; /* Prefer wake_affine over balance flags */
5727
		if (cpu != prev_cpu && wake_affine(affine_sd, p, prev_cpu, sync))
M
Mike Galbraith 已提交
5728
			new_cpu = cpu;
5729
	}
5730

M
Mike Galbraith 已提交
5731 5732
	if (!sd) {
		if (sd_flag & SD_BALANCE_WAKE) /* XXX always ? */
5733
			new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
M
Mike Galbraith 已提交
5734 5735

	} else while (sd) {
5736
		struct sched_group *group;
5737
		int weight;
5738

5739
		if (!(sd->flags & sd_flag)) {
5740 5741 5742
			sd = sd->child;
			continue;
		}
5743

5744
		group = find_idlest_group(sd, p, cpu, sd_flag);
5745 5746 5747 5748
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
5749

5750
		new_cpu = find_idlest_cpu(group, p, cpu);
5751 5752 5753 5754
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
5755
		}
5756 5757 5758

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
5759
		weight = sd->span_weight;
5760 5761
		sd = NULL;
		for_each_domain(cpu, tmp) {
5762
			if (weight <= tmp->span_weight)
5763
				break;
5764
			if (tmp->flags & sd_flag)
5765 5766 5767
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
5768
	}
5769
	rcu_read_unlock();
5770

5771
	return new_cpu;
5772
}
5773 5774 5775 5776

/*
 * 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
5777
 * previous cpu. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
5778
 */
5779
static void migrate_task_rq_fair(struct task_struct *p)
5780
{
5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806
	/*
	 * As blocked tasks retain absolute vruntime the migration needs to
	 * deal with this by subtracting the old and adding the new
	 * min_vruntime -- the latter is done by enqueue_entity() when placing
	 * the task on the new runqueue.
	 */
	if (p->state == TASK_WAKING) {
		struct sched_entity *se = &p->se;
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
		u64 min_vruntime;

#ifndef CONFIG_64BIT
		u64 min_vruntime_copy;

		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

		se->vruntime -= min_vruntime;
	}

5807
	/*
5808 5809 5810 5811 5812
	 * 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.
5813
	 */
5814 5815 5816 5817
	remove_entity_load_avg(&p->se);

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
5818 5819

	/* We have migrated, no longer consider this task hot */
5820
	p->se.exec_start = 0;
5821
}
5822 5823 5824 5825 5826

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
5827 5828
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
5829 5830
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
5831 5832 5833 5834
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
5835 5836
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
5837 5838 5839 5840 5841 5842 5843 5844 5845
	 *
	 * 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.
5846
	 */
5847
	return calc_delta_fair(gran, se);
5848 5849
}

5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871
/*
 * 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
5872
	gran = wakeup_gran(curr, se);
5873 5874 5875 5876 5877 5878
	if (vdiff > gran)
		return 1;

	return 0;
}

5879 5880
static void set_last_buddy(struct sched_entity *se)
{
5881 5882 5883 5884 5885
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
5886 5887 5888 5889
}

static void set_next_buddy(struct sched_entity *se)
{
5890 5891 5892 5893 5894
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
5895 5896
}

5897 5898
static void set_skip_buddy(struct sched_entity *se)
{
5899 5900
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
5901 5902
}

5903 5904 5905
/*
 * Preempt the current task with a newly woken task if needed:
 */
5906
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
5907 5908
{
	struct task_struct *curr = rq->curr;
5909
	struct sched_entity *se = &curr->se, *pse = &p->se;
5910
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
5911
	int scale = cfs_rq->nr_running >= sched_nr_latency;
5912
	int next_buddy_marked = 0;
5913

I
Ingo Molnar 已提交
5914 5915 5916
	if (unlikely(se == pse))
		return;

5917
	/*
5918
	 * This is possible from callers such as attach_tasks(), in which we
5919 5920 5921 5922 5923 5924 5925
	 * 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;

5926
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
5927
		set_next_buddy(pse);
5928 5929
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
5930

5931 5932 5933
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
5934 5935 5936 5937 5938 5939
	 *
	 * 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.
5940 5941 5942 5943
	 */
	if (test_tsk_need_resched(curr))
		return;

5944 5945 5946 5947 5948
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

5949
	/*
5950 5951
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
5952
	 */
5953
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
5954
		return;
5955

5956
	find_matching_se(&se, &pse);
5957
	update_curr(cfs_rq_of(se));
5958
	BUG_ON(!pse);
5959 5960 5961 5962 5963 5964 5965
	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);
5966
		goto preempt;
5967
	}
5968

5969
	return;
5970

5971
preempt:
5972
	resched_curr(rq);
5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986
	/*
	 * 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);
5987 5988
}

5989
static struct task_struct *
5990
pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
5991 5992 5993
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
5994
	struct task_struct *p;
5995
	int new_tasks;
5996

5997
again:
5998 5999
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
6000
		goto idle;
6001

6002
	if (prev->sched_class != &fair_sched_class)
6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021
		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.
		 */
6022 6023 6024 6025 6026
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
6027

6028 6029 6030 6031 6032 6033 6034 6035 6036
			/*
			 * 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;
		}
6037 6038 6039 6040 6041 6042 6043 6044 6045 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 6074 6075 6076

		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
6077

6078
	if (!cfs_rq->nr_running)
6079
		goto idle;
6080

6081
	put_prev_task(rq, prev);
6082

6083
	do {
6084
		se = pick_next_entity(cfs_rq, NULL);
6085
		set_next_entity(cfs_rq, se);
6086 6087 6088
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
6089
	p = task_of(se);
6090

6091 6092
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
6093 6094

	return p;
6095 6096

idle:
6097 6098 6099 6100 6101 6102
	/*
	 * 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.
	 */
6103
	lockdep_unpin_lock(&rq->lock, cookie);
6104
	new_tasks = idle_balance(rq);
6105
	lockdep_repin_lock(&rq->lock, cookie);
6106 6107 6108 6109 6110
	/*
	 * 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.
	 */
6111
	if (new_tasks < 0)
6112 6113
		return RETRY_TASK;

6114
	if (new_tasks > 0)
6115 6116 6117
		goto again;

	return NULL;
6118 6119 6120 6121 6122
}

/*
 * Account for a descheduled task:
 */
6123
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
6124 6125 6126 6127 6128 6129
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
6130
		put_prev_entity(cfs_rq, se);
6131 6132 6133
	}
}

6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158
/*
 * 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);
6159 6160 6161 6162 6163
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
6164
		rq_clock_skip_update(rq, true);
6165 6166 6167 6168 6169
	}

	set_skip_buddy(se);
}

6170 6171 6172 6173
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

6174 6175
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
6176 6177 6178 6179 6180 6181 6182 6183 6184 6185
		return false;

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

	yield_task_fair(rq);

	return true;
}

6186
#ifdef CONFIG_SMP
6187
/**************************************************
P
Peter Zijlstra 已提交
6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203
 * 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
6204
 * is derived from the nice value as per sched_prio_to_weight[].
P
Peter Zijlstra 已提交
6205 6206 6207 6208 6209 6210
 *
 * 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)
 *
6211
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
6212 6213 6214 6215 6216 6217
 * 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):
 *
6218
 *   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
6219 6220 6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256
 *
 * 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:
 *
6257
 *             log_2 n
P
Peter Zijlstra 已提交
6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302
 *   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.]
6303
 */
6304

6305 6306
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

6307 6308
enum fbq_type { regular, remote, all };

6309
#define LBF_ALL_PINNED	0x01
6310
#define LBF_NEED_BREAK	0x02
6311 6312
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
6313 6314 6315 6316 6317

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
6318
	int			src_cpu;
6319 6320 6321 6322

	int			dst_cpu;
	struct rq		*dst_rq;

6323 6324
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
6325
	enum cpu_idle_type	idle;
6326
	long			imbalance;
6327 6328 6329
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

6330
	unsigned int		flags;
6331 6332 6333 6334

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
6335 6336

	enum fbq_type		fbq_type;
6337
	struct list_head	tasks;
6338 6339
};

6340 6341 6342
/*
 * Is this task likely cache-hot:
 */
6343
static int task_hot(struct task_struct *p, struct lb_env *env)
6344 6345 6346
{
	s64 delta;

6347 6348
	lockdep_assert_held(&env->src_rq->lock);

6349 6350 6351 6352 6353 6354 6355 6356 6357
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
6358
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
6359 6360 6361 6362 6363 6364 6365 6366 6367
			(&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;

6368
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
6369 6370 6371 6372

	return delta < (s64)sysctl_sched_migration_cost;
}

6373
#ifdef CONFIG_NUMA_BALANCING
6374
/*
6375 6376 6377
 * 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.
6378
 */
6379
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
6380
{
6381
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
6382
	unsigned long src_faults, dst_faults;
6383 6384
	int src_nid, dst_nid;

6385
	if (!static_branch_likely(&sched_numa_balancing))
6386 6387
		return -1;

6388
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
6389
		return -1;
6390 6391 6392 6393

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

6394
	if (src_nid == dst_nid)
6395
		return -1;
6396

6397 6398 6399 6400 6401 6402 6403
	/* 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;
	}
6404

6405 6406
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
6407
		return 0;
6408

6409 6410 6411 6412 6413 6414
	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);
6415 6416
	}

6417
	return dst_faults < src_faults;
6418 6419
}

6420
#else
6421
static inline int migrate_degrades_locality(struct task_struct *p,
6422 6423
					     struct lb_env *env)
{
6424
	return -1;
6425
}
6426 6427
#endif

6428 6429 6430 6431
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
6432
int can_migrate_task(struct task_struct *p, struct lb_env *env)
6433
{
6434
	int tsk_cache_hot;
6435 6436 6437

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

6438 6439
	/*
	 * We do not migrate tasks that are:
6440
	 * 1) throttled_lb_pair, or
6441
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
6442 6443
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
6444
	 */
6445 6446 6447
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

6448
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
6449
		int cpu;
6450

6451
		schedstat_inc(p->se.statistics.nr_failed_migrations_affine);
6452

6453 6454
		env->flags |= LBF_SOME_PINNED;

6455 6456 6457 6458 6459 6460 6461 6462
		/*
		 * 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.
		 */
6463
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
6464 6465
			return 0;

6466 6467 6468
		/* 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))) {
6469
				env->flags |= LBF_DST_PINNED;
6470 6471 6472
				env->new_dst_cpu = cpu;
				break;
			}
6473
		}
6474

6475 6476
		return 0;
	}
6477 6478

	/* Record that we found atleast one task that could run on dst_cpu */
6479
	env->flags &= ~LBF_ALL_PINNED;
6480

6481
	if (task_running(env->src_rq, p)) {
6482
		schedstat_inc(p->se.statistics.nr_failed_migrations_running);
6483 6484 6485 6486 6487
		return 0;
	}

	/*
	 * Aggressive migration if:
6488 6489 6490
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
6491
	 */
6492 6493 6494
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
6495

6496
	if (tsk_cache_hot <= 0 ||
6497
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
6498
		if (tsk_cache_hot == 1) {
6499 6500
			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
			schedstat_inc(p->se.statistics.nr_forced_migrations);
6501
		}
6502 6503 6504
		return 1;
	}

6505
	schedstat_inc(p->se.statistics.nr_failed_migrations_hot);
Z
Zhang Hang 已提交
6506
	return 0;
6507 6508
}

6509
/*
6510 6511 6512 6513 6514 6515 6516
 * 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;
6517
	deactivate_task(env->src_rq, p, 0);
6518 6519 6520
	set_task_cpu(p, env->dst_cpu);
}

6521
/*
6522
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
6523 6524
 * part of active balancing operations within "domain".
 *
6525
 * Returns a task if successful and NULL otherwise.
6526
 */
6527
static struct task_struct *detach_one_task(struct lb_env *env)
6528 6529 6530
{
	struct task_struct *p, *n;

6531 6532
	lockdep_assert_held(&env->src_rq->lock);

6533 6534 6535
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
6536

6537
		detach_task(p, env);
6538

6539
		/*
6540
		 * Right now, this is only the second place where
6541
		 * lb_gained[env->idle] is updated (other is detach_tasks)
6542
		 * so we can safely collect stats here rather than
6543
		 * inside detach_tasks().
6544
		 */
6545
		schedstat_inc(env->sd->lb_gained[env->idle]);
6546
		return p;
6547
	}
6548
	return NULL;
6549 6550
}

6551 6552
static const unsigned int sched_nr_migrate_break = 32;

6553
/*
6554 6555
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
6556
 *
6557
 * Returns number of detached tasks if successful and 0 otherwise.
6558
 */
6559
static int detach_tasks(struct lb_env *env)
6560
{
6561 6562
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
6563
	unsigned long load;
6564 6565 6566
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
6567

6568
	if (env->imbalance <= 0)
6569
		return 0;
6570

6571
	while (!list_empty(tasks)) {
6572 6573 6574 6575 6576 6577 6578
		/*
		 * 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;

6579
		p = list_first_entry(tasks, struct task_struct, se.group_node);
6580

6581 6582
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
6583
		if (env->loop > env->loop_max)
6584
			break;
6585 6586

		/* take a breather every nr_migrate tasks */
6587
		if (env->loop > env->loop_break) {
6588
			env->loop_break += sched_nr_migrate_break;
6589
			env->flags |= LBF_NEED_BREAK;
6590
			break;
6591
		}
6592

6593
		if (!can_migrate_task(p, env))
6594 6595 6596
			goto next;

		load = task_h_load(p);
6597

6598
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
6599 6600
			goto next;

6601
		if ((load / 2) > env->imbalance)
6602
			goto next;
6603

6604 6605 6606 6607
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
6608
		env->imbalance -= load;
6609 6610

#ifdef CONFIG_PREEMPT
6611 6612
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
6613
		 * kernels will stop after the first task is detached to minimize
6614 6615
		 * the critical section.
		 */
6616
		if (env->idle == CPU_NEWLY_IDLE)
6617
			break;
6618 6619
#endif

6620 6621 6622 6623
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
6624
		if (env->imbalance <= 0)
6625
			break;
6626 6627 6628

		continue;
next:
6629
		list_move_tail(&p->se.group_node, tasks);
6630
	}
6631

6632
	/*
6633 6634 6635
	 * 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().
6636
	 */
6637
	schedstat_add(env->sd->lb_gained[env->idle], detached);
6638

6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650
	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);
6651
	p->on_rq = TASK_ON_RQ_QUEUED;
6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679
	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);
6680

6681 6682 6683 6684
		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
6685 6686
}

P
Peter Zijlstra 已提交
6687
#ifdef CONFIG_FAIR_GROUP_SCHED
6688
static void update_blocked_averages(int cpu)
6689 6690
{
	struct rq *rq = cpu_rq(cpu);
6691 6692
	struct cfs_rq *cfs_rq;
	unsigned long flags;
6693

6694 6695
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
6696

6697 6698 6699 6700
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
6701
	for_each_leaf_cfs_rq(rq, cfs_rq) {
6702 6703 6704
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
6705

6706
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, true))
6707 6708
			update_tg_load_avg(cfs_rq, 0);
	}
6709
	raw_spin_unlock_irqrestore(&rq->lock, flags);
6710 6711
}

6712
/*
6713
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
6714 6715 6716
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
6717
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
6718
{
6719 6720
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
6721
	unsigned long now = jiffies;
6722
	unsigned long load;
6723

6724
	if (cfs_rq->last_h_load_update == now)
6725 6726
		return;

6727 6728 6729 6730 6731 6732 6733
	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;
	}
6734

6735
	if (!se) {
6736
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
6737 6738 6739 6740 6741
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
6742 6743
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
6744 6745 6746 6747
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
6748 6749
}

6750
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
6751
{
6752
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
6753

6754
	update_cfs_rq_h_load(cfs_rq);
6755
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
6756
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
6757 6758
}
#else
6759
static inline void update_blocked_averages(int cpu)
6760
{
6761 6762 6763 6764 6765 6766
	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);
6767
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, true);
6768
	raw_spin_unlock_irqrestore(&rq->lock, flags);
6769 6770
}

6771
static unsigned long task_h_load(struct task_struct *p)
6772
{
6773
	return p->se.avg.load_avg;
6774
}
P
Peter Zijlstra 已提交
6775
#endif
6776 6777

/********** Helpers for find_busiest_group ************************/
6778 6779 6780 6781 6782 6783 6784

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

6785 6786 6787 6788 6789 6790 6791
/*
 * 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
6792
	unsigned long load_per_task;
6793
	unsigned long group_capacity;
6794
	unsigned long group_util; /* Total utilization of the group */
6795 6796 6797
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
6798
	enum group_type group_type;
6799
	int group_no_capacity;
6800 6801 6802 6803
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
6804 6805
};

J
Joonsoo Kim 已提交
6806 6807 6808 6809 6810 6811 6812 6813
/*
 * 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 */
6814
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
6815 6816 6817
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
6818
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
6819 6820
};

6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832
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,
6833
		.total_capacity = 0UL,
6834 6835
		.busiest_stat = {
			.avg_load = 0UL,
6836 6837
			.sum_nr_running = 0,
			.group_type = group_other,
6838 6839 6840 6841
		},
	};
}

6842 6843 6844
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
6845
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
6846 6847
 *
 * Return: The load index.
6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869
 */
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;
}

6870
static unsigned long scale_rt_capacity(int cpu)
6871 6872
{
	struct rq *rq = cpu_rq(cpu);
6873
	u64 total, used, age_stamp, avg;
6874
	s64 delta;
6875

6876 6877 6878 6879
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
6880 6881
	age_stamp = READ_ONCE(rq->age_stamp);
	avg = READ_ONCE(rq->rt_avg);
6882
	delta = __rq_clock_broken(rq) - age_stamp;
6883

6884 6885 6886 6887
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
6888

6889
	used = div_u64(avg, total);
6890

6891 6892
	if (likely(used < SCHED_CAPACITY_SCALE))
		return SCHED_CAPACITY_SCALE - used;
6893

6894
	return 1;
6895 6896
}

6897
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
6898
{
6899
	unsigned long capacity = arch_scale_cpu_capacity(sd, cpu);
6900 6901
	struct sched_group *sdg = sd->groups;

6902
	cpu_rq(cpu)->cpu_capacity_orig = capacity;
6903

6904
	capacity *= scale_rt_capacity(cpu);
6905
	capacity >>= SCHED_CAPACITY_SHIFT;
6906

6907 6908
	if (!capacity)
		capacity = 1;
6909

6910 6911
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
6912
	sdg->sgc->min_capacity = capacity;
6913 6914
}

6915
void update_group_capacity(struct sched_domain *sd, int cpu)
6916 6917 6918
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
6919
	unsigned long capacity, min_capacity;
6920 6921 6922 6923
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
6924
	sdg->sgc->next_update = jiffies + interval;
6925 6926

	if (!child) {
6927
		update_cpu_capacity(sd, cpu);
6928 6929 6930
		return;
	}

6931
	capacity = 0;
6932
	min_capacity = ULONG_MAX;
6933

P
Peter Zijlstra 已提交
6934 6935 6936 6937 6938 6939
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

6940
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
6941
			struct sched_group_capacity *sgc;
6942
			struct rq *rq = cpu_rq(cpu);
6943

6944
			/*
6945
			 * build_sched_domains() -> init_sched_groups_capacity()
6946 6947 6948
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
6949 6950
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
6951
			 *
6952
			 * This avoids capacity from being 0 and
6953 6954 6955
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
6956
				capacity += capacity_of(cpu);
6957 6958 6959
			} else {
				sgc = rq->sd->groups->sgc;
				capacity += sgc->capacity;
6960
			}
6961

6962
			min_capacity = min(capacity, min_capacity);
6963
		}
P
Peter Zijlstra 已提交
6964 6965 6966 6967
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
6968
		 */
P
Peter Zijlstra 已提交
6969 6970 6971

		group = child->groups;
		do {
6972 6973 6974 6975
			struct sched_group_capacity *sgc = group->sgc;

			capacity += sgc->capacity;
			min_capacity = min(sgc->min_capacity, min_capacity);
P
Peter Zijlstra 已提交
6976 6977 6978
			group = group->next;
		} while (group != child->groups);
	}
6979

6980
	sdg->sgc->capacity = capacity;
6981
	sdg->sgc->min_capacity = min_capacity;
6982 6983
}

6984
/*
6985 6986 6987
 * 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
6988 6989
 */
static inline int
6990
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
6991
{
6992 6993
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
6994 6995
}

6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011
/*
 * 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
7012 7013
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
7014 7015
 *
 * When this is so detected; this group becomes a candidate for busiest; see
7016
 * update_sd_pick_busiest(). And calculate_imbalance() and
7017
 * find_busiest_group() avoid some of the usual balance conditions to allow it
7018 7019 7020 7021 7022 7023 7024
 * 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.
 */

7025
static inline int sg_imbalanced(struct sched_group *group)
7026
{
7027
	return group->sgc->imbalance;
7028 7029
}

7030
/*
7031 7032 7033
 * 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
7034 7035
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
7036 7037 7038 7039 7040
 * 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.
7041
 */
7042 7043
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
7044
{
7045 7046
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
7047

7048
	if ((sgs->group_capacity * 100) >
7049
			(sgs->group_util * env->sd->imbalance_pct))
7050
		return true;
7051

7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064 7065 7066 7067
	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;
7068

7069
	if ((sgs->group_capacity * 100) <
7070
			(sgs->group_util * env->sd->imbalance_pct))
7071
		return true;
7072

7073
	return false;
7074 7075
}

7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086
/*
 * group_smaller_cpu_capacity: Returns true if sched_group sg has smaller
 * per-CPU capacity than sched_group ref.
 */
static inline bool
group_smaller_cpu_capacity(struct sched_group *sg, struct sched_group *ref)
{
	return sg->sgc->min_capacity * capacity_margin <
						ref->sgc->min_capacity * 1024;
}

7087 7088 7089
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
7090
{
7091
	if (sgs->group_no_capacity)
7092 7093 7094 7095 7096 7097 7098 7099
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

7100 7101
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
7102
 * @env: The load balancing environment.
7103 7104 7105 7106
 * @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.
7107
 * @overload: Indicate more than one runnable task for any CPU.
7108
 */
7109 7110
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
7111 7112
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
7113
{
7114
	unsigned long load;
7115
	int i, nr_running;
7116

7117 7118
	memset(sgs, 0, sizeof(*sgs));

7119
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
7120 7121 7122
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
7123
		if (local_group)
7124
			load = target_load(i, load_idx);
7125
		else
7126 7127 7128
			load = source_load(i, load_idx);

		sgs->group_load += load;
7129
		sgs->group_util += cpu_util(i);
7130
		sgs->sum_nr_running += rq->cfs.h_nr_running;
7131

7132 7133
		nr_running = rq->nr_running;
		if (nr_running > 1)
7134 7135
			*overload = true;

7136 7137 7138 7139
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
7140
		sgs->sum_weighted_load += weighted_cpuload(i);
7141 7142 7143 7144
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
7145
			sgs->idle_cpus++;
7146 7147
	}

7148 7149
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
7150
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
7151

7152
	if (sgs->sum_nr_running)
7153
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
7154

7155
	sgs->group_weight = group->group_weight;
7156

7157
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
7158
	sgs->group_type = group_classify(group, sgs);
7159 7160
}

7161 7162
/**
 * update_sd_pick_busiest - return 1 on busiest group
7163
 * @env: The load balancing environment.
7164 7165
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
7166
 * @sgs: sched_group statistics
7167 7168 7169
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
7170 7171 7172
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
7173
 */
7174
static bool update_sd_pick_busiest(struct lb_env *env,
7175 7176
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
7177
				   struct sg_lb_stats *sgs)
7178
{
7179
	struct sg_lb_stats *busiest = &sds->busiest_stat;
7180

7181
	if (sgs->group_type > busiest->group_type)
7182 7183
		return true;

7184 7185 7186 7187 7188 7189
	if (sgs->group_type < busiest->group_type)
		return false;

	if (sgs->avg_load <= busiest->avg_load)
		return false;

7190 7191 7192 7193 7194 7195 7196 7197 7198 7199 7200 7201 7202 7203
	if (!(env->sd->flags & SD_ASYM_CPUCAPACITY))
		goto asym_packing;

	/*
	 * Candidate sg has no more than one task per CPU and
	 * has higher per-CPU capacity. Migrating tasks to less
	 * capable CPUs may harm throughput. Maximize throughput,
	 * power/energy consequences are not considered.
	 */
	if (sgs->sum_nr_running <= sgs->group_weight &&
	    group_smaller_cpu_capacity(sds->local, sg))
		return false;

asym_packing:
7204 7205
	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
7206 7207
		return true;

7208 7209 7210
	/* No ASYM_PACKING if target cpu is already busy */
	if (env->idle == CPU_NOT_IDLE)
		return true;
7211 7212 7213 7214 7215
	/*
	 * 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.
	 */
7216
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
7217 7218 7219
		if (!sds->busiest)
			return true;

7220 7221
		/* Prefer to move from highest possible cpu's work */
		if (group_first_cpu(sds->busiest) < group_first_cpu(sg))
7222 7223 7224 7225 7226 7227
			return true;
	}

	return false;
}

7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238 7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253 7254 7255 7256 7257
#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 */

7258
/**
7259
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
7260
 * @env: The load balancing environment.
7261 7262
 * @sds: variable to hold the statistics for this sched_domain.
 */
7263
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
7264
{
7265 7266
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
7267
	struct sg_lb_stats tmp_sgs;
7268
	int load_idx, prefer_sibling = 0;
7269
	bool overload = false;
7270 7271 7272 7273

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

7274
	load_idx = get_sd_load_idx(env->sd, env->idle);
7275 7276

	do {
J
Joonsoo Kim 已提交
7277
		struct sg_lb_stats *sgs = &tmp_sgs;
7278 7279
		int local_group;

7280
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
7281 7282 7283
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
7284 7285

			if (env->idle != CPU_NEWLY_IDLE ||
7286 7287
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
7288
		}
7289

7290 7291
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
7292

7293 7294 7295
		if (local_group)
			goto next_group;

7296 7297
		/*
		 * In case the child domain prefers tasks go to siblings
7298
		 * first, lower the sg capacity so that we'll try
7299 7300
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
7301 7302 7303 7304
		 * 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).
7305
		 */
7306
		if (prefer_sibling && sds->local &&
7307 7308 7309
		    group_has_capacity(env, &sds->local_stat) &&
		    (sgs->sum_nr_running > 1)) {
			sgs->group_no_capacity = 1;
7310
			sgs->group_type = group_classify(sg, sgs);
7311
		}
7312

7313
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
7314
			sds->busiest = sg;
J
Joonsoo Kim 已提交
7315
			sds->busiest_stat = *sgs;
7316 7317
		}

7318 7319 7320
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
7321
		sds->total_capacity += sgs->group_capacity;
7322

7323
		sg = sg->next;
7324
	} while (sg != env->sd->groups);
7325 7326 7327

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
7328 7329 7330 7331 7332 7333 7334

	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;
	}

7335 7336 7337 7338 7339 7340 7341 7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353
}

/**
 * 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.
 *
7354
 * Return: 1 when packing is required and a task should be moved to
7355 7356
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
7357
 * @env: The load balancing environment.
7358 7359
 * @sds: Statistics of the sched_domain which is to be packed
 */
7360
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
7361 7362 7363
{
	int busiest_cpu;

7364
	if (!(env->sd->flags & SD_ASYM_PACKING))
7365 7366
		return 0;

7367 7368 7369
	if (env->idle == CPU_NOT_IDLE)
		return 0;

7370 7371 7372 7373
	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
7374
	if (env->dst_cpu > busiest_cpu)
7375 7376
		return 0;

7377
	env->imbalance = DIV_ROUND_CLOSEST(
7378
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
7379
		SCHED_CAPACITY_SCALE);
7380

7381
	return 1;
7382 7383 7384 7385 7386 7387
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
7388
 * @env: The load balancing environment.
7389 7390
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
7391 7392
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
7393
{
7394
	unsigned long tmp, capa_now = 0, capa_move = 0;
7395
	unsigned int imbn = 2;
7396
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
7397
	struct sg_lb_stats *local, *busiest;
7398

J
Joonsoo Kim 已提交
7399 7400
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
7401

J
Joonsoo Kim 已提交
7402 7403 7404 7405
	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;
7406

J
Joonsoo Kim 已提交
7407
	scaled_busy_load_per_task =
7408
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
7409
		busiest->group_capacity;
J
Joonsoo Kim 已提交
7410

7411 7412
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
7413
		env->imbalance = busiest->load_per_task;
7414 7415 7416 7417 7418
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
7419
	 * however we may be able to increase total CPU capacity used by
7420 7421 7422
	 * moving them.
	 */

7423
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
7424
			min(busiest->load_per_task, busiest->avg_load);
7425
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
7426
			min(local->load_per_task, local->avg_load);
7427
	capa_now /= SCHED_CAPACITY_SCALE;
7428 7429

	/* Amount of load we'd subtract */
7430
	if (busiest->avg_load > scaled_busy_load_per_task) {
7431
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
7432
			    min(busiest->load_per_task,
7433
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
7434
	}
7435 7436

	/* Amount of load we'd add */
7437
	if (busiest->avg_load * busiest->group_capacity <
7438
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
7439 7440
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
7441
	} else {
7442
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
7443
		      local->group_capacity;
J
Joonsoo Kim 已提交
7444
	}
7445
	capa_move += local->group_capacity *
7446
		    min(local->load_per_task, local->avg_load + tmp);
7447
	capa_move /= SCHED_CAPACITY_SCALE;
7448 7449

	/* Move if we gain throughput */
7450
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
7451
		env->imbalance = busiest->load_per_task;
7452 7453 7454 7455 7456
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
7457
 * @env: load balance environment
7458 7459
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
7460
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
7461
{
7462
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
7463 7464 7465 7466
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
7467

7468
	if (busiest->group_type == group_imbalanced) {
7469 7470 7471 7472
		/*
		 * 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
7473 7474
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
7475 7476
	}

7477
	/*
7478 7479 7480 7481
	 * Avg load of busiest sg can be less and avg load of local sg can
	 * be greater than avg load across all sgs of sd because avg load
	 * factors in sg capacity and sgs with smaller group_type are
	 * skipped when updating the busiest sg:
7482
	 */
7483 7484
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
7485 7486
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
7487 7488
	}

7489 7490 7491 7492 7493
	/*
	 * If there aren't any idle cpus, avoid creating some.
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
7494
		load_above_capacity = busiest->sum_nr_running * SCHED_CAPACITY_SCALE;
7495
		if (load_above_capacity > busiest->group_capacity) {
7496
			load_above_capacity -= busiest->group_capacity;
7497
			load_above_capacity *= scale_load_down(NICE_0_LOAD);
7498 7499
			load_above_capacity /= busiest->group_capacity;
		} else
7500
			load_above_capacity = ~0UL;
7501 7502 7503 7504 7505 7506
	}

	/*
	 * 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,
7507 7508
	 * we also don't want to reduce the group load below the group
	 * capacity. Thus we look for the minimum possible imbalance.
7509
	 */
7510
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
7511 7512

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
7513
	env->imbalance = min(
7514 7515
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
7516
	) / SCHED_CAPACITY_SCALE;
7517 7518 7519

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
7520
	 * there is no guarantee that any tasks will be moved so we'll have
7521 7522 7523
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
7524
	if (env->imbalance < busiest->load_per_task)
7525
		return fix_small_imbalance(env, sds);
7526
}
7527

7528 7529 7530 7531
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
7532
 * if there is an imbalance.
7533 7534 7535 7536
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
7537
 * @env: The load balancing environment.
7538
 *
7539
 * Return:	- The busiest group if imbalance exists.
7540
 */
J
Joonsoo Kim 已提交
7541
static struct sched_group *find_busiest_group(struct lb_env *env)
7542
{
J
Joonsoo Kim 已提交
7543
	struct sg_lb_stats *local, *busiest;
7544 7545
	struct sd_lb_stats sds;

7546
	init_sd_lb_stats(&sds);
7547 7548 7549 7550 7551

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
7552
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
7553 7554
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
7555

7556
	/* ASYM feature bypasses nice load balance check */
7557
	if (check_asym_packing(env, &sds))
7558 7559
		return sds.busiest;

7560
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
7561
	if (!sds.busiest || busiest->sum_nr_running == 0)
7562 7563
		goto out_balanced;

7564 7565
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
7566

P
Peter Zijlstra 已提交
7567 7568
	/*
	 * If the busiest group is imbalanced the below checks don't
7569
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
7570 7571
	 * isn't true due to cpus_allowed constraints and the like.
	 */
7572
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
7573 7574
		goto force_balance;

7575
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
7576 7577
	if (env->idle == CPU_NEWLY_IDLE && group_has_capacity(env, local) &&
	    busiest->group_no_capacity)
7578 7579
		goto force_balance;

7580
	/*
7581
	 * If the local group is busier than the selected busiest group
7582 7583
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
7584
	if (local->avg_load >= busiest->avg_load)
7585 7586
		goto out_balanced;

7587 7588 7589 7590
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
7591
	if (local->avg_load >= sds.avg_load)
7592 7593
		goto out_balanced;

7594
	if (env->idle == CPU_IDLE) {
7595
		/*
7596 7597 7598 7599 7600
		 * 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
7601
		 */
7602 7603
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
7604
			goto out_balanced;
7605 7606 7607 7608 7609
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
7610 7611
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
7612
			goto out_balanced;
7613
	}
7614

7615
force_balance:
7616
	/* Looks like there is an imbalance. Compute it */
7617
	calculate_imbalance(env, &sds);
7618 7619 7620
	return sds.busiest;

out_balanced:
7621
	env->imbalance = 0;
7622 7623 7624 7625 7626 7627
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
7628
static struct rq *find_busiest_queue(struct lb_env *env,
7629
				     struct sched_group *group)
7630 7631
{
	struct rq *busiest = NULL, *rq;
7632
	unsigned long busiest_load = 0, busiest_capacity = 1;
7633 7634
	int i;

7635
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
7636
		unsigned long capacity, wl;
7637 7638 7639 7640
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
7641

7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655 7656 7657 7658 7659 7660 7661 7662 7663
		/*
		 * 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;

7664
		capacity = capacity_of(i);
7665

7666
		wl = weighted_cpuload(i);
7667

7668 7669
		/*
		 * When comparing with imbalance, use weighted_cpuload()
7670
		 * which is not scaled with the cpu capacity.
7671
		 */
7672 7673 7674

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
7675 7676
			continue;

7677 7678
		/*
		 * For the load comparisons with the other cpu's, consider
7679 7680 7681
		 * 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.
7682
		 *
7683
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
7684
		 * multiplication to rid ourselves of the division works out
7685 7686
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
7687
		 */
7688
		if (wl * busiest_capacity > busiest_load * capacity) {
7689
			busiest_load = wl;
7690
			busiest_capacity = capacity;
7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701 7702 7703
			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

7704
static int need_active_balance(struct lb_env *env)
7705
{
7706 7707 7708
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
7709 7710 7711 7712 7713 7714

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
7715
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
7716
			return 1;
7717 7718
	}

7719 7720 7721 7722 7723 7724 7725 7726 7727 7728 7729 7730 7731
	/*
	 * 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;
	}

7732 7733 7734
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

7735 7736
static int active_load_balance_cpu_stop(void *data);

7737 7738 7739 7740 7741 7742 7743 7744 7745 7746 7747 7748 7749 7750 7751 7752 7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767
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.
	 */
7768
	return balance_cpu == env->dst_cpu;
7769 7770
}

7771 7772 7773 7774 7775 7776
/*
 * 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,
7777
			int *continue_balancing)
7778
{
7779
	int ld_moved, cur_ld_moved, active_balance = 0;
7780
	struct sched_domain *sd_parent = sd->parent;
7781 7782 7783
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
7784
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
7785

7786 7787
	struct lb_env env = {
		.sd		= sd,
7788 7789
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
7790
		.dst_grpmask    = sched_group_cpus(sd->groups),
7791
		.idle		= idle,
7792
		.loop_break	= sched_nr_migrate_break,
7793
		.cpus		= cpus,
7794
		.fbq_type	= all,
7795
		.tasks		= LIST_HEAD_INIT(env.tasks),
7796 7797
	};

7798 7799 7800 7801
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
7802
	if (idle == CPU_NEWLY_IDLE)
7803 7804
		env.dst_grpmask = NULL;

7805 7806
	cpumask_copy(cpus, cpu_active_mask);

7807
	schedstat_inc(sd->lb_count[idle]);
7808 7809

redo:
7810 7811
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
7812
		goto out_balanced;
7813
	}
7814

7815
	group = find_busiest_group(&env);
7816
	if (!group) {
7817
		schedstat_inc(sd->lb_nobusyg[idle]);
7818 7819 7820
		goto out_balanced;
	}

7821
	busiest = find_busiest_queue(&env, group);
7822
	if (!busiest) {
7823
		schedstat_inc(sd->lb_nobusyq[idle]);
7824 7825 7826
		goto out_balanced;
	}

7827
	BUG_ON(busiest == env.dst_rq);
7828

7829
	schedstat_add(sd->lb_imbalance[idle], env.imbalance);
7830

7831 7832 7833
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

7834 7835 7836 7837 7838 7839 7840 7841
	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.
		 */
7842
		env.flags |= LBF_ALL_PINNED;
7843
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
7844

7845
more_balance:
7846
		raw_spin_lock_irqsave(&busiest->lock, flags);
7847 7848 7849 7850 7851

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
7852
		cur_ld_moved = detach_tasks(&env);
7853 7854

		/*
7855 7856 7857 7858 7859
		 * 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.
7860
		 */
7861 7862 7863 7864 7865 7866 7867 7868

		raw_spin_unlock(&busiest->lock);

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

7869
		local_irq_restore(flags);
7870

7871 7872 7873 7874 7875
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

7876 7877 7878 7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894
		/*
		 * 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.
		 */
7895
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
7896

7897 7898 7899
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

7900
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
7901
			env.dst_cpu	 = env.new_dst_cpu;
7902
			env.flags	&= ~LBF_DST_PINNED;
7903 7904
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
7905

7906 7907 7908 7909 7910 7911
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
7912

7913 7914 7915 7916
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
7917
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
7918

7919
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
7920 7921 7922
				*group_imbalance = 1;
		}

7923
		/* All tasks on this runqueue were pinned by CPU affinity */
7924
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
7925
			cpumask_clear_cpu(cpu_of(busiest), cpus);
7926 7927 7928
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
7929
				goto redo;
7930
			}
7931
			goto out_all_pinned;
7932 7933 7934 7935
		}
	}

	if (!ld_moved) {
7936
		schedstat_inc(sd->lb_failed[idle]);
7937 7938 7939 7940 7941 7942 7943 7944
		/*
		 * 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++;
7945

7946
		if (need_active_balance(&env)) {
7947 7948
			raw_spin_lock_irqsave(&busiest->lock, flags);

7949 7950 7951
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
7952 7953
			 */
			if (!cpumask_test_cpu(this_cpu,
7954
					tsk_cpus_allowed(busiest->curr))) {
7955 7956
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
7957
				env.flags |= LBF_ALL_PINNED;
7958 7959 7960
				goto out_one_pinned;
			}

7961 7962 7963 7964 7965
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
7966 7967 7968 7969 7970 7971
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
7972

7973
			if (active_balance) {
7974 7975 7976
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
7977
			}
7978

7979
			/* We've kicked active balancing, force task migration. */
7980 7981 7982 7983 7984 7985 7986 7987 7988 7989 7990 7991 7992
			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
7993
		 * detach_tasks).
7994 7995 7996 7997 7998 7999 8000 8001
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018
	/*
	 * 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.
	 */
8019
	schedstat_inc(sd->lb_balanced[idle]);
8020 8021 8022 8023 8024

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
8025
	if (((env.flags & LBF_ALL_PINNED) &&
8026
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
8027 8028 8029
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

8030
	ld_moved = 0;
8031 8032 8033 8034
out:
	return ld_moved;
}

8035 8036 8037 8038 8039 8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050
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
8051
update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
8052 8053 8054
{
	unsigned long interval, next;

8055 8056
	/* used by idle balance, so cpu_busy = 0 */
	interval = get_sd_balance_interval(sd, 0);
8057 8058 8059 8060 8061 8062
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

8063 8064 8065 8066
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
8067
static int idle_balance(struct rq *this_rq)
8068
{
8069 8070
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
8071 8072
	struct sched_domain *sd;
	int pulled_task = 0;
8073
	u64 curr_cost = 0;
8074

8075 8076 8077 8078 8079 8080
	/*
	 * 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);

8081 8082
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
8083 8084 8085
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
8086
			update_next_balance(sd, &next_balance);
8087 8088
		rcu_read_unlock();

8089
		goto out;
8090
	}
8091

8092 8093
	raw_spin_unlock(&this_rq->lock);

8094
	update_blocked_averages(this_cpu);
8095
	rcu_read_lock();
8096
	for_each_domain(this_cpu, sd) {
8097
		int continue_balancing = 1;
8098
		u64 t0, domain_cost;
8099 8100 8101 8102

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

8103
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
8104
			update_next_balance(sd, &next_balance);
8105
			break;
8106
		}
8107

8108
		if (sd->flags & SD_BALANCE_NEWIDLE) {
8109 8110
			t0 = sched_clock_cpu(this_cpu);

8111
			pulled_task = load_balance(this_cpu, this_rq,
8112 8113
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
8114 8115 8116 8117 8118 8119

			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;
8120
		}
8121

8122
		update_next_balance(sd, &next_balance);
8123 8124 8125 8126 8127 8128

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
8129 8130
			break;
	}
8131
	rcu_read_unlock();
8132 8133 8134

	raw_spin_lock(&this_rq->lock);

8135 8136 8137
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

8138
	/*
8139 8140 8141
	 * 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.
8142
	 */
8143
	if (this_rq->cfs.h_nr_running && !pulled_task)
8144
		pulled_task = 1;
8145

8146 8147 8148
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
8149
		this_rq->next_balance = next_balance;
8150

8151
	/* Is there a task of a high priority class? */
8152
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
8153 8154
		pulled_task = -1;

8155
	if (pulled_task)
8156 8157
		this_rq->idle_stamp = 0;

8158
	return pulled_task;
8159 8160 8161
}

/*
8162 8163 8164 8165
 * 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.
8166
 */
8167
static int active_load_balance_cpu_stop(void *data)
8168
{
8169 8170
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
8171
	int target_cpu = busiest_rq->push_cpu;
8172
	struct rq *target_rq = cpu_rq(target_cpu);
8173
	struct sched_domain *sd;
8174
	struct task_struct *p = NULL;
8175 8176 8177 8178 8179 8180 8181

	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;
8182 8183 8184

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
8185
		goto out_unlock;
8186 8187 8188 8189 8190 8191 8192 8193 8194

	/*
	 * 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. */
8195
	rcu_read_lock();
8196 8197 8198 8199 8200 8201 8202
	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)) {
8203 8204
		struct lb_env env = {
			.sd		= sd,
8205 8206 8207 8208
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
8209 8210 8211
			.idle		= CPU_IDLE,
		};

8212
		schedstat_inc(sd->alb_count);
8213

8214
		p = detach_one_task(&env);
8215
		if (p) {
8216
			schedstat_inc(sd->alb_pushed);
8217 8218 8219
			/* Active balancing done, reset the failure counter. */
			sd->nr_balance_failed = 0;
		} else {
8220
			schedstat_inc(sd->alb_failed);
8221
		}
8222
	}
8223
	rcu_read_unlock();
8224 8225
out_unlock:
	busiest_rq->active_balance = 0;
8226 8227 8228 8229 8230 8231 8232
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

8233
	return 0;
8234 8235
}

8236 8237 8238 8239 8240
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

8241
#ifdef CONFIG_NO_HZ_COMMON
8242 8243 8244 8245 8246 8247
/*
 * 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.
 */
8248
static struct {
8249
	cpumask_var_t idle_cpus_mask;
8250
	atomic_t nr_cpus;
8251 8252
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
8253

8254
static inline int find_new_ilb(void)
8255
{
8256
	int ilb = cpumask_first(nohz.idle_cpus_mask);
8257

8258 8259 8260 8261
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
8262 8263
}

8264 8265 8266 8267 8268
/*
 * 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).
 */
8269
static void nohz_balancer_kick(void)
8270 8271 8272 8273 8274
{
	int ilb_cpu;

	nohz.next_balance++;

8275
	ilb_cpu = find_new_ilb();
8276

8277 8278
	if (ilb_cpu >= nr_cpu_ids)
		return;
8279

8280
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
8281 8282 8283 8284 8285 8286 8287 8288
		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);
8289 8290 8291
	return;
}

8292
void nohz_balance_exit_idle(unsigned int cpu)
8293 8294
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
8295 8296 8297 8298 8299 8300 8301
		/*
		 * 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);
		}
8302 8303 8304 8305
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

8306 8307 8308
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
8309
	int cpu = smp_processor_id();
8310 8311

	rcu_read_lock();
8312
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
8313 8314 8315 8316 8317

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

8318
	atomic_inc(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
8319
unlock:
8320 8321 8322 8323 8324 8325
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
8326
	int cpu = smp_processor_id();
8327 8328

	rcu_read_lock();
8329
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
8330 8331 8332 8333 8334

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

8335
	atomic_dec(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
8336
unlock:
8337 8338 8339
	rcu_read_unlock();
}

8340
/*
8341
 * This routine will record that the cpu is going idle with tick stopped.
8342
 * This info will be used in performing idle load balancing in the future.
8343
 */
8344
void nohz_balance_enter_idle(int cpu)
8345
{
8346 8347 8348 8349 8350 8351
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

8352 8353
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
8354

8355 8356 8357 8358 8359 8360
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

8361 8362 8363
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
8364 8365 8366 8367 8368
}
#endif

static DEFINE_SPINLOCK(balancing);

8369 8370 8371 8372
/*
 * 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.
 */
8373
void update_max_interval(void)
8374 8375 8376 8377
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

8378 8379 8380 8381
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
8382
 * Balancing parameters are set up in init_sched_domains.
8383
 */
8384
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
8385
{
8386
	int continue_balancing = 1;
8387
	int cpu = rq->cpu;
8388
	unsigned long interval;
8389
	struct sched_domain *sd;
8390 8391 8392
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
8393 8394
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
8395

8396
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
8397

8398
	rcu_read_lock();
8399
	for_each_domain(cpu, sd) {
8400 8401 8402 8403 8404 8405 8406 8407 8408 8409 8410 8411
		/*
		 * 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;

8412 8413 8414
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

8415 8416 8417 8418 8419 8420 8421 8422 8423 8424 8425
		/*
		 * 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;
		}

8426
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
8427 8428 8429 8430 8431 8432 8433 8434

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

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
8435
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
8436
				/*
8437
				 * The LBF_DST_PINNED logic could have changed
8438 8439
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
8440
				 */
8441
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
8442 8443
			}
			sd->last_balance = jiffies;
8444
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
8445 8446 8447 8448 8449 8450 8451 8452
		}
		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;
		}
8453 8454
	}
	if (need_decay) {
8455
		/*
8456 8457
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
8458
		 */
8459 8460
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
8461
	}
8462
	rcu_read_unlock();
8463 8464 8465 8466 8467 8468

	/*
	 * 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.
	 */
8469
	if (likely(update_next_balance)) {
8470
		rq->next_balance = next_balance;
8471 8472 8473 8474 8475 8476 8477 8478 8479 8480 8481 8482 8483 8484

#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
	}
8485 8486
}

8487
#ifdef CONFIG_NO_HZ_COMMON
8488
/*
8489
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
8490 8491
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
8492
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
8493
{
8494
	int this_cpu = this_rq->cpu;
8495 8496
	struct rq *rq;
	int balance_cpu;
8497 8498 8499
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
8500

8501 8502 8503
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
8504 8505

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
8506
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
8507 8508 8509 8510 8511 8512 8513
			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.
		 */
8514
		if (need_resched())
8515 8516
			break;

V
Vincent Guittot 已提交
8517 8518
		rq = cpu_rq(balance_cpu);

8519 8520 8521 8522 8523 8524 8525
		/*
		 * 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);
8526
			cpu_load_update_idle(rq);
8527 8528 8529
			raw_spin_unlock_irq(&rq->lock);
			rebalance_domains(rq, CPU_IDLE);
		}
8530

8531 8532 8533 8534
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
8535
	}
8536 8537 8538 8539 8540 8541 8542 8543

	/*
	 * 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;
8544 8545
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
8546 8547 8548
}

/*
8549
 * Current heuristic for kicking the idle load balancer in the presence
8550
 * of an idle cpu in the system.
8551
 *   - This rq has more than one task.
8552 8553 8554 8555
 *   - 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.
8556 8557
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
8558
 */
8559
static inline bool nohz_kick_needed(struct rq *rq)
8560 8561
{
	unsigned long now = jiffies;
8562
	struct sched_domain_shared *sds;
8563
	struct sched_domain *sd;
8564
	int nr_busy, cpu = rq->cpu;
8565
	bool kick = false;
8566

8567
	if (unlikely(rq->idle_balance))
8568
		return false;
8569

8570 8571 8572 8573
       /*
	* 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.
	*/
8574
	set_cpu_sd_state_busy();
8575
	nohz_balance_exit_idle(cpu);
8576 8577 8578 8579 8580 8581

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
8582
		return false;
8583 8584

	if (time_before(now, nohz.next_balance))
8585
		return false;
8586

8587
	if (rq->nr_running >= 2)
8588
		return true;
8589

8590
	rcu_read_lock();
8591 8592 8593 8594 8595 8596 8597
	sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
	if (sds) {
		/*
		 * XXX: write a coherent comment on why we do this.
		 * See also: http://lkml.kernel.org/r/20111202010832.602203411@sbsiddha-desk.sc.intel.com
		 */
		nr_busy = atomic_read(&sds->nr_busy_cpus);
8598 8599 8600 8601 8602
		if (nr_busy > 1) {
			kick = true;
			goto unlock;
		}

8603
	}
8604

8605 8606 8607 8608 8609 8610 8611 8612
	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
			kick = true;
			goto unlock;
		}
	}
8613

8614
	sd = rcu_dereference(per_cpu(sd_asym, cpu));
8615
	if (sd && (cpumask_first_and(nohz.idle_cpus_mask,
8616 8617 8618 8619
				  sched_domain_span(sd)) < cpu)) {
		kick = true;
		goto unlock;
	}
8620

8621
unlock:
8622
	rcu_read_unlock();
8623
	return kick;
8624 8625
}
#else
8626
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
8627 8628 8629 8630 8631 8632
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
8633
static __latent_entropy void run_rebalance_domains(struct softirq_action *h)
8634
{
8635
	struct rq *this_rq = this_rq();
8636
	enum cpu_idle_type idle = this_rq->idle_balance ?
8637 8638 8639
						CPU_IDLE : CPU_NOT_IDLE;

	/*
8640
	 * If this cpu has a pending nohz_balance_kick, then do the
8641
	 * balancing on behalf of the other idle cpus whose ticks are
8642 8643 8644 8645
	 * 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.
8646
	 */
8647
	nohz_idle_balance(this_rq, idle);
8648
	rebalance_domains(this_rq, idle);
8649 8650 8651 8652 8653
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
8654
void trigger_load_balance(struct rq *rq)
8655 8656
{
	/* Don't need to rebalance while attached to NULL domain */
8657 8658 8659 8660
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
8661
		raise_softirq(SCHED_SOFTIRQ);
8662
#ifdef CONFIG_NO_HZ_COMMON
8663
	if (nohz_kick_needed(rq))
8664
		nohz_balancer_kick();
8665
#endif
8666 8667
}

8668 8669 8670
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
8671 8672

	update_runtime_enabled(rq);
8673 8674 8675 8676 8677
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
8678 8679 8680

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
8681 8682
}

8683
#endif /* CONFIG_SMP */
8684

8685 8686 8687
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
8688
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
8689 8690 8691 8692 8693 8694
{
	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
8695
		entity_tick(cfs_rq, se, queued);
8696
	}
8697

8698
	if (static_branch_unlikely(&sched_numa_balancing))
8699
		task_tick_numa(rq, curr);
8700 8701 8702
}

/*
P
Peter Zijlstra 已提交
8703 8704 8705
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
8706
 */
P
Peter Zijlstra 已提交
8707
static void task_fork_fair(struct task_struct *p)
8708
{
8709 8710
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
P
Peter Zijlstra 已提交
8711
	struct rq *rq = this_rq();
8712

8713
	raw_spin_lock(&rq->lock);
8714 8715
	update_rq_clock(rq);

8716 8717
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;
8718 8719
	if (curr) {
		update_curr(cfs_rq);
8720
		se->vruntime = curr->vruntime;
8721
	}
8722
	place_entity(cfs_rq, se, 1);
8723

P
Peter Zijlstra 已提交
8724
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
8725
		/*
8726 8727 8728
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
8729
		swap(curr->vruntime, se->vruntime);
8730
		resched_curr(rq);
8731
	}
8732

8733
	se->vruntime -= cfs_rq->min_vruntime;
8734
	raw_spin_unlock(&rq->lock);
8735 8736
}

8737 8738 8739 8740
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
8741 8742
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
8743
{
8744
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
8745 8746
		return;

8747 8748 8749 8750 8751
	/*
	 * 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
8752
	if (rq->curr == p) {
8753
		if (p->prio > oldprio)
8754
			resched_curr(rq);
8755
	} else
8756
		check_preempt_curr(rq, p, 0);
8757 8758
}

8759
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
8760 8761 8762 8763
{
	struct sched_entity *se = &p->se;

	/*
8764 8765 8766 8767 8768 8769 8770 8771 8772 8773
	 * 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
8774
	 *
8775 8776 8777 8778
	 * - 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
8779
	 */
8780 8781 8782 8783 8784 8785 8786 8787 8788 8789
	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);
8790
	u64 now = cfs_rq_clock_task(cfs_rq);
8791 8792

	if (!vruntime_normalized(p)) {
P
Peter Zijlstra 已提交
8793 8794 8795 8796 8797 8798 8799
		/*
		 * 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;
	}
8800

8801
	/* Catch up with the cfs_rq and remove our load when we leave */
8802
	update_cfs_rq_load_avg(now, cfs_rq, false);
8803
	detach_entity_load_avg(cfs_rq, se);
8804
	update_tg_load_avg(cfs_rq, false);
P
Peter Zijlstra 已提交
8805 8806
}

8807
static void attach_task_cfs_rq(struct task_struct *p)
8808
{
8809
	struct sched_entity *se = &p->se;
8810
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
8811
	u64 now = cfs_rq_clock_task(cfs_rq);
8812 8813

#ifdef CONFIG_FAIR_GROUP_SCHED
8814 8815 8816 8817 8818 8819
	/*
	 * 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
8820

8821
	/* Synchronize task with its cfs_rq */
8822
	update_cfs_rq_load_avg(now, cfs_rq, false);
8823
	attach_entity_load_avg(cfs_rq, se);
8824
	update_tg_load_avg(cfs_rq, false);
8825 8826 8827 8828

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
8829

8830 8831 8832 8833 8834 8835 8836 8837
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);
8838

8839
	if (task_on_rq_queued(p)) {
8840
		/*
8841 8842 8843
		 * 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.
8844
		 */
8845 8846 8847 8848
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
8849
	}
8850 8851
}

8852 8853 8854 8855 8856 8857 8858 8859 8860
/* 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;

8861 8862 8863 8864 8865 8866 8867
	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);
	}
8868 8869
}

8870 8871 8872 8873 8874 8875 8876
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
8877
#ifdef CONFIG_SMP
8878 8879
	atomic_long_set(&cfs_rq->removed_load_avg, 0);
	atomic_long_set(&cfs_rq->removed_util_avg, 0);
8880
#endif
8881 8882
}

P
Peter Zijlstra 已提交
8883
#ifdef CONFIG_FAIR_GROUP_SCHED
8884 8885 8886 8887 8888 8889 8890 8891
static void task_set_group_fair(struct task_struct *p)
{
	struct sched_entity *se = &p->se;

	set_task_rq(p, task_cpu(p));
	se->depth = se->parent ? se->parent->depth + 1 : 0;
}

8892
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
8893
{
8894
	detach_task_cfs_rq(p);
8895
	set_task_rq(p, task_cpu(p));
8896 8897 8898 8899 8900

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
8901
	attach_task_cfs_rq(p);
P
Peter Zijlstra 已提交
8902
}
8903

8904 8905 8906 8907 8908 8909 8910 8911 8912 8913 8914 8915 8916
static void task_change_group_fair(struct task_struct *p, int type)
{
	switch (type) {
	case TASK_SET_GROUP:
		task_set_group_fair(p);
		break;

	case TASK_MOVE_GROUP:
		task_move_group_fair(p);
		break;
	}
}

8917 8918 8919 8920 8921 8922 8923 8924 8925
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]);
8926
		if (tg->se)
8927 8928 8929 8930 8931 8932 8933 8934 8935 8936
			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 sched_entity *se;
8937
	struct cfs_rq *cfs_rq;
8938 8939 8940 8941 8942 8943 8944 8945 8946 8947 8948 8949 8950 8951 8952 8953 8954 8955 8956 8957 8958 8959 8960 8961 8962 8963
	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]);
8964
		init_entity_runnable_average(se);
8965 8966 8967 8968 8969 8970 8971 8972 8973 8974
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

8975 8976 8977 8978 8979 8980 8981 8982 8983 8984 8985 8986
void online_fair_sched_group(struct task_group *tg)
{
	struct sched_entity *se;
	struct rq *rq;
	int i;

	for_each_possible_cpu(i) {
		rq = cpu_rq(i);
		se = tg->se[i];

		raw_spin_lock_irq(&rq->lock);
		post_init_entity_util_avg(se);
8987
		sync_throttle(tg, i);
8988 8989 8990 8991
		raw_spin_unlock_irq(&rq->lock);
	}
}

8992
void unregister_fair_sched_group(struct task_group *tg)
8993 8994
{
	unsigned long flags;
8995 8996
	struct rq *rq;
	int cpu;
8997

8998 8999 9000
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
9001

9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9013 9014
		/*
		 * 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);
	}
9015 9016 9017 9018 9019 9020 9021 9022 9023 9024 9025 9026 9027 9028 9029 9030 9031 9032 9033
}

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
9034
	if (!parent) {
9035
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
9036 9037
		se->depth = 0;
	} else {
9038
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
9039 9040
		se->depth = parent->depth + 1;
	}
9041 9042

	se->my_q = cfs_rq;
9043 9044
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
9045 9046 9047 9048 9049 9050 9051 9052 9053 9054 9055 9056 9057 9058 9059 9060 9061 9062 9063 9064 9065 9066 9067 9068 9069 9070 9071 9072 9073 9074
	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);
9075 9076 9077

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
9078
		for_each_sched_entity(se)
9079 9080 9081 9082 9083 9084 9085 9086 9087 9088 9089 9090 9091 9092 9093 9094 9095
			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;
}

9096 9097
void online_fair_sched_group(struct task_group *tg) { }

9098
void unregister_fair_sched_group(struct task_group *tg) { }
9099 9100 9101

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
9102

9103
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
9104 9105 9106 9107 9108 9109 9110 9111 9112
{
	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)
9113
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
9114 9115 9116 9117

	return rr_interval;
}

9118 9119 9120
/*
 * All the scheduling class methods:
 */
9121
const struct sched_class fair_sched_class = {
9122
	.next			= &idle_sched_class,
9123 9124 9125
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
9126
	.yield_to_task		= yield_to_task_fair,
9127

I
Ingo Molnar 已提交
9128
	.check_preempt_curr	= check_preempt_wakeup,
9129 9130 9131 9132

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

9133
#ifdef CONFIG_SMP
L
Li Zefan 已提交
9134
	.select_task_rq		= select_task_rq_fair,
9135
	.migrate_task_rq	= migrate_task_rq_fair,
9136

9137 9138
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
9139

9140
	.task_dead		= task_dead_fair,
9141
	.set_cpus_allowed	= set_cpus_allowed_common,
9142
#endif
9143

9144
	.set_curr_task          = set_curr_task_fair,
9145
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
9146
	.task_fork		= task_fork_fair,
9147 9148

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
9149
	.switched_from		= switched_from_fair,
9150
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
9151

9152 9153
	.get_rr_interval	= get_rr_interval_fair,

9154 9155
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
9156
#ifdef CONFIG_FAIR_GROUP_SCHED
9157
	.task_change_group	= task_change_group_fair,
P
Peter Zijlstra 已提交
9158
#endif
9159 9160 9161
};

#ifdef CONFIG_SCHED_DEBUG
9162
void print_cfs_stats(struct seq_file *m, int cpu)
9163 9164 9165
{
	struct cfs_rq *cfs_rq;

9166
	rcu_read_lock();
9167
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
9168
		print_cfs_rq(m, cpu, cfs_rq);
9169
	rcu_read_unlock();
9170
}
9171 9172 9173 9174 9175 9176 9177 9178 9179 9180 9181 9182 9183 9184 9185 9186 9187 9188 9189 9190 9191

#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 */
9192 9193 9194 9195 9196 9197

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

9198
#ifdef CONFIG_NO_HZ_COMMON
9199
	nohz.next_balance = jiffies;
9200 9201 9202 9203 9204
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
#endif
#endif /* SMP */

}