fair.c 265.3 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * 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>
19 20
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
21
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
22
 */
23
#include "sched.h"
24 25 26

#include <trace/events/sched.h>

27
/*
28
 * Targeted preemption latency for CPU-bound tasks:
29
 *
30
 * NOTE: this latency value is not the same as the concept of
I
Ingo Molnar 已提交
31 32 33
 * 'timeslice length' - timeslices in CFS are of variable length
 * and have no persistent notion like in traditional, time-slice
 * based scheduling concepts.
34
 *
I
Ingo Molnar 已提交
35 36
 * (to see the precise effective timeslice length of your workload,
 *  run vmstat and monitor the context-switches (cs) field)
37 38
 *
 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
39
 */
40 41
unsigned int sysctl_sched_latency			= 6000000ULL;
unsigned int normalized_sysctl_sched_latency		= 6000000ULL;
42

43 44 45 46
/*
 * The initial- and re-scaling of tunables is configurable
 *
 * Options are:
47 48 49 50 51 52
 *
 *   SCHED_TUNABLESCALING_NONE - unscaled, always *1
 *   SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
 *   SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
 *
 * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
53
 */
54
enum sched_tunable_scaling sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG;
55

56
/*
57
 * Minimal preemption granularity for CPU-bound tasks:
58
 *
59
 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
60
 */
61 62
unsigned int sysctl_sched_min_granularity		= 750000ULL;
unsigned int normalized_sysctl_sched_min_granularity	= 750000ULL;
63 64

/*
65
 * This value is kept at sysctl_sched_latency/sysctl_sched_min_granularity
66
 */
67
static unsigned int sched_nr_latency = 8;
68 69

/*
70
 * After fork, child runs first. If set to 0 (default) then
71
 * parent will (try to) run first.
72
 */
73
unsigned int sysctl_sched_child_runs_first __read_mostly;
74 75 76 77 78 79 80

/*
 * SCHED_OTHER wake-up granularity.
 *
 * This option delays the preemption effects of decoupled workloads
 * and reduces their over-scheduling. Synchronous workloads will still
 * have immediate wakeup/sleep latencies.
81 82
 *
 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
83
 */
84 85
unsigned int sysctl_sched_wakeup_granularity		= 1000000UL;
unsigned int normalized_sysctl_sched_wakeup_granularity	= 1000000UL;
86

87
const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
88

T
Tim Chen 已提交
89 90
#ifdef CONFIG_SMP
/*
91
 * For asym packing, by default the lower numbered CPU has higher priority.
T
Tim Chen 已提交
92 93 94 95 96 97 98
 */
int __weak arch_asym_cpu_priority(int cpu)
{
	return -cpu;
}
#endif

99 100 101 102 103 104 105 106 107
#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.
 *
108 109 110
 * (default: 5 msec, units: microseconds)
 */
unsigned int sysctl_sched_cfs_bandwidth_slice		= 5000UL;
111 112
#endif

113 114
/*
 * The margin used when comparing utilization with CPU capacity:
115
 * util * margin < capacity * 1024
116 117
 *
 * (default: ~20%)
118
 */
119
unsigned int capacity_margin				= 1280;
120

121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
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;
}

139 140 141 142 143 144 145 146 147
/*
 * 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:
 */
148
static unsigned int get_update_sysctl_factor(void)
149
{
150
	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
	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();
}

186
#define WMULT_CONST	(~0U)
187 188
#define WMULT_SHIFT	32

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204
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;
}
205 206

/*
207 208 209 210
 * delta_exec * weight / lw.weight
 *   OR
 * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT
 *
211
 * Either weight := NICE_0_LOAD and lw \e sched_prio_to_wmult[], in which case
212 213 214 215 216
 * 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.
217
 */
218
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
219
{
220 221
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
222

223
	__update_inv_weight(lw);
224

225 226 227 228 229
	if (unlikely(fact >> 32)) {
		while (fact >> 32) {
			fact >>= 1;
			shift--;
		}
230 231
	}

232 233
	/* hint to use a 32x32->64 mul */
	fact = (u64)(u32)fact * lw->inv_weight;
234

235 236 237 238
	while (fact >> 32) {
		fact >>= 1;
		shift--;
	}
239

240
	return mul_u64_u32_shr(delta_exec, fact, shift);
241 242 243 244
}


const struct sched_class fair_sched_class;
245

246 247 248 249
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

250
#ifdef CONFIG_FAIR_GROUP_SCHED
251

252
/* cpu runqueue to which this cfs_rq is attached */
253 254
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
255
	return cfs_rq->rq;
256 257
}

258 259
static inline struct task_struct *task_of(struct sched_entity *se)
{
260
	SCHED_WARN_ON(!entity_is_task(se));
261 262 263
	return container_of(se, struct task_struct, se);
}

P
Peter Zijlstra 已提交
264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
/* 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;
}

285 286 287
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
288 289
		struct rq *rq = rq_of(cfs_rq);
		int cpu = cpu_of(rq);
290 291 292
		/*
		 * Ensure we either appear before our parent (if already
		 * enqueued) or force our parent to appear after us when it is
293 294 295 296 297
		 * enqueued. The fact that we always enqueue bottom-up
		 * reduces this to two cases and a special case for the root
		 * cfs_rq. Furthermore, it also means that we will always reset
		 * tmp_alone_branch either when the branch is connected
		 * to a tree or when we reach the beg of the tree
298 299
		 */
		if (cfs_rq->tg->parent &&
300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319
		    cfs_rq->tg->parent->cfs_rq[cpu]->on_list) {
			/*
			 * If parent is already on the list, we add the child
			 * just before. Thanks to circular linked property of
			 * the list, this means to put the child at the tail
			 * of the list that starts by parent.
			 */
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
				&(cfs_rq->tg->parent->cfs_rq[cpu]->leaf_cfs_rq_list));
			/*
			 * The branch is now connected to its tree so we can
			 * reset tmp_alone_branch to the beginning of the
			 * list.
			 */
			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
		} else if (!cfs_rq->tg->parent) {
			/*
			 * cfs rq without parent should be put
			 * at the tail of the list.
			 */
320
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
				&rq->leaf_cfs_rq_list);
			/*
			 * We have reach the beg of a tree so we can reset
			 * tmp_alone_branch to the beginning of the list.
			 */
			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
		} else {
			/*
			 * The parent has not already been added so we want to
			 * make sure that it will be put after us.
			 * tmp_alone_branch points to the beg of the branch
			 * where we will add parent.
			 */
			list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
				rq->tmp_alone_branch);
			/*
			 * update tmp_alone_branch to points to the new beg
			 * of the branch
			 */
			rq->tmp_alone_branch = &cfs_rq->leaf_cfs_rq_list;
341
		}
342 343 344 345 346 347 348 349 350 351 352 353 354

		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
355
/* Iterate thr' all leaf cfs_rq's on a runqueue */
356 357 358
#define for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos)			\
	list_for_each_entry_safe(cfs_rq, pos, &rq->leaf_cfs_rq_list,	\
				 leaf_cfs_rq_list)
P
Peter Zijlstra 已提交
359 360

/* Do the two (enqueued) entities belong to the same group ? */
P
Peter Zijlstra 已提交
361
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
362 363 364
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
365
		return se->cfs_rq;
P
Peter Zijlstra 已提交
366

P
Peter Zijlstra 已提交
367
	return NULL;
P
Peter Zijlstra 已提交
368 369 370 371 372 373 374
}

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

375 376 377 378 379 380 381 382 383 384 385 386 387
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
388 389
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406

	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);
	}
}

407 408 409 410 411 412
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
413

414 415 416
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
417 418 419
}


P
Peter Zijlstra 已提交
420 421
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
422

P
Peter Zijlstra 已提交
423
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
424
{
P
Peter Zijlstra 已提交
425
	return &task_rq(p)->cfs;
426 427
}

P
Peter Zijlstra 已提交
428 429 430 431 432 433 434 435 436 437 438 439 440 441
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;
}

442 443 444 445 446 447 448 449
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)
{
}

450 451
#define for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos)	\
		for (cfs_rq = &rq->cfs, pos = NULL; cfs_rq; cfs_rq = pos)
P
Peter Zijlstra 已提交
452 453 454 455 456 457

static inline struct sched_entity *parent_entity(struct sched_entity *se)
{
	return NULL;
}

458 459 460 461 462
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
463 464
#endif	/* CONFIG_FAIR_GROUP_SCHED */

465
static __always_inline
466
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
467 468 469 470 471

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

472
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
473
{
474
	s64 delta = (s64)(vruntime - max_vruntime);
475
	if (delta > 0)
476
		max_vruntime = vruntime;
477

478
	return max_vruntime;
479 480
}

481
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
482 483 484 485 486 487 488 489
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

490 491 492 493 494 495
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

496 497
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
498
	struct sched_entity *curr = cfs_rq->curr;
499
	struct rb_node *leftmost = rb_first_cached(&cfs_rq->tasks_timeline);
500

501 502
	u64 vruntime = cfs_rq->min_vruntime;

503 504 505 506 507 508
	if (curr) {
		if (curr->on_rq)
			vruntime = curr->vruntime;
		else
			curr = NULL;
	}
509

510 511 512
	if (leftmost) { /* non-empty tree */
		struct sched_entity *se;
		se = rb_entry(leftmost, struct sched_entity, run_node);
513

514
		if (!curr)
515 516 517 518 519
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

520
	/* ensure we never gain time by being placed backwards. */
521
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
522 523 524 525
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
526 527
}

528 529 530
/*
 * Enqueue an entity into the rb-tree:
 */
531
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
532
{
533
	struct rb_node **link = &cfs_rq->tasks_timeline.rb_root.rb_node;
534 535
	struct rb_node *parent = NULL;
	struct sched_entity *entry;
536
	bool leftmost = true;
537 538 539 540 541 542 543 544 545 546 547

	/*
	 * 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.
		 */
548
		if (entity_before(se, entry)) {
549 550 551
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
552
			leftmost = false;
553 554 555 556
		}
	}

	rb_link_node(&se->run_node, parent, link);
557 558
	rb_insert_color_cached(&se->run_node,
			       &cfs_rq->tasks_timeline, leftmost);
559 560
}

561
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
562
{
563
	rb_erase_cached(&se->run_node, &cfs_rq->tasks_timeline);
564 565
}

566
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
567
{
568
	struct rb_node *left = rb_first_cached(&cfs_rq->tasks_timeline);
569 570 571 572 573

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
574 575
}

576 577 578 579 580 581 582 583 584 585 586
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
587
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
588
{
589
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline.rb_root);
590

591 592
	if (!last)
		return NULL;
593 594

	return rb_entry(last, struct sched_entity, run_node);
595 596
}

597 598 599 600
/**************************************************************
 * Scheduling class statistics methods:
 */

601
int sched_proc_update_handler(struct ctl_table *table, int write,
602
		void __user *buffer, size_t *lenp,
603 604
		loff_t *ppos)
{
605
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
606
	unsigned int factor = get_update_sysctl_factor();
607 608 609 610 611 612 613

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

614 615 616 617 618 619 620
#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

621 622 623
	return 0;
}
#endif
624

625
/*
626
 * delta /= w
627
 */
628
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
629
{
630
	if (unlikely(se->load.weight != NICE_0_LOAD))
631
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
632 633 634 635

	return delta;
}

636 637 638
/*
 * The idea is to set a period in which each task runs once.
 *
639
 * When there are too many tasks (sched_nr_latency) we have to stretch
640 641 642 643
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
644 645
static u64 __sched_period(unsigned long nr_running)
{
646 647 648 649
	if (unlikely(nr_running > sched_nr_latency))
		return nr_running * sysctl_sched_min_granularity;
	else
		return sysctl_sched_latency;
650 651
}

652 653 654 655
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
656
 * s = p*P[w/rw]
657
 */
P
Peter Zijlstra 已提交
658
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
659
{
M
Mike Galbraith 已提交
660
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
661

M
Mike Galbraith 已提交
662
	for_each_sched_entity(se) {
L
Lin Ming 已提交
663
		struct load_weight *load;
664
		struct load_weight lw;
L
Lin Ming 已提交
665 666 667

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
668

M
Mike Galbraith 已提交
669
		if (unlikely(!se->on_rq)) {
670
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
671 672 673 674

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
675
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
676 677
	}
	return slice;
678 679
}

680
/*
A
Andrei Epure 已提交
681
 * We calculate the vruntime slice of a to-be-inserted task.
682
 *
683
 * vs = s/w
684
 */
685
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
686
{
687
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
688 689
}

690
#ifdef CONFIG_SMP
691
#include "pelt.h"
692 693
#include "sched-pelt.h"

694
static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu);
695 696
static unsigned long task_h_load(struct task_struct *p);

697 698
/* Give new sched_entity start runnable values to heavy its load in infant time */
void init_entity_runnable_average(struct sched_entity *se)
699
{
700
	struct sched_avg *sa = &se->avg;
701

702 703
	memset(sa, 0, sizeof(*sa));

704 705 706 707 708 709 710
	/*
	 * 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))
711 712
		sa->runnable_load_avg = sa->load_avg = scale_load_down(se->load.weight);

713 714
	se->runnable_weight = se->load.weight;

715
	/* when this task enqueue'ed, it will contribute to its cfs_rq's load_avg */
716
}
717

718
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
719
static void attach_entity_cfs_rq(struct sched_entity *se);
720

721 722 723 724 725 726 727 728 729 730 731 732 733
/*
 * 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:
 *
734
 *   util_avg_cap = (cpu_scale - cfs_rq->avg.util_avg) / 2^n
735
 *
736
 * where n denotes the nth task and cpu_scale the CPU capacity.
737
 *
738 739
 * For example, for a CPU with 1024 of capacity, a simplest series from
 * the beginning would be like:
740 741 742 743 744 745 746 747 748 749 750
 *
 *  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;
751 752
	long cpu_scale = arch_scale_cpu_capacity(NULL, cpu_of(rq_of(cfs_rq)));
	long cap = (long)(cpu_scale - cfs_rq->avg.util_avg) / 2;
753 754 755 756 757 758 759 760 761 762 763 764

	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;
		}
	}
765 766 767 768 769 770 771

	if (entity_is_task(se)) {
		struct task_struct *p = task_of(se);
		if (p->sched_class != &fair_sched_class) {
			/*
			 * For !fair tasks do:
			 *
772
			update_cfs_rq_load_avg(now, cfs_rq);
773
			attach_entity_load_avg(cfs_rq, se, 0);
774 775 776 777 778
			switched_from_fair(rq, p);
			 *
			 * such that the next switched_to_fair() has the
			 * expected state.
			 */
779
			se->avg.last_update_time = cfs_rq_clock_task(cfs_rq);
780 781 782 783
			return;
		}
	}

784
	attach_entity_cfs_rq(se);
785 786
}

787
#else /* !CONFIG_SMP */
788
void init_entity_runnable_average(struct sched_entity *se)
789 790
{
}
791 792 793
void post_init_entity_util_avg(struct sched_entity *se)
{
}
794 795 796
static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
{
}
797
#endif /* CONFIG_SMP */
798

799
/*
800
 * Update the current task's runtime statistics.
801
 */
802
static void update_curr(struct cfs_rq *cfs_rq)
803
{
804
	struct sched_entity *curr = cfs_rq->curr;
805
	u64 now = rq_clock_task(rq_of(cfs_rq));
806
	u64 delta_exec;
807 808 809 810

	if (unlikely(!curr))
		return;

811 812
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
813
		return;
814

I
Ingo Molnar 已提交
815
	curr->exec_start = now;
816

817 818 819 820
	schedstat_set(curr->statistics.exec_max,
		      max(delta_exec, curr->statistics.exec_max));

	curr->sum_exec_runtime += delta_exec;
821
	schedstat_add(cfs_rq->exec_clock, delta_exec);
822 823 824 825

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

826 827 828
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

829
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
830
		cgroup_account_cputime(curtask, delta_exec);
831
		account_group_exec_runtime(curtask, delta_exec);
832
	}
833 834

	account_cfs_rq_runtime(cfs_rq, delta_exec);
835 836
}

837 838 839 840 841
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

842
static inline void
843
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
844
{
845 846 847 848 849 850 851
	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);
852 853

	if (entity_is_task(se) && task_on_rq_migrating(task_of(se)) &&
854 855
	    likely(wait_start > prev_wait_start))
		wait_start -= prev_wait_start;
856

857
	__schedstat_set(se->statistics.wait_start, wait_start);
858 859
}

860
static inline void
861 862 863
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *p;
864 865
	u64 delta;

866 867 868 869
	if (!schedstat_enabled())
		return;

	delta = rq_clock(rq_of(cfs_rq)) - schedstat_val(se->statistics.wait_start);
870 871 872 873 874 875 876 877 878

	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.
			 */
879
			__schedstat_set(se->statistics.wait_start, delta);
880 881 882 883 884
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

885
	__schedstat_set(se->statistics.wait_max,
886
		      max(schedstat_val(se->statistics.wait_max), delta));
887 888 889
	__schedstat_inc(se->statistics.wait_count);
	__schedstat_add(se->statistics.wait_sum, delta);
	__schedstat_set(se->statistics.wait_start, 0);
890 891
}

892
static inline void
893 894 895
update_stats_enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *tsk = NULL;
896 897 898 899 900 901 902
	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);
903 904 905 906

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

907 908
	if (sleep_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - sleep_start;
909 910 911 912

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

913
		if (unlikely(delta > schedstat_val(se->statistics.sleep_max)))
914
			__schedstat_set(se->statistics.sleep_max, delta);
915

916 917
		__schedstat_set(se->statistics.sleep_start, 0);
		__schedstat_add(se->statistics.sum_sleep_runtime, delta);
918 919 920 921 922 923

		if (tsk) {
			account_scheduler_latency(tsk, delta >> 10, 1);
			trace_sched_stat_sleep(tsk, delta);
		}
	}
924 925
	if (block_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - block_start;
926 927 928 929

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

930
		if (unlikely(delta > schedstat_val(se->statistics.block_max)))
931
			__schedstat_set(se->statistics.block_max, delta);
932

933 934
		__schedstat_set(se->statistics.block_start, 0);
		__schedstat_add(se->statistics.sum_sleep_runtime, delta);
935 936 937

		if (tsk) {
			if (tsk->in_iowait) {
938 939
				__schedstat_add(se->statistics.iowait_sum, delta);
				__schedstat_inc(se->statistics.iowait_count);
940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
				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);
		}
	}
958 959
}

960 961 962
/*
 * Task is being enqueued - update stats:
 */
963
static inline void
964
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
965
{
966 967 968
	if (!schedstat_enabled())
		return;

969 970 971 972
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
973
	if (se != cfs_rq->curr)
974
		update_stats_wait_start(cfs_rq, se);
975 976 977

	if (flags & ENQUEUE_WAKEUP)
		update_stats_enqueue_sleeper(cfs_rq, se);
978 979 980
}

static inline void
981
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
982
{
983 984 985 986

	if (!schedstat_enabled())
		return;

987 988 989 990
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
991
	if (se != cfs_rq->curr)
992
		update_stats_wait_end(cfs_rq, se);
993

994 995
	if ((flags & DEQUEUE_SLEEP) && entity_is_task(se)) {
		struct task_struct *tsk = task_of(se);
996

997
		if (tsk->state & TASK_INTERRUPTIBLE)
998
			__schedstat_set(se->statistics.sleep_start,
999 1000
				      rq_clock(rq_of(cfs_rq)));
		if (tsk->state & TASK_UNINTERRUPTIBLE)
1001
			__schedstat_set(se->statistics.block_start,
1002
				      rq_clock(rq_of(cfs_rq)));
1003 1004 1005
	}
}

1006 1007 1008 1009
/*
 * We are picking a new current task - update its stats:
 */
static inline void
1010
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
1011 1012 1013 1014
{
	/*
	 * We are starting a new run period:
	 */
1015
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
1016 1017 1018 1019 1020 1021
}

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

1022 1023
#ifdef CONFIG_NUMA_BALANCING
/*
1024 1025 1026
 * 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.
1027
 */
1028 1029
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
1030 1031 1032

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
1033

1034 1035 1036
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
	pid_t gid;
	int active_nodes;

	struct rcu_head rcu;
	unsigned long total_faults;
	unsigned long max_faults_cpu;
	/*
	 * 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.
	 */
	unsigned long *faults_cpu;
	unsigned long faults[0];
};

static inline unsigned long group_faults_priv(struct numa_group *ng);
static inline unsigned long group_faults_shared(struct numa_group *ng);

1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
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)
{
1084
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
1085 1086 1087
	unsigned int scan, floor;
	unsigned int windows = 1;

1088 1089
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
1090 1091 1092 1093 1094 1095
	floor = 1000 / windows;

	scan = sysctl_numa_balancing_scan_period_min / task_nr_scan_windows(p);
	return max_t(unsigned int, floor, scan);
}

1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114
static unsigned int task_scan_start(struct task_struct *p)
{
	unsigned long smin = task_scan_min(p);
	unsigned long period = smin;

	/* Scale the maximum scan period with the amount of shared memory. */
	if (p->numa_group) {
		struct numa_group *ng = p->numa_group;
		unsigned long shared = group_faults_shared(ng);
		unsigned long private = group_faults_priv(ng);

		period *= atomic_read(&ng->refcount);
		period *= shared + 1;
		period /= private + shared + 1;
	}

	return max(smin, period);
}

1115 1116
static unsigned int task_scan_max(struct task_struct *p)
{
1117 1118
	unsigned long smin = task_scan_min(p);
	unsigned long smax;
1119 1120 1121

	/* Watch for min being lower than max due to floor calculations */
	smax = sysctl_numa_balancing_scan_period_max / task_nr_scan_windows(p);
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136

	/* Scale the maximum scan period with the amount of shared memory. */
	if (p->numa_group) {
		struct numa_group *ng = p->numa_group;
		unsigned long shared = group_faults_shared(ng);
		unsigned long private = group_faults_priv(ng);
		unsigned long period = smax;

		period *= atomic_read(&ng->refcount);
		period *= shared + 1;
		period /= private + shared + 1;

		smax = max(smax, period);
	}

1137 1138 1139
	return max(smin, smax);
}

1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180
void init_numa_balancing(unsigned long clone_flags, struct task_struct *p)
{
	int mm_users = 0;
	struct mm_struct *mm = p->mm;

	if (mm) {
		mm_users = atomic_read(&mm->mm_users);
		if (mm_users == 1) {
			mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
			mm->numa_scan_seq = 0;
		}
	}
	p->node_stamp			= 0;
	p->numa_scan_seq		= mm ? mm->numa_scan_seq : 0;
	p->numa_scan_period		= sysctl_numa_balancing_scan_delay;
	p->numa_work.next		= &p->numa_work;
	p->numa_faults			= NULL;
	p->numa_group			= NULL;
	p->last_task_numa_placement	= 0;
	p->last_sum_exec_runtime	= 0;

	/* New address space, reset the preferred nid */
	if (!(clone_flags & CLONE_VM)) {
		p->numa_preferred_nid = -1;
		return;
	}

	/*
	 * New thread, keep existing numa_preferred_nid which should be copied
	 * already by arch_dup_task_struct but stagger when scans start.
	 */
	if (mm) {
		unsigned int delay;

		delay = min_t(unsigned int, task_scan_max(current),
			current->numa_scan_period * mm_users * NSEC_PER_MSEC);
		delay += 2 * TICK_NSEC;
		p->node_stamp = delay;
	}
}

1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
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));
}

1193 1194 1195 1196 1197 1198 1199 1200 1201
/* 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)

1202 1203 1204 1205 1206
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

1207
/*
1208
 * The averaged statistics, shared & private, memory & CPU,
1209 1210 1211 1212 1213
 * 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)
1214
{
1215
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
1216 1217 1218 1219
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
1220
	if (!p->numa_faults)
1221 1222
		return 0;

1223 1224
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
1225 1226
}

1227 1228 1229 1230 1231
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

1232 1233
	return p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 1)];
1234 1235
}

1236 1237
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
1238 1239
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
1240 1241
}

1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265
static inline unsigned long group_faults_priv(struct numa_group *ng)
{
	unsigned long faults = 0;
	int node;

	for_each_online_node(node) {
		faults += ng->faults[task_faults_idx(NUMA_MEM, node, 1)];
	}

	return faults;
}

static inline unsigned long group_faults_shared(struct numa_group *ng)
{
	unsigned long faults = 0;
	int node;

	for_each_online_node(node) {
		faults += ng->faults[task_faults_idx(NUMA_MEM, node, 0)];
	}

	return faults;
}

1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277
/*
 * 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;
}

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 1314
/* 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 &&
1315
					dist >= maxdist)
1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342
			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;
}

1343 1344 1345 1346 1347 1348
/*
 * 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.
 */
1349 1350
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1351
{
1352
	unsigned long faults, total_faults;
1353

1354
	if (!p->numa_faults)
1355 1356 1357 1358 1359 1360 1361
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1362
	faults = task_faults(p, nid);
1363 1364
	faults += score_nearby_nodes(p, nid, dist, true);

1365
	return 1000 * faults / total_faults;
1366 1367
}

1368 1369
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1370
{
1371 1372 1373 1374 1375 1376 1377 1378
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1379 1380
		return 0;

1381
	faults = group_faults(p, nid);
1382 1383
	faults += score_nearby_nodes(p, nid, dist, false);

1384
	return 1000 * faults / total_faults;
1385 1386
}

1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426
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;

	/*
1427 1428
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1429
	 */
1430 1431
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1432 1433 1434
		return true;

	/*
1435 1436 1437 1438 1439 1440
	 * 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)
1441
	 */
1442 1443
	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;
1444 1445
}

1446
static unsigned long weighted_cpuload(struct rq *rq);
1447 1448
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1449
static unsigned long capacity_of(int cpu);
1450

1451
/* Cached statistics for all CPUs within a node */
1452 1453
struct numa_stats {
	unsigned long load;
1454 1455

	/* Total compute capacity of CPUs on a node */
1456
	unsigned long compute_capacity;
1457

1458
	unsigned int nr_running;
1459
};
1460

1461 1462 1463 1464 1465
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1466 1467
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1468 1469 1470 1471 1472 1473

	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;
1474
		ns->load += weighted_cpuload(rq);
1475
		ns->compute_capacity += capacity_of(cpu);
1476 1477

		cpus++;
1478 1479
	}

1480 1481 1482 1483 1484
	/*
	 * 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.
	 *
1485
	 * We'll detect a huge imbalance and bail there.
1486 1487 1488 1489
	 */
	if (!cpus)
		return;

1490 1491 1492 1493
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, ns->compute_capacity);
	capacity = cpus / smt; /* cores */

1494
	capacity = min_t(unsigned, capacity,
1495
		DIV_ROUND_CLOSEST(ns->compute_capacity, SCHED_CAPACITY_SCALE));
1496 1497
}

1498 1499
struct task_numa_env {
	struct task_struct *p;
1500

1501 1502
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1503

1504
	struct numa_stats src_stats, dst_stats;
1505

1506
	int imbalance_pct;
1507
	int dist;
1508 1509 1510

	struct task_struct *best_task;
	long best_imp;
1511 1512 1513
	int best_cpu;
};

1514 1515 1516 1517 1518
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);
1519 1520
	if (p)
		get_task_struct(p);
1521 1522 1523 1524 1525 1526

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

1527
static bool load_too_imbalanced(long src_load, long dst_load,
1528 1529
				struct task_numa_env *env)
{
1530 1531
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542
	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;
1543

1544
	imb = abs(dst_load * src_capacity - src_load * dst_capacity);
1545

1546
	orig_src_load = env->src_stats.load;
1547
	orig_dst_load = env->dst_stats.load;
1548

1549
	old_imb = abs(orig_dst_load * src_capacity - orig_src_load * dst_capacity);
1550 1551 1552

	/* Would this change make things worse? */
	return (imb > old_imb);
1553 1554
}

1555 1556 1557 1558 1559 1560
/*
 * 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
 */
1561
static void task_numa_compare(struct task_numa_env *env,
1562
			      long taskimp, long groupimp, bool maymove)
1563 1564 1565
{
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1566
	long src_load, dst_load;
1567
	long load;
1568
	long imp = env->p->numa_group ? groupimp : taskimp;
1569
	long moveimp = imp;
1570
	int dist = env->dist;
1571 1572

	rcu_read_lock();
1573 1574
	cur = task_rcu_dereference(&dst_rq->curr);
	if (cur && ((cur->flags & PF_EXITING) || is_idle_task(cur)))
1575 1576
		cur = NULL;

1577 1578 1579 1580 1581 1582 1583
	/*
	 * 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;

1584 1585 1586 1587 1588 1589 1590
	if (!cur) {
		if (maymove || imp > env->best_imp)
			goto assign;
		else
			goto unlock;
	}

1591 1592 1593 1594
	/*
	 * "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
1595
	 * the value is, the more remote accesses that would be expected to
1596 1597
	 * be incurred if the tasks were swapped.
	 */
1598 1599 1600
	/* Skip this swap candidate if cannot move to the source cpu */
	if (!cpumask_test_cpu(env->src_cpu, &cur->cpus_allowed))
		goto unlock;
1601

1602 1603 1604 1605 1606 1607 1608
	/*
	 * If dst and source tasks are in the same NUMA group, or not
	 * in any group then look only at task weights.
	 */
	if (cur->numa_group == env->p->numa_group) {
		imp = taskimp + task_weight(cur, env->src_nid, dist) -
		      task_weight(cur, env->dst_nid, dist);
1609
		/*
1610 1611
		 * Add some hysteresis to prevent swapping the
		 * tasks within a group over tiny differences.
1612
		 */
1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625
		if (cur->numa_group)
			imp -= imp / 16;
	} else {
		/*
		 * 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 && env->p->numa_group)
			imp += group_weight(cur, env->src_nid, dist) -
			       group_weight(cur, env->dst_nid, dist);
		else
			imp += task_weight(cur, env->src_nid, dist) -
			       task_weight(cur, env->dst_nid, dist);
1626 1627
	}

1628
	if (imp <= env->best_imp)
1629 1630
		goto unlock;

1631 1632 1633
	if (maymove && moveimp > imp && moveimp > env->best_imp) {
		imp = moveimp - 1;
		cur = NULL;
1634
		goto assign;
1635
	}
1636 1637 1638 1639

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
1640 1641 1642 1643
	load = task_h_load(env->p) - task_h_load(cur);
	if (!load)
		goto assign;

1644 1645
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1646

1647
	if (load_too_imbalanced(src_load, dst_load, env))
1648 1649
		goto unlock;

1650
assign:
1651 1652 1653 1654
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
1655 1656
	if (!cur) {
		/*
1657
		 * select_idle_siblings() uses an per-CPU cpumask that
1658 1659 1660
		 * can be used from IRQ context.
		 */
		local_irq_disable();
1661 1662
		env->dst_cpu = select_idle_sibling(env->p, env->src_cpu,
						   env->dst_cpu);
1663 1664
		local_irq_enable();
	}
1665

1666 1667 1668 1669 1670
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1671 1672
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1673
{
1674 1675
	long src_load, dst_load, load;
	bool maymove = false;
1676 1677
	int cpu;

1678 1679 1680 1681 1682 1683 1684 1685 1686 1687
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;

	/*
	 * If the improvement from just moving env->p direction is better
	 * than swapping tasks around, check if a move is possible.
	 */
	maymove = !load_too_imbalanced(src_load, dst_load, env);

1688 1689
	for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
		/* Skip this CPU if the source task cannot migrate */
1690
		if (!cpumask_test_cpu(cpu, &env->p->cpus_allowed))
1691 1692 1693
			continue;

		env->dst_cpu = cpu;
1694
		task_numa_compare(env, taskimp, groupimp, maymove);
1695 1696 1697
	}
}

1698 1699 1700 1701
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1702

1703
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1704
		.src_nid = task_node(p),
1705 1706 1707 1708 1709

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1710
		.best_cpu = -1,
1711 1712
	};
	struct sched_domain *sd;
1713
	unsigned long taskweight, groupweight;
1714
	int nid, ret, dist;
1715
	long taskimp, groupimp;
1716

1717
	/*
1718 1719 1720 1721 1722 1723
	 * 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.
1724 1725
	 */
	rcu_read_lock();
1726
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1727 1728
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1729 1730
	rcu_read_unlock();

1731 1732 1733 1734 1735 1736 1737
	/*
	 * 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)) {
1738
		sched_setnuma(p, task_node(p));
1739 1740 1741
		return -EINVAL;
	}

1742
	env.dst_nid = p->numa_preferred_nid;
1743 1744 1745 1746 1747 1748
	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;
1749
	update_numa_stats(&env.dst_stats, env.dst_nid);
1750

1751
	/* Try to find a spot on the preferred nid. */
1752
	task_numa_find_cpu(&env, taskimp, groupimp);
1753

1754 1755 1756 1757 1758 1759 1760
	/*
	 * 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.
	 */
1761
	if (env.best_cpu == -1 || (p->numa_group && p->numa_group->active_nodes > 1)) {
1762 1763 1764
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1765

1766
			dist = node_distance(env.src_nid, env.dst_nid);
1767 1768 1769 1770 1771
			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);
			}
1772

1773
			/* Only consider nodes where both task and groups benefit */
1774 1775
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1776
			if (taskimp < 0 && groupimp < 0)
1777 1778
				continue;

1779
			env.dist = dist;
1780 1781
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1782
			task_numa_find_cpu(&env, taskimp, groupimp);
1783 1784 1785
		}
	}

1786 1787 1788 1789 1790 1791 1792 1793
	/*
	 * 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.
	 */
1794 1795 1796 1797
	if (p->numa_group) {
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
1798
			nid = cpu_to_node(env.best_cpu);
1799

1800 1801
		if (nid != p->numa_preferred_nid)
			sched_setnuma(p, nid);
1802 1803 1804 1805 1806
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1807

1808 1809 1810 1811
	/*
	 * Reset the scan period if the task is being rescheduled on an
	 * alternative node to recheck if the tasks is now properly placed.
	 */
1812
	p->numa_scan_period = task_scan_start(p);
1813

1814
	if (env.best_task == NULL) {
1815 1816 1817
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1818 1819 1820
		return ret;
	}

1821 1822
	ret = migrate_swap(p, env.best_task, env.best_cpu, env.src_cpu);

1823 1824
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1825 1826
	put_task_struct(env.best_task);
	return ret;
1827 1828
}

1829 1830 1831
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1832 1833
	unsigned long interval = HZ;

1834
	/* This task has no NUMA fault statistics yet */
1835
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1836 1837
		return;

1838
	/* Periodically retry migrating the task to the preferred node */
1839
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
1840
	p->numa_migrate_retry = jiffies + interval;
1841 1842

	/* Success if task is already running on preferred CPU */
1843
	if (task_node(p) == p->numa_preferred_nid)
1844 1845 1846
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1847
	task_numa_migrate(p);
1848 1849
}

1850
/*
1851
 * Find out how many nodes on the workload is actively running on. Do this by
1852 1853 1854 1855
 * 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.
 */
1856
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1857 1858
{
	unsigned long faults, max_faults = 0;
1859
	int nid, active_nodes = 0;
1860 1861 1862 1863 1864 1865 1866 1867 1868

	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);
1869 1870
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1871
	}
1872 1873 1874

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1875 1876
}

1877 1878 1879
/*
 * 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
1880 1881 1882
 * 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.
1883 1884
 */
#define NUMA_PERIOD_SLOTS 10
1885
#define NUMA_PERIOD_THRESHOLD 7
1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896

/*
 * 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;
1897
	int lr_ratio, ps_ratio;
1898 1899 1900 1901 1902 1903 1904 1905
	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
1906 1907 1908
	 * 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
1909
	 */
1910
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926
		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);
1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945
	lr_ratio = (local * NUMA_PERIOD_SLOTS) / (local + remote);
	ps_ratio = (private * NUMA_PERIOD_SLOTS) / (private + shared);

	if (ps_ratio >= NUMA_PERIOD_THRESHOLD) {
		/*
		 * Most memory accesses are local. There is no need to
		 * do fast NUMA scanning, since memory is already local.
		 */
		int slot = ps_ratio - NUMA_PERIOD_THRESHOLD;
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else if (lr_ratio >= NUMA_PERIOD_THRESHOLD) {
		/*
		 * Most memory accesses are shared with other tasks.
		 * There is no point in continuing fast NUMA scanning,
		 * since other tasks may just move the memory elsewhere.
		 */
		int slot = lr_ratio - NUMA_PERIOD_THRESHOLD;
1946 1947 1948 1949 1950
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else {
		/*
1951 1952 1953
		 * Private memory faults exceed (SLOTS-THRESHOLD)/SLOTS,
		 * yet they are not on the local NUMA node. Speed up
		 * NUMA scanning to get the memory moved over.
1954
		 */
1955 1956
		int ratio = max(lr_ratio, ps_ratio);
		diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot;
1957 1958 1959 1960 1961 1962 1963
	}

	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));
}

1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981
/*
 * 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 {
1982
		delta = p->se.avg.load_sum;
1983
		*period = LOAD_AVG_MAX;
1984 1985 1986 1987 1988 1989 1990 1991
	}

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

	return delta;
}

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038
/*
 * 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;
2039
		nodemask_t max_group = NODE_MASK_NONE;
2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072
		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. */
2073 2074
		if (!max_faults)
			break;
2075 2076 2077 2078 2079
		nodes = max_group;
	}
	return nid;
}

2080 2081
static void task_numa_placement(struct task_struct *p)
{
2082 2083
	int seq, nid, max_nid = -1;
	unsigned long max_faults = 0;
2084
	unsigned long fault_types[2] = { 0, 0 };
2085 2086
	unsigned long total_faults;
	u64 runtime, period;
2087
	spinlock_t *group_lock = NULL;
2088

2089 2090 2091 2092 2093
	/*
	 * 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:
	 */
2094
	seq = READ_ONCE(p->mm->numa_scan_seq);
2095 2096 2097
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
2098
	p->numa_scan_period_max = task_scan_max(p);
2099

2100 2101 2102 2103
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

2104 2105 2106
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
2107
		spin_lock_irq(group_lock);
2108 2109
	}

2110 2111
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
2112 2113
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
2114
		unsigned long faults = 0, group_faults = 0;
2115
		int priv;
2116

2117
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
2118
			long diff, f_diff, f_weight;
2119

2120 2121 2122 2123
			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);
2124

2125
			/* Decay existing window, copy faults since last scan */
2126 2127 2128
			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;
2129

2130 2131 2132 2133 2134 2135 2136 2137
			/*
			 * 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);
2138
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
2139
				   (total_faults + 1);
2140 2141
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
2142

2143 2144 2145
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
2146
			p->total_numa_faults += diff;
2147
			if (p->numa_group) {
2148 2149 2150 2151 2152 2153 2154 2155 2156
				/*
				 * 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;
2157
				p->numa_group->total_faults += diff;
2158
				group_faults += p->numa_group->faults[mem_idx];
2159
			}
2160 2161
		}

2162 2163 2164 2165 2166 2167 2168
		if (!p->numa_group) {
			if (faults > max_faults) {
				max_faults = faults;
				max_nid = nid;
			}
		} else if (group_faults > max_faults) {
			max_faults = group_faults;
2169 2170
			max_nid = nid;
		}
2171 2172
	}

2173
	if (p->numa_group) {
2174
		numa_group_count_active_nodes(p->numa_group);
2175
		spin_unlock_irq(group_lock);
2176
		max_nid = preferred_group_nid(p, max_nid);
2177 2178
	}

2179 2180 2181 2182
	if (max_faults) {
		/* Set the new preferred node */
		if (max_nid != p->numa_preferred_nid)
			sched_setnuma(p, max_nid);
2183
	}
2184 2185

	update_task_scan_period(p, fault_types[0], fault_types[1]);
2186 2187
}

2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198
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);
}

2199 2200
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2201 2202 2203 2204 2205 2206 2207 2208 2209
{
	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) +
2210
				    4*nr_node_ids*sizeof(unsigned long);
2211 2212 2213 2214 2215 2216

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

		atomic_set(&grp->refcount, 1);
2217 2218
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2219
		spin_lock_init(&grp->lock);
2220
		grp->gid = p->pid;
2221
		/* Second half of the array tracks nids where faults happen */
2222 2223
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2224

2225
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2226
			grp->faults[i] = p->numa_faults[i];
2227

2228
		grp->total_faults = p->total_numa_faults;
2229

2230 2231 2232 2233 2234
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2235
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2236 2237

	if (!cpupid_match_pid(tsk, cpupid))
2238
		goto no_join;
2239 2240 2241

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2242
		goto no_join;
2243 2244 2245

	my_grp = p->numa_group;
	if (grp == my_grp)
2246
		goto no_join;
2247 2248 2249 2250 2251 2252

	/*
	 * 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)
2253
		goto no_join;
2254 2255 2256 2257 2258

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2259
		goto no_join;
2260

2261 2262 2263 2264 2265 2266 2267
	/* 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;
2268

2269 2270 2271
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2272
	if (join && !get_numa_group(grp))
2273
		goto no_join;
2274 2275 2276 2277 2278 2279

	rcu_read_unlock();

	if (!join)
		return;

2280 2281
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2282

2283
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2284 2285
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2286
	}
2287 2288
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2289 2290 2291 2292 2293

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

	spin_unlock(&my_grp->lock);
2294
	spin_unlock_irq(&grp->lock);
2295 2296 2297 2298

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2299 2300 2301 2302 2303
	return;

no_join:
	rcu_read_unlock();
	return;
2304 2305 2306 2307 2308
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2309
	void *numa_faults = p->numa_faults;
2310 2311
	unsigned long flags;
	int i;
2312 2313

	if (grp) {
2314
		spin_lock_irqsave(&grp->lock, flags);
2315
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2316
			grp->faults[i] -= p->numa_faults[i];
2317
		grp->total_faults -= p->total_numa_faults;
2318

2319
		grp->nr_tasks--;
2320
		spin_unlock_irqrestore(&grp->lock, flags);
2321
		RCU_INIT_POINTER(p->numa_group, NULL);
2322 2323 2324
		put_numa_group(grp);
	}

2325
	p->numa_faults = NULL;
2326
	kfree(numa_faults);
2327 2328
}

2329 2330 2331
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2332
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2333 2334
{
	struct task_struct *p = current;
2335
	bool migrated = flags & TNF_MIGRATED;
2336
	int cpu_node = task_node(current);
2337
	int local = !!(flags & TNF_FAULT_LOCAL);
2338
	struct numa_group *ng;
2339
	int priv;
2340

2341
	if (!static_branch_likely(&sched_numa_balancing))
2342 2343
		return;

2344 2345 2346 2347
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2348
	/* Allocate buffer to track faults on a per-node basis */
2349 2350
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2351
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2352

2353 2354
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2355
			return;
2356

2357
		p->total_numa_faults = 0;
2358
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2359
	}
2360

2361 2362 2363 2364 2365 2366 2367 2368
	/*
	 * 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);
2369
		if (!priv && !(flags & TNF_NO_GROUP))
2370
			task_numa_group(p, last_cpupid, flags, &priv);
2371 2372
	}

2373 2374 2375 2376 2377 2378
	/*
	 * 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.
	 */
2379 2380 2381 2382
	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))
2383 2384
		local = 1;

2385 2386 2387 2388
	/*
	 * Retry task to preferred node migration periodically, in case it
	 * case it previously failed, or the scheduler moved us.
	 */
2389 2390
	if (time_after(jiffies, p->numa_migrate_retry)) {
		task_numa_placement(p);
2391
		numa_migrate_preferred(p);
2392
	}
2393

I
Ingo Molnar 已提交
2394 2395
	if (migrated)
		p->numa_pages_migrated += pages;
2396 2397
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2398

2399 2400
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2401
	p->numa_faults_locality[local] += pages;
2402 2403
}

2404 2405
static void reset_ptenuma_scan(struct task_struct *p)
{
2406 2407 2408 2409 2410 2411 2412 2413
	/*
	 * 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:
	 */
2414
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2415 2416 2417
	p->mm->numa_scan_offset = 0;
}

2418 2419 2420 2421 2422 2423 2424 2425 2426
/*
 * 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;
2427
	u64 runtime = p->se.sum_exec_runtime;
2428
	struct vm_area_struct *vma;
2429
	unsigned long start, end;
2430
	unsigned long nr_pte_updates = 0;
2431
	long pages, virtpages;
2432

2433
	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446

	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;

2447
	if (!mm->numa_next_scan) {
2448 2449
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2450 2451
	}

2452 2453 2454 2455 2456 2457 2458
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2459 2460
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
2461
		p->numa_scan_period = task_scan_start(p);
2462
	}
2463

2464
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2465 2466 2467
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2468 2469 2470 2471 2472 2473
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2474 2475 2476
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2477
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2478 2479
	if (!pages)
		return;
2480

2481

2482 2483
	if (!down_read_trylock(&mm->mmap_sem))
		return;
2484
	vma = find_vma(mm, start);
2485 2486
	if (!vma) {
		reset_ptenuma_scan(p);
2487
		start = 0;
2488 2489
		vma = mm->mmap;
	}
2490
	for (; vma; vma = vma->vm_next) {
2491
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2492
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2493
			continue;
2494
		}
2495

2496 2497 2498 2499 2500 2501 2502 2503 2504 2505
		/*
		 * 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
2506 2507 2508 2509 2510 2511
		/*
		 * 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;
2512

2513 2514 2515 2516
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2517
			nr_pte_updates = change_prot_numa(vma, start, end);
2518 2519

			/*
2520 2521 2522 2523 2524 2525
			 * 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.
2526 2527 2528
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2529
			virtpages -= (end - start) >> PAGE_SHIFT;
2530

2531
			start = end;
2532
			if (pages <= 0 || virtpages <= 0)
2533
				goto out;
2534 2535

			cond_resched();
2536
		} while (end != vma->vm_end);
2537
	}
2538

2539
out:
2540
	/*
P
Peter Zijlstra 已提交
2541 2542 2543 2544
	 * 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.
2545 2546
	 */
	if (vma)
2547
		mm->numa_scan_offset = start;
2548 2549 2550
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561

	/*
	 * 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;
	}
2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586
}

/*
 * 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;

2587
	if (now > curr->node_stamp + period) {
2588
		if (!curr->node_stamp)
2589
			curr->numa_scan_period = task_scan_start(curr);
2590
		curr->node_stamp += period;
2591 2592 2593 2594 2595 2596 2597

		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);
		}
	}
}
2598

2599 2600 2601 2602
#else
static void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
}
2603 2604 2605 2606 2607 2608 2609 2610

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)
{
}
2611

2612 2613
#endif /* CONFIG_NUMA_BALANCING */

2614 2615 2616 2617
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2618
	if (!parent_entity(se))
2619
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2620
#ifdef CONFIG_SMP
2621 2622 2623 2624 2625 2626
	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);
	}
2627
#endif
2628 2629 2630 2631 2632 2633 2634
	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);
2635
	if (!parent_entity(se))
2636
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2637
#ifdef CONFIG_SMP
2638 2639
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2640
		list_del_init(&se->group_node);
2641
	}
2642
#endif
2643 2644 2645
	cfs_rq->nr_running--;
}

2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686
/*
 * Signed add 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 add_positive(_ptr, _val) do {                           \
	typeof(_ptr) ptr = (_ptr);                              \
	typeof(_val) val = (_val);                              \
	typeof(*ptr) res, var = READ_ONCE(*ptr);                \
								\
	res = var + val;                                        \
								\
	if (val < 0 && res > var)                               \
		res = 0;                                        \
								\
	WRITE_ONCE(*ptr, res);                                  \
} while (0)

/*
 * 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)

#ifdef CONFIG_SMP
static inline void
enqueue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2687 2688 2689 2690
	cfs_rq->runnable_weight += se->runnable_weight;

	cfs_rq->avg.runnable_load_avg += se->avg.runnable_load_avg;
	cfs_rq->avg.runnable_load_sum += se_runnable(se) * se->avg.runnable_load_sum;
2691 2692 2693 2694 2695
}

static inline void
dequeue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2696 2697 2698 2699 2700
	cfs_rq->runnable_weight -= se->runnable_weight;

	sub_positive(&cfs_rq->avg.runnable_load_avg, se->avg.runnable_load_avg);
	sub_positive(&cfs_rq->avg.runnable_load_sum,
		     se_runnable(se) * se->avg.runnable_load_sum);
2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726
}

static inline void
enqueue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	cfs_rq->avg.load_avg += se->avg.load_avg;
	cfs_rq->avg.load_sum += se_weight(se) * se->avg.load_sum;
}

static inline void
dequeue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	sub_positive(&cfs_rq->avg.load_avg, se->avg.load_avg);
	sub_positive(&cfs_rq->avg.load_sum, se_weight(se) * se->avg.load_sum);
}
#else
static inline void
enqueue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
static inline void
dequeue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
static inline void
enqueue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
static inline void
dequeue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) { }
#endif

2727
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
2728
			    unsigned long weight, unsigned long runnable)
2729 2730 2731 2732 2733 2734 2735 2736 2737 2738
{
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
		account_entity_dequeue(cfs_rq, se);
		dequeue_runnable_load_avg(cfs_rq, se);
	}
	dequeue_load_avg(cfs_rq, se);

2739
	se->runnable_weight = runnable;
2740 2741 2742
	update_load_set(&se->load, weight);

#ifdef CONFIG_SMP
2743 2744 2745 2746 2747 2748 2749
	do {
		u32 divider = LOAD_AVG_MAX - 1024 + se->avg.period_contrib;

		se->avg.load_avg = div_u64(se_weight(se) * se->avg.load_sum, divider);
		se->avg.runnable_load_avg =
			div_u64(se_runnable(se) * se->avg.runnable_load_sum, divider);
	} while (0);
2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765
#endif

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

void reweight_task(struct task_struct *p, int prio)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	struct load_weight *load = &se->load;
	unsigned long weight = scale_load(sched_prio_to_weight[prio]);

2766
	reweight_entity(cfs_rq, se, weight, weight);
2767 2768 2769
	load->inv_weight = sched_prio_to_wmult[prio];
}

2770
#ifdef CONFIG_FAIR_GROUP_SCHED
2771
#ifdef CONFIG_SMP
2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809
/*
 * All this does is approximate the hierarchical proportion which includes that
 * global sum we all love to hate.
 *
 * That is, the weight of a group entity, is the proportional share of the
 * group weight based on the group runqueue weights. That is:
 *
 *                     tg->weight * grq->load.weight
 *   ge->load.weight = -----------------------------               (1)
 *			  \Sum grq->load.weight
 *
 * Now, because computing that sum is prohibitively expensive to compute (been
 * there, done that) we approximate it with this average stuff. The average
 * moves slower and therefore the approximation is cheaper and more stable.
 *
 * So instead of the above, we substitute:
 *
 *   grq->load.weight -> grq->avg.load_avg                         (2)
 *
 * which yields the following:
 *
 *                     tg->weight * grq->avg.load_avg
 *   ge->load.weight = ------------------------------              (3)
 *				tg->load_avg
 *
 * Where: tg->load_avg ~= \Sum grq->avg.load_avg
 *
 * That is shares_avg, and it is right (given the approximation (2)).
 *
 * The problem with it is that because the average is slow -- it was designed
 * to be exactly that of course -- this leads to transients in boundary
 * conditions. In specific, the case where the group was idle and we start the
 * one task. It takes time for our CPU's grq->avg.load_avg to build up,
 * yielding bad latency etc..
 *
 * Now, in that special case (1) reduces to:
 *
 *                     tg->weight * grq->load.weight
2810
 *   ge->load.weight = ----------------------------- = tg->weight   (4)
2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823
 *			    grp->load.weight
 *
 * That is, the sum collapses because all other CPUs are idle; the UP scenario.
 *
 * So what we do is modify our approximation (3) to approach (4) in the (near)
 * UP case, like:
 *
 *   ge->load.weight =
 *
 *              tg->weight * grq->load.weight
 *     ---------------------------------------------------         (5)
 *     tg->load_avg - grq->avg.load_avg + grq->load.weight
 *
2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835
 * But because grq->load.weight can drop to 0, resulting in a divide by zero,
 * we need to use grq->avg.load_avg as its lower bound, which then gives:
 *
 *
 *                     tg->weight * grq->load.weight
 *   ge->load.weight = -----------------------------		   (6)
 *				tg_load_avg'
 *
 * Where:
 *
 *   tg_load_avg' = tg->load_avg - grq->avg.load_avg +
 *                  max(grq->load.weight, grq->avg.load_avg)
2836 2837 2838 2839 2840 2841 2842 2843 2844
 *
 * And that is shares_weight and is icky. In the (near) UP case it approaches
 * (4) while in the normal case it approaches (3). It consistently
 * overestimates the ge->load.weight and therefore:
 *
 *   \Sum ge->load.weight >= tg->weight
 *
 * hence icky!
 */
2845
static long calc_group_shares(struct cfs_rq *cfs_rq)
2846
{
2847 2848 2849 2850
	long tg_weight, tg_shares, load, shares;
	struct task_group *tg = cfs_rq->tg;

	tg_shares = READ_ONCE(tg->shares);
2851

2852
	load = max(scale_load_down(cfs_rq->load.weight), cfs_rq->avg.load_avg);
2853

2854
	tg_weight = atomic_long_read(&tg->load_avg);
2855

2856 2857 2858
	/* Ensure tg_weight >= load */
	tg_weight -= cfs_rq->tg_load_avg_contrib;
	tg_weight += load;
2859

2860
	shares = (tg_shares * load);
2861 2862
	if (tg_weight)
		shares /= tg_weight;
2863

2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875
	/*
	 * MIN_SHARES has to be unscaled here to support per-CPU partitioning
	 * of a group with small tg->shares value. It is a floor value which is
	 * assigned as a minimum load.weight to the sched_entity representing
	 * the group on a CPU.
	 *
	 * E.g. on 64-bit for a group with tg->shares of scale_load(15)=15*1024
	 * on an 8-core system with 8 tasks each runnable on one CPU shares has
	 * to be 15*1024*1/8=1920 instead of scale_load(MIN_SHARES)=2*1024. In
	 * case no task is runnable on a CPU MIN_SHARES=2 should be returned
	 * instead of 0.
	 */
2876
	return clamp_t(long, shares, MIN_SHARES, tg_shares);
2877
}
2878 2879

/*
2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904
 * This calculates the effective runnable weight for a group entity based on
 * the group entity weight calculated above.
 *
 * Because of the above approximation (2), our group entity weight is
 * an load_avg based ratio (3). This means that it includes blocked load and
 * does not represent the runnable weight.
 *
 * Approximate the group entity's runnable weight per ratio from the group
 * runqueue:
 *
 *					     grq->avg.runnable_load_avg
 *   ge->runnable_weight = ge->load.weight * -------------------------- (7)
 *						 grq->avg.load_avg
 *
 * However, analogous to above, since the avg numbers are slow, this leads to
 * transients in the from-idle case. Instead we use:
 *
 *   ge->runnable_weight = ge->load.weight *
 *
 *		max(grq->avg.runnable_load_avg, grq->runnable_weight)
 *		-----------------------------------------------------	(8)
 *		      max(grq->avg.load_avg, grq->load.weight)
 *
 * Where these max() serve both to use the 'instant' values to fix the slow
 * from-idle and avoid the /0 on to-idle, similar to (6).
2905 2906 2907
 */
static long calc_group_runnable(struct cfs_rq *cfs_rq, long shares)
{
2908 2909 2910 2911 2912 2913 2914
	long runnable, load_avg;

	load_avg = max(cfs_rq->avg.load_avg,
		       scale_load_down(cfs_rq->load.weight));

	runnable = max(cfs_rq->avg.runnable_load_avg,
		       scale_load_down(cfs_rq->runnable_weight));
2915 2916 2917 2918

	runnable *= shares;
	if (load_avg)
		runnable /= load_avg;
2919

2920 2921
	return clamp_t(long, runnable, MIN_SHARES, shares);
}
2922
#endif /* CONFIG_SMP */
2923

2924 2925
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2926 2927 2928 2929 2930
/*
 * Recomputes the group entity based on the current state of its group
 * runqueue.
 */
static void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
2931
{
2932 2933
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);
	long shares, runnable;
P
Peter Zijlstra 已提交
2934

2935
	if (!gcfs_rq)
2936 2937
		return;

2938
	if (throttled_hierarchy(gcfs_rq))
P
Peter Zijlstra 已提交
2939
		return;
2940

2941
#ifndef CONFIG_SMP
2942
	runnable = shares = READ_ONCE(gcfs_rq->tg->shares);
2943 2944

	if (likely(se->load.weight == shares))
2945
		return;
2946
#else
2947 2948
	shares   = calc_group_shares(gcfs_rq);
	runnable = calc_group_runnable(gcfs_rq, shares);
2949
#endif
P
Peter Zijlstra 已提交
2950

2951
	reweight_entity(cfs_rq_of(se), se, shares, runnable);
P
Peter Zijlstra 已提交
2952
}
2953

P
Peter Zijlstra 已提交
2954
#else /* CONFIG_FAIR_GROUP_SCHED */
2955
static inline void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
2956 2957 2958 2959
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2960
static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq, int flags)
2961
{
2962 2963
	struct rq *rq = rq_of(cfs_rq);

2964
	if (&rq->cfs == cfs_rq || (flags & SCHED_CPUFREQ_MIGRATION)) {
2965 2966 2967
		/*
		 * There are a few boundary cases this might miss but it should
		 * get called often enough that that should (hopefully) not be
2968
		 * a real problem.
2969 2970 2971 2972 2973 2974 2975 2976 2977 2978
		 *
		 * 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().
		 */
2979
		cpufreq_update_util(rq, flags);
2980 2981 2982
	}
}

2983
#ifdef CONFIG_SMP
2984
#ifdef CONFIG_FAIR_GROUP_SCHED
2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997
/**
 * 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'.
 *
2998
 * Updating tg's load_avg is necessary before update_cfs_share().
2999
 */
3000
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
3001
{
3002
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
3003

3004 3005 3006 3007 3008 3009
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

3010 3011 3012
	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;
3013
	}
3014
}
3015

3016
/*
3017
 * Called within set_task_rq() right before setting a task's CPU. The
3018 3019 3020 3021 3022 3023
 * 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)
{
3024 3025 3026
	u64 p_last_update_time;
	u64 n_last_update_time;

3027 3028 3029 3030 3031 3032 3033 3034 3035 3036
	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.
	 */
3037 3038
	if (!(se->avg.last_update_time && prev))
		return;
3039 3040

#ifndef CONFIG_64BIT
3041
	{
3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055
		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);
3056
	}
3057
#else
3058 3059
	p_last_update_time = prev->avg.last_update_time;
	n_last_update_time = next->avg.last_update_time;
3060
#endif
3061 3062
	__update_load_avg_blocked_se(p_last_update_time, cpu_of(rq_of(prev)), se);
	se->avg.last_update_time = n_last_update_time;
3063
}
3064

3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075

/*
 * When on migration a sched_entity joins/leaves the PELT hierarchy, we need to
 * propagate its contribution. The key to this propagation is the invariant
 * that for each group:
 *
 *   ge->avg == grq->avg						(1)
 *
 * _IFF_ we look at the pure running and runnable sums. Because they
 * represent the very same entity, just at different points in the hierarchy.
 *
3076 3077 3078
 * Per the above update_tg_cfs_util() is trivial and simply copies the running
 * sum over (but still wrong, because the group entity and group rq do not have
 * their PELT windows aligned).
3079 3080 3081 3082 3083 3084 3085 3086
 *
 * However, update_tg_cfs_runnable() is more complex. So we have:
 *
 *   ge->avg.load_avg = ge->load.weight * ge->avg.runnable_avg		(2)
 *
 * And since, like util, the runnable part should be directly transferable,
 * the following would _appear_ to be the straight forward approach:
 *
3087
 *   grq->avg.load_avg = grq->load.weight * grq->avg.runnable_avg	(3)
3088 3089 3090
 *
 * And per (1) we have:
 *
3091
 *   ge->avg.runnable_avg == grq->avg.runnable_avg
3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109
 *
 * Which gives:
 *
 *                      ge->load.weight * grq->avg.load_avg
 *   ge->avg.load_avg = -----------------------------------		(4)
 *                               grq->load.weight
 *
 * Except that is wrong!
 *
 * Because while for entities historical weight is not important and we
 * really only care about our future and therefore can consider a pure
 * runnable sum, runqueues can NOT do this.
 *
 * We specifically want runqueues to have a load_avg that includes
 * historical weights. Those represent the blocked load, the load we expect
 * to (shortly) return to us. This only works by keeping the weights as
 * integral part of the sum. We therefore cannot decompose as per (3).
 *
3110 3111 3112 3113 3114 3115
 * Another reason this doesn't work is that runnable isn't a 0-sum entity.
 * Imagine a rq with 2 tasks that each are runnable 2/3 of the time. Then the
 * rq itself is runnable anywhere between 2/3 and 1 depending on how the
 * runnable section of these tasks overlap (or not). If they were to perfectly
 * align the rq as a whole would be runnable 2/3 of the time. If however we
 * always have at least 1 runnable task, the rq as a whole is always runnable.
3116
 *
3117
 * So we'll have to approximate.. :/
3118
 *
3119
 * Given the constraint:
3120
 *
3121
 *   ge->avg.running_sum <= ge->avg.runnable_sum <= LOAD_AVG_MAX
3122
 *
3123 3124
 * We can construct a rule that adds runnable to a rq by assuming minimal
 * overlap.
3125
 *
3126
 * On removal, we'll assume each task is equally runnable; which yields:
3127
 *
3128
 *   grq->avg.runnable_sum = grq->avg.load_sum / grq->load.weight
3129
 *
3130
 * XXX: only do this for the part of runnable > running ?
3131 3132 3133
 *
 */

3134
static inline void
3135
update_tg_cfs_util(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3136 3137 3138 3139 3140 3141 3142
{
	long delta = gcfs_rq->avg.util_avg - se->avg.util_avg;

	/* Nothing to update */
	if (!delta)
		return;

3143 3144 3145 3146 3147 3148 3149 3150
	/*
	 * The relation between sum and avg is:
	 *
	 *   LOAD_AVG_MAX - 1024 + sa->period_contrib
	 *
	 * however, the PELT windows are not aligned between grq and gse.
	 */

3151 3152 3153 3154 3155 3156 3157 3158 3159 3160
	/* Set new sched_entity's utilization */
	se->avg.util_avg = gcfs_rq->avg.util_avg;
	se->avg.util_sum = se->avg.util_avg * LOAD_AVG_MAX;

	/* Update parent cfs_rq utilization */
	add_positive(&cfs_rq->avg.util_avg, delta);
	cfs_rq->avg.util_sum = cfs_rq->avg.util_avg * LOAD_AVG_MAX;
}

static inline void
3161
update_tg_cfs_runnable(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3162
{
3163 3164 3165 3166
	long delta_avg, running_sum, runnable_sum = gcfs_rq->prop_runnable_sum;
	unsigned long runnable_load_avg, load_avg;
	u64 runnable_load_sum, load_sum = 0;
	s64 delta_sum;
3167

3168 3169
	if (!runnable_sum)
		return;
3170

3171
	gcfs_rq->prop_runnable_sum = 0;
3172

3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195
	if (runnable_sum >= 0) {
		/*
		 * Add runnable; clip at LOAD_AVG_MAX. Reflects that until
		 * the CPU is saturated running == runnable.
		 */
		runnable_sum += se->avg.load_sum;
		runnable_sum = min(runnable_sum, (long)LOAD_AVG_MAX);
	} else {
		/*
		 * Estimate the new unweighted runnable_sum of the gcfs_rq by
		 * assuming all tasks are equally runnable.
		 */
		if (scale_load_down(gcfs_rq->load.weight)) {
			load_sum = div_s64(gcfs_rq->avg.load_sum,
				scale_load_down(gcfs_rq->load.weight));
		}

		/* But make sure to not inflate se's runnable */
		runnable_sum = min(se->avg.load_sum, load_sum);
	}

	/*
	 * runnable_sum can't be lower than running_sum
3196
	 * As running sum is scale with CPU capacity wehreas the runnable sum
3197 3198 3199 3200 3201 3202
	 * is not we rescale running_sum 1st
	 */
	running_sum = se->avg.util_sum /
		arch_scale_cpu_capacity(NULL, cpu_of(rq_of(cfs_rq)));
	runnable_sum = max(runnable_sum, running_sum);

3203 3204
	load_sum = (s64)se_weight(se) * runnable_sum;
	load_avg = div_s64(load_sum, LOAD_AVG_MAX);
3205

3206 3207
	delta_sum = load_sum - (s64)se_weight(se) * se->avg.load_sum;
	delta_avg = load_avg - se->avg.load_avg;
3208

3209 3210 3211 3212
	se->avg.load_sum = runnable_sum;
	se->avg.load_avg = load_avg;
	add_positive(&cfs_rq->avg.load_avg, delta_avg);
	add_positive(&cfs_rq->avg.load_sum, delta_sum);
3213

3214 3215
	runnable_load_sum = (s64)se_runnable(se) * runnable_sum;
	runnable_load_avg = div_s64(runnable_load_sum, LOAD_AVG_MAX);
3216 3217
	delta_sum = runnable_load_sum - se_weight(se) * se->avg.runnable_load_sum;
	delta_avg = runnable_load_avg - se->avg.runnable_load_avg;
3218

3219 3220
	se->avg.runnable_load_sum = runnable_sum;
	se->avg.runnable_load_avg = runnable_load_avg;
3221

3222
	if (se->on_rq) {
3223 3224
		add_positive(&cfs_rq->avg.runnable_load_avg, delta_avg);
		add_positive(&cfs_rq->avg.runnable_load_sum, delta_sum);
3225 3226 3227
	}
}

3228
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum)
3229
{
3230 3231
	cfs_rq->propagate = 1;
	cfs_rq->prop_runnable_sum += runnable_sum;
3232 3233 3234 3235 3236
}

/* Update task and its cfs_rq load average */
static inline int propagate_entity_load_avg(struct sched_entity *se)
{
3237
	struct cfs_rq *cfs_rq, *gcfs_rq;
3238 3239 3240 3241

	if (entity_is_task(se))
		return 0;

3242 3243
	gcfs_rq = group_cfs_rq(se);
	if (!gcfs_rq->propagate)
3244 3245
		return 0;

3246 3247
	gcfs_rq->propagate = 0;

3248 3249
	cfs_rq = cfs_rq_of(se);

3250
	add_tg_cfs_propagate(cfs_rq, gcfs_rq->prop_runnable_sum);
3251

3252 3253
	update_tg_cfs_util(cfs_rq, se, gcfs_rq);
	update_tg_cfs_runnable(cfs_rq, se, gcfs_rq);
3254 3255 3256 3257

	return 1;
}

3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276
/*
 * Check if we need to update the load and the utilization of a blocked
 * group_entity:
 */
static inline bool skip_blocked_update(struct sched_entity *se)
{
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);

	/*
	 * If sched_entity still have not zero load or utilization, we have to
	 * decay it:
	 */
	if (se->avg.load_avg || se->avg.util_avg)
		return false;

	/*
	 * If there is a pending propagation, we have to update the load and
	 * the utilization of the sched_entity:
	 */
3277
	if (gcfs_rq->propagate)
3278 3279 3280 3281 3282 3283 3284 3285 3286 3287
		return false;

	/*
	 * Otherwise, the load and the utilization of the sched_entity is
	 * already zero and there is no pending propagation, so it will be a
	 * waste of time to try to decay it:
	 */
	return true;
}

3288
#else /* CONFIG_FAIR_GROUP_SCHED */
3289

3290
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
3291 3292 3293 3294 3295 3296

static inline int propagate_entity_load_avg(struct sched_entity *se)
{
	return 0;
}

3297
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum) {}
3298

3299
#endif /* CONFIG_FAIR_GROUP_SCHED */
3300

3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311
/**
 * 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
 *
 * 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.
 *
3312 3313 3314 3315
 * 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.
3316
 */
3317
static inline int
3318
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
3319
{
3320
	unsigned long removed_load = 0, removed_util = 0, removed_runnable_sum = 0;
3321
	struct sched_avg *sa = &cfs_rq->avg;
3322
	int decayed = 0;
3323

3324 3325
	if (cfs_rq->removed.nr) {
		unsigned long r;
3326
		u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;
3327 3328 3329 3330

		raw_spin_lock(&cfs_rq->removed.lock);
		swap(cfs_rq->removed.util_avg, removed_util);
		swap(cfs_rq->removed.load_avg, removed_load);
3331
		swap(cfs_rq->removed.runnable_sum, removed_runnable_sum);
3332 3333 3334 3335
		cfs_rq->removed.nr = 0;
		raw_spin_unlock(&cfs_rq->removed.lock);

		r = removed_load;
3336
		sub_positive(&sa->load_avg, r);
3337
		sub_positive(&sa->load_sum, r * divider);
3338

3339
		r = removed_util;
3340
		sub_positive(&sa->util_avg, r);
3341
		sub_positive(&sa->util_sum, r * divider);
3342

3343
		add_tg_cfs_propagate(cfs_rq, -(long)removed_runnable_sum);
3344 3345

		decayed = 1;
3346
	}
3347

3348
	decayed |= __update_load_avg_cfs_rq(now, cpu_of(rq_of(cfs_rq)), cfs_rq);
3349

3350 3351 3352 3353
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
3354

3355
	if (decayed)
3356
		cfs_rq_util_change(cfs_rq, 0);
3357

3358
	return decayed;
3359 3360
}

3361 3362 3363 3364
/**
 * 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
3365
 * @flags: migration hints
3366 3367 3368 3369
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3370
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3371
{
3372 3373 3374 3375 3376 3377 3378 3379 3380
	u32 divider = LOAD_AVG_MAX - 1024 + cfs_rq->avg.period_contrib;

	/*
	 * When we attach the @se to the @cfs_rq, we must align the decay
	 * window because without that, really weird and wonderful things can
	 * happen.
	 *
	 * XXX illustrate
	 */
3381
	se->avg.last_update_time = cfs_rq->avg.last_update_time;
3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399
	se->avg.period_contrib = cfs_rq->avg.period_contrib;

	/*
	 * Hell(o) Nasty stuff.. we need to recompute _sum based on the new
	 * period_contrib. This isn't strictly correct, but since we're
	 * entirely outside of the PELT hierarchy, nobody cares if we truncate
	 * _sum a little.
	 */
	se->avg.util_sum = se->avg.util_avg * divider;

	se->avg.load_sum = divider;
	if (se_weight(se)) {
		se->avg.load_sum =
			div_u64(se->avg.load_avg * se->avg.load_sum, se_weight(se));
	}

	se->avg.runnable_load_sum = se->avg.load_sum;

3400
	enqueue_load_avg(cfs_rq, se);
3401 3402
	cfs_rq->avg.util_avg += se->avg.util_avg;
	cfs_rq->avg.util_sum += se->avg.util_sum;
3403 3404

	add_tg_cfs_propagate(cfs_rq, se->avg.load_sum);
3405

3406
	cfs_rq_util_change(cfs_rq, flags);
3407 3408
}

3409 3410 3411 3412 3413 3414 3415 3416
/**
 * 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.
 */
3417 3418
static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3419
	dequeue_load_avg(cfs_rq, se);
3420 3421
	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
3422 3423

	add_tg_cfs_propagate(cfs_rq, -se->avg.load_sum);
3424

3425
	cfs_rq_util_change(cfs_rq, 0);
3426 3427
}

3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454
/*
 * Optional action to be done while updating the load average
 */
#define UPDATE_TG	0x1
#define SKIP_AGE_LOAD	0x2
#define DO_ATTACH	0x4

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

	/*
	 * 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
	 */
	if (se->avg.last_update_time && !(flags & SKIP_AGE_LOAD))
		__update_load_avg_se(now, cpu, cfs_rq, se);

	decayed  = update_cfs_rq_load_avg(now, cfs_rq);
	decayed |= propagate_entity_load_avg(se);

	if (!se->avg.last_update_time && (flags & DO_ATTACH)) {

3455 3456 3457 3458 3459 3460 3461 3462
		/*
		 * DO_ATTACH means we're here from enqueue_entity().
		 * !last_update_time means we've passed through
		 * migrate_task_rq_fair() indicating we migrated.
		 *
		 * IOW we're enqueueing a task on a new CPU.
		 */
		attach_entity_load_avg(cfs_rq, se, SCHED_CPUFREQ_MIGRATION);
3463 3464 3465 3466 3467 3468
		update_tg_load_avg(cfs_rq, 0);

	} else if (decayed && (flags & UPDATE_TG))
		update_tg_load_avg(cfs_rq, 0);
}

3469
#ifndef CONFIG_64BIT
3470 3471
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
3472
	u64 last_update_time_copy;
3473
	u64 last_update_time;
3474

3475 3476 3477 3478 3479
	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);
3480 3481 3482

	return last_update_time;
}
3483
#else
3484 3485 3486 3487
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
3488 3489
#endif

3490 3491 3492 3493 3494 3495 3496 3497 3498 3499
/*
 * 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);
3500
	__update_load_avg_blocked_se(last_update_time, cpu_of(rq_of(cfs_rq)), se);
3501 3502
}

3503 3504 3505 3506 3507 3508 3509
/*
 * 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);
3510
	unsigned long flags;
3511 3512

	/*
3513 3514 3515 3516 3517 3518 3519
	 * 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.
3520 3521
	 */

3522
	sync_entity_load_avg(se);
3523 3524 3525 3526 3527

	raw_spin_lock_irqsave(&cfs_rq->removed.lock, flags);
	++cfs_rq->removed.nr;
	cfs_rq->removed.util_avg	+= se->avg.util_avg;
	cfs_rq->removed.load_avg	+= se->avg.load_avg;
3528
	cfs_rq->removed.runnable_sum	+= se->avg.load_sum; /* == runnable_sum */
3529
	raw_spin_unlock_irqrestore(&cfs_rq->removed.lock, flags);
3530
}
3531

3532 3533
static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq)
{
3534
	return cfs_rq->avg.runnable_load_avg;
3535 3536 3537 3538 3539 3540 3541
}

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

3542
static int idle_balance(struct rq *this_rq, struct rq_flags *rf);
3543

3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570
static inline unsigned long task_util(struct task_struct *p)
{
	return READ_ONCE(p->se.avg.util_avg);
}

static inline unsigned long _task_util_est(struct task_struct *p)
{
	struct util_est ue = READ_ONCE(p->se.avg.util_est);

	return max(ue.ewma, ue.enqueued);
}

static inline unsigned long task_util_est(struct task_struct *p)
{
	return max(task_util(p), _task_util_est(p));
}

static inline void util_est_enqueue(struct cfs_rq *cfs_rq,
				    struct task_struct *p)
{
	unsigned int enqueued;

	if (!sched_feat(UTIL_EST))
		return;

	/* Update root cfs_rq's estimated utilization */
	enqueued  = cfs_rq->avg.util_est.enqueued;
3571
	enqueued += (_task_util_est(p) | UTIL_AVG_UNCHANGED);
3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596
	WRITE_ONCE(cfs_rq->avg.util_est.enqueued, enqueued);
}

/*
 * Check if a (signed) value is within a specified (unsigned) margin,
 * based on the observation that:
 *
 *     abs(x) < y := (unsigned)(x + y - 1) < (2 * y - 1)
 *
 * NOTE: this only works when value + maring < INT_MAX.
 */
static inline bool within_margin(int value, int margin)
{
	return ((unsigned int)(value + margin - 1) < (2 * margin - 1));
}

static void
util_est_dequeue(struct cfs_rq *cfs_rq, struct task_struct *p, bool task_sleep)
{
	long last_ewma_diff;
	struct util_est ue;

	if (!sched_feat(UTIL_EST))
		return;

3597 3598 3599 3600
	/* Update root cfs_rq's estimated utilization */
	ue.enqueued  = cfs_rq->avg.util_est.enqueued;
	ue.enqueued -= min_t(unsigned int, ue.enqueued,
			     (_task_util_est(p) | UTIL_AVG_UNCHANGED));
3601 3602 3603 3604 3605 3606 3607 3608 3609
	WRITE_ONCE(cfs_rq->avg.util_est.enqueued, ue.enqueued);

	/*
	 * Skip update of task's estimated utilization when the task has not
	 * yet completed an activation, e.g. being migrated.
	 */
	if (!task_sleep)
		return;

3610 3611 3612 3613 3614 3615 3616 3617
	/*
	 * If the PELT values haven't changed since enqueue time,
	 * skip the util_est update.
	 */
	ue = p->se.avg.util_est;
	if (ue.enqueued & UTIL_AVG_UNCHANGED)
		return;

3618 3619 3620 3621
	/*
	 * Skip update of task's estimated utilization when its EWMA is
	 * already ~1% close to its last activation value.
	 */
3622
	ue.enqueued = (task_util(p) | UTIL_AVG_UNCHANGED);
3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649
	last_ewma_diff = ue.enqueued - ue.ewma;
	if (within_margin(last_ewma_diff, (SCHED_CAPACITY_SCALE / 100)))
		return;

	/*
	 * Update Task's estimated utilization
	 *
	 * When *p completes an activation we can consolidate another sample
	 * of the task size. This is done by storing the current PELT value
	 * as ue.enqueued and by using this value to update the Exponential
	 * Weighted Moving Average (EWMA):
	 *
	 *  ewma(t) = w *  task_util(p) + (1-w) * ewma(t-1)
	 *          = w *  task_util(p) +         ewma(t-1)  - w * ewma(t-1)
	 *          = w * (task_util(p) -         ewma(t-1)) +     ewma(t-1)
	 *          = w * (      last_ewma_diff            ) +     ewma(t-1)
	 *          = w * (last_ewma_diff  +  ewma(t-1) / w)
	 *
	 * Where 'w' is the weight of new samples, which is configured to be
	 * 0.25, thus making w=1/4 ( >>= UTIL_EST_WEIGHT_SHIFT)
	 */
	ue.ewma <<= UTIL_EST_WEIGHT_SHIFT;
	ue.ewma  += last_ewma_diff;
	ue.ewma >>= UTIL_EST_WEIGHT_SHIFT;
	WRITE_ONCE(p->se.avg.util_est, ue);
}

3650 3651
#else /* CONFIG_SMP */

3652 3653
#define UPDATE_TG	0x0
#define SKIP_AGE_LOAD	0x0
3654
#define DO_ATTACH	0x0
3655

3656
static inline void update_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int not_used1)
3657
{
3658
	cfs_rq_util_change(cfs_rq, 0);
3659 3660
}

3661
static inline void remove_entity_load_avg(struct sched_entity *se) {}
3662

3663
static inline void
3664
attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) {}
3665 3666 3667
static inline void
detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}

3668
static inline int idle_balance(struct rq *rq, struct rq_flags *rf)
3669 3670 3671 3672
{
	return 0;
}

3673 3674 3675 3676 3677 3678 3679
static inline void
util_est_enqueue(struct cfs_rq *cfs_rq, struct task_struct *p) {}

static inline void
util_est_dequeue(struct cfs_rq *cfs_rq, struct task_struct *p,
		 bool task_sleep) {}

3680
#endif /* CONFIG_SMP */
3681

P
Peter Zijlstra 已提交
3682 3683 3684 3685 3686 3687 3688 3689 3690
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)
3691
		schedstat_inc(cfs_rq->nr_spread_over);
P
Peter Zijlstra 已提交
3692 3693 3694
#endif
}

3695 3696 3697
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3698
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3699

3700 3701 3702 3703 3704 3705
	/*
	 * 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
3706
	if (initial && sched_feat(START_DEBIT))
3707
		vruntime += sched_vslice(cfs_rq, se);
3708

3709
	/* sleeps up to a single latency don't count. */
3710
	if (!initial) {
3711
		unsigned long thresh = sysctl_sched_latency;
3712

3713 3714 3715 3716 3717 3718
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3719

3720
		vruntime -= thresh;
3721 3722
	}

3723
	/* ensure we never gain time by being placed backwards. */
3724
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3725 3726
}

3727 3728
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740
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())  {
3741
		printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
3742
			     "stat_blocked and stat_runtime require the "
3743
			     "kernel parameter schedstats=enable or "
3744 3745 3746 3747 3748
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767

/*
 * 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)
 *
3768
 *	->migrate_task_rq_fair() (p->state == TASK_WAKING)
3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779
 *	  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.
 */

3780
static void
3781
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3782
{
3783 3784 3785
	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
	bool curr = cfs_rq->curr == se;

3786
	/*
3787 3788
	 * If we're the current task, we must renormalise before calling
	 * update_curr().
3789
	 */
3790
	if (renorm && curr)
3791 3792
		se->vruntime += cfs_rq->min_vruntime;

3793 3794
	update_curr(cfs_rq);

3795
	/*
3796 3797 3798 3799
	 * 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.
3800
	 */
3801 3802 3803
	if (renorm && !curr)
		se->vruntime += cfs_rq->min_vruntime;

3804 3805 3806 3807 3808 3809 3810 3811
	/*
	 * When enqueuing a sched_entity, we must:
	 *   - Update loads to have both entity and cfs_rq synced with now.
	 *   - Add its load to cfs_rq->runnable_avg
	 *   - For group_entity, update its weight to reflect the new share of
	 *     its group cfs_rq
	 *   - Add its new weight to cfs_rq->load.weight
	 */
3812
	update_load_avg(cfs_rq, se, UPDATE_TG | DO_ATTACH);
3813
	update_cfs_group(se);
3814
	enqueue_runnable_load_avg(cfs_rq, se);
3815
	account_entity_enqueue(cfs_rq, se);
3816

3817
	if (flags & ENQUEUE_WAKEUP)
3818
		place_entity(cfs_rq, se, 0);
3819

3820
	check_schedstat_required();
3821 3822
	update_stats_enqueue(cfs_rq, se, flags);
	check_spread(cfs_rq, se);
3823
	if (!curr)
3824
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3825
	se->on_rq = 1;
3826

3827
	if (cfs_rq->nr_running == 1) {
3828
		list_add_leaf_cfs_rq(cfs_rq);
3829 3830
		check_enqueue_throttle(cfs_rq);
	}
3831 3832
}

3833
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3834
{
3835 3836
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3837
		if (cfs_rq->last != se)
3838
			break;
3839 3840

		cfs_rq->last = NULL;
3841 3842
	}
}
P
Peter Zijlstra 已提交
3843

3844 3845 3846 3847
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3848
		if (cfs_rq->next != se)
3849
			break;
3850 3851

		cfs_rq->next = NULL;
3852
	}
P
Peter Zijlstra 已提交
3853 3854
}

3855 3856 3857 3858
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3859
		if (cfs_rq->skip != se)
3860
			break;
3861 3862

		cfs_rq->skip = NULL;
3863 3864 3865
	}
}

P
Peter Zijlstra 已提交
3866 3867
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3868 3869 3870 3871 3872
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
3873 3874 3875

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
3876 3877
}

3878
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3879

3880
static void
3881
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3882
{
3883 3884 3885 3886
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
3887 3888 3889 3890 3891 3892 3893 3894 3895

	/*
	 * When dequeuing a sched_entity, we must:
	 *   - Update loads to have both entity and cfs_rq synced with now.
	 *   - Substract its load from the cfs_rq->runnable_avg.
	 *   - Substract its previous weight from cfs_rq->load.weight.
	 *   - For group entity, update its weight to reflect the new share
	 *     of its group cfs_rq.
	 */
3896
	update_load_avg(cfs_rq, se, UPDATE_TG);
3897
	dequeue_runnable_load_avg(cfs_rq, se);
3898

3899
	update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
3900

P
Peter Zijlstra 已提交
3901
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3902

3903
	if (se != cfs_rq->curr)
3904
		__dequeue_entity(cfs_rq, se);
3905
	se->on_rq = 0;
3906
	account_entity_dequeue(cfs_rq, se);
3907 3908

	/*
3909 3910 3911 3912
	 * 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.
3913
	 */
3914
	if (!(flags & DEQUEUE_SLEEP))
3915
		se->vruntime -= cfs_rq->min_vruntime;
3916

3917 3918 3919
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

3920
	update_cfs_group(se);
3921 3922 3923 3924 3925 3926 3927 3928 3929

	/*
	 * 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);
3930 3931 3932 3933 3934
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
3935
static void
I
Ingo Molnar 已提交
3936
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3937
{
3938
	unsigned long ideal_runtime, delta_exec;
3939 3940
	struct sched_entity *se;
	s64 delta;
3941

P
Peter Zijlstra 已提交
3942
	ideal_runtime = sched_slice(cfs_rq, curr);
3943
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
3944
	if (delta_exec > ideal_runtime) {
3945
		resched_curr(rq_of(cfs_rq));
3946 3947 3948 3949 3950
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961
		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;

3962 3963
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
3964

3965 3966
	if (delta < 0)
		return;
3967

3968
	if (delta > ideal_runtime)
3969
		resched_curr(rq_of(cfs_rq));
3970 3971
}

3972
static void
3973
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
3974
{
3975 3976 3977 3978 3979 3980 3981
	/* '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.
		 */
3982
		update_stats_wait_end(cfs_rq, se);
3983
		__dequeue_entity(cfs_rq, se);
3984
		update_load_avg(cfs_rq, se, UPDATE_TG);
3985 3986
	}

3987
	update_stats_curr_start(cfs_rq, se);
3988
	cfs_rq->curr = se;
3989

I
Ingo Molnar 已提交
3990 3991 3992 3993 3994
	/*
	 * 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):
	 */
3995
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
3996 3997 3998
		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
3999
	}
4000

4001
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
4002 4003
}

4004 4005 4006
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

4007 4008 4009 4010 4011 4012 4013
/*
 * 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
 */
4014 4015
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4016
{
4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027
	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 */
4028

4029 4030 4031 4032 4033
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
4034 4035 4036 4037 4038 4039 4040 4041 4042 4043
		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;
		}

4044 4045 4046
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
4047

4048 4049 4050 4051 4052 4053
	/*
	 * 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;

4054 4055 4056 4057 4058 4059
	/*
	 * 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;

4060
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4061 4062

	return se;
4063 4064
}

4065
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4066

4067
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
4068 4069 4070 4071 4072 4073
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
4074
		update_curr(cfs_rq);
4075

4076 4077 4078
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

4079
	check_spread(cfs_rq, prev);
4080

4081
	if (prev->on_rq) {
4082
		update_stats_wait_start(cfs_rq, prev);
4083 4084
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
4085
		/* in !on_rq case, update occurred at dequeue */
4086
		update_load_avg(cfs_rq, prev, 0);
4087
	}
4088
	cfs_rq->curr = NULL;
4089 4090
}

P
Peter Zijlstra 已提交
4091 4092
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
4093 4094
{
	/*
4095
	 * Update run-time statistics of the 'current'.
4096
	 */
4097
	update_curr(cfs_rq);
4098

4099 4100 4101
	/*
	 * Ensure that runnable average is periodically updated.
	 */
4102
	update_load_avg(cfs_rq, curr, UPDATE_TG);
4103
	update_cfs_group(curr);
4104

P
Peter Zijlstra 已提交
4105 4106 4107 4108 4109
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
4110
	if (queued) {
4111
		resched_curr(rq_of(cfs_rq));
4112 4113
		return;
	}
P
Peter Zijlstra 已提交
4114 4115 4116 4117 4118 4119 4120 4121
	/*
	 * 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
4122
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
4123
		check_preempt_tick(cfs_rq, curr);
4124 4125
}

4126 4127 4128 4129 4130 4131

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

#ifdef CONFIG_CFS_BANDWIDTH
4132 4133

#ifdef HAVE_JUMP_LABEL
4134
static struct static_key __cfs_bandwidth_used;
4135 4136 4137

static inline bool cfs_bandwidth_used(void)
{
4138
	return static_key_false(&__cfs_bandwidth_used);
4139 4140
}

4141
void cfs_bandwidth_usage_inc(void)
4142
{
4143
	static_key_slow_inc_cpuslocked(&__cfs_bandwidth_used);
4144 4145 4146 4147
}

void cfs_bandwidth_usage_dec(void)
{
4148
	static_key_slow_dec_cpuslocked(&__cfs_bandwidth_used);
4149 4150 4151 4152 4153 4154 4155
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

4156 4157
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
4158 4159
#endif /* HAVE_JUMP_LABEL */

4160 4161 4162 4163 4164 4165 4166 4167
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
4168 4169 4170 4171 4172 4173

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

P
Paul Turner 已提交
4174 4175 4176 4177 4178 4179 4180
/*
 * 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
 */
4181
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
4182 4183 4184 4185 4186 4187 4188 4189 4190
{
	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);
4191
	cfs_b->expires_seq++;
P
Paul Turner 已提交
4192 4193
}

4194 4195 4196 4197 4198
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

4199 4200 4201 4202
/* 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))
4203
		return cfs_rq->throttled_clock_task - cfs_rq->throttled_clock_task_time;
4204

4205
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
4206 4207
}

4208 4209
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4210 4211 4212
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
4213
	u64 amount = 0, min_amount, expires;
4214
	int expires_seq;
4215 4216 4217 4218 4219 4220 4221

	/* 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;
4222
	else {
P
Peter Zijlstra 已提交
4223
		start_cfs_bandwidth(cfs_b);
4224 4225 4226 4227 4228 4229

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
4230
	}
4231
	expires_seq = cfs_b->expires_seq;
P
Paul Turner 已提交
4232
	expires = cfs_b->runtime_expires;
4233 4234 4235
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
4236 4237 4238 4239 4240
	/*
	 * we may have advanced our local expiration to account for allowed
	 * spread between our sched_clock and the one on which runtime was
	 * issued.
	 */
4241 4242
	if (cfs_rq->expires_seq != expires_seq) {
		cfs_rq->expires_seq = expires_seq;
P
Paul Turner 已提交
4243
		cfs_rq->runtime_expires = expires;
4244
	}
4245 4246

	return cfs_rq->runtime_remaining > 0;
4247 4248
}

P
Paul Turner 已提交
4249 4250 4251 4252 4253
/*
 * 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)
4254
{
P
Paul Turner 已提交
4255 4256 4257
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
4258
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
4259 4260
		return;

P
Paul Turner 已提交
4261 4262 4263 4264 4265 4266 4267 4268 4269
	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
4270
	 * whether the global deadline(cfs_b->expires_seq) has advanced.
P
Paul Turner 已提交
4271
	 */
4272
	if (cfs_rq->expires_seq == cfs_b->expires_seq) {
P
Paul Turner 已提交
4273 4274 4275 4276 4277 4278 4279 4280
		/* 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;
	}
}

4281
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
4282 4283
{
	/* dock delta_exec before expiring quota (as it could span periods) */
4284
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
4285 4286 4287
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
4288 4289
		return;

4290 4291 4292 4293 4294
	/*
	 * 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))
4295
		resched_curr(rq_of(cfs_rq));
4296 4297
}

4298
static __always_inline
4299
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
4300
{
4301
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
4302 4303 4304 4305 4306
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

4307 4308
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
4309
	return cfs_bandwidth_used() && cfs_rq->throttled;
4310 4311
}

4312 4313 4314
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
4315
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341
}

/*
 * 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);
}

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) {
4342
		/* adjust cfs_rq_clock_task() */
4343
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
4344
					     cfs_rq->throttled_clock_task;
4345 4346 4347 4348 4349 4350 4351 4352 4353 4354
	}

	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)];

4355 4356
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
4357
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
4358 4359 4360 4361 4362
	cfs_rq->throttle_count++;

	return 0;
}

4363
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
4364 4365 4366 4367 4368
{
	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
4369
	bool empty;
4370 4371 4372

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

4373
	/* freeze hierarchy runnable averages while throttled */
4374 4375 4376
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393

	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)
4394
		sub_nr_running(rq, task_delta);
4395 4396

	cfs_rq->throttled = 1;
4397
	cfs_rq->throttled_clock = rq_clock(rq);
4398
	raw_spin_lock(&cfs_b->lock);
4399
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4400

4401 4402 4403 4404 4405
	/*
	 * 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
4406 4407 4408 4409 4410 4411 4412 4413

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

4414 4415 4416
	raw_spin_unlock(&cfs_b->lock);
}

4417
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
4418 4419 4420 4421 4422 4423 4424
{
	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;

4425
	se = cfs_rq->tg->se[cpu_of(rq)];
4426 4427

	cfs_rq->throttled = 0;
4428 4429 4430

	update_rq_clock(rq);

4431
	raw_spin_lock(&cfs_b->lock);
4432
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
4433 4434 4435
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

4436 4437 4438
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456
	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)
4457
		add_nr_running(rq, task_delta);
4458

4459
	/* Determine whether we need to wake up potentially idle CPU: */
4460
	if (rq->curr == rq->idle && rq->cfs.nr_running)
4461
		resched_curr(rq);
4462 4463 4464 4465 4466 4467
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
4468 4469
	u64 runtime;
	u64 starting_runtime = remaining;
4470 4471 4472 4473 4474

	rcu_read_lock();
	list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
				throttled_list) {
		struct rq *rq = rq_of(cfs_rq);
4475
		struct rq_flags rf;
4476

4477
		rq_lock(rq, &rf);
4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493
		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:
4494
		rq_unlock(rq, &rf);
4495 4496 4497 4498 4499 4500

		if (!remaining)
			break;
	}
	rcu_read_unlock();

4501
	return starting_runtime - remaining;
4502 4503
}

4504 4505 4506 4507 4508 4509 4510 4511
/*
 * 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)
{
4512
	u64 runtime, runtime_expires;
4513
	int throttled;
4514 4515 4516

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
4517
		goto out_deactivate;
4518

4519
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4520
	cfs_b->nr_periods += overrun;
4521

4522 4523 4524 4525 4526 4527
	/*
	 * 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
4528 4529 4530

	__refill_cfs_bandwidth_runtime(cfs_b);

4531 4532 4533
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
4534
		return 0;
4535 4536
	}

4537 4538 4539
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

4540 4541 4542
	runtime_expires = cfs_b->runtime_expires;

	/*
4543 4544 4545 4546 4547
	 * 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.
4548
	 */
4549 4550
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
4551 4552 4553 4554 4555 4556 4557
		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);
4558 4559

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4560
	}
4561

4562 4563 4564 4565 4566 4567 4568
	/*
	 * 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;
4569

4570 4571 4572 4573
	return 0;

out_deactivate:
	return 1;
4574
}
4575

4576 4577 4578 4579 4580 4581 4582
/* 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;

4583 4584 4585 4586
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
4587
 * hrtimer base being cleared by hrtimer_start. In the case of
4588 4589
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614
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
4615 4616 4617
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646
}

/* 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)
{
4647 4648 4649
	if (!cfs_bandwidth_used())
		return;

4650
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665
		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 */
4666 4667 4668
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
4669
		return;
4670
	}
4671

4672
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
4673
		runtime = cfs_b->runtime;
4674

4675 4676 4677 4678 4679 4680 4681 4682 4683 4684
	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)
4685
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4686 4687 4688
	raw_spin_unlock(&cfs_b->lock);
}

4689 4690 4691 4692 4693 4694 4695
/*
 * 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)
{
4696 4697 4698
	if (!cfs_bandwidth_used())
		return;

4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712
	/* 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);
}

4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726
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;
4727
	cfs_rq->throttled_clock_task = rq_clock_task(cpu_rq(cpu));
4728 4729
}

4730
/* conditionally throttle active cfs_rq's from put_prev_entity() */
4731
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4732
{
4733
	if (!cfs_bandwidth_used())
4734
		return false;
4735

4736
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4737
		return false;
4738 4739 4740 4741 4742 4743

	/*
	 * 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))
4744
		return true;
4745 4746

	throttle_cfs_rq(cfs_rq);
4747
	return true;
4748
}
4749 4750 4751 4752 4753

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
4754

4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766
	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;

4767
	raw_spin_lock(&cfs_b->lock);
4768
	for (;;) {
P
Peter Zijlstra 已提交
4769
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4770 4771 4772 4773 4774
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4775 4776
	if (idle)
		cfs_b->period_active = 0;
4777
	raw_spin_unlock(&cfs_b->lock);
4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789

	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
4790
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801
	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
4802
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4803
{
4804 4805
	u64 overrun;

P
Peter Zijlstra 已提交
4806
	lockdep_assert_held(&cfs_b->lock);
4807

4808 4809 4810 4811 4812 4813 4814 4815
	if (cfs_b->period_active)
		return;

	cfs_b->period_active = 1;
	overrun = hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
	cfs_b->runtime_expires += (overrun + 1) * ktime_to_ns(cfs_b->period);
	cfs_b->expires_seq++;
	hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
4816 4817 4818 4819
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4820 4821 4822 4823
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4824 4825 4826 4827
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4828
/*
4829
 * Both these CPU hotplug callbacks race against unregister_fair_sched_group()
4830 4831 4832 4833 4834 4835
 *
 * The race is harmless, since modifying bandwidth settings of unhooked group
 * bits doesn't do much.
 */

/* cpu online calback */
4836 4837
static void __maybe_unused update_runtime_enabled(struct rq *rq)
{
4838
	struct task_group *tg;
4839

4840 4841 4842 4843 4844 4845
	lockdep_assert_held(&rq->lock);

	rcu_read_lock();
	list_for_each_entry_rcu(tg, &task_groups, list) {
		struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
		struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
4846 4847 4848 4849 4850

		raw_spin_lock(&cfs_b->lock);
		cfs_rq->runtime_enabled = cfs_b->quota != RUNTIME_INF;
		raw_spin_unlock(&cfs_b->lock);
	}
4851
	rcu_read_unlock();
4852 4853
}

4854
/* cpu offline callback */
4855
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4856
{
4857 4858 4859 4860 4861 4862 4863
	struct task_group *tg;

	lockdep_assert_held(&rq->lock);

	rcu_read_lock();
	list_for_each_entry_rcu(tg, &task_groups, list) {
		struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
4864 4865 4866 4867 4868 4869 4870 4871

		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
4872
		cfs_rq->runtime_remaining = 1;
4873
		/*
4874
		 * Offline rq is schedulable till CPU is completely disabled
4875 4876 4877 4878
		 * in take_cpu_down(), so we prevent new cfs throttling here.
		 */
		cfs_rq->runtime_enabled = 0;

4879 4880 4881
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
4882
	rcu_read_unlock();
4883 4884 4885
}

#else /* CONFIG_CFS_BANDWIDTH */
4886 4887
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
4888
	return rq_clock_task(rq_of(cfs_rq));
4889 4890
}

4891
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
4892
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
4893
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
4894
static inline void sync_throttle(struct task_group *tg, int cpu) {}
4895
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
4896 4897 4898 4899 4900

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911

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;
}
4912 4913 4914 4915 4916

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) {}
4917 4918
#endif

4919 4920 4921 4922 4923
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) {}
4924
static inline void update_runtime_enabled(struct rq *rq) {}
4925
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
4926 4927 4928

#endif /* CONFIG_CFS_BANDWIDTH */

4929 4930 4931 4932
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
4933 4934 4935 4936 4937 4938
#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);

4939
	SCHED_WARN_ON(task_rq(p) != rq);
P
Peter Zijlstra 已提交
4940

4941
	if (rq->cfs.h_nr_running > 1) {
P
Peter Zijlstra 已提交
4942 4943 4944 4945 4946 4947
		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)
4948
				resched_curr(rq);
P
Peter Zijlstra 已提交
4949 4950
			return;
		}
4951
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
4952 4953
	}
}
4954 4955 4956 4957 4958 4959 4960 4961 4962 4963

/*
 * 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;

4964
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
4965 4966 4967 4968 4969
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
4970
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
4971 4972 4973 4974
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
4975 4976 4977 4978

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
4979 4980
#endif

4981 4982 4983 4984 4985
/*
 * 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:
 */
4986
static void
4987
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4988 4989
{
	struct cfs_rq *cfs_rq;
4990
	struct sched_entity *se = &p->se;
4991

4992 4993 4994 4995 4996 4997 4998 4999
	/*
	 * The code below (indirectly) updates schedutil which looks at
	 * the cfs_rq utilization to select a frequency.
	 * Let's add the task's estimated utilization to the cfs_rq's
	 * estimated utilization, before we update schedutil.
	 */
	util_est_enqueue(&rq->cfs, p);

5000 5001 5002 5003 5004 5005
	/*
	 * 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)
5006
		cpufreq_update_util(rq, SCHED_CPUFREQ_IOWAIT);
5007

5008
	for_each_sched_entity(se) {
5009
		if (se->on_rq)
5010 5011
			break;
		cfs_rq = cfs_rq_of(se);
5012
		enqueue_entity(cfs_rq, se, flags);
5013 5014 5015 5016 5017 5018

		/*
		 * 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.
5019
		 */
5020 5021
		if (cfs_rq_throttled(cfs_rq))
			break;
5022
		cfs_rq->h_nr_running++;
5023

5024
		flags = ENQUEUE_WAKEUP;
5025
	}
P
Peter Zijlstra 已提交
5026

P
Peter Zijlstra 已提交
5027
	for_each_sched_entity(se) {
5028
		cfs_rq = cfs_rq_of(se);
5029
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
5030

5031 5032 5033
		if (cfs_rq_throttled(cfs_rq))
			break;

5034
		update_load_avg(cfs_rq, se, UPDATE_TG);
5035
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5036 5037
	}

Y
Yuyang Du 已提交
5038
	if (!se)
5039
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
5040

5041
	hrtick_update(rq);
5042 5043
}

5044 5045
static void set_next_buddy(struct sched_entity *se);

5046 5047 5048 5049 5050
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
5051
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5052 5053
{
	struct cfs_rq *cfs_rq;
5054
	struct sched_entity *se = &p->se;
5055
	int task_sleep = flags & DEQUEUE_SLEEP;
5056 5057 5058

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5059
		dequeue_entity(cfs_rq, se, flags);
5060 5061 5062 5063 5064 5065 5066 5067 5068

		/*
		 * 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;
5069
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5070

5071
		/* Don't dequeue parent if it has other entities besides us */
5072
		if (cfs_rq->load.weight) {
5073 5074
			/* Avoid re-evaluating load for this entity: */
			se = parent_entity(se);
5075 5076 5077 5078
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
5079 5080
			if (task_sleep && se && !throttled_hierarchy(cfs_rq))
				set_next_buddy(se);
5081
			break;
5082
		}
5083
		flags |= DEQUEUE_SLEEP;
5084
	}
P
Peter Zijlstra 已提交
5085

P
Peter Zijlstra 已提交
5086
	for_each_sched_entity(se) {
5087
		cfs_rq = cfs_rq_of(se);
5088
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5089

5090 5091 5092
		if (cfs_rq_throttled(cfs_rq))
			break;

5093
		update_load_avg(cfs_rq, se, UPDATE_TG);
5094
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5095 5096
	}

Y
Yuyang Du 已提交
5097
	if (!se)
5098
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
5099

5100
	util_est_dequeue(&rq->cfs, p, task_sleep);
5101
	hrtick_update(rq);
5102 5103
}

5104
#ifdef CONFIG_SMP
5105 5106 5107 5108 5109

/* 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);

5110
#ifdef CONFIG_NO_HZ_COMMON
5111 5112 5113 5114 5115
/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
5116
 * The exact cpuload calculated at every tick would be:
5117
 *
5118 5119
 *   load' = (1 - 1/2^i) * load + (1/2^i) * cur_load
 *
5120 5121
 * 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:
5122 5123 5124
 *
 *   load_n   = (1 - 1/2^i)^n * load_0
 *   load_n+1 = (1 - 1/2^i)   * load_n + (1/2^i) * cur_load
5125 5126 5127
 *
 * decay_load_missed() below does efficient calculation of
 *
5128 5129 5130 5131 5132 5133
 *   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())
5134
 *
5135
 * The calculation is approximated on a 128 point scale.
5136 5137
 */
#define DEGRADE_SHIFT		7
5138 5139 5140 5141 5142 5143 5144 5145 5146

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 }
};
5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175

/*
 * 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;
}
5176 5177 5178 5179

static struct {
	cpumask_var_t idle_cpus_mask;
	atomic_t nr_cpus;
5180
	int has_blocked;		/* Idle CPUS has blocked load */
5181
	unsigned long next_balance;     /* in jiffy units */
5182
	unsigned long next_blocked;	/* Next update of blocked load in jiffies */
5183 5184
} nohz ____cacheline_aligned;

5185
#endif /* CONFIG_NO_HZ_COMMON */
5186

5187
/**
5188
 * __cpu_load_update - update the rq->cpu_load[] statistics
5189 5190 5191 5192
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 *
5193
 * Update rq->cpu_load[] statistics. This function is usually called every
5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219
 * 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
5220
 * term.
5221
 */
5222 5223
static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
			    unsigned long pending_updates)
5224
{
5225
	unsigned long __maybe_unused tickless_load = this_rq->cpu_load[0];
5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236
	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 */

5237
		old_load = this_rq->cpu_load[i];
5238
#ifdef CONFIG_NO_HZ_COMMON
5239
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
5240 5241 5242 5243 5244 5245 5246 5247 5248
		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;
		}
5249
#endif
5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262
		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;
	}
}

5263
/* Used instead of source_load when we know the type == 0 */
5264
static unsigned long weighted_cpuload(struct rq *rq)
5265
{
5266
	return cfs_rq_runnable_load_avg(&rq->cfs);
5267 5268
}

5269
#ifdef CONFIG_NO_HZ_COMMON
5270 5271
/*
 * There is no sane way to deal with nohz on smp when using jiffies because the
5272
 * CPU doing the jiffies update might drift wrt the CPU doing the jiffy reading
5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286
 * 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)
5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297
{
	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.
		 */
5298
		cpu_load_update(this_rq, load, pending_updates);
5299 5300 5301
	}
}

5302 5303 5304 5305
/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
5306
static void cpu_load_update_idle(struct rq *this_rq)
5307 5308 5309 5310
{
	/*
	 * bail if there's load or we're actually up-to-date.
	 */
5311
	if (weighted_cpuload(this_rq))
5312 5313
		return;

5314
	cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), 0);
5315 5316 5317
}

/*
5318 5319 5320 5321
 * 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.
5322
 */
5323
void cpu_load_update_nohz_start(void)
5324 5325
{
	struct rq *this_rq = this_rq();
5326 5327 5328 5329 5330 5331

	/*
	 * 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.
	 */
5332
	this_rq->cpu_load[0] = weighted_cpuload(this_rq);
5333 5334 5335 5336 5337 5338 5339
}

/*
 * Account the tickless load in the end of a nohz frame.
 */
void cpu_load_update_nohz_stop(void)
{
5340
	unsigned long curr_jiffies = READ_ONCE(jiffies);
5341 5342
	struct rq *this_rq = this_rq();
	unsigned long load;
5343
	struct rq_flags rf;
5344 5345 5346 5347

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

5348
	load = weighted_cpuload(this_rq);
5349
	rq_lock(this_rq, &rf);
5350
	update_rq_clock(this_rq);
5351
	cpu_load_update_nohz(this_rq, curr_jiffies, load);
5352
	rq_unlock(this_rq, &rf);
5353
}
5354 5355 5356 5357 5358 5359 5360 5361
#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)
{
5362
#ifdef CONFIG_NO_HZ_COMMON
5363 5364
	/* See the mess around cpu_load_update_nohz(). */
	this_rq->last_load_update_tick = READ_ONCE(jiffies);
5365
#endif
5366 5367
	cpu_load_update(this_rq, load, 1);
}
5368 5369 5370 5371

/*
 * Called from scheduler_tick()
 */
5372
void cpu_load_update_active(struct rq *this_rq)
5373
{
5374
	unsigned long load = weighted_cpuload(this_rq);
5375 5376 5377 5378 5379

	if (tick_nohz_tick_stopped())
		cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), load);
	else
		cpu_load_update_periodic(this_rq, load);
5380 5381
}

5382
/*
5383
 * Return a low guess at the load of a migration-source CPU weighted
5384 5385 5386 5387 5388 5389 5390 5391
 * 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);
5392
	unsigned long total = weighted_cpuload(rq);
5393 5394 5395 5396 5397 5398 5399 5400

	if (type == 0 || !sched_feat(LB_BIAS))
		return total;

	return min(rq->cpu_load[type-1], total);
}

/*
5401
 * Return a high guess at the load of a migration-target CPU weighted
5402 5403 5404 5405 5406
 * according to the scheduling class and "nice" value.
 */
static unsigned long target_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
5407
	unsigned long total = weighted_cpuload(rq);
5408 5409 5410 5411 5412 5413 5414

	if (type == 0 || !sched_feat(LB_BIAS))
		return total;

	return max(rq->cpu_load[type-1], total);
}

5415
static unsigned long capacity_of(int cpu)
5416
{
5417
	return cpu_rq(cpu)->cpu_capacity;
5418 5419
}

5420 5421 5422 5423 5424
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

5425 5426 5427
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5428
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
5429
	unsigned long load_avg = weighted_cpuload(rq);
5430 5431

	if (nr_running)
5432
		return load_avg / nr_running;
5433 5434 5435 5436

	return 0;
}

P
Peter Zijlstra 已提交
5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453
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
5454 5455
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
P
Peter Zijlstra 已提交
5456
 *
M
Mike Galbraith 已提交
5457
 * A waker of many should wake a different task than the one last awakened
P
Peter Zijlstra 已提交
5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469
 * 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
5470
 */
5471 5472
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
5473 5474
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
5475
	int factor = this_cpu_read(sd_llc_size);
5476

M
Mike Galbraith 已提交
5477 5478 5479 5480 5481
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
5482 5483
}

5484
/*
5485 5486 5487
 * The purpose of wake_affine() is to quickly determine on which CPU we can run
 * soonest. For the purpose of speed we only consider the waking and previous
 * CPU.
5488
 *
5489 5490
 * wake_affine_idle() - only considers 'now', it check if the waking CPU is
 *			cache-affine and is (or	will be) idle.
5491 5492 5493 5494
 *
 * wake_affine_weight() - considers the weight to reflect the average
 *			  scheduling latency of the CPUs. This seems to work
 *			  for the overloaded case.
5495
 */
5496
static int
5497
wake_affine_idle(int this_cpu, int prev_cpu, int sync)
5498
{
5499 5500 5501 5502 5503
	/*
	 * If this_cpu is idle, it implies the wakeup is from interrupt
	 * context. Only allow the move if cache is shared. Otherwise an
	 * interrupt intensive workload could force all tasks onto one
	 * node depending on the IO topology or IRQ affinity settings.
5504 5505 5506 5507 5508 5509
	 *
	 * If the prev_cpu is idle and cache affine then avoid a migration.
	 * There is no guarantee that the cache hot data from an interrupt
	 * is more important than cache hot data on the prev_cpu and from
	 * a cpufreq perspective, it's better to have higher utilisation
	 * on one CPU.
5510
	 */
5511 5512
	if (available_idle_cpu(this_cpu) && cpus_share_cache(this_cpu, prev_cpu))
		return available_idle_cpu(prev_cpu) ? prev_cpu : this_cpu;
5513

5514
	if (sync && cpu_rq(this_cpu)->nr_running == 1)
5515
		return this_cpu;
5516

5517
	return nr_cpumask_bits;
5518 5519
}

5520
static int
5521 5522
wake_affine_weight(struct sched_domain *sd, struct task_struct *p,
		   int this_cpu, int prev_cpu, int sync)
5523 5524 5525 5526
{
	s64 this_eff_load, prev_eff_load;
	unsigned long task_load;

5527
	this_eff_load = target_load(this_cpu, sd->wake_idx);
5528 5529 5530 5531

	if (sync) {
		unsigned long current_load = task_h_load(current);

5532
		if (current_load > this_eff_load)
5533
			return this_cpu;
5534

5535
		this_eff_load -= current_load;
5536 5537 5538 5539
	}

	task_load = task_h_load(p);

5540 5541 5542 5543
	this_eff_load += task_load;
	if (sched_feat(WA_BIAS))
		this_eff_load *= 100;
	this_eff_load *= capacity_of(prev_cpu);
5544

5545
	prev_eff_load = source_load(prev_cpu, sd->wake_idx);
5546 5547 5548 5549
	prev_eff_load -= task_load;
	if (sched_feat(WA_BIAS))
		prev_eff_load *= 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
5550

5551 5552 5553 5554 5555 5556 5557 5558 5559 5560
	/*
	 * If sync, adjust the weight of prev_eff_load such that if
	 * prev_eff == this_eff that select_idle_sibling() will consider
	 * stacking the wakee on top of the waker if no other CPU is
	 * idle.
	 */
	if (sync)
		prev_eff_load += 1;

	return this_eff_load < prev_eff_load ? this_cpu : nr_cpumask_bits;
5561 5562
}

5563
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
5564
		       int this_cpu, int prev_cpu, int sync)
5565
{
5566
	int target = nr_cpumask_bits;
5567

5568
	if (sched_feat(WA_IDLE))
5569
		target = wake_affine_idle(this_cpu, prev_cpu, sync);
5570

5571 5572
	if (sched_feat(WA_WEIGHT) && target == nr_cpumask_bits)
		target = wake_affine_weight(sd, p, this_cpu, prev_cpu, sync);
5573

5574
	schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts);
5575 5576
	if (target == nr_cpumask_bits)
		return prev_cpu;
5577

5578 5579 5580
	schedstat_inc(sd->ttwu_move_affine);
	schedstat_inc(p->se.statistics.nr_wakeups_affine);
	return target;
5581 5582
}

5583
static unsigned long cpu_util_wake(int cpu, struct task_struct *p);
5584 5585 5586

static unsigned long capacity_spare_wake(int cpu, struct task_struct *p)
{
5587
	return max_t(long, capacity_of(cpu) - cpu_util_wake(cpu, p), 0);
5588 5589
}

5590 5591 5592
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
5593 5594
 *
 * Assumes p is allowed on at least one CPU in sd.
5595 5596
 */
static struct sched_group *
P
Peter Zijlstra 已提交
5597
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
5598
		  int this_cpu, int sd_flag)
5599
{
5600
	struct sched_group *idlest = NULL, *group = sd->groups;
5601
	struct sched_group *most_spare_sg = NULL;
5602 5603 5604
	unsigned long min_runnable_load = ULONG_MAX;
	unsigned long this_runnable_load = ULONG_MAX;
	unsigned long min_avg_load = ULONG_MAX, this_avg_load = ULONG_MAX;
5605
	unsigned long most_spare = 0, this_spare = 0;
5606
	int load_idx = sd->forkexec_idx;
5607 5608 5609
	int imbalance_scale = 100 + (sd->imbalance_pct-100)/2;
	unsigned long imbalance = scale_load_down(NICE_0_LOAD) *
				(sd->imbalance_pct-100) / 100;
5610

5611 5612 5613
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

5614
	do {
5615 5616
		unsigned long load, avg_load, runnable_load;
		unsigned long spare_cap, max_spare_cap;
5617 5618
		int local_group;
		int i;
5619

5620
		/* Skip over this group if it has no CPUs allowed */
5621
		if (!cpumask_intersects(sched_group_span(group),
5622
					&p->cpus_allowed))
5623 5624 5625
			continue;

		local_group = cpumask_test_cpu(this_cpu,
5626
					       sched_group_span(group));
5627

5628 5629 5630 5631
		/*
		 * Tally up the load of all CPUs in the group and find
		 * the group containing the CPU with most spare capacity.
		 */
5632
		avg_load = 0;
5633
		runnable_load = 0;
5634
		max_spare_cap = 0;
5635

5636
		for_each_cpu(i, sched_group_span(group)) {
5637
			/* Bias balancing toward CPUs of our domain */
5638 5639 5640 5641 5642
			if (local_group)
				load = source_load(i, load_idx);
			else
				load = target_load(i, load_idx);

5643 5644 5645
			runnable_load += load;

			avg_load += cfs_rq_load_avg(&cpu_rq(i)->cfs);
5646 5647 5648 5649 5650

			spare_cap = capacity_spare_wake(i, p);

			if (spare_cap > max_spare_cap)
				max_spare_cap = spare_cap;
5651 5652
		}

5653
		/* Adjust by relative CPU capacity of the group */
5654 5655 5656 5657
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
		runnable_load = (runnable_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
5658 5659

		if (local_group) {
5660 5661
			this_runnable_load = runnable_load;
			this_avg_load = avg_load;
5662 5663
			this_spare = max_spare_cap;
		} else {
5664 5665 5666
			if (min_runnable_load > (runnable_load + imbalance)) {
				/*
				 * The runnable load is significantly smaller
5667
				 * so we can pick this new CPU:
5668 5669 5670 5671 5672 5673 5674 5675
				 */
				min_runnable_load = runnable_load;
				min_avg_load = avg_load;
				idlest = group;
			} else if ((runnable_load < (min_runnable_load + imbalance)) &&
				   (100*min_avg_load > imbalance_scale*avg_load)) {
				/*
				 * The runnable loads are close so take the
5676
				 * blocked load into account through avg_load:
5677 5678
				 */
				min_avg_load = avg_load;
5679 5680 5681 5682 5683 5684 5685
				idlest = group;
			}

			if (most_spare < max_spare_cap) {
				most_spare = max_spare_cap;
				most_spare_sg = group;
			}
5686 5687 5688
		}
	} while (group = group->next, group != sd->groups);

5689 5690 5691 5692 5693 5694
	/*
	 * 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.
5695 5696 5697 5698
	 *
	 * Spare capacity can't be used for fork because the utilization has
	 * not been set yet, we must first select a rq to compute the initial
	 * utilization.
5699
	 */
5700 5701 5702
	if (sd_flag & SD_BALANCE_FORK)
		goto skip_spare;

5703
	if (this_spare > task_util(p) / 2 &&
5704
	    imbalance_scale*this_spare > 100*most_spare)
5705
		return NULL;
5706 5707

	if (most_spare > task_util(p) / 2)
5708 5709
		return most_spare_sg;

5710
skip_spare:
5711 5712 5713
	if (!idlest)
		return NULL;

5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725
	/*
	 * When comparing groups across NUMA domains, it's possible for the
	 * local domain to be very lightly loaded relative to the remote
	 * domains but "imbalance" skews the comparison making remote CPUs
	 * look much more favourable. When considering cross-domain, add
	 * imbalance to the runnable load on the remote node and consider
	 * staying local.
	 */
	if ((sd->flags & SD_NUMA) &&
	    min_runnable_load + imbalance >= this_runnable_load)
		return NULL;

5726
	if (min_runnable_load > (this_runnable_load + imbalance))
5727
		return NULL;
5728 5729 5730 5731 5732

	if ((this_runnable_load < (min_runnable_load + imbalance)) &&
	     (100*this_avg_load < imbalance_scale*min_avg_load))
		return NULL;

5733 5734 5735 5736
	return idlest;
}

/*
5737
 * find_idlest_group_cpu - find the idlest CPU among the CPUs in the group.
5738 5739
 */
static int
5740
find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
5741 5742
{
	unsigned long load, min_load = ULONG_MAX;
5743 5744 5745 5746
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
5747 5748
	int i;

5749 5750
	/* Check if we have any choice: */
	if (group->group_weight == 1)
5751
		return cpumask_first(sched_group_span(group));
5752

5753
	/* Traverse only the allowed CPUs */
5754
	for_each_cpu_and(i, sched_group_span(group), &p->cpus_allowed) {
5755
		if (available_idle_cpu(i)) {
5756 5757 5758 5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776
			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;
			}
5777
		} else if (shallowest_idle_cpu == -1) {
5778
			load = weighted_cpuload(cpu_rq(i));
5779
			if (load < min_load) {
5780 5781 5782
				min_load = load;
				least_loaded_cpu = i;
			}
5783 5784 5785
		}
	}

5786
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
5787
}
5788

5789 5790 5791
static inline int find_idlest_cpu(struct sched_domain *sd, struct task_struct *p,
				  int cpu, int prev_cpu, int sd_flag)
{
5792
	int new_cpu = cpu;
5793

5794 5795 5796
	if (!cpumask_intersects(sched_domain_span(sd), &p->cpus_allowed))
		return prev_cpu;

5797 5798 5799 5800 5801 5802 5803
	/*
	 * We need task's util for capacity_spare_wake, sync it up to prev_cpu's
	 * last_update_time.
	 */
	if (!(sd_flag & SD_BALANCE_FORK))
		sync_entity_load_avg(&p->se);

5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820
	while (sd) {
		struct sched_group *group;
		struct sched_domain *tmp;
		int weight;

		if (!(sd->flags & sd_flag)) {
			sd = sd->child;
			continue;
		}

		group = find_idlest_group(sd, p, cpu, sd_flag);
		if (!group) {
			sd = sd->child;
			continue;
		}

		new_cpu = find_idlest_group_cpu(group, p, cpu);
5821
		if (new_cpu == cpu) {
5822
			/* Now try balancing at a lower domain level of 'cpu': */
5823 5824 5825 5826
			sd = sd->child;
			continue;
		}

5827
		/* Now try balancing at a lower domain level of 'new_cpu': */
5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840 5841
		cpu = new_cpu;
		weight = sd->span_weight;
		sd = NULL;
		for_each_domain(cpu, tmp) {
			if (weight <= tmp->span_weight)
				break;
			if (tmp->flags & sd_flag)
				sd = tmp;
		}
	}

	return new_cpu;
}

5842
#ifdef CONFIG_SCHED_SMT
5843
DEFINE_STATIC_KEY_FALSE(sched_smt_present);
5844 5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871

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
5872
void __update_idle_core(struct rq *rq)
5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884
{
	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;

5885
		if (!available_idle_cpu(cpu))
5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901
			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);
5902
	int core, cpu;
5903

P
Peter Zijlstra 已提交
5904 5905 5906
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5907 5908 5909
	if (!test_idle_cores(target, false))
		return -1;

5910
	cpumask_and(cpus, sched_domain_span(sd), &p->cpus_allowed);
5911

5912
	for_each_cpu_wrap(core, cpus, target) {
5913 5914 5915 5916
		bool idle = true;

		for_each_cpu(cpu, cpu_smt_mask(core)) {
			cpumask_clear_cpu(cpu, cpus);
5917
			if (!available_idle_cpu(cpu))
5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939
				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
5940 5941 5942
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5943
	for_each_cpu(cpu, cpu_smt_mask(target)) {
5944
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
5945
			continue;
5946
		if (available_idle_cpu(cpu))
5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970
			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).
5971
 */
5972 5973
static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int target)
{
5974
	struct sched_domain *this_sd;
5975
	u64 avg_cost, avg_idle;
5976 5977
	u64 time, cost;
	s64 delta;
5978
	int cpu, nr = INT_MAX;
5979

5980 5981 5982 5983
	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
	if (!this_sd)
		return -1;

5984 5985 5986 5987
	/*
	 * Due to large variance we need a large fuzz factor; hackbench in
	 * particularly is sensitive here.
	 */
5988 5989 5990 5991
	avg_idle = this_rq()->avg_idle / 512;
	avg_cost = this_sd->avg_scan_cost + 1;

	if (sched_feat(SIS_AVG_CPU) && avg_idle < avg_cost)
5992 5993
		return -1;

5994 5995 5996 5997 5998 5999 6000 6001
	if (sched_feat(SIS_PROP)) {
		u64 span_avg = sd->span_weight * avg_idle;
		if (span_avg > 4*avg_cost)
			nr = div_u64(span_avg, avg_cost);
		else
			nr = 4;
	}

6002 6003
	time = local_clock();

6004
	for_each_cpu_wrap(cpu, sched_domain_span(sd), target) {
6005 6006
		if (!--nr)
			return -1;
6007
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6008
			continue;
6009
		if (available_idle_cpu(cpu))
6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022
			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.
6023
 */
6024
static int select_idle_sibling(struct task_struct *p, int prev, int target)
6025
{
6026
	struct sched_domain *sd;
6027
	int i, recent_used_cpu;
6028

6029
	if (available_idle_cpu(target))
6030
		return target;
6031 6032

	/*
6033
	 * If the previous CPU is cache affine and idle, don't be stupid:
6034
	 */
6035
	if (prev != target && cpus_share_cache(prev, target) && available_idle_cpu(prev))
6036
		return prev;
6037

6038
	/* Check a recently used CPU as a potential idle candidate: */
6039 6040 6041 6042
	recent_used_cpu = p->recent_used_cpu;
	if (recent_used_cpu != prev &&
	    recent_used_cpu != target &&
	    cpus_share_cache(recent_used_cpu, target) &&
6043
	    available_idle_cpu(recent_used_cpu) &&
6044 6045 6046
	    cpumask_test_cpu(p->recent_used_cpu, &p->cpus_allowed)) {
		/*
		 * Replace recent_used_cpu with prev as it is a potential
6047
		 * candidate for the next wake:
6048 6049 6050 6051 6052
		 */
		p->recent_used_cpu = prev;
		return recent_used_cpu;
	}

6053
	sd = rcu_dereference(per_cpu(sd_llc, target));
6054 6055
	if (!sd)
		return target;
6056

6057 6058 6059
	i = select_idle_core(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
6060

6061 6062 6063 6064 6065 6066 6067
	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;
6068

6069 6070
	return target;
}
6071

6072 6073 6074 6075 6076 6077 6078
/**
 * Amount of capacity of a CPU that is (estimated to be) used by CFS tasks
 * @cpu: the CPU to get the utilization of
 *
 * The unit of the return value must be the one of capacity so we can compare
 * the utilization with the capacity of the CPU that is available for CFS task
 * (ie cpu_capacity).
6079 6080 6081 6082 6083 6084 6085 6086 6087 6088
 *
 * 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.
 *
6089 6090 6091 6092 6093 6094 6095 6096
 * The estimated utilization of a CPU is defined to be the maximum between its
 * cfs_rq.avg.util_avg and the sum of the estimated utilization of the tasks
 * currently RUNNABLE on that CPU.
 * This allows to properly represent the expected utilization of a CPU which
 * has just got a big task running since a long sleep period. At the same time
 * however it preserves the benefits of the "blocked utilization" in
 * describing the potential for other tasks waking up on the same CPU.
 *
6097 6098 6099 6100 6101 6102 6103 6104 6105 6106
 * 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).
6107 6108
 *
 * Return: the (estimated) utilization for the specified CPU
6109
 */
6110
static inline unsigned long cpu_util(int cpu)
6111
{
6112 6113 6114 6115 6116 6117 6118 6119
	struct cfs_rq *cfs_rq;
	unsigned int util;

	cfs_rq = &cpu_rq(cpu)->cfs;
	util = READ_ONCE(cfs_rq->avg.util_avg);

	if (sched_feat(UTIL_EST))
		util = max(util, READ_ONCE(cfs_rq->avg.util_est.enqueued));
6120

6121
	return min_t(unsigned long, util, capacity_orig_of(cpu));
6122
}
6123

6124
/*
6125
 * cpu_util_wake: Compute CPU utilization with any contributions from
6126 6127
 * the waking task p removed.
 */
6128
static unsigned long cpu_util_wake(int cpu, struct task_struct *p)
6129
{
6130 6131
	struct cfs_rq *cfs_rq;
	unsigned int util;
6132 6133

	/* Task has no contribution or is new */
6134
	if (cpu != task_cpu(p) || !READ_ONCE(p->se.avg.last_update_time))
6135 6136
		return cpu_util(cpu);

6137 6138 6139 6140 6141
	cfs_rq = &cpu_rq(cpu)->cfs;
	util = READ_ONCE(cfs_rq->avg.util_avg);

	/* Discount task's blocked util from CPU's util */
	util -= min_t(unsigned int, util, task_util(p));
6142

6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177
	/*
	 * Covered cases:
	 *
	 * a) if *p is the only task sleeping on this CPU, then:
	 *      cpu_util (== task_util) > util_est (== 0)
	 *    and thus we return:
	 *      cpu_util_wake = (cpu_util - task_util) = 0
	 *
	 * b) if other tasks are SLEEPING on this CPU, which is now exiting
	 *    IDLE, then:
	 *      cpu_util >= task_util
	 *      cpu_util > util_est (== 0)
	 *    and thus we discount *p's blocked utilization to return:
	 *      cpu_util_wake = (cpu_util - task_util) >= 0
	 *
	 * c) if other tasks are RUNNABLE on that CPU and
	 *      util_est > cpu_util
	 *    then we use util_est since it returns a more restrictive
	 *    estimation of the spare capacity on that CPU, by just
	 *    considering the expected utilization of tasks already
	 *    runnable on that CPU.
	 *
	 * Cases a) and b) are covered by the above code, while case c) is
	 * covered by the following code when estimated utilization is
	 * enabled.
	 */
	if (sched_feat(UTIL_EST))
		util = max(util, READ_ONCE(cfs_rq->avg.util_est.enqueued));

	/*
	 * Utilization (estimated) can exceed the CPU capacity, thus let's
	 * clamp to the maximum CPU capacity to ensure consistency with
	 * the cpu_util call.
	 */
	return min_t(unsigned long, util, capacity_orig_of(cpu));
6178 6179
}

6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197
/*
 * 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;

6198 6199 6200
	/* Bring task utilization in sync with prev_cpu */
	sync_entity_load_avg(&p->se);

6201 6202 6203
	return min_cap * 1024 < task_util(p) * capacity_margin;
}

6204
/*
6205 6206 6207
 * 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.
6208
 *
6209 6210
 * 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.
6211
 *
6212
 * Returns the target CPU number.
6213 6214 6215
 *
 * preempt must be disabled.
 */
6216
static int
6217
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
6218
{
6219
	struct sched_domain *tmp, *sd = NULL;
6220
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
6221
	int new_cpu = prev_cpu;
6222
	int want_affine = 0;
6223
	int sync = (wake_flags & WF_SYNC) && !(current->flags & PF_EXITING);
6224

P
Peter Zijlstra 已提交
6225 6226
	if (sd_flag & SD_BALANCE_WAKE) {
		record_wakee(p);
6227
		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu)
6228
			      && cpumask_test_cpu(cpu, &p->cpus_allowed);
P
Peter Zijlstra 已提交
6229
	}
6230

6231
	rcu_read_lock();
6232
	for_each_domain(cpu, tmp) {
6233
		if (!(tmp->flags & SD_LOAD_BALANCE))
M
Mike Galbraith 已提交
6234
			break;
6235

6236
		/*
6237
		 * If both 'cpu' and 'prev_cpu' are part of this domain,
6238
		 * cpu is a valid SD_WAKE_AFFINE target.
6239
		 */
6240 6241
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
6242 6243 6244 6245
			if (cpu != prev_cpu)
				new_cpu = wake_affine(tmp, p, cpu, prev_cpu, sync);

			sd = NULL; /* Prefer wake_affine over balance flags */
6246
			break;
6247
		}
6248

6249
		if (tmp->flags & sd_flag)
6250
			sd = tmp;
M
Mike Galbraith 已提交
6251 6252
		else if (!want_affine)
			break;
6253 6254
	}

6255 6256
	if (unlikely(sd)) {
		/* Slow path */
6257
		new_cpu = find_idlest_cpu(sd, p, cpu, prev_cpu, sd_flag);
6258 6259 6260 6261 6262 6263 6264
	} else if (sd_flag & SD_BALANCE_WAKE) { /* XXX always ? */
		/* Fast path */

		new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);

		if (want_affine)
			current->recent_used_cpu = cpu;
6265
	}
6266
	rcu_read_unlock();
6267

6268
	return new_cpu;
6269
}
6270

6271 6272
static void detach_entity_cfs_rq(struct sched_entity *se);

6273
/*
6274
 * Called immediately before a task is migrated to a new CPU; task_cpu(p) and
6275
 * cfs_rq_of(p) references at time of call are still valid and identify the
6276
 * previous CPU. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
6277
 */
6278
static void migrate_task_rq_fair(struct task_struct *p)
6279
{
6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305
	/*
	 * 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;
	}

6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324
	if (p->on_rq == TASK_ON_RQ_MIGRATING) {
		/*
		 * In case of TASK_ON_RQ_MIGRATING we in fact hold the 'old'
		 * rq->lock and can modify state directly.
		 */
		lockdep_assert_held(&task_rq(p)->lock);
		detach_entity_cfs_rq(&p->se);

	} else {
		/*
		 * 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.
		 */
		remove_entity_load_avg(&p->se);
	}
6325 6326 6327

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
6328 6329

	/* We have migrated, no longer consider this task hot */
6330
	p->se.exec_start = 0;
6331
}
6332 6333 6334 6335 6336

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
6337 6338
#endif /* CONFIG_SMP */

6339
static unsigned long wakeup_gran(struct sched_entity *se)
6340 6341 6342 6343
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
6344 6345
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
6346 6347 6348 6349 6350 6351 6352 6353 6354
	 *
	 * 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.
6355
	 */
6356
	return calc_delta_fair(gran, se);
6357 6358
}

6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380
/*
 * 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;

6381
	gran = wakeup_gran(se);
6382 6383 6384 6385 6386 6387
	if (vdiff > gran)
		return 1;

	return 0;
}

6388 6389
static void set_last_buddy(struct sched_entity *se)
{
6390 6391 6392
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6393 6394 6395
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6396
		cfs_rq_of(se)->last = se;
6397
	}
6398 6399 6400 6401
}

static void set_next_buddy(struct sched_entity *se)
{
6402 6403 6404
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6405 6406 6407
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6408
		cfs_rq_of(se)->next = se;
6409
	}
6410 6411
}

6412 6413
static void set_skip_buddy(struct sched_entity *se)
{
6414 6415
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
6416 6417
}

6418 6419 6420
/*
 * Preempt the current task with a newly woken task if needed:
 */
6421
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
6422 6423
{
	struct task_struct *curr = rq->curr;
6424
	struct sched_entity *se = &curr->se, *pse = &p->se;
6425
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
6426
	int scale = cfs_rq->nr_running >= sched_nr_latency;
6427
	int next_buddy_marked = 0;
6428

I
Ingo Molnar 已提交
6429 6430 6431
	if (unlikely(se == pse))
		return;

6432
	/*
6433
	 * This is possible from callers such as attach_tasks(), in which we
6434 6435 6436 6437 6438 6439 6440
	 * 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;

6441
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
6442
		set_next_buddy(pse);
6443 6444
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
6445

6446 6447 6448
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
6449 6450 6451 6452 6453 6454
	 *
	 * 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.
6455 6456 6457 6458
	 */
	if (test_tsk_need_resched(curr))
		return;

6459 6460 6461 6462 6463
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

6464
	/*
6465 6466
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
6467
	 */
6468
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
6469
		return;
6470

6471
	find_matching_se(&se, &pse);
6472
	update_curr(cfs_rq_of(se));
6473
	BUG_ON(!pse);
6474 6475 6476 6477 6478 6479 6480
	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);
6481
		goto preempt;
6482
	}
6483

6484
	return;
6485

6486
preempt:
6487
	resched_curr(rq);
6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501
	/*
	 * 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);
6502 6503
}

6504
static struct task_struct *
6505
pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
6506 6507 6508
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
6509
	struct task_struct *p;
6510
	int new_tasks;
6511

6512
again:
6513
	if (!cfs_rq->nr_running)
6514
		goto idle;
6515

6516
#ifdef CONFIG_FAIR_GROUP_SCHED
6517
	if (prev->sched_class != &fair_sched_class)
6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535 6536
		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.
		 */
6537 6538 6539 6540 6541
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
6542

6543 6544 6545
			/*
			 * This call to check_cfs_rq_runtime() will do the
			 * throttle and dequeue its entity in the parent(s).
6546
			 * Therefore the nr_running test will indeed
6547 6548
			 * be correct.
			 */
6549 6550 6551 6552 6553 6554
			if (unlikely(check_cfs_rq_runtime(cfs_rq))) {
				cfs_rq = &rq->cfs;

				if (!cfs_rq->nr_running)
					goto idle;

6555
				goto simple;
6556
			}
6557
		}
6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590

		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);
	}

6591
	goto done;
6592 6593
simple:
#endif
6594

6595
	put_prev_task(rq, prev);
6596

6597
	do {
6598
		se = pick_next_entity(cfs_rq, NULL);
6599
		set_next_entity(cfs_rq, se);
6600 6601 6602
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
6603
	p = task_of(se);
6604

6605
done: __maybe_unused;
6606 6607 6608 6609 6610 6611 6612 6613 6614
#ifdef CONFIG_SMP
	/*
	 * Move the next running task to the front of
	 * the list, so our cfs_tasks list becomes MRU
	 * one.
	 */
	list_move(&p->se.group_node, &rq->cfs_tasks);
#endif

6615 6616
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
6617 6618

	return p;
6619 6620

idle:
6621 6622
	new_tasks = idle_balance(rq, rf);

6623 6624 6625 6626 6627
	/*
	 * 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.
	 */
6628
	if (new_tasks < 0)
6629 6630
		return RETRY_TASK;

6631
	if (new_tasks > 0)
6632 6633 6634
		goto again;

	return NULL;
6635 6636 6637 6638 6639
}

/*
 * Account for a descheduled task:
 */
6640
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
6641 6642 6643 6644 6645 6646
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
6647
		put_prev_entity(cfs_rq, se);
6648 6649 6650
	}
}

6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675
/*
 * 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);
6676 6677 6678 6679 6680
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
6681
		rq_clock_skip_update(rq);
6682 6683 6684 6685 6686
	}

	set_skip_buddy(se);
}

6687 6688 6689 6690
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

6691 6692
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
6693 6694 6695 6696 6697 6698 6699 6700 6701 6702
		return false;

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

	yield_task_fair(rq);

	return true;
}

6703
#ifdef CONFIG_SMP
6704
/**************************************************
P
Peter Zijlstra 已提交
6705 6706 6707 6708 6709
 * Fair scheduling class load-balancing methods.
 *
 * BASICS
 *
 * The purpose of load-balancing is to achieve the same basic fairness the
6710
 * per-CPU scheduler provides, namely provide a proportional amount of compute
P
Peter Zijlstra 已提交
6711 6712 6713 6714
 * 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)
 *
6715
 * Where W_i,n is the n-th weight average for CPU i. The instantaneous weight
P
Peter Zijlstra 已提交
6716 6717 6718 6719
 * W_i,0 is defined as:
 *
 *   W_i,0 = \Sum_j w_i,j                                             (2)
 *
6720
 * Where w_i,j is the weight of the j-th runnable task on CPU i. This weight
6721
 * is derived from the nice value as per sched_prio_to_weight[].
P
Peter Zijlstra 已提交
6722 6723 6724 6725 6726 6727
 *
 * 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)
 *
6728
 * C_i is the compute capacity of CPU i, typically it is the
P
Peter Zijlstra 已提交
6729 6730 6731 6732 6733 6734
 * 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):
 *
6735
 *   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
6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748
 *
 * 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)
6749
 * for all i,j solution, we create a tree of CPUs that follows the hardware
P
Peter Zijlstra 已提交
6750
 * topology where each level pairs two lower groups (or better). This results
6751
 * in O(log n) layers. Furthermore we reduce the number of CPUs going up the
P
Peter Zijlstra 已提交
6752
 * tree to only the first of the previous level and we decrease the frequency
6753
 * of load-balance at each level inv. proportional to the number of CPUs in
P
Peter Zijlstra 已提交
6754 6755 6756 6757 6758 6759 6760 6761
 * 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
6762
 *         |         |     `- number of CPUs doing load-balance
P
Peter Zijlstra 已提交
6763 6764 6765 6766 6767 6768 6769
 *         |         `- 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
6770
 * to every other CPU in at most O(log n) steps:
P
Peter Zijlstra 已提交
6771 6772 6773
 *
 * The adjacency matrix of the resulting graph is given by:
 *
6774
 *             log_2 n
P
Peter Zijlstra 已提交
6775 6776 6777 6778 6779 6780 6781
 *   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)
 *
6782
 * Showing there's indeed a path between every CPU in at most O(log n) steps.
P
Peter Zijlstra 已提交
6783 6784 6785 6786 6787 6788 6789 6790 6791
 * 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
6792
 * balancing is more aggressive and has the newly idle CPU iterate up the domain
P
Peter Zijlstra 已提交
6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812
 * 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)
 *
6813
 * w_i,j,k is the weight of the j-th runnable task in the k-th cgroup on CPU i.
P
Peter Zijlstra 已提交
6814 6815 6816 6817 6818 6819
 *
 * 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.]
6820
 */
6821

6822 6823
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

6824 6825
enum fbq_type { regular, remote, all };

6826
#define LBF_ALL_PINNED	0x01
6827
#define LBF_NEED_BREAK	0x02
6828 6829
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
6830
#define LBF_NOHZ_STATS	0x10
6831
#define LBF_NOHZ_AGAIN	0x20
6832 6833 6834 6835 6836

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
6837
	int			src_cpu;
6838 6839 6840 6841

	int			dst_cpu;
	struct rq		*dst_rq;

6842 6843
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
6844
	enum cpu_idle_type	idle;
6845
	long			imbalance;
6846 6847 6848
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

6849
	unsigned int		flags;
6850 6851 6852 6853

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
6854 6855

	enum fbq_type		fbq_type;
6856
	struct list_head	tasks;
6857 6858
};

6859 6860 6861
/*
 * Is this task likely cache-hot:
 */
6862
static int task_hot(struct task_struct *p, struct lb_env *env)
6863 6864 6865
{
	s64 delta;

6866 6867
	lockdep_assert_held(&env->src_rq->lock);

6868 6869 6870 6871 6872 6873 6874 6875 6876
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
6877
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
6878 6879 6880 6881 6882 6883 6884 6885 6886
			(&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;

6887
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
6888 6889 6890 6891

	return delta < (s64)sysctl_sched_migration_cost;
}

6892
#ifdef CONFIG_NUMA_BALANCING
6893
/*
6894 6895 6896
 * 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.
6897
 */
6898
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
6899
{
6900
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
6901 6902
	unsigned long src_weight, dst_weight;
	int src_nid, dst_nid, dist;
6903

6904
	if (!static_branch_likely(&sched_numa_balancing))
6905 6906
		return -1;

6907
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
6908
		return -1;
6909 6910 6911 6912

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

6913
	if (src_nid == dst_nid)
6914
		return -1;
6915

6916 6917 6918 6919 6920 6921 6922
	/* 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;
	}
6923

6924 6925
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
6926
		return 0;
6927

6928
	/* Leaving a core idle is often worse than degrading locality. */
6929
	if (env->idle == CPU_IDLE)
6930 6931
		return -1;

6932
	dist = node_distance(src_nid, dst_nid);
6933
	if (numa_group) {
6934 6935
		src_weight = group_weight(p, src_nid, dist);
		dst_weight = group_weight(p, dst_nid, dist);
6936
	} else {
6937 6938
		src_weight = task_weight(p, src_nid, dist);
		dst_weight = task_weight(p, dst_nid, dist);
6939 6940
	}

6941
	return dst_weight < src_weight;
6942 6943
}

6944
#else
6945
static inline int migrate_degrades_locality(struct task_struct *p,
6946 6947
					     struct lb_env *env)
{
6948
	return -1;
6949
}
6950 6951
#endif

6952 6953 6954 6955
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
6956
int can_migrate_task(struct task_struct *p, struct lb_env *env)
6957
{
6958
	int tsk_cache_hot;
6959 6960 6961

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

6962 6963
	/*
	 * We do not migrate tasks that are:
6964
	 * 1) throttled_lb_pair, or
6965
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
6966 6967
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
6968
	 */
6969 6970 6971
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

6972
	if (!cpumask_test_cpu(env->dst_cpu, &p->cpus_allowed)) {
6973
		int cpu;
6974

6975
		schedstat_inc(p->se.statistics.nr_failed_migrations_affine);
6976

6977 6978
		env->flags |= LBF_SOME_PINNED;

6979
		/*
6980
		 * Remember if this task can be migrated to any other CPU in
6981 6982 6983
		 * our sched_group. We may want to revisit it if we couldn't
		 * meet load balance goals by pulling other tasks on src_cpu.
		 *
6984 6985
		 * Avoid computing new_dst_cpu for NEWLY_IDLE or if we have
		 * already computed one in current iteration.
6986
		 */
6987
		if (env->idle == CPU_NEWLY_IDLE || (env->flags & LBF_DST_PINNED))
6988 6989
			return 0;

6990
		/* Prevent to re-select dst_cpu via env's CPUs: */
6991
		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
6992
			if (cpumask_test_cpu(cpu, &p->cpus_allowed)) {
6993
				env->flags |= LBF_DST_PINNED;
6994 6995 6996
				env->new_dst_cpu = cpu;
				break;
			}
6997
		}
6998

6999 7000
		return 0;
	}
7001 7002

	/* Record that we found atleast one task that could run on dst_cpu */
7003
	env->flags &= ~LBF_ALL_PINNED;
7004

7005
	if (task_running(env->src_rq, p)) {
7006
		schedstat_inc(p->se.statistics.nr_failed_migrations_running);
7007 7008 7009 7010 7011
		return 0;
	}

	/*
	 * Aggressive migration if:
7012 7013 7014
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
7015
	 */
7016 7017 7018
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
7019

7020
	if (tsk_cache_hot <= 0 ||
7021
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
7022
		if (tsk_cache_hot == 1) {
7023 7024
			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
			schedstat_inc(p->se.statistics.nr_forced_migrations);
7025
		}
7026 7027 7028
		return 1;
	}

7029
	schedstat_inc(p->se.statistics.nr_failed_migrations_hot);
Z
Zhang Hang 已提交
7030
	return 0;
7031 7032
}

7033
/*
7034 7035 7036 7037 7038 7039 7040
 * 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;
7041
	deactivate_task(env->src_rq, p, DEQUEUE_NOCLOCK);
7042 7043 7044
	set_task_cpu(p, env->dst_cpu);
}

7045
/*
7046
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
7047 7048
 * part of active balancing operations within "domain".
 *
7049
 * Returns a task if successful and NULL otherwise.
7050
 */
7051
static struct task_struct *detach_one_task(struct lb_env *env)
7052
{
7053
	struct task_struct *p;
7054

7055 7056
	lockdep_assert_held(&env->src_rq->lock);

7057 7058
	list_for_each_entry_reverse(p,
			&env->src_rq->cfs_tasks, se.group_node) {
7059 7060
		if (!can_migrate_task(p, env))
			continue;
7061

7062
		detach_task(p, env);
7063

7064
		/*
7065
		 * Right now, this is only the second place where
7066
		 * lb_gained[env->idle] is updated (other is detach_tasks)
7067
		 * so we can safely collect stats here rather than
7068
		 * inside detach_tasks().
7069
		 */
7070
		schedstat_inc(env->sd->lb_gained[env->idle]);
7071
		return p;
7072
	}
7073
	return NULL;
7074 7075
}

7076 7077
static const unsigned int sched_nr_migrate_break = 32;

7078
/*
7079 7080
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
7081
 *
7082
 * Returns number of detached tasks if successful and 0 otherwise.
7083
 */
7084
static int detach_tasks(struct lb_env *env)
7085
{
7086 7087
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
7088
	unsigned long load;
7089 7090 7091
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
7092

7093
	if (env->imbalance <= 0)
7094
		return 0;
7095

7096
	while (!list_empty(tasks)) {
7097 7098 7099 7100 7101 7102 7103
		/*
		 * 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;

7104
		p = list_last_entry(tasks, struct task_struct, se.group_node);
7105

7106 7107
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
7108
		if (env->loop > env->loop_max)
7109
			break;
7110 7111

		/* take a breather every nr_migrate tasks */
7112
		if (env->loop > env->loop_break) {
7113
			env->loop_break += sched_nr_migrate_break;
7114
			env->flags |= LBF_NEED_BREAK;
7115
			break;
7116
		}
7117

7118
		if (!can_migrate_task(p, env))
7119 7120 7121
			goto next;

		load = task_h_load(p);
7122

7123
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
7124 7125
			goto next;

7126
		if ((load / 2) > env->imbalance)
7127
			goto next;
7128

7129 7130 7131 7132
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
7133
		env->imbalance -= load;
7134 7135

#ifdef CONFIG_PREEMPT
7136 7137
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
7138
		 * kernels will stop after the first task is detached to minimize
7139 7140
		 * the critical section.
		 */
7141
		if (env->idle == CPU_NEWLY_IDLE)
7142
			break;
7143 7144
#endif

7145 7146 7147 7148
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
7149
		if (env->imbalance <= 0)
7150
			break;
7151 7152 7153

		continue;
next:
7154
		list_move(&p->se.group_node, tasks);
7155
	}
7156

7157
	/*
7158 7159 7160
	 * 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().
7161
	 */
7162
	schedstat_add(env->sd->lb_gained[env->idle], detached);
7163

7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174
	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);
7175
	activate_task(rq, p, ENQUEUE_NOCLOCK);
7176
	p->on_rq = TASK_ON_RQ_QUEUED;
7177 7178 7179 7180 7181 7182 7183 7184 7185
	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)
{
7186 7187 7188
	struct rq_flags rf;

	rq_lock(rq, &rf);
7189
	update_rq_clock(rq);
7190
	attach_task(rq, p);
7191
	rq_unlock(rq, &rf);
7192 7193 7194 7195 7196 7197 7198 7199 7200 7201
}

/*
 * 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;
7202
	struct rq_flags rf;
7203

7204
	rq_lock(env->dst_rq, &rf);
7205
	update_rq_clock(env->dst_rq);
7206 7207 7208 7209

	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
		list_del_init(&p->se.group_node);
7210

7211 7212 7213
		attach_task(env->dst_rq, p);
	}

7214
	rq_unlock(env->dst_rq, &rf);
7215 7216
}

7217 7218 7219 7220 7221 7222 7223 7224 7225 7226 7227
static inline bool cfs_rq_has_blocked(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->avg.load_avg)
		return true;

	if (cfs_rq->avg.util_avg)
		return true;

	return false;
}

7228
static inline bool others_have_blocked(struct rq *rq)
7229 7230 7231 7232
{
	if (READ_ONCE(rq->avg_rt.util_avg))
		return true;

7233 7234 7235
	if (READ_ONCE(rq->avg_dl.util_avg))
		return true;

7236 7237 7238 7239 7240
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
	if (READ_ONCE(rq->avg_irq.util_avg))
		return true;
#endif

7241 7242 7243
	return false;
}

7244 7245
#ifdef CONFIG_FAIR_GROUP_SCHED

7246 7247 7248 7249 7250 7251 7252 7253 7254 7255 7256
static inline bool cfs_rq_is_decayed(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->load.weight)
		return false;

	if (cfs_rq->avg.load_sum)
		return false;

	if (cfs_rq->avg.util_sum)
		return false;

7257
	if (cfs_rq->avg.runnable_load_sum)
7258 7259 7260 7261 7262
		return false;

	return true;
}

7263
static void update_blocked_averages(int cpu)
7264 7265
{
	struct rq *rq = cpu_rq(cpu);
7266
	struct cfs_rq *cfs_rq, *pos;
7267
	const struct sched_class *curr_class;
7268
	struct rq_flags rf;
7269
	bool done = true;
7270

7271
	rq_lock_irqsave(rq, &rf);
7272
	update_rq_clock(rq);
7273

7274 7275 7276 7277
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
7278
	for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos) {
7279 7280
		struct sched_entity *se;

7281 7282 7283
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
7284

7285
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq))
7286
			update_tg_load_avg(cfs_rq, 0);
7287

7288 7289 7290
		/* Propagate pending load changes to the parent, if any: */
		se = cfs_rq->tg->se[cpu];
		if (se && !skip_blocked_update(se))
7291
			update_load_avg(cfs_rq_of(se), se, 0);
7292 7293 7294 7295 7296 7297 7298

		/*
		 * There can be a lot of idle CPU cgroups.  Don't let fully
		 * decayed cfs_rqs linger on the list.
		 */
		if (cfs_rq_is_decayed(cfs_rq))
			list_del_leaf_cfs_rq(cfs_rq);
7299 7300 7301

		/* Don't need periodic decay once load/util_avg are null */
		if (cfs_rq_has_blocked(cfs_rq))
7302
			done = false;
7303
	}
7304 7305 7306 7307

	curr_class = rq->curr->sched_class;
	update_rt_rq_load_avg(rq_clock_task(rq), rq, curr_class == &rt_sched_class);
	update_dl_rq_load_avg(rq_clock_task(rq), rq, curr_class == &dl_sched_class);
7308
	update_irq_load_avg(rq, 0);
7309
	/* Don't need periodic decay once load/util_avg are null */
7310
	if (others_have_blocked(rq))
7311
		done = false;
7312 7313 7314

#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7315 7316
	if (done)
		rq->has_blocked_load = 0;
7317
#endif
7318
	rq_unlock_irqrestore(rq, &rf);
7319 7320
}

7321
/*
7322
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
7323 7324 7325
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
7326
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
7327
{
7328 7329
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
7330
	unsigned long now = jiffies;
7331
	unsigned long load;
7332

7333
	if (cfs_rq->last_h_load_update == now)
7334 7335
		return;

7336 7337 7338 7339 7340 7341 7342
	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;
	}
7343

7344
	if (!se) {
7345
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
7346 7347 7348 7349 7350
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
7351 7352
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
7353 7354 7355 7356
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
7357 7358
}

7359
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
7360
{
7361
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
7362

7363
	update_cfs_rq_h_load(cfs_rq);
7364
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
7365
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
7366 7367
}
#else
7368
static inline void update_blocked_averages(int cpu)
7369
{
7370 7371
	struct rq *rq = cpu_rq(cpu);
	struct cfs_rq *cfs_rq = &rq->cfs;
7372
	const struct sched_class *curr_class;
7373
	struct rq_flags rf;
7374

7375
	rq_lock_irqsave(rq, &rf);
7376
	update_rq_clock(rq);
7377
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq);
7378 7379 7380 7381

	curr_class = rq->curr->sched_class;
	update_rt_rq_load_avg(rq_clock_task(rq), rq, curr_class == &rt_sched_class);
	update_dl_rq_load_avg(rq_clock_task(rq), rq, curr_class == &dl_sched_class);
7382
	update_irq_load_avg(rq, 0);
7383 7384
#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7385
	if (!cfs_rq_has_blocked(cfs_rq) && !others_have_blocked(rq))
7386
		rq->has_blocked_load = 0;
7387
#endif
7388
	rq_unlock_irqrestore(rq, &rf);
7389 7390
}

7391
static unsigned long task_h_load(struct task_struct *p)
7392
{
7393
	return p->se.avg.load_avg;
7394
}
P
Peter Zijlstra 已提交
7395
#endif
7396 7397

/********** Helpers for find_busiest_group ************************/
7398 7399 7400 7401 7402 7403 7404

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

7405 7406 7407 7408 7409 7410 7411
/*
 * 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
7412
	unsigned long load_per_task;
7413
	unsigned long group_capacity;
7414
	unsigned long group_util; /* Total utilization of the group */
7415 7416 7417
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
7418
	enum group_type group_type;
7419
	int group_no_capacity;
7420 7421 7422 7423
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
7424 7425
};

J
Joonsoo Kim 已提交
7426 7427 7428 7429 7430 7431 7432
/*
 * 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 */
7433
	unsigned long total_running;
J
Joonsoo Kim 已提交
7434
	unsigned long total_load;	/* Total load of all groups in sd */
7435
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
7436 7437 7438
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
7439
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
7440 7441
};

7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452
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,
7453
		.total_running = 0UL,
7454
		.total_load = 0UL,
7455
		.total_capacity = 0UL,
7456 7457
		.busiest_stat = {
			.avg_load = 0UL,
7458 7459
			.sum_nr_running = 0,
			.group_type = group_other,
7460 7461 7462 7463
		},
	};
}

7464 7465 7466
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
7467
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
7468 7469
 *
 * Return: The load index.
7470 7471 7472 7473 7474 7475 7476 7477 7478 7479 7480 7481 7482 7483 7484 7485 7486 7487 7488 7489 7490 7491
 */
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;
}

7492
static unsigned long scale_rt_capacity(struct sched_domain *sd, int cpu)
7493 7494
{
	struct rq *rq = cpu_rq(cpu);
7495
	unsigned long max = arch_scale_cpu_capacity(sd, cpu);
7496 7497
	unsigned long used, free;
	unsigned long irq;
7498

7499
	irq = cpu_util_irq(rq);
7500

7501 7502
	if (unlikely(irq >= max))
		return 1;
7503

7504 7505
	used = READ_ONCE(rq->avg_rt.util_avg);
	used += READ_ONCE(rq->avg_dl.util_avg);
7506

7507 7508
	if (unlikely(used >= max))
		return 1;
7509

7510
	free = max - used;
7511 7512

	return scale_irq_capacity(free, irq, max);
7513 7514
}

7515
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
7516
{
7517
	unsigned long capacity = scale_rt_capacity(sd, cpu);
7518 7519
	struct sched_group *sdg = sd->groups;

7520
	cpu_rq(cpu)->cpu_capacity_orig = arch_scale_cpu_capacity(sd, cpu);
7521

7522 7523
	if (!capacity)
		capacity = 1;
7524

7525 7526
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
7527
	sdg->sgc->min_capacity = capacity;
7528 7529
}

7530
void update_group_capacity(struct sched_domain *sd, int cpu)
7531 7532 7533
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
7534
	unsigned long capacity, min_capacity;
7535 7536 7537 7538
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
7539
	sdg->sgc->next_update = jiffies + interval;
7540 7541

	if (!child) {
7542
		update_cpu_capacity(sd, cpu);
7543 7544 7545
		return;
	}

7546
	capacity = 0;
7547
	min_capacity = ULONG_MAX;
7548

P
Peter Zijlstra 已提交
7549 7550 7551 7552 7553 7554
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

7555
		for_each_cpu(cpu, sched_group_span(sdg)) {
7556
			struct sched_group_capacity *sgc;
7557
			struct rq *rq = cpu_rq(cpu);
7558

7559
			/*
7560
			 * build_sched_domains() -> init_sched_groups_capacity()
7561 7562 7563
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
7564 7565
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
7566
			 *
7567
			 * This avoids capacity from being 0 and
7568 7569 7570
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
7571
				capacity += capacity_of(cpu);
7572 7573 7574
			} else {
				sgc = rq->sd->groups->sgc;
				capacity += sgc->capacity;
7575
			}
7576

7577
			min_capacity = min(capacity, min_capacity);
7578
		}
P
Peter Zijlstra 已提交
7579 7580 7581 7582
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
7583
		 */
P
Peter Zijlstra 已提交
7584 7585 7586

		group = child->groups;
		do {
7587 7588 7589 7590
			struct sched_group_capacity *sgc = group->sgc;

			capacity += sgc->capacity;
			min_capacity = min(sgc->min_capacity, min_capacity);
P
Peter Zijlstra 已提交
7591 7592 7593
			group = group->next;
		} while (group != child->groups);
	}
7594

7595
	sdg->sgc->capacity = capacity;
7596
	sdg->sgc->min_capacity = min_capacity;
7597 7598
}

7599
/*
7600 7601 7602
 * 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
7603 7604
 */
static inline int
7605
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
7606
{
7607 7608
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
7609 7610
}

7611 7612
/*
 * Group imbalance indicates (and tries to solve) the problem where balancing
7613
 * groups is inadequate due to ->cpus_allowed constraints.
7614
 *
7615 7616
 * 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.
7617 7618
 * Something like:
 *
7619 7620
 *	{ 0 1 2 3 } { 4 5 6 7 }
 *	        *     * * *
7621 7622 7623
 *
 * 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
7624
 * cpu 3 and leave one of the CPUs in the second group unused.
7625 7626
 *
 * The current solution to this issue is detecting the skew in the first group
7627 7628
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
7629 7630
 *
 * When this is so detected; this group becomes a candidate for busiest; see
7631
 * update_sd_pick_busiest(). And calculate_imbalance() and
7632
 * find_busiest_group() avoid some of the usual balance conditions to allow it
7633 7634 7635 7636 7637 7638 7639
 * 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.
 */

7640
static inline int sg_imbalanced(struct sched_group *group)
7641
{
7642
	return group->sgc->imbalance;
7643 7644
}

7645
/*
7646 7647 7648
 * 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
7649 7650
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
7651 7652 7653 7654 7655
 * 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.
7656
 */
7657 7658
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
7659
{
7660 7661
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
7662

7663
	if ((sgs->group_capacity * 100) >
7664
			(sgs->group_util * env->sd->imbalance_pct))
7665
		return true;
7666

7667 7668 7669 7670 7671 7672 7673 7674 7675 7676 7677 7678 7679 7680 7681 7682
	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;
7683

7684
	if ((sgs->group_capacity * 100) <
7685
			(sgs->group_util * env->sd->imbalance_pct))
7686
		return true;
7687

7688
	return false;
7689 7690
}

7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701
/*
 * 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;
}

7702 7703 7704
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
7705
{
7706
	if (sgs->group_no_capacity)
7707 7708 7709 7710 7711 7712 7713 7714
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

7715
static bool update_nohz_stats(struct rq *rq, bool force)
7716 7717 7718 7719
{
#ifdef CONFIG_NO_HZ_COMMON
	unsigned int cpu = rq->cpu;

7720 7721 7722
	if (!rq->has_blocked_load)
		return false;

7723
	if (!cpumask_test_cpu(cpu, nohz.idle_cpus_mask))
7724
		return false;
7725

7726
	if (!force && !time_after(jiffies, rq->last_blocked_load_update_tick))
7727
		return true;
7728 7729

	update_blocked_averages(cpu);
7730 7731 7732 7733

	return rq->has_blocked_load;
#else
	return false;
7734 7735 7736
#endif
}

7737 7738
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
7739
 * @env: The load balancing environment.
7740 7741 7742 7743
 * @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.
7744
 * @overload: Indicate more than one runnable task for any CPU.
7745
 */
7746 7747
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
7748 7749
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
7750
{
7751
	unsigned long load;
7752
	int i, nr_running;
7753

7754 7755
	memset(sgs, 0, sizeof(*sgs));

7756
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
7757 7758
		struct rq *rq = cpu_rq(i);

7759
		if ((env->flags & LBF_NOHZ_STATS) && update_nohz_stats(rq, false))
7760
			env->flags |= LBF_NOHZ_AGAIN;
7761

7762
		/* Bias balancing toward CPUs of our domain: */
7763
		if (local_group)
7764
			load = target_load(i, load_idx);
7765
		else
7766 7767 7768
			load = source_load(i, load_idx);

		sgs->group_load += load;
7769
		sgs->group_util += cpu_util(i);
7770
		sgs->sum_nr_running += rq->cfs.h_nr_running;
7771

7772 7773
		nr_running = rq->nr_running;
		if (nr_running > 1)
7774 7775
			*overload = true;

7776 7777 7778 7779
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
7780
		sgs->sum_weighted_load += weighted_cpuload(rq);
7781 7782 7783 7784
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
7785
			sgs->idle_cpus++;
7786 7787
	}

7788 7789
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
7790
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
7791

7792
	if (sgs->sum_nr_running)
7793
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
7794

7795
	sgs->group_weight = group->group_weight;
7796

7797
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
7798
	sgs->group_type = group_classify(group, sgs);
7799 7800
}

7801 7802
/**
 * update_sd_pick_busiest - return 1 on busiest group
7803
 * @env: The load balancing environment.
7804 7805
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
7806
 * @sgs: sched_group statistics
7807 7808 7809
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
7810 7811 7812
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
7813
 */
7814
static bool update_sd_pick_busiest(struct lb_env *env,
7815 7816
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
7817
				   struct sg_lb_stats *sgs)
7818
{
7819
	struct sg_lb_stats *busiest = &sds->busiest_stat;
7820

7821
	if (sgs->group_type > busiest->group_type)
7822 7823
		return true;

7824 7825 7826 7827 7828 7829
	if (sgs->group_type < busiest->group_type)
		return false;

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

7830 7831 7832 7833 7834 7835 7836 7837 7838 7839 7840 7841 7842 7843
	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:
7844 7845
	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
7846 7847
		return true;

7848
	/* No ASYM_PACKING if target CPU is already busy */
7849 7850
	if (env->idle == CPU_NOT_IDLE)
		return true;
7851
	/*
T
Tim Chen 已提交
7852 7853 7854
	 * ASYM_PACKING needs to move all the work to the highest
	 * prority CPUs in the group, therefore mark all groups
	 * of lower priority than ourself as busy.
7855
	 */
T
Tim Chen 已提交
7856 7857
	if (sgs->sum_nr_running &&
	    sched_asym_prefer(env->dst_cpu, sg->asym_prefer_cpu)) {
7858 7859 7860
		if (!sds->busiest)
			return true;

7861
		/* Prefer to move from lowest priority CPU's work */
T
Tim Chen 已提交
7862 7863
		if (sched_asym_prefer(sds->busiest->asym_prefer_cpu,
				      sg->asym_prefer_cpu))
7864 7865 7866 7867 7868 7869
			return true;
	}

	return false;
}

7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897 7898 7899
#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 */

7900
/**
7901
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
7902
 * @env: The load balancing environment.
7903 7904
 * @sds: variable to hold the statistics for this sched_domain.
 */
7905
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
7906
{
7907 7908
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
7909
	struct sg_lb_stats *local = &sds->local_stat;
J
Joonsoo Kim 已提交
7910
	struct sg_lb_stats tmp_sgs;
7911
	int load_idx, prefer_sibling = 0;
7912
	bool overload = false;
7913 7914 7915 7916

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

7917
#ifdef CONFIG_NO_HZ_COMMON
7918
	if (env->idle == CPU_NEWLY_IDLE && READ_ONCE(nohz.has_blocked))
7919 7920 7921
		env->flags |= LBF_NOHZ_STATS;
#endif

7922
	load_idx = get_sd_load_idx(env->sd, env->idle);
7923 7924

	do {
J
Joonsoo Kim 已提交
7925
		struct sg_lb_stats *sgs = &tmp_sgs;
7926 7927
		int local_group;

7928
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_span(sg));
J
Joonsoo Kim 已提交
7929 7930
		if (local_group) {
			sds->local = sg;
7931
			sgs = local;
7932 7933

			if (env->idle != CPU_NEWLY_IDLE ||
7934 7935
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
7936
		}
7937

7938 7939
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
7940

7941 7942 7943
		if (local_group)
			goto next_group;

7944 7945
		/*
		 * In case the child domain prefers tasks go to siblings
7946
		 * first, lower the sg capacity so that we'll try
7947 7948
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
7949 7950 7951 7952
		 * 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).
7953
		 */
7954
		if (prefer_sibling && sds->local &&
7955 7956
		    group_has_capacity(env, local) &&
		    (sgs->sum_nr_running > local->sum_nr_running + 1)) {
7957
			sgs->group_no_capacity = 1;
7958
			sgs->group_type = group_classify(sg, sgs);
7959
		}
7960

7961
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
7962
			sds->busiest = sg;
J
Joonsoo Kim 已提交
7963
			sds->busiest_stat = *sgs;
7964 7965
		}

7966 7967
next_group:
		/* Now, start updating sd_lb_stats */
7968
		sds->total_running += sgs->sum_nr_running;
7969
		sds->total_load += sgs->group_load;
7970
		sds->total_capacity += sgs->group_capacity;
7971

7972
		sg = sg->next;
7973
	} while (sg != env->sd->groups);
7974

7975 7976 7977 7978 7979 7980 7981 7982 7983
#ifdef CONFIG_NO_HZ_COMMON
	if ((env->flags & LBF_NOHZ_AGAIN) &&
	    cpumask_subset(nohz.idle_cpus_mask, sched_domain_span(env->sd))) {

		WRITE_ONCE(nohz.next_blocked,
			   jiffies + msecs_to_jiffies(LOAD_AVG_PERIOD));
	}
#endif

7984 7985
	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
7986 7987 7988 7989 7990 7991

	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;
	}
7992 7993 7994 7995
}

/**
 * check_asym_packing - Check to see if the group is packed into the
7996
 *			sched domain.
7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010
 *
 * 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.
 *
8011
 * Return: 1 when packing is required and a task should be moved to
8012
 * this CPU.  The amount of the imbalance is returned in env->imbalance.
8013
 *
8014
 * @env: The load balancing environment.
8015 8016
 * @sds: Statistics of the sched_domain which is to be packed
 */
8017
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
8018 8019 8020
{
	int busiest_cpu;

8021
	if (!(env->sd->flags & SD_ASYM_PACKING))
8022 8023
		return 0;

8024 8025 8026
	if (env->idle == CPU_NOT_IDLE)
		return 0;

8027 8028 8029
	if (!sds->busiest)
		return 0;

T
Tim Chen 已提交
8030 8031
	busiest_cpu = sds->busiest->asym_prefer_cpu;
	if (sched_asym_prefer(busiest_cpu, env->dst_cpu))
8032 8033
		return 0;

8034
	env->imbalance = DIV_ROUND_CLOSEST(
8035
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
8036
		SCHED_CAPACITY_SCALE);
8037

8038
	return 1;
8039 8040 8041 8042 8043 8044
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
8045
 * @env: The load balancing environment.
8046 8047
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
8048 8049
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8050
{
8051
	unsigned long tmp, capa_now = 0, capa_move = 0;
8052
	unsigned int imbn = 2;
8053
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
8054
	struct sg_lb_stats *local, *busiest;
8055

J
Joonsoo Kim 已提交
8056 8057
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8058

J
Joonsoo Kim 已提交
8059 8060 8061 8062
	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;
8063

J
Joonsoo Kim 已提交
8064
	scaled_busy_load_per_task =
8065
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8066
		busiest->group_capacity;
J
Joonsoo Kim 已提交
8067

8068 8069
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
8070
		env->imbalance = busiest->load_per_task;
8071 8072 8073 8074 8075
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
8076
	 * however we may be able to increase total CPU capacity used by
8077 8078 8079
	 * moving them.
	 */

8080
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
8081
			min(busiest->load_per_task, busiest->avg_load);
8082
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
8083
			min(local->load_per_task, local->avg_load);
8084
	capa_now /= SCHED_CAPACITY_SCALE;
8085 8086

	/* Amount of load we'd subtract */
8087
	if (busiest->avg_load > scaled_busy_load_per_task) {
8088
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
8089
			    min(busiest->load_per_task,
8090
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
8091
	}
8092 8093

	/* Amount of load we'd add */
8094
	if (busiest->avg_load * busiest->group_capacity <
8095
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
8096 8097
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
8098
	} else {
8099
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8100
		      local->group_capacity;
J
Joonsoo Kim 已提交
8101
	}
8102
	capa_move += local->group_capacity *
8103
		    min(local->load_per_task, local->avg_load + tmp);
8104
	capa_move /= SCHED_CAPACITY_SCALE;
8105 8106

	/* Move if we gain throughput */
8107
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
8108
		env->imbalance = busiest->load_per_task;
8109 8110 8111 8112 8113
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
8114
 * @env: load balance environment
8115 8116
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
8117
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8118
{
8119
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
8120 8121 8122 8123
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8124

8125
	if (busiest->group_type == group_imbalanced) {
8126 8127
		/*
		 * In the group_imb case we cannot rely on group-wide averages
8128
		 * to ensure CPU-load equilibrium, look at wider averages. XXX
8129
		 */
J
Joonsoo Kim 已提交
8130 8131
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
8132 8133
	}

8134
	/*
8135 8136 8137 8138
	 * 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:
8139
	 */
8140 8141
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
8142 8143
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
8144 8145
	}

8146
	/*
8147
	 * If there aren't any idle CPUs, avoid creating some.
8148 8149 8150
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
8151
		load_above_capacity = busiest->sum_nr_running * SCHED_CAPACITY_SCALE;
8152
		if (load_above_capacity > busiest->group_capacity) {
8153
			load_above_capacity -= busiest->group_capacity;
8154
			load_above_capacity *= scale_load_down(NICE_0_LOAD);
8155 8156
			load_above_capacity /= busiest->group_capacity;
		} else
8157
			load_above_capacity = ~0UL;
8158 8159 8160
	}

	/*
8161
	 * We're trying to get all the CPUs to the average_load, so we don't
8162
	 * want to push ourselves above the average load, nor do we wish to
8163
	 * reduce the max loaded CPU below the average load. At the same time,
8164 8165
	 * we also don't want to reduce the group load below the group
	 * capacity. Thus we look for the minimum possible imbalance.
8166
	 */
8167
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
8168 8169

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
8170
	env->imbalance = min(
8171 8172
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
8173
	) / SCHED_CAPACITY_SCALE;
8174 8175 8176

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
8177
	 * there is no guarantee that any tasks will be moved so we'll have
8178 8179 8180
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
8181
	if (env->imbalance < busiest->load_per_task)
8182
		return fix_small_imbalance(env, sds);
8183
}
8184

8185 8186 8187 8188
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
8189
 * if there is an imbalance.
8190 8191 8192 8193
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
8194
 * @env: The load balancing environment.
8195
 *
8196
 * Return:	- The busiest group if imbalance exists.
8197
 */
J
Joonsoo Kim 已提交
8198
static struct sched_group *find_busiest_group(struct lb_env *env)
8199
{
J
Joonsoo Kim 已提交
8200
	struct sg_lb_stats *local, *busiest;
8201 8202
	struct sd_lb_stats sds;

8203
	init_sd_lb_stats(&sds);
8204 8205 8206 8207 8208

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
8209
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
8210 8211
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
8212

8213
	/* ASYM feature bypasses nice load balance check */
8214
	if (check_asym_packing(env, &sds))
8215 8216
		return sds.busiest;

8217
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
8218
	if (!sds.busiest || busiest->sum_nr_running == 0)
8219 8220
		goto out_balanced;

8221
	/* XXX broken for overlapping NUMA groups */
8222 8223
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
8224

P
Peter Zijlstra 已提交
8225 8226
	/*
	 * If the busiest group is imbalanced the below checks don't
8227
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
8228 8229
	 * isn't true due to cpus_allowed constraints and the like.
	 */
8230
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
8231 8232
		goto force_balance;

8233 8234 8235 8236 8237
	/*
	 * When dst_cpu is idle, prevent SMP nice and/or asymmetric group
	 * capacities from resulting in underutilization due to avg_load.
	 */
	if (env->idle != CPU_NOT_IDLE && group_has_capacity(env, local) &&
8238
	    busiest->group_no_capacity)
8239 8240
		goto force_balance;

8241
	/*
8242
	 * If the local group is busier than the selected busiest group
8243 8244
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
8245
	if (local->avg_load >= busiest->avg_load)
8246 8247
		goto out_balanced;

8248 8249 8250 8251
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
8252
	if (local->avg_load >= sds.avg_load)
8253 8254
		goto out_balanced;

8255
	if (env->idle == CPU_IDLE) {
8256
		/*
8257
		 * This CPU is idle. If the busiest group is not overloaded
8258
		 * and there is no imbalance between this and busiest group
8259
		 * wrt idle CPUs, it is balanced. The imbalance becomes
8260 8261
		 * significant if the diff is greater than 1 otherwise we
		 * might end up to just move the imbalance on another group
8262
		 */
8263 8264
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
8265
			goto out_balanced;
8266 8267 8268 8269 8270
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
8271 8272
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
8273
			goto out_balanced;
8274
	}
8275

8276
force_balance:
8277
	/* Looks like there is an imbalance. Compute it */
8278
	calculate_imbalance(env, &sds);
8279
	return env->imbalance ? sds.busiest : NULL;
8280 8281

out_balanced:
8282
	env->imbalance = 0;
8283 8284 8285 8286
	return NULL;
}

/*
8287
 * find_busiest_queue - find the busiest runqueue among the CPUs in the group.
8288
 */
8289
static struct rq *find_busiest_queue(struct lb_env *env,
8290
				     struct sched_group *group)
8291 8292
{
	struct rq *busiest = NULL, *rq;
8293
	unsigned long busiest_load = 0, busiest_capacity = 1;
8294 8295
	int i;

8296
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
8297
		unsigned long capacity, wl;
8298 8299 8300 8301
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
8302

8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315 8316 8317 8318 8319 8320 8321 8322 8323 8324
		/*
		 * 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;

8325
		capacity = capacity_of(i);
8326

8327
		wl = weighted_cpuload(rq);
8328

8329 8330
		/*
		 * When comparing with imbalance, use weighted_cpuload()
8331
		 * which is not scaled with the CPU capacity.
8332
		 */
8333 8334 8335

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
8336 8337
			continue;

8338
		/*
8339 8340 8341
		 * For the load comparisons with the other CPU's, consider
		 * the weighted_cpuload() scaled with the CPU capacity, so
		 * that the load can be moved away from the CPU that is
8342
		 * potentially running at a lower capacity.
8343
		 *
8344
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
8345
		 * multiplication to rid ourselves of the division works out
8346 8347
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
8348
		 */
8349
		if (wl * busiest_capacity > busiest_load * capacity) {
8350
			busiest_load = wl;
8351
			busiest_capacity = capacity;
8352 8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364
			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

8365
static int need_active_balance(struct lb_env *env)
8366
{
8367 8368 8369
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
8370 8371 8372

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
T
Tim Chen 已提交
8373 8374
		 * lower priority CPUs in order to pack all tasks in the
		 * highest priority CPUs.
8375
		 */
T
Tim Chen 已提交
8376 8377
		if ((sd->flags & SD_ASYM_PACKING) &&
		    sched_asym_prefer(env->dst_cpu, env->src_cpu))
8378
			return 1;
8379 8380
	}

8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393
	/*
	 * 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;
	}

8394 8395 8396
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

8397 8398
static int active_load_balance_cpu_stop(void *data);

8399 8400 8401 8402 8403
static int should_we_balance(struct lb_env *env)
{
	struct sched_group *sg = env->sd->groups;
	int cpu, balance_cpu = -1;

8404 8405 8406 8407 8408 8409 8410
	/*
	 * Ensure the balancing environment is consistent; can happen
	 * when the softirq triggers 'during' hotplug.
	 */
	if (!cpumask_test_cpu(env->dst_cpu, env->cpus))
		return 0;

8411
	/*
8412
	 * In the newly idle case, we will allow all the CPUs
8413 8414 8415 8416 8417
	 * to do the newly idle load balance.
	 */
	if (env->idle == CPU_NEWLY_IDLE)
		return 1;

8418
	/* Try to find first idle CPU */
8419
	for_each_cpu_and(cpu, group_balance_mask(sg), env->cpus) {
8420
		if (!idle_cpu(cpu))
8421 8422 8423 8424 8425 8426 8427 8428 8429 8430
			continue;

		balance_cpu = cpu;
		break;
	}

	if (balance_cpu == -1)
		balance_cpu = group_balance_cpu(sg);

	/*
8431
	 * First idle CPU or the first CPU(busiest) in this sched group
8432 8433
	 * is eligible for doing load balancing at this and above domains.
	 */
8434
	return balance_cpu == env->dst_cpu;
8435 8436
}

8437 8438 8439 8440 8441 8442
/*
 * 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,
8443
			int *continue_balancing)
8444
{
8445
	int ld_moved, cur_ld_moved, active_balance = 0;
8446
	struct sched_domain *sd_parent = sd->parent;
8447 8448
	struct sched_group *group;
	struct rq *busiest;
8449
	struct rq_flags rf;
8450
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
8451

8452 8453
	struct lb_env env = {
		.sd		= sd,
8454 8455
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
8456
		.dst_grpmask    = sched_group_span(sd->groups),
8457
		.idle		= idle,
8458
		.loop_break	= sched_nr_migrate_break,
8459
		.cpus		= cpus,
8460
		.fbq_type	= all,
8461
		.tasks		= LIST_HEAD_INIT(env.tasks),
8462 8463
	};

8464
	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
8465

8466
	schedstat_inc(sd->lb_count[idle]);
8467 8468

redo:
8469 8470
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
8471
		goto out_balanced;
8472
	}
8473

8474
	group = find_busiest_group(&env);
8475
	if (!group) {
8476
		schedstat_inc(sd->lb_nobusyg[idle]);
8477 8478 8479
		goto out_balanced;
	}

8480
	busiest = find_busiest_queue(&env, group);
8481
	if (!busiest) {
8482
		schedstat_inc(sd->lb_nobusyq[idle]);
8483 8484 8485
		goto out_balanced;
	}

8486
	BUG_ON(busiest == env.dst_rq);
8487

8488
	schedstat_add(sd->lb_imbalance[idle], env.imbalance);
8489

8490 8491 8492
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

8493 8494 8495 8496 8497 8498 8499 8500
	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.
		 */
8501
		env.flags |= LBF_ALL_PINNED;
8502
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
8503

8504
more_balance:
8505
		rq_lock_irqsave(busiest, &rf);
8506
		update_rq_clock(busiest);
8507 8508 8509 8510 8511

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
8512
		cur_ld_moved = detach_tasks(&env);
8513 8514

		/*
8515 8516 8517 8518 8519
		 * 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.
8520
		 */
8521

8522
		rq_unlock(busiest, &rf);
8523 8524 8525 8526 8527 8528

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

8529
		local_irq_restore(rf.flags);
8530

8531 8532 8533 8534 8535
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

8536 8537 8538 8539
		/*
		 * 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
8540
		 * iterate on same src_cpu is dependent on number of CPUs in our
8541 8542 8543 8544 8545 8546 8547 8548 8549 8550 8551 8552 8553 8554
		 * 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.
		 */
8555
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
8556

8557
			/* Prevent to re-select dst_cpu via env's CPUs */
8558 8559
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

8560
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
8561
			env.dst_cpu	 = env.new_dst_cpu;
8562
			env.flags	&= ~LBF_DST_PINNED;
8563 8564
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
8565

8566 8567 8568 8569 8570 8571
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
8572

8573 8574 8575 8576
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
8577
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
8578

8579
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
8580 8581 8582
				*group_imbalance = 1;
		}

8583
		/* All tasks on this runqueue were pinned by CPU affinity */
8584
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
8585
			cpumask_clear_cpu(cpu_of(busiest), cpus);
8586 8587 8588 8589 8590 8591 8592 8593 8594
			/*
			 * Attempting to continue load balancing at the current
			 * sched_domain level only makes sense if there are
			 * active CPUs remaining as possible busiest CPUs to
			 * pull load from which are not contained within the
			 * destination group that is receiving any migrated
			 * load.
			 */
			if (!cpumask_subset(cpus, env.dst_grpmask)) {
8595 8596
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
8597
				goto redo;
8598
			}
8599
			goto out_all_pinned;
8600 8601 8602 8603
		}
	}

	if (!ld_moved) {
8604
		schedstat_inc(sd->lb_failed[idle]);
8605 8606 8607 8608 8609 8610 8611 8612
		/*
		 * 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++;
8613

8614
		if (need_active_balance(&env)) {
8615 8616
			unsigned long flags;

8617 8618
			raw_spin_lock_irqsave(&busiest->lock, flags);

8619 8620 8621 8622
			/*
			 * Don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest CPU can't be
			 * moved to this_cpu:
8623
			 */
8624
			if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) {
8625 8626
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
8627
				env.flags |= LBF_ALL_PINNED;
8628 8629 8630
				goto out_one_pinned;
			}

8631 8632 8633 8634 8635
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
8636 8637 8638 8639 8640 8641
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
8642

8643
			if (active_balance) {
8644 8645 8646
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
8647
			}
8648

8649
			/* We've kicked active balancing, force task migration. */
8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662
			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
8663
		 * detach_tasks).
8664 8665 8666 8667 8668 8669 8670 8671
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
8672 8673 8674 8675 8676 8677 8678 8679 8680 8681 8682 8683 8684 8685 8686 8687 8688
	/*
	 * 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.
	 */
8689
	schedstat_inc(sd->lb_balanced[idle]);
8690 8691 8692 8693 8694

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
8695
	if (((env.flags & LBF_ALL_PINNED) &&
8696
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
8697 8698 8699
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

8700
	ld_moved = 0;
8701 8702 8703 8704
out:
	return ld_moved;
}

8705 8706 8707 8708 8709 8710 8711 8712 8713 8714 8715 8716 8717 8718 8719 8720
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
8721
update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
8722 8723 8724
{
	unsigned long interval, next;

8725 8726
	/* used by idle balance, so cpu_busy = 0 */
	interval = get_sd_balance_interval(sd, 0);
8727 8728 8729 8730 8731 8732
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

8733
/*
8734
 * active_load_balance_cpu_stop is run by the CPU stopper. It pushes
8735 8736 8737
 * 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.
8738
 */
8739
static int active_load_balance_cpu_stop(void *data)
8740
{
8741 8742
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
8743
	int target_cpu = busiest_rq->push_cpu;
8744
	struct rq *target_rq = cpu_rq(target_cpu);
8745
	struct sched_domain *sd;
8746
	struct task_struct *p = NULL;
8747
	struct rq_flags rf;
8748

8749
	rq_lock_irq(busiest_rq, &rf);
8750 8751 8752 8753 8754 8755 8756
	/*
	 * Between queueing the stop-work and running it is a hole in which
	 * CPUs can become inactive. We should not move tasks from or to
	 * inactive CPUs.
	 */
	if (!cpu_active(busiest_cpu) || !cpu_active(target_cpu))
		goto out_unlock;
8757

8758
	/* Make sure the requested CPU hasn't gone down in the meantime: */
8759 8760 8761
	if (unlikely(busiest_cpu != smp_processor_id() ||
		     !busiest_rq->active_balance))
		goto out_unlock;
8762 8763 8764

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
8765
		goto out_unlock;
8766 8767 8768 8769

	/*
	 * This condition is "impossible", if it occurs
	 * we need to fix it. Originally reported by
8770
	 * Bjorn Helgaas on a 128-CPU setup.
8771 8772 8773 8774
	 */
	BUG_ON(busiest_rq == target_rq);

	/* Search for an sd spanning us and the target CPU. */
8775
	rcu_read_lock();
8776 8777 8778 8779 8780 8781 8782
	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)) {
8783 8784
		struct lb_env env = {
			.sd		= sd,
8785 8786 8787 8788
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
8789
			.idle		= CPU_IDLE,
8790 8791 8792 8793 8794 8795 8796
			/*
			 * can_migrate_task() doesn't need to compute new_dst_cpu
			 * for active balancing. Since we have CPU_IDLE, but no
			 * @dst_grpmask we need to make that test go away with lying
			 * about DST_PINNED.
			 */
			.flags		= LBF_DST_PINNED,
8797 8798
		};

8799
		schedstat_inc(sd->alb_count);
8800
		update_rq_clock(busiest_rq);
8801

8802
		p = detach_one_task(&env);
8803
		if (p) {
8804
			schedstat_inc(sd->alb_pushed);
8805 8806 8807
			/* Active balancing done, reset the failure counter. */
			sd->nr_balance_failed = 0;
		} else {
8808
			schedstat_inc(sd->alb_failed);
8809
		}
8810
	}
8811
	rcu_read_unlock();
8812 8813
out_unlock:
	busiest_rq->active_balance = 0;
8814
	rq_unlock(busiest_rq, &rf);
8815 8816 8817 8818 8819 8820

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

8821
	return 0;
8822 8823
}

8824 8825 8826 8827 8828 8829 8830 8831 8832 8833 8834 8835 8836 8837 8838 8839 8840 8841 8842 8843 8844 8845 8846 8847 8848 8849 8850 8851 8852 8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863 8864 8865 8866 8867 8868 8869 8870 8871 8872 8873 8874 8875 8876 8877 8878 8879 8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890 8891 8892 8893 8894 8895 8896 8897 8898 8899 8900 8901 8902 8903 8904 8905 8906 8907 8908 8909 8910 8911 8912 8913 8914 8915 8916 8917 8918 8919 8920 8921 8922 8923 8924 8925 8926 8927 8928 8929 8930 8931 8932 8933 8934 8935 8936 8937 8938 8939 8940 8941
static DEFINE_SPINLOCK(balancing);

/*
 * 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.
 */
void update_max_interval(void)
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

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

	rcu_read_lock();
	for_each_domain(cpu, sd) {
		/*
		 * 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;

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

		/*
		 * 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;
		}

		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);

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

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
				/*
				 * The LBF_DST_PINNED logic could have changed
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
				 */
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
			}
			sd->last_balance = jiffies;
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
		}
		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;
		}
	}
	if (need_decay) {
		/*
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
		 */
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
	}
	rcu_read_unlock();

	/*
	 * next_balance will be updated only when there is a need.
	 * When the cpu is attached to null domain for ex, it will not be
	 * updated.
	 */
	if (likely(update_next_balance)) {
		rq->next_balance = next_balance;

#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
	}
}

8942 8943 8944 8945 8946
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

8947
#ifdef CONFIG_NO_HZ_COMMON
8948 8949 8950 8951 8952 8953
/*
 * 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.
 */
8954

8955
static inline int find_new_ilb(void)
8956
{
8957
	int ilb = cpumask_first(nohz.idle_cpus_mask);
8958

8959 8960 8961 8962
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
8963 8964
}

8965 8966 8967 8968 8969
/*
 * 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).
 */
8970
static void kick_ilb(unsigned int flags)
8971 8972 8973 8974 8975
{
	int ilb_cpu;

	nohz.next_balance++;

8976
	ilb_cpu = find_new_ilb();
8977

8978 8979
	if (ilb_cpu >= nr_cpu_ids)
		return;
8980

8981
	flags = atomic_fetch_or(flags, nohz_flags(ilb_cpu));
P
Peter Zijlstra 已提交
8982
	if (flags & NOHZ_KICK_MASK)
8983
		return;
8984

8985 8986
	/*
	 * Use smp_send_reschedule() instead of resched_cpu().
8987
	 * This way we generate a sched IPI on the target CPU which
8988 8989 8990 8991
	 * is idle. And the softirq performing nohz idle load balance
	 * will be run before returning from the IPI.
	 */
	smp_send_reschedule(ilb_cpu);
8992 8993 8994 8995 8996 8997 8998 8999 9000 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010
}

/*
 * Current heuristic for kicking the idle load balancer in the presence
 * of an idle cpu in the system.
 *   - This rq has more than one task.
 *   - 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.
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
 */
static void nohz_balancer_kick(struct rq *rq)
{
	unsigned long now = jiffies;
	struct sched_domain_shared *sds;
	struct sched_domain *sd;
	int nr_busy, i, cpu = rq->cpu;
9011
	unsigned int flags = 0;
9012 9013 9014 9015 9016 9017 9018 9019

	if (unlikely(rq->idle_balance))
		return;

	/*
	 * 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.
	 */
9020
	nohz_balance_exit_idle(rq);
9021 9022 9023 9024 9025 9026 9027 9028

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

9029 9030
	if (READ_ONCE(nohz.has_blocked) &&
	    time_after(now, READ_ONCE(nohz.next_blocked)))
9031 9032
		flags = NOHZ_STATS_KICK;

9033
	if (time_before(now, nohz.next_balance))
9034
		goto out;
9035 9036

	if (rq->nr_running >= 2) {
9037
		flags = NOHZ_KICK_MASK;
9038 9039 9040 9041 9042 9043 9044 9045 9046 9047 9048 9049
		goto out;
	}

	rcu_read_lock();
	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);
		if (nr_busy > 1) {
9050
			flags = NOHZ_KICK_MASK;
9051 9052 9053 9054 9055 9056 9057 9058 9059
			goto unlock;
		}

	}

	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
9060
			flags = NOHZ_KICK_MASK;
9061 9062 9063 9064 9065 9066 9067 9068 9069 9070 9071 9072
			goto unlock;
		}
	}

	sd = rcu_dereference(per_cpu(sd_asym, cpu));
	if (sd) {
		for_each_cpu(i, sched_domain_span(sd)) {
			if (i == cpu ||
			    !cpumask_test_cpu(i, nohz.idle_cpus_mask))
				continue;

			if (sched_asym_prefer(i, cpu)) {
9073
				flags = NOHZ_KICK_MASK;
9074 9075 9076 9077 9078 9079 9080
				goto unlock;
			}
		}
	}
unlock:
	rcu_read_unlock();
out:
9081 9082
	if (flags)
		kick_ilb(flags);
9083 9084
}

9085
static void set_cpu_sd_state_busy(int cpu)
9086
{
9087
	struct sched_domain *sd;
9088

9089 9090
	rcu_read_lock();
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
9091

9092 9093 9094 9095 9096 9097 9098
	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

	atomic_inc(&sd->shared->nr_busy_cpus);
unlock:
	rcu_read_unlock();
9099 9100
}

9101 9102 9103 9104 9105 9106 9107 9108 9109 9110 9111 9112 9113 9114 9115
void nohz_balance_exit_idle(struct rq *rq)
{
	SCHED_WARN_ON(rq != this_rq());

	if (likely(!rq->nohz_tick_stopped))
		return;

	rq->nohz_tick_stopped = 0;
	cpumask_clear_cpu(rq->cpu, nohz.idle_cpus_mask);
	atomic_dec(&nohz.nr_cpus);

	set_cpu_sd_state_busy(rq->cpu);
}

static void set_cpu_sd_state_idle(int cpu)
9116 9117 9118 9119
{
	struct sched_domain *sd;

	rcu_read_lock();
9120
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
9121 9122 9123 9124 9125

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

9126
	atomic_dec(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
9127
unlock:
9128 9129 9130
	rcu_read_unlock();
}

9131
/*
9132
 * This routine will record that the CPU is going idle with tick stopped.
9133
 * This info will be used in performing idle load balancing in the future.
9134
 */
9135
void nohz_balance_enter_idle(int cpu)
9136
{
9137 9138 9139 9140
	struct rq *rq = cpu_rq(cpu);

	SCHED_WARN_ON(cpu != smp_processor_id());

9141
	/* If this CPU is going down, then nothing needs to be done: */
9142 9143 9144
	if (!cpu_active(cpu))
		return;

9145
	/* Spare idle load balancing on CPUs that don't want to be disturbed: */
9146
	if (!housekeeping_cpu(cpu, HK_FLAG_SCHED))
9147 9148
		return;

9149 9150 9151 9152 9153 9154 9155 9156 9157 9158 9159 9160 9161
	/*
	 * Can be set safely without rq->lock held
	 * If a clear happens, it will have evaluated last additions because
	 * rq->lock is held during the check and the clear
	 */
	rq->has_blocked_load = 1;

	/*
	 * The tick is still stopped but load could have been added in the
	 * meantime. We set the nohz.has_blocked flag to trig a check of the
	 * *_avg. The CPU is already part of nohz.idle_cpus_mask so the clear
	 * of nohz.has_blocked can only happen after checking the new load
	 */
9162
	if (rq->nohz_tick_stopped)
9163
		goto out;
9164

9165
	/* If we're a completely isolated CPU, we don't play: */
9166
	if (on_null_domain(rq))
9167 9168
		return;

9169 9170
	rq->nohz_tick_stopped = 1;

9171 9172
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
9173

9174 9175 9176 9177 9178 9179 9180
	/*
	 * Ensures that if nohz_idle_balance() fails to observe our
	 * @idle_cpus_mask store, it must observe the @has_blocked
	 * store.
	 */
	smp_mb__after_atomic();

9181
	set_cpu_sd_state_idle(cpu);
9182 9183 9184 9185 9186 9187 9188

out:
	/*
	 * Each time a cpu enter idle, we assume that it has blocked load and
	 * enable the periodic update of the load of idle cpus
	 */
	WRITE_ONCE(nohz.has_blocked, 1);
9189 9190 9191
}

/*
9192 9193 9194 9195 9196
 * Internal function that runs load balance for all idle cpus. The load balance
 * can be a simple update of blocked load or a complete load balance with
 * tasks movement depending of flags.
 * The function returns false if the loop has stopped before running
 * through all idle CPUs.
9197
 */
9198 9199
static bool _nohz_idle_balance(struct rq *this_rq, unsigned int flags,
			       enum cpu_idle_type idle)
9200
{
9201
	/* Earliest time when we have to do rebalance again */
9202 9203
	unsigned long now = jiffies;
	unsigned long next_balance = now + 60*HZ;
9204
	bool has_blocked_load = false;
9205
	int update_next_balance = 0;
P
Peter Zijlstra 已提交
9206 9207
	int this_cpu = this_rq->cpu;
	int balance_cpu;
9208
	int ret = false;
P
Peter Zijlstra 已提交
9209
	struct rq *rq;
9210

P
Peter Zijlstra 已提交
9211
	SCHED_WARN_ON((flags & NOHZ_KICK_MASK) == NOHZ_BALANCE_KICK);
9212

9213 9214 9215 9216 9217 9218 9219 9220 9221 9222 9223 9224 9225 9226 9227 9228
	/*
	 * We assume there will be no idle load after this update and clear
	 * the has_blocked flag. If a cpu enters idle in the mean time, it will
	 * set the has_blocked flag and trig another update of idle load.
	 * Because a cpu that becomes idle, is added to idle_cpus_mask before
	 * setting the flag, we are sure to not clear the state and not
	 * check the load of an idle cpu.
	 */
	WRITE_ONCE(nohz.has_blocked, 0);

	/*
	 * Ensures that if we miss the CPU, we must see the has_blocked
	 * store from nohz_balance_enter_idle().
	 */
	smp_mb();

9229
	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
9230
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
9231 9232 9233
			continue;

		/*
9234 9235
		 * If this CPU gets work to do, stop the load balancing
		 * work being done for other CPUs. Next load
9236 9237
		 * balancing owner will pick it up.
		 */
9238 9239 9240 9241
		if (need_resched()) {
			has_blocked_load = true;
			goto abort;
		}
9242

V
Vincent Guittot 已提交
9243 9244
		rq = cpu_rq(balance_cpu);

9245
		has_blocked_load |= update_nohz_stats(rq, true);
9246

9247 9248 9249 9250 9251
		/*
		 * If time for next balance is due,
		 * do the balance.
		 */
		if (time_after_eq(jiffies, rq->next_balance)) {
9252 9253
			struct rq_flags rf;

9254
			rq_lock_irqsave(rq, &rf);
9255
			update_rq_clock(rq);
9256
			cpu_load_update_idle(rq);
9257
			rq_unlock_irqrestore(rq, &rf);
9258

P
Peter Zijlstra 已提交
9259 9260
			if (flags & NOHZ_BALANCE_KICK)
				rebalance_domains(rq, CPU_IDLE);
9261
		}
9262

9263 9264 9265 9266
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
9267
	}
9268

9269 9270 9271 9272 9273 9274
	/* Newly idle CPU doesn't need an update */
	if (idle != CPU_NEWLY_IDLE) {
		update_blocked_averages(this_cpu);
		has_blocked_load |= this_rq->has_blocked_load;
	}

P
Peter Zijlstra 已提交
9275 9276 9277
	if (flags & NOHZ_BALANCE_KICK)
		rebalance_domains(this_rq, CPU_IDLE);

9278 9279 9280
	WRITE_ONCE(nohz.next_blocked,
		now + msecs_to_jiffies(LOAD_AVG_PERIOD));

9281 9282 9283
	/* The full idle balance loop has been done */
	ret = true;

9284 9285 9286 9287
abort:
	/* There is still blocked load, enable periodic update */
	if (has_blocked_load)
		WRITE_ONCE(nohz.has_blocked, 1);
9288

9289 9290 9291 9292 9293 9294 9295
	/*
	 * 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;
P
Peter Zijlstra 已提交
9296

9297 9298 9299 9300 9301 9302 9303 9304 9305 9306 9307 9308 9309 9310 9311 9312 9313 9314 9315 9316 9317 9318 9319 9320 9321 9322 9323 9324 9325
	return ret;
}

/*
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
static bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
{
	int this_cpu = this_rq->cpu;
	unsigned int flags;

	if (!(atomic_read(nohz_flags(this_cpu)) & NOHZ_KICK_MASK))
		return false;

	if (idle != CPU_IDLE) {
		atomic_andnot(NOHZ_KICK_MASK, nohz_flags(this_cpu));
		return false;
	}

	/*
	 * barrier, pairs with nohz_balance_enter_idle(), ensures ...
	 */
	flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(this_cpu));
	if (!(flags & NOHZ_KICK_MASK))
		return false;

	_nohz_idle_balance(this_rq, flags, idle);

P
Peter Zijlstra 已提交
9326
	return true;
9327
}
9328 9329 9330 9331 9332 9333 9334 9335 9336 9337 9338 9339 9340 9341 9342 9343 9344 9345 9346 9347 9348 9349 9350 9351 9352 9353 9354 9355 9356 9357 9358 9359 9360

static void nohz_newidle_balance(struct rq *this_rq)
{
	int this_cpu = this_rq->cpu;

	/*
	 * This CPU doesn't want to be disturbed by scheduler
	 * housekeeping
	 */
	if (!housekeeping_cpu(this_cpu, HK_FLAG_SCHED))
		return;

	/* Will wake up very soon. No time for doing anything else*/
	if (this_rq->avg_idle < sysctl_sched_migration_cost)
		return;

	/* Don't need to update blocked load of idle CPUs*/
	if (!READ_ONCE(nohz.has_blocked) ||
	    time_before(jiffies, READ_ONCE(nohz.next_blocked)))
		return;

	raw_spin_unlock(&this_rq->lock);
	/*
	 * This CPU is going to be idle and blocked load of idle CPUs
	 * need to be updated. Run the ilb locally as it is a good
	 * candidate for ilb instead of waking up another idle CPU.
	 * Kick an normal ilb if we failed to do the update.
	 */
	if (!_nohz_idle_balance(this_rq, NOHZ_STATS_KICK, CPU_NEWLY_IDLE))
		kick_ilb(NOHZ_STATS_KICK);
	raw_spin_lock(&this_rq->lock);
}

9361 9362 9363
#else /* !CONFIG_NO_HZ_COMMON */
static inline void nohz_balancer_kick(struct rq *rq) { }

9364
static inline bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
P
Peter Zijlstra 已提交
9365 9366 9367
{
	return false;
}
9368 9369

static inline void nohz_newidle_balance(struct rq *this_rq) { }
9370
#endif /* CONFIG_NO_HZ_COMMON */
9371

P
Peter Zijlstra 已提交
9372 9373 9374 9375 9376 9377 9378 9379 9380 9381 9382 9383 9384 9385 9386 9387 9388 9389 9390 9391 9392 9393 9394 9395 9396 9397 9398 9399 9400 9401 9402 9403 9404 9405
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
static int idle_balance(struct rq *this_rq, struct rq_flags *rf)
{
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
	struct sched_domain *sd;
	int pulled_task = 0;
	u64 curr_cost = 0;

	/*
	 * 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);

	/*
	 * Do not pull tasks towards !active CPUs...
	 */
	if (!cpu_active(this_cpu))
		return 0;

	/*
	 * 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.
	 */
	rq_unpin_lock(this_rq, rf);

	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
9406

P
Peter Zijlstra 已提交
9407 9408 9409 9410 9411 9412
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, &next_balance);
		rcu_read_unlock();

9413 9414
		nohz_newidle_balance(this_rq);

P
Peter Zijlstra 已提交
9415 9416 9417 9418 9419 9420 9421 9422 9423 9424 9425 9426 9427 9428 9429 9430 9431 9432 9433 9434 9435 9436 9437 9438 9439 9440 9441 9442 9443 9444 9445 9446 9447 9448 9449 9450 9451 9452 9453 9454 9455 9456 9457 9458 9459 9460 9461 9462 9463
		goto out;
	}

	raw_spin_unlock(&this_rq->lock);

	update_blocked_averages(this_cpu);
	rcu_read_lock();
	for_each_domain(this_cpu, sd) {
		int continue_balancing = 1;
		u64 t0, domain_cost;

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

		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
			update_next_balance(sd, &next_balance);
			break;
		}

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

			pulled_task = load_balance(this_cpu, this_rq,
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);

			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;
		}

		update_next_balance(sd, &next_balance);

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

	raw_spin_lock(&this_rq->lock);

	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

9464
out:
P
Peter Zijlstra 已提交
9465 9466 9467 9468 9469 9470 9471 9472 9473 9474 9475 9476 9477 9478 9479 9480 9481 9482 9483 9484 9485 9486 9487 9488
	/*
	 * 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.
	 */
	if (this_rq->cfs.h_nr_running && !pulled_task)
		pulled_task = 1;

	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
		this_rq->next_balance = next_balance;

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

	if (pulled_task)
		this_rq->idle_stamp = 0;

	rq_repin_lock(this_rq, rf);

	return pulled_task;
}

9489 9490 9491 9492
/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
9493
static __latent_entropy void run_rebalance_domains(struct softirq_action *h)
9494
{
9495
	struct rq *this_rq = this_rq();
9496
	enum cpu_idle_type idle = this_rq->idle_balance ?
9497 9498 9499
						CPU_IDLE : CPU_NOT_IDLE;

	/*
9500 9501
	 * If this CPU has a pending nohz_balance_kick, then do the
	 * balancing on behalf of the other idle CPUs whose ticks are
9502
	 * stopped. Do nohz_idle_balance *before* rebalance_domains to
9503
	 * give the idle CPUs a chance to load balance. Else we may
9504 9505
	 * load balance only within the local sched_domain hierarchy
	 * and abort nohz_idle_balance altogether if we pull some load.
9506
	 */
P
Peter Zijlstra 已提交
9507 9508 9509 9510 9511
	if (nohz_idle_balance(this_rq, idle))
		return;

	/* normal load balance */
	update_blocked_averages(this_rq->cpu);
9512
	rebalance_domains(this_rq, idle);
9513 9514 9515 9516 9517
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
9518
void trigger_load_balance(struct rq *rq)
9519 9520
{
	/* Don't need to rebalance while attached to NULL domain */
9521 9522 9523 9524
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
9525
		raise_softirq(SCHED_SOFTIRQ);
9526 9527

	nohz_balancer_kick(rq);
9528 9529
}

9530 9531 9532
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
9533 9534

	update_runtime_enabled(rq);
9535 9536 9537 9538 9539
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
9540 9541 9542

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
9543 9544
}

9545
#endif /* CONFIG_SMP */
9546

9547
/*
9548 9549 9550 9551 9552 9553
 * scheduler tick hitting a task of our scheduling class.
 *
 * NOTE: This function can be called remotely by the tick offload that
 * goes along full dynticks. Therefore no local assumption can be made
 * and everything must be accessed through the @rq and @curr passed in
 * parameters.
9554
 */
P
Peter Zijlstra 已提交
9555
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
9556 9557 9558 9559 9560 9561
{
	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
9562
		entity_tick(cfs_rq, se, queued);
9563
	}
9564

9565
	if (static_branch_unlikely(&sched_numa_balancing))
9566
		task_tick_numa(rq, curr);
9567 9568 9569
}

/*
P
Peter Zijlstra 已提交
9570 9571 9572
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
9573
 */
P
Peter Zijlstra 已提交
9574
static void task_fork_fair(struct task_struct *p)
9575
{
9576 9577
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
P
Peter Zijlstra 已提交
9578
	struct rq *rq = this_rq();
9579
	struct rq_flags rf;
9580

9581
	rq_lock(rq, &rf);
9582 9583
	update_rq_clock(rq);

9584 9585
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;
9586 9587
	if (curr) {
		update_curr(cfs_rq);
9588
		se->vruntime = curr->vruntime;
9589
	}
9590
	place_entity(cfs_rq, se, 1);
9591

P
Peter Zijlstra 已提交
9592
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
9593
		/*
9594 9595 9596
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
9597
		swap(curr->vruntime, se->vruntime);
9598
		resched_curr(rq);
9599
	}
9600

9601
	se->vruntime -= cfs_rq->min_vruntime;
9602
	rq_unlock(rq, &rf);
9603 9604
}

9605 9606 9607 9608
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
9609 9610
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
9611
{
9612
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
9613 9614
		return;

9615 9616 9617 9618 9619
	/*
	 * 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
9620
	if (rq->curr == p) {
9621
		if (p->prio > oldprio)
9622
			resched_curr(rq);
9623
	} else
9624
		check_preempt_curr(rq, p, 0);
9625 9626
}

9627
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
9628 9629 9630 9631
{
	struct sched_entity *se = &p->se;

	/*
9632 9633 9634 9635 9636 9637 9638 9639 9640 9641
	 * 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
9642
	 *
9643 9644 9645 9646
	 * - 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
9647
	 */
9648 9649
	if (!se->sum_exec_runtime ||
	    (p->state == TASK_WAKING && p->sched_remote_wakeup))
9650 9651 9652 9653 9654
		return true;

	return false;
}

9655 9656 9657 9658 9659 9660 9661 9662 9663 9664 9665 9666 9667 9668 9669 9670 9671 9672
#ifdef CONFIG_FAIR_GROUP_SCHED
/*
 * Propagate the changes of the sched_entity across the tg tree to make it
 * visible to the root
 */
static void propagate_entity_cfs_rq(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq;

	/* Start to propagate at parent */
	se = se->parent;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);

		if (cfs_rq_throttled(cfs_rq))
			break;

9673
		update_load_avg(cfs_rq, se, UPDATE_TG);
9674 9675 9676 9677 9678 9679
	}
}
#else
static void propagate_entity_cfs_rq(struct sched_entity *se) { }
#endif

9680
static void detach_entity_cfs_rq(struct sched_entity *se)
9681 9682 9683
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

9684
	/* Catch up with the cfs_rq and remove our load when we leave */
9685
	update_load_avg(cfs_rq, se, 0);
9686
	detach_entity_load_avg(cfs_rq, se);
9687
	update_tg_load_avg(cfs_rq, false);
9688
	propagate_entity_cfs_rq(se);
P
Peter Zijlstra 已提交
9689 9690
}

9691
static void attach_entity_cfs_rq(struct sched_entity *se)
9692
{
9693
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
9694 9695

#ifdef CONFIG_FAIR_GROUP_SCHED
9696 9697 9698 9699 9700 9701
	/*
	 * 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
9702

9703
	/* Synchronize entity with its cfs_rq */
9704
	update_load_avg(cfs_rq, se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
9705
	attach_entity_load_avg(cfs_rq, se, 0);
9706
	update_tg_load_avg(cfs_rq, false);
9707
	propagate_entity_cfs_rq(se);
9708 9709 9710 9711 9712 9713 9714 9715 9716 9717 9718 9719 9720 9721 9722 9723 9724 9725 9726 9727 9728 9729 9730 9731 9732
}

static void detach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	if (!vruntime_normalized(p)) {
		/*
		 * 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;
	}

	detach_entity_cfs_rq(se);
}

static void attach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	attach_entity_cfs_rq(se);
9733 9734 9735 9736

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
9737

9738 9739 9740 9741 9742 9743 9744 9745
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);
9746

9747
	if (task_on_rq_queued(p)) {
9748
		/*
9749 9750 9751
		 * 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.
9752
		 */
9753 9754 9755 9756
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
9757
	}
9758 9759
}

9760 9761 9762 9763 9764 9765 9766 9767 9768
/* 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;

9769 9770 9771 9772 9773 9774 9775
	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);
	}
9776 9777
}

9778 9779
void init_cfs_rq(struct cfs_rq *cfs_rq)
{
9780
	cfs_rq->tasks_timeline = RB_ROOT_CACHED;
9781 9782 9783 9784
	cfs_rq->min_vruntime = (u64)(-(1LL << 20));
#ifndef CONFIG_64BIT
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
9785
#ifdef CONFIG_SMP
9786
	raw_spin_lock_init(&cfs_rq->removed.lock);
9787
#endif
9788 9789
}

P
Peter Zijlstra 已提交
9790
#ifdef CONFIG_FAIR_GROUP_SCHED
9791 9792 9793 9794 9795 9796 9797 9798
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;
}

9799
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
9800
{
9801
	detach_task_cfs_rq(p);
9802
	set_task_rq(p, task_cpu(p));
9803 9804 9805 9806 9807

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
9808
	attach_task_cfs_rq(p);
P
Peter Zijlstra 已提交
9809
}
9810

9811 9812 9813 9814 9815 9816 9817 9818 9819 9820 9821 9822 9823
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;
	}
}

9824 9825 9826 9827 9828 9829 9830 9831 9832
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]);
9833
		if (tg->se)
9834 9835 9836 9837 9838 9839 9840 9841 9842 9843
			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;
9844
	struct cfs_rq *cfs_rq;
9845 9846
	int i;

K
Kees Cook 已提交
9847
	tg->cfs_rq = kcalloc(nr_cpu_ids, sizeof(cfs_rq), GFP_KERNEL);
9848 9849
	if (!tg->cfs_rq)
		goto err;
K
Kees Cook 已提交
9850
	tg->se = kcalloc(nr_cpu_ids, sizeof(se), GFP_KERNEL);
9851 9852 9853 9854 9855 9856 9857 9858 9859 9860 9861 9862 9863 9864 9865 9866 9867 9868 9869 9870
	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]);
9871
		init_entity_runnable_average(se);
9872 9873 9874 9875 9876 9877 9878 9879 9880 9881
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

9882 9883 9884 9885 9886 9887 9888 9889 9890 9891 9892
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);
9893
		update_rq_clock(rq);
9894
		attach_entity_cfs_rq(se);
9895
		sync_throttle(tg, i);
9896 9897 9898 9899
		raw_spin_unlock_irq(&rq->lock);
	}
}

9900
void unregister_fair_sched_group(struct task_group *tg)
9901 9902
{
	unsigned long flags;
9903 9904
	struct rq *rq;
	int cpu;
9905

9906 9907 9908
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
9909

9910 9911 9912 9913 9914 9915 9916 9917 9918 9919 9920 9921 9922
		/*
		 * 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);
	}
9923 9924 9925 9926 9927 9928 9929 9930 9931 9932 9933 9934 9935 9936 9937 9938 9939 9940 9941
}

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
9942
	if (!parent) {
9943
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
9944 9945
		se->depth = 0;
	} else {
9946
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
9947 9948
		se->depth = parent->depth + 1;
	}
9949 9950

	se->my_q = cfs_rq;
9951 9952
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
9953 9954 9955 9956 9957 9958 9959 9960 9961 9962 9963 9964 9965 9966 9967 9968 9969 9970 9971 9972 9973 9974 9975 9976
	se->parent = parent;
}

static DEFINE_MUTEX(shares_mutex);

int sched_group_set_shares(struct task_group *tg, unsigned long shares)
{
	int i;

	/*
	 * 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);
9977 9978
		struct sched_entity *se = tg->se[i];
		struct rq_flags rf;
9979 9980

		/* Propagate contribution to hierarchy */
9981
		rq_lock_irqsave(rq, &rf);
9982
		update_rq_clock(rq);
9983
		for_each_sched_entity(se) {
9984
			update_load_avg(cfs_rq_of(se), se, UPDATE_TG);
9985
			update_cfs_group(se);
9986
		}
9987
		rq_unlock_irqrestore(rq, &rf);
9988 9989 9990 9991 9992 9993 9994 9995 9996 9997 9998 9999 10000 10001 10002
	}

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;
}

10003 10004
void online_fair_sched_group(struct task_group *tg) { }

10005
void unregister_fair_sched_group(struct task_group *tg) { }
10006 10007 10008

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
10009

10010
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
10011 10012 10013 10014 10015 10016 10017 10018 10019
{
	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)
10020
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
10021 10022 10023 10024

	return rr_interval;
}

10025 10026 10027
/*
 * All the scheduling class methods:
 */
10028
const struct sched_class fair_sched_class = {
10029
	.next			= &idle_sched_class,
10030 10031 10032
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
10033
	.yield_to_task		= yield_to_task_fair,
10034

I
Ingo Molnar 已提交
10035
	.check_preempt_curr	= check_preempt_wakeup,
10036 10037 10038 10039

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

10040
#ifdef CONFIG_SMP
L
Li Zefan 已提交
10041
	.select_task_rq		= select_task_rq_fair,
10042
	.migrate_task_rq	= migrate_task_rq_fair,
10043

10044 10045
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
10046

10047
	.task_dead		= task_dead_fair,
10048
	.set_cpus_allowed	= set_cpus_allowed_common,
10049
#endif
10050

10051
	.set_curr_task          = set_curr_task_fair,
10052
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
10053
	.task_fork		= task_fork_fair,
10054 10055

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
10056
	.switched_from		= switched_from_fair,
10057
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
10058

10059 10060
	.get_rr_interval	= get_rr_interval_fair,

10061 10062
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
10063
#ifdef CONFIG_FAIR_GROUP_SCHED
10064
	.task_change_group	= task_change_group_fair,
P
Peter Zijlstra 已提交
10065
#endif
10066 10067 10068
};

#ifdef CONFIG_SCHED_DEBUG
10069
void print_cfs_stats(struct seq_file *m, int cpu)
10070
{
10071
	struct cfs_rq *cfs_rq, *pos;
10072

10073
	rcu_read_lock();
10074
	for_each_leaf_cfs_rq_safe(cpu_rq(cpu), cfs_rq, pos)
10075
		print_cfs_rq(m, cpu, cfs_rq);
10076
	rcu_read_unlock();
10077
}
10078 10079 10080 10081 10082 10083 10084 10085 10086 10087 10088 10089 10090 10091 10092 10093 10094 10095 10096 10097 10098

#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 */
10099 10100 10101 10102 10103 10104

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

10105
#ifdef CONFIG_NO_HZ_COMMON
10106
	nohz.next_balance = jiffies;
10107
	nohz.next_blocked = jiffies;
10108 10109 10110 10111 10112
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
#endif
#endif /* SMP */

}