fair.c 273.7 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;
	}
}

355 356 357
/* Iterate through all leaf cfs_rq's on a runqueue: */
#define for_each_leaf_cfs_rq(rq, cfs_rq) \
	list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)
P
Peter Zijlstra 已提交
358 359

/* Do the two (enqueued) entities belong to the same group ? */
P
Peter Zijlstra 已提交
360
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
361 362 363
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
364
		return se->cfs_rq;
P
Peter Zijlstra 已提交
365

P
Peter Zijlstra 已提交
366
	return NULL;
P
Peter Zijlstra 已提交
367 368 369 370 371 372 373
}

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

374 375 376 377 378 379 380 381 382 383 384 385 386
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
387 388
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405

	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);
	}
}

406 407 408 409 410 411
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
412

413 414 415
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
416 417 418
}


P
Peter Zijlstra 已提交
419 420
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
421

P
Peter Zijlstra 已提交
422
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
423
{
P
Peter Zijlstra 已提交
424
	return &task_rq(p)->cfs;
425 426
}

P
Peter Zijlstra 已提交
427 428 429 430 431 432 433 434 435 436 437 438 439 440
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;
}

441 442 443 444 445 446 447 448
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)
{
}

449 450
#define for_each_leaf_cfs_rq(rq, cfs_rq)	\
		for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)
P
Peter Zijlstra 已提交
451 452 453 454 455 456

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

457 458 459 460 461
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
462 463
#endif	/* CONFIG_FAIR_GROUP_SCHED */

464
static __always_inline
465
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
466 467 468 469 470

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

471
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
472
{
473
	s64 delta = (s64)(vruntime - max_vruntime);
474
	if (delta > 0)
475
		max_vruntime = vruntime;
476

477
	return max_vruntime;
478 479
}

480
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
481 482 483 484 485 486 487 488
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

489 490 491 492 493 494
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

495 496
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
497
	struct sched_entity *curr = cfs_rq->curr;
498
	struct rb_node *leftmost = rb_first_cached(&cfs_rq->tasks_timeline);
499

500 501
	u64 vruntime = cfs_rq->min_vruntime;

502 503 504 505 506 507
	if (curr) {
		if (curr->on_rq)
			vruntime = curr->vruntime;
		else
			curr = NULL;
	}
508

509 510 511
	if (leftmost) { /* non-empty tree */
		struct sched_entity *se;
		se = rb_entry(leftmost, struct sched_entity, run_node);
512

513
		if (!curr)
514 515 516 517 518
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

519
	/* ensure we never gain time by being placed backwards. */
520
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
521 522 523 524
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
525 526
}

527 528 529
/*
 * Enqueue an entity into the rb-tree:
 */
530
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
531
{
532
	struct rb_node **link = &cfs_rq->tasks_timeline.rb_root.rb_node;
533 534
	struct rb_node *parent = NULL;
	struct sched_entity *entry;
535
	bool leftmost = true;
536 537 538 539 540 541 542 543 544 545 546

	/*
	 * 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.
		 */
547
		if (entity_before(se, entry)) {
548 549 550
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
551
			leftmost = false;
552 553 554 555
		}
	}

	rb_link_node(&se->run_node, parent, link);
556 557
	rb_insert_color_cached(&se->run_node,
			       &cfs_rq->tasks_timeline, leftmost);
558 559
}

560
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
561
{
562
	rb_erase_cached(&se->run_node, &cfs_rq->tasks_timeline);
563 564
}

565
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
566
{
567
	struct rb_node *left = rb_first_cached(&cfs_rq->tasks_timeline);
568 569 570 571 572

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
573 574
}

575 576 577 578 579 580 581 582 583 584 585
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
586
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
587
{
588
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline.rb_root);
589

590 591
	if (!last)
		return NULL;
592 593

	return rb_entry(last, struct sched_entity, run_node);
594 595
}

596 597 598 599
/**************************************************************
 * Scheduling class statistics methods:
 */

600
int sched_proc_update_handler(struct ctl_table *table, int write,
601
		void __user *buffer, size_t *lenp,
602 603
		loff_t *ppos)
{
604
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
605
	unsigned int factor = get_update_sysctl_factor();
606 607 608 609 610 611 612

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

613 614 615 616 617 618 619
#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

620 621 622
	return 0;
}
#endif
623

624
/*
625
 * delta /= w
626
 */
627
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
628
{
629
	if (unlikely(se->load.weight != NICE_0_LOAD))
630
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
631 632 633 634

	return delta;
}

635 636 637
/*
 * The idea is to set a period in which each task runs once.
 *
638
 * When there are too many tasks (sched_nr_latency) we have to stretch
639 640 641 642
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
643 644
static u64 __sched_period(unsigned long nr_running)
{
645 646 647 648
	if (unlikely(nr_running > sched_nr_latency))
		return nr_running * sysctl_sched_min_granularity;
	else
		return sysctl_sched_latency;
649 650
}

651 652 653 654
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
655
 * s = p*P[w/rw]
656
 */
P
Peter Zijlstra 已提交
657
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
658
{
M
Mike Galbraith 已提交
659
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
660

M
Mike Galbraith 已提交
661
	for_each_sched_entity(se) {
L
Lin Ming 已提交
662
		struct load_weight *load;
663
		struct load_weight lw;
L
Lin Ming 已提交
664 665 666

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
667

M
Mike Galbraith 已提交
668
		if (unlikely(!se->on_rq)) {
669
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
670 671 672 673

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
674
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
675 676
	}
	return slice;
677 678
}

679
/*
A
Andrei Epure 已提交
680
 * We calculate the vruntime slice of a to-be-inserted task.
681
 *
682
 * vs = s/w
683
 */
684
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
685
{
686
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
687 688
}

689
#ifdef CONFIG_SMP
690
#include "pelt.h"
691 692
#include "sched-pelt.h"

693
static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu);
694
static unsigned long task_h_load(struct task_struct *p);
695
static unsigned long task_h_load_static(struct task_struct *p);
696

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

799 800 801 802 803 804 805 806 807
static inline void
update_exec_raw(struct cfs_rq *cfs_rq, struct sched_entity *curr)
{
	u64 now = rq_clock(rq_of(cfs_rq));

	curr->sum_exec_raw += now - curr->exec_start_raw;
	curr->exec_start_raw = now;
}

808
/*
809
 * Update the current task's runtime statistics.
810
 */
811
static void update_curr(struct cfs_rq *cfs_rq)
812
{
813
	struct sched_entity *curr = cfs_rq->curr;
814
	u64 now = rq_clock_task(rq_of(cfs_rq));
815
	u64 delta_exec;
816 817 818 819

	if (unlikely(!curr))
		return;

820 821
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
822
		return;
823

I
Ingo Molnar 已提交
824
	curr->exec_start = now;
825

826 827 828 829
	schedstat_set(curr->statistics.exec_max,
		      max(delta_exec, curr->statistics.exec_max));

	curr->sum_exec_runtime += delta_exec;
830
	schedstat_add(cfs_rq->exec_clock, delta_exec);
831 832 833 834

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

835 836 837
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

838
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
839
		cgroup_account_cputime(curtask, delta_exec);
840
		account_group_exec_runtime(curtask, delta_exec);
841
	}
842 843

	account_cfs_rq_runtime(cfs_rq, delta_exec);
844
	update_exec_raw(cfs_rq, curr);
845 846
}

847 848 849 850 851
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

852
static inline void
853
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
854
{
855 856 857 858 859 860 861
	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);
862 863

	if (entity_is_task(se) && task_on_rq_migrating(task_of(se)) &&
864 865
	    likely(wait_start > prev_wait_start))
		wait_start -= prev_wait_start;
866

867
	__schedstat_set(se->statistics.wait_start, wait_start);
868 869
}

870
static inline void
871 872 873
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *p;
874 875
	u64 delta;

876 877 878 879
	if (!schedstat_enabled())
		return;

	delta = rq_clock(rq_of(cfs_rq)) - schedstat_val(se->statistics.wait_start);
880 881 882 883 884 885 886 887 888

	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.
			 */
889
			__schedstat_set(se->statistics.wait_start, delta);
890 891 892 893 894
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

895
	__schedstat_set(se->statistics.wait_max,
896
		      max(schedstat_val(se->statistics.wait_max), delta));
897 898 899
	__schedstat_inc(se->statistics.wait_count);
	__schedstat_add(se->statistics.wait_sum, delta);
	__schedstat_set(se->statistics.wait_start, 0);
900 901
}

902
static inline void
903 904 905
update_stats_enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *tsk = NULL;
906 907 908 909 910 911 912
	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);
913 914 915 916

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

917 918
	if (sleep_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - sleep_start;
919 920 921 922

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

923
		if (unlikely(delta > schedstat_val(se->statistics.sleep_max)))
924
			__schedstat_set(se->statistics.sleep_max, delta);
925

926 927
		__schedstat_set(se->statistics.sleep_start, 0);
		__schedstat_add(se->statistics.sum_sleep_runtime, delta);
928 929 930 931 932 933

		if (tsk) {
			account_scheduler_latency(tsk, delta >> 10, 1);
			trace_sched_stat_sleep(tsk, delta);
		}
	}
934 935
	if (block_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - block_start;
936 937 938 939

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

940
		if (unlikely(delta > schedstat_val(se->statistics.block_max)))
941
			__schedstat_set(se->statistics.block_max, delta);
942

943 944
		__schedstat_set(se->statistics.block_start, 0);
		__schedstat_add(se->statistics.sum_sleep_runtime, delta);
945 946 947

		if (tsk) {
			if (tsk->in_iowait) {
948 949
				__schedstat_add(se->statistics.iowait_sum, delta);
				__schedstat_inc(se->statistics.iowait_count);
950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967
				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);
		}
	}
968 969
}

970 971 972
/*
 * Task is being enqueued - update stats:
 */
973
static inline void
974
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
975
{
976 977 978
	if (!schedstat_enabled())
		return;

979 980 981 982
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
983
	if (se != cfs_rq->curr)
984
		update_stats_wait_start(cfs_rq, se);
985 986 987

	if (flags & ENQUEUE_WAKEUP)
		update_stats_enqueue_sleeper(cfs_rq, se);
988 989 990
}

static inline void
991
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
992
{
993 994 995 996

	if (!schedstat_enabled())
		return;

997 998 999 1000
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
1001
	if (se != cfs_rq->curr)
1002
		update_stats_wait_end(cfs_rq, se);
1003

1004 1005
	if ((flags & DEQUEUE_SLEEP) && entity_is_task(se)) {
		struct task_struct *tsk = task_of(se);
1006

1007
		if (tsk->state & TASK_INTERRUPTIBLE)
1008
			__schedstat_set(se->statistics.sleep_start,
1009 1010
				      rq_clock(rq_of(cfs_rq)));
		if (tsk->state & TASK_UNINTERRUPTIBLE)
1011
			__schedstat_set(se->statistics.block_start,
1012
				      rq_clock(rq_of(cfs_rq)));
1013 1014 1015
	}
}

1016 1017 1018 1019
/*
 * We are picking a new current task - update its stats:
 */
static inline void
1020
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
1021 1022 1023 1024
{
	/*
	 * We are starting a new run period:
	 */
1025
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
1026
	se->exec_start_raw = rq_clock_task(rq_of(cfs_rq));
1027 1028 1029 1030 1031 1032
}

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

1033 1034
#ifdef CONFIG_NUMA_BALANCING
/*
1035 1036 1037
 * 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.
1038
 */
1039 1040
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
1041 1042 1043

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
1044

1045 1046 1047
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067
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];
};

1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082
/*
 * For functions that can be called in multiple contexts that permit reading
 * ->numa_group (see struct task_struct for locking rules).
 */
static struct numa_group *deref_task_numa_group(struct task_struct *p)
{
	return rcu_dereference_check(p->numa_group, p == current ||
		(lockdep_is_held(&task_rq(p)->lock) && !READ_ONCE(p->on_cpu)));
}

static struct numa_group *deref_curr_numa_group(struct task_struct *p)
{
	return rcu_dereference_protected(p->numa_group, p == current);
}

1083 1084 1085
static inline unsigned long group_faults_priv(struct numa_group *ng);
static inline unsigned long group_faults_shared(struct numa_group *ng);

1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109
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)
{
1110
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
1111 1112 1113
	unsigned int scan, floor;
	unsigned int windows = 1;

1114 1115
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
1116 1117 1118 1119 1120 1121
	floor = 1000 / windows;

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

1122 1123 1124 1125
static unsigned int task_scan_start(struct task_struct *p)
{
	unsigned long smin = task_scan_min(p);
	unsigned long period = smin;
1126
	struct numa_group *ng;
1127 1128

	/* Scale the maximum scan period with the amount of shared memory. */
1129 1130 1131
	rcu_read_lock();
	ng = rcu_dereference(p->numa_group);
	if (ng) {
1132 1133 1134 1135 1136 1137 1138
		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;
	}
1139
	rcu_read_unlock();
1140 1141 1142 1143

	return max(smin, period);
}

1144 1145
static unsigned int task_scan_max(struct task_struct *p)
{
1146 1147
	unsigned long smin = task_scan_min(p);
	unsigned long smax;
1148
	struct numa_group *ng;
1149 1150 1151

	/* Watch for min being lower than max due to floor calculations */
	smax = sysctl_numa_balancing_scan_period_max / task_nr_scan_windows(p);
1152 1153

	/* Scale the maximum scan period with the amount of shared memory. */
1154 1155
	ng = deref_curr_numa_group(p);
	if (ng) {
1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
		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);
	}

1167 1168 1169
	return max(smin, smax);
}

1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186
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;
1187
	RCU_INIT_POINTER(p->numa_group, NULL);
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210
	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;
	}
}

1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
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));
}

1223 1224 1225 1226 1227 1228 1229 1230 1231
/* 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)

1232 1233
pid_t task_numa_group_id(struct task_struct *p)
{
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243
	struct numa_group *ng;
	pid_t gid = 0;

	rcu_read_lock();
	ng = rcu_dereference(p->numa_group);
	if (ng)
		gid = ng->gid;
	rcu_read_unlock();

	return gid;
1244 1245
}

1246
/*
1247
 * The averaged statistics, shared & private, memory & CPU,
1248 1249 1250 1251 1252
 * 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)
1253
{
1254
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
1255 1256 1257 1258
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
1259
	if (!p->numa_faults)
1260 1261
		return 0;

1262 1263
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
1264 1265
}

1266 1267
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
1268 1269 1270
	struct numa_group *ng = deref_task_numa_group(p);

	if (!ng)
1271 1272
		return 0;

1273 1274
	return ng->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		ng->faults[task_faults_idx(NUMA_MEM, nid, 1)];
1275 1276
}

1277 1278
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
1279 1280
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
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
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;
}

1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318
/*
 * 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;
}

1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355
/* 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 &&
1356
					dist >= maxdist)
1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383
			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;
}

1384 1385 1386 1387 1388 1389
/*
 * 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.
 */
1390 1391
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1392
{
1393
	unsigned long faults, total_faults;
1394

1395
	if (!p->numa_faults)
1396 1397 1398 1399 1400 1401 1402
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1403
	faults = task_faults(p, nid);
1404 1405
	faults += score_nearby_nodes(p, nid, dist, true);

1406
	return 1000 * faults / total_faults;
1407 1408
}

1409 1410
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1411
{
1412
	struct numa_group *ng = deref_task_numa_group(p);
1413 1414
	unsigned long faults, total_faults;

1415
	if (!ng)
1416 1417
		return 0;

1418
	total_faults = ng->total_faults;
1419 1420

	if (!total_faults)
1421 1422
		return 0;

1423
	faults = group_faults(p, nid);
1424 1425
	faults += score_nearby_nodes(p, nid, dist, false);

1426
	return 1000 * faults / total_faults;
1427 1428
}

1429 1430 1431
bool should_numa_migrate_memory(struct task_struct *p, struct page * page,
				int src_nid, int dst_cpu)
{
1432
	struct numa_group *ng = deref_curr_numa_group(p);
1433 1434 1435 1436
	int dst_nid = cpu_to_node(dst_cpu);
	int last_cpupid, this_cpupid;

	this_cpupid = cpu_pid_to_cpupid(dst_cpu, current->pid);
1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447
	last_cpupid = page_cpupid_xchg_last(page, this_cpupid);

	/*
	 * Allow first faults or private faults to migrate immediately early in
	 * the lifetime of a task. The magic number 4 is based on waiting for
	 * two full passes of the "multi-stage node selection" test that is
	 * executed below.
	 */
	if ((p->numa_preferred_nid == -1 || p->numa_scan_seq <= 4) &&
	    (cpupid_pid_unset(last_cpupid) || cpupid_match_pid(p, last_cpupid)))
		return true;
1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478

	/*
	 * 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.
	 */
	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;

	/*
1479 1480
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1481
	 */
1482 1483
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1484 1485 1486
		return true;

	/*
1487 1488 1489 1490 1491 1492
	 * 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)
1493
	 */
1494 1495
	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;
1496 1497
}

1498
static unsigned long weighted_cpuload(struct rq *rq);
1499 1500
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1501
static unsigned long capacity_of(int cpu);
1502

1503
/* Cached statistics for all CPUs within a node */
1504 1505
struct numa_stats {
	unsigned long load;
1506 1507

	/* Total compute capacity of CPUs on a node */
1508
	unsigned long compute_capacity;
1509

1510
	unsigned int nr_running;
1511
};
1512

1513 1514 1515 1516 1517
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1518 1519
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1520 1521 1522 1523 1524 1525

	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;
1526
		ns->load += weighted_cpuload(rq);
1527
		ns->compute_capacity += capacity_of(cpu);
1528 1529

		cpus++;
1530 1531
	}

1532 1533 1534 1535 1536
	/*
	 * 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.
	 *
1537
	 * We'll detect a huge imbalance and bail there.
1538 1539 1540 1541
	 */
	if (!cpus)
		return;

1542 1543 1544 1545
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, ns->compute_capacity);
	capacity = cpus / smt; /* cores */

1546
	capacity = min_t(unsigned, capacity,
1547
		DIV_ROUND_CLOSEST(ns->compute_capacity, SCHED_CAPACITY_SCALE));
1548 1549
}

1550 1551
struct task_numa_env {
	struct task_struct *p;
1552

1553 1554
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1555

1556
	struct numa_stats src_stats, dst_stats;
1557

1558
	int imbalance_pct;
1559
	int dist;
1560 1561 1562

	struct task_struct *best_task;
	long best_imp;
1563 1564 1565
	int best_cpu;
};

1566 1567 1568
static void task_numa_assign(struct task_numa_env *env,
			     struct task_struct *p, long imp)
{
1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583
	struct rq *rq = cpu_rq(env->dst_cpu);

	/* Bail out if run-queue part of active NUMA balance. */
	if (xchg(&rq->numa_migrate_on, 1))
		return;

	/*
	 * Clear previous best_cpu/rq numa-migrate flag, since task now
	 * found a better CPU to move/swap.
	 */
	if (env->best_cpu != -1) {
		rq = cpu_rq(env->best_cpu);
		WRITE_ONCE(rq->numa_migrate_on, 0);
	}

1584 1585
	if (env->best_task)
		put_task_struct(env->best_task);
1586 1587
	if (p)
		get_task_struct(p);
1588 1589 1590 1591 1592 1593

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

1594
static bool load_too_imbalanced(long src_load, long dst_load,
1595 1596
				struct task_numa_env *env)
{
1597 1598
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609
	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;
1610

1611
	imb = abs(dst_load * src_capacity - src_load * dst_capacity);
1612

1613
	orig_src_load = env->src_stats.load;
1614
	orig_dst_load = env->dst_stats.load;
1615

1616
	old_imb = abs(orig_dst_load * src_capacity - orig_src_load * dst_capacity);
1617 1618 1619

	/* Would this change make things worse? */
	return (imb > old_imb);
1620 1621
}

1622 1623 1624 1625 1626 1627 1628
/*
 * Maximum NUMA importance can be 1998 (2*999);
 * SMALLIMP @ 30 would be close to 1998/64.
 * Used to deter task migration.
 */
#define SMALLIMP	30

1629 1630 1631 1632 1633 1634
/*
 * 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
 */
1635
static void task_numa_compare(struct task_numa_env *env,
1636
			      long taskimp, long groupimp, bool maymove)
1637
{
1638
	struct numa_group *cur_ng, *p_ng = deref_curr_numa_group(env->p);
1639
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
1640
	long imp = p_ng ? groupimp : taskimp;
1641
	struct task_struct *cur;
1642
	long src_load, dst_load;
1643
	int dist = env->dist;
1644 1645
	long moveimp = imp;
	long load;
1646

1647 1648 1649
	if (READ_ONCE(dst_rq->numa_migrate_on))
		return;

1650
	rcu_read_lock();
1651 1652
	cur = task_rcu_dereference(&dst_rq->curr);
	if (cur && ((cur->flags & PF_EXITING) || is_idle_task(cur)))
1653 1654
		cur = NULL;

1655 1656 1657 1658 1659 1660 1661
	/*
	 * 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;

1662
	if (!cur) {
1663
		if (maymove && moveimp >= env->best_imp)
1664 1665 1666 1667 1668
			goto assign;
		else
			goto unlock;
	}

1669 1670 1671 1672
	/*
	 * "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
1673
	 * the value is, the more remote accesses that would be expected to
1674 1675
	 * be incurred if the tasks were swapped.
	 */
1676 1677 1678
	/* Skip this swap candidate if cannot move to the source cpu */
	if (!cpumask_test_cpu(env->src_cpu, &cur->cpus_allowed))
		goto unlock;
1679

1680 1681 1682 1683
	/*
	 * If dst and source tasks are in the same NUMA group, or not
	 * in any group then look only at task weights.
	 */
1684 1685
	cur_ng = rcu_dereference(cur->numa_group);
	if (cur_ng == p_ng) {
1686 1687
		imp = taskimp + task_weight(cur, env->src_nid, dist) -
		      task_weight(cur, env->dst_nid, dist);
1688
		/*
1689 1690
		 * Add some hysteresis to prevent swapping the
		 * tasks within a group over tiny differences.
1691
		 */
1692
		if (cur_ng)
1693 1694 1695 1696 1697 1698
			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.
		 */
1699
		if (cur_ng && p_ng)
1700 1701 1702 1703 1704
			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);
1705 1706
	}

1707
	if (maymove && moveimp > imp && moveimp > env->best_imp) {
1708
		imp = moveimp;
1709
		cur = NULL;
1710
		goto assign;
1711
	}
1712

1713 1714 1715 1716 1717 1718 1719 1720 1721
	/*
	 * If the NUMA importance is less than SMALLIMP,
	 * task migration might only result in ping pong
	 * of tasks and also hurt performance due to cache
	 * misses.
	 */
	if (imp < SMALLIMP || imp <= env->best_imp + SMALLIMP / 2)
		goto unlock;

1722 1723 1724
	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
1725 1726 1727 1728
	load = task_h_load(env->p) - task_h_load(cur);
	if (!load)
		goto assign;

1729 1730
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1731

1732
	if (load_too_imbalanced(src_load, dst_load, env))
1733 1734
		goto unlock;

1735
assign:
1736 1737 1738 1739
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
1740 1741
	if (!cur) {
		/*
1742
		 * select_idle_siblings() uses an per-CPU cpumask that
1743 1744 1745
		 * can be used from IRQ context.
		 */
		local_irq_disable();
1746 1747
		env->dst_cpu = select_idle_sibling(env->p, env->src_cpu,
						   env->dst_cpu);
1748 1749
		local_irq_enable();
	}
1750

1751 1752 1753 1754 1755
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1756 1757
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1758
{
1759 1760
	long src_load, dst_load, load;
	bool maymove = false;
1761 1762
	int cpu;

1763 1764 1765 1766 1767 1768 1769 1770 1771 1772
	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);

1773 1774
	for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
		/* Skip this CPU if the source task cannot migrate */
1775
		if (!cpumask_test_cpu(cpu, &env->p->cpus_allowed))
1776 1777 1778
			continue;

		env->dst_cpu = cpu;
1779
		task_numa_compare(env, taskimp, groupimp, maymove);
1780 1781 1782
	}
}

1783 1784 1785 1786
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1787

1788
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1789
		.src_nid = task_node(p),
1790 1791 1792 1793 1794

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1795
		.best_cpu = -1,
1796
	};
1797
	unsigned long taskweight, groupweight;
1798
	struct sched_domain *sd;
1799 1800
	long taskimp, groupimp;
	struct numa_group *ng;
1801
	struct rq *best_rq;
1802
	int nid, ret, dist;
1803

1804
	/*
1805 1806 1807 1808 1809 1810
	 * 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.
1811 1812
	 */
	rcu_read_lock();
1813
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1814 1815
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1816 1817
	rcu_read_unlock();

1818 1819 1820 1821 1822 1823 1824
	/*
	 * 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)) {
1825
		sched_setnuma(p, task_node(p));
1826 1827 1828
		return -EINVAL;
	}

1829
	env.dst_nid = p->numa_preferred_nid;
1830 1831 1832 1833 1834 1835
	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;
1836
	update_numa_stats(&env.dst_stats, env.dst_nid);
1837

1838
	/* Try to find a spot on the preferred nid. */
1839
	task_numa_find_cpu(&env, taskimp, groupimp);
1840

1841 1842 1843 1844 1845 1846 1847
	/*
	 * 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.
	 */
1848 1849
	ng = deref_curr_numa_group(p);
	if (env.best_cpu == -1 || (ng && ng->active_nodes > 1)) {
1850 1851 1852
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1853

1854
			dist = node_distance(env.src_nid, env.dst_nid);
1855 1856 1857 1858 1859
			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);
			}
1860

1861
			/* Only consider nodes where both task and groups benefit */
1862 1863
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1864
			if (taskimp < 0 && groupimp < 0)
1865 1866
				continue;

1867
			env.dist = dist;
1868 1869
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1870
			task_numa_find_cpu(&env, taskimp, groupimp);
1871 1872 1873
		}
	}

1874 1875 1876 1877 1878 1879 1880 1881
	/*
	 * 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.
	 */
1882
	if (ng) {
1883 1884 1885
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
1886
			nid = cpu_to_node(env.best_cpu);
1887

1888 1889
		if (nid != p->numa_preferred_nid)
			sched_setnuma(p, nid);
1890 1891 1892 1893 1894
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1895

1896
	best_rq = cpu_rq(env.best_cpu);
1897
	if (env.best_task == NULL) {
1898
		ret = migrate_task_to(p, env.best_cpu);
1899
		WRITE_ONCE(best_rq->numa_migrate_on, 0);
1900 1901
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1902 1903 1904
		return ret;
	}

1905
	ret = migrate_swap(p, env.best_task, env.best_cpu, env.src_cpu);
1906
	WRITE_ONCE(best_rq->numa_migrate_on, 0);
1907

1908 1909
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1910 1911
	put_task_struct(env.best_task);
	return ret;
1912 1913
}

1914 1915 1916
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1917 1918
	unsigned long interval = HZ;

1919
	/* This task has no NUMA fault statistics yet */
1920
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1921 1922
		return;

1923
	/* Periodically retry migrating the task to the preferred node */
1924
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
1925
	p->numa_migrate_retry = jiffies + interval;
1926 1927

	/* Success if task is already running on preferred CPU */
1928
	if (task_node(p) == p->numa_preferred_nid)
1929 1930 1931
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1932
	task_numa_migrate(p);
1933 1934
}

1935
/*
1936
 * Find out how many nodes on the workload is actively running on. Do this by
1937 1938 1939 1940
 * 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.
 */
1941
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1942 1943
{
	unsigned long faults, max_faults = 0;
1944
	int nid, active_nodes = 0;
1945 1946 1947 1948 1949 1950 1951 1952 1953

	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);
1954 1955
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1956
	}
1957 1958 1959

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1960 1961
}

1962 1963 1964
/*
 * 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
1965 1966 1967
 * 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.
1968 1969
 */
#define NUMA_PERIOD_SLOTS 10
1970
#define NUMA_PERIOD_THRESHOLD 7
1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981

/*
 * 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;
1982
	int lr_ratio, ps_ratio;
1983 1984 1985 1986 1987 1988 1989 1990
	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
1991 1992 1993
	 * 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
1994
	 */
1995
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
		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);
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030
	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;
2031 2032 2033 2034 2035
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else {
		/*
2036 2037 2038
		 * 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.
2039
		 */
2040 2041
		int ratio = max(lr_ratio, ps_ratio);
		diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot;
2042 2043 2044 2045 2046 2047 2048
	}

	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));
}

2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065
/*
 * 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;
2066 2067 2068 2069

		/* Avoid time going backwards, prevent potential divide error: */
		if (unlikely((s64)*period < 0))
			*period = 0;
2070
	} else {
2071
		delta = p->se.avg.load_sum;
2072
		*period = LOAD_AVG_MAX;
2073 2074 2075 2076 2077 2078 2079 2080
	}

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

	return delta;
}

2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127
/*
 * 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;
2128
		nodemask_t max_group = NODE_MASK_NONE;
2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161
		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. */
2162 2163
		if (!max_faults)
			break;
2164 2165 2166 2167 2168
		nodes = max_group;
	}
	return nid;
}

2169 2170
static void task_numa_placement(struct task_struct *p)
{
2171 2172
	int seq, nid, max_nid = -1;
	unsigned long max_faults = 0;
2173
	unsigned long fault_types[2] = { 0, 0 };
2174 2175
	unsigned long total_faults;
	u64 runtime, period;
2176
	spinlock_t *group_lock = NULL;
2177
	struct numa_group *ng;
2178

2179 2180 2181 2182 2183
	/*
	 * 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:
	 */
2184
	seq = READ_ONCE(p->mm->numa_scan_seq);
2185 2186 2187
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
2188
	p->numa_scan_period_max = task_scan_max(p);
2189

2190 2191 2192 2193
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

2194
	/* If the task is part of a group prevent parallel updates to group stats */
2195 2196 2197
	ng = deref_curr_numa_group(p);
	if (ng) {
		group_lock = &ng->lock;
2198
		spin_lock_irq(group_lock);
2199 2200
	}

2201 2202
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
2203 2204
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
2205
		unsigned long faults = 0, group_faults = 0;
2206
		int priv;
2207

2208
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
2209
			long diff, f_diff, f_weight;
2210

2211 2212 2213 2214
			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);
2215

2216
			/* Decay existing window, copy faults since last scan */
2217 2218 2219
			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;
2220

2221 2222 2223 2224 2225 2226 2227 2228
			/*
			 * 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);
2229
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
2230
				   (total_faults + 1);
2231 2232
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
2233

2234 2235 2236
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
2237
			p->total_numa_faults += diff;
2238
			if (ng) {
2239 2240 2241 2242 2243 2244 2245
				/*
				 * 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.
				 */
2246 2247 2248 2249
				ng->faults[mem_idx] += diff;
				ng->faults_cpu[mem_idx] += f_diff;
				ng->total_faults += diff;
				group_faults += ng->faults[mem_idx];
2250
			}
2251 2252
		}

2253
		if (!ng) {
2254 2255 2256 2257 2258 2259
			if (faults > max_faults) {
				max_faults = faults;
				max_nid = nid;
			}
		} else if (group_faults > max_faults) {
			max_faults = group_faults;
2260 2261
			max_nid = nid;
		}
2262 2263
	}

2264 2265
	if (ng) {
		numa_group_count_active_nodes(ng);
2266
		spin_unlock_irq(group_lock);
2267
		max_nid = preferred_group_nid(p, max_nid);
2268 2269
	}

2270 2271 2272 2273
	if (max_faults) {
		/* Set the new preferred node */
		if (max_nid != p->numa_preferred_nid)
			sched_setnuma(p, max_nid);
2274
	}
2275 2276

	update_task_scan_period(p, fault_types[0], fault_types[1]);
2277 2278
}

2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289
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);
}

2290 2291
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2292 2293 2294 2295 2296 2297 2298
{
	struct numa_group *grp, *my_grp;
	struct task_struct *tsk;
	bool join = false;
	int cpu = cpupid_to_cpu(cpupid);
	int i;

2299
	if (unlikely(!deref_curr_numa_group(p))) {
2300
		unsigned int size = sizeof(struct numa_group) +
2301
				    4*nr_node_ids*sizeof(unsigned long);
2302 2303 2304 2305 2306 2307

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

		atomic_set(&grp->refcount, 1);
2308 2309
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2310
		spin_lock_init(&grp->lock);
2311
		grp->gid = p->pid;
2312
		/* Second half of the array tracks nids where faults happen */
2313 2314
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2315

2316
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2317
			grp->faults[i] = p->numa_faults[i];
2318

2319
		grp->total_faults = p->total_numa_faults;
2320

2321 2322 2323 2324 2325
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2326
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2327 2328

	if (!cpupid_match_pid(tsk, cpupid))
2329
		goto no_join;
2330 2331 2332

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2333
		goto no_join;
2334

2335
	my_grp = deref_curr_numa_group(p);
2336
	if (grp == my_grp)
2337
		goto no_join;
2338 2339 2340 2341 2342 2343

	/*
	 * 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)
2344
		goto no_join;
2345 2346 2347 2348 2349

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2350
		goto no_join;
2351

2352 2353 2354 2355 2356 2357 2358
	/* 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;
2359

2360 2361 2362
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2363
	if (join && !get_numa_group(grp))
2364
		goto no_join;
2365 2366 2367 2368 2369 2370

	rcu_read_unlock();

	if (!join)
		return;

2371 2372
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2373

2374
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2375 2376
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2377
	}
2378 2379
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2380 2381 2382 2383 2384

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

	spin_unlock(&my_grp->lock);
2385
	spin_unlock_irq(&grp->lock);
2386 2387 2388 2389

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2390 2391 2392 2393 2394
	return;

no_join:
	rcu_read_unlock();
	return;
2395 2396
}

2397 2398 2399 2400 2401 2402 2403 2404
/*
 * Get rid of NUMA staticstics associated with a task (either current or dead).
 * If @final is set, the task is dead and has reached refcount zero, so we can
 * safely free all relevant data structures. Otherwise, there might be
 * concurrent reads from places like load balancing and procfs, and we should
 * reset the data back to default state without freeing ->numa_faults.
 */
void task_numa_free(struct task_struct *p, bool final)
2405
{
2406 2407
	/* safe: p either is current or is being freed by current */
	struct numa_group *grp = rcu_dereference_raw(p->numa_group);
2408
	unsigned long *numa_faults = p->numa_faults;
2409 2410
	unsigned long flags;
	int i;
2411

2412 2413 2414
	if (!numa_faults)
		return;

2415
	if (grp) {
2416
		spin_lock_irqsave(&grp->lock, flags);
2417
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2418
			grp->faults[i] -= p->numa_faults[i];
2419
		grp->total_faults -= p->total_numa_faults;
2420

2421
		grp->nr_tasks--;
2422
		spin_unlock_irqrestore(&grp->lock, flags);
2423
		RCU_INIT_POINTER(p->numa_group, NULL);
2424 2425 2426
		put_numa_group(grp);
	}

2427 2428 2429 2430 2431 2432 2433 2434
	if (final) {
		p->numa_faults = NULL;
		kfree(numa_faults);
	} else {
		p->total_numa_faults = 0;
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
			numa_faults[i] = 0;
	}
2435 2436
}

2437 2438 2439
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2440
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2441 2442
{
	struct task_struct *p = current;
2443
	bool migrated = flags & TNF_MIGRATED;
2444
	int cpu_node = task_node(current);
2445
	int local = !!(flags & TNF_FAULT_LOCAL);
2446
	struct numa_group *ng;
2447
	int priv;
2448

2449
	if (!static_branch_likely(&sched_numa_balancing))
2450 2451
		return;

2452 2453 2454 2455
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2456
	/* Allocate buffer to track faults on a per-node basis */
2457 2458
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2459
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2460

2461 2462
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2463
			return;
2464

2465
		p->total_numa_faults = 0;
2466
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2467
	}
2468

2469 2470 2471 2472 2473 2474 2475 2476
	/*
	 * 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);
2477
		if (!priv && !(flags & TNF_NO_GROUP))
2478
			task_numa_group(p, last_cpupid, flags, &priv);
2479 2480
	}

2481 2482 2483 2484 2485 2486
	/*
	 * 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.
	 */
2487
	ng = deref_curr_numa_group(p);
2488 2489 2490
	if (!priv && !local && ng && ng->active_nodes > 1 &&
				numa_is_active_node(cpu_node, ng) &&
				numa_is_active_node(mem_node, ng))
2491 2492
		local = 1;

2493 2494 2495 2496
	/*
	 * Retry task to preferred node migration periodically, in case it
	 * case it previously failed, or the scheduler moved us.
	 */
2497 2498
	if (time_after(jiffies, p->numa_migrate_retry)) {
		task_numa_placement(p);
2499
		numa_migrate_preferred(p);
2500
	}
2501

I
Ingo Molnar 已提交
2502 2503
	if (migrated)
		p->numa_pages_migrated += pages;
2504 2505
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2506

2507 2508
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2509
	p->numa_faults_locality[local] += pages;
2510 2511
}

2512 2513
static void reset_ptenuma_scan(struct task_struct *p)
{
2514 2515 2516 2517 2518 2519 2520 2521
	/*
	 * 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:
	 */
2522
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2523 2524 2525
	p->mm->numa_scan_offset = 0;
}

2526 2527 2528 2529 2530 2531 2532 2533 2534
/*
 * 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;
2535
	u64 runtime = p->se.sum_exec_runtime;
2536
	struct vm_area_struct *vma;
2537
	unsigned long start, end;
2538
	unsigned long nr_pte_updates = 0;
2539
	long pages, virtpages;
2540

2541
	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554

	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;

2555
	if (!mm->numa_next_scan) {
2556 2557
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2558 2559
	}

2560 2561 2562 2563 2564 2565 2566
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2567 2568
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
2569
		p->numa_scan_period = task_scan_start(p);
2570
	}
2571

2572
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2573 2574 2575
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2576 2577 2578 2579 2580 2581
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2582 2583 2584
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2585
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2586 2587
	if (!pages)
		return;
2588

2589

2590 2591
	if (!down_read_trylock(&mm->mmap_sem))
		return;
2592
	vma = find_vma(mm, start);
2593 2594
	if (!vma) {
		reset_ptenuma_scan(p);
2595
		start = 0;
2596 2597
		vma = mm->mmap;
	}
2598
	for (; vma; vma = vma->vm_next) {
2599
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2600
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2601
			continue;
2602
		}
2603

2604 2605 2606 2607 2608 2609 2610 2611 2612 2613
		/*
		 * 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
2614 2615 2616 2617 2618 2619
		/*
		 * 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;
2620

2621 2622 2623 2624
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2625
			nr_pte_updates = change_prot_numa(vma, start, end);
2626 2627

			/*
2628 2629 2630 2631 2632 2633
			 * 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.
2634 2635 2636
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2637
			virtpages -= (end - start) >> PAGE_SHIFT;
2638

2639
			start = end;
2640
			if (pages <= 0 || virtpages <= 0)
2641
				goto out;
2642 2643

			cond_resched();
2644
		} while (end != vma->vm_end);
2645
	}
2646

2647
out:
2648
	/*
P
Peter Zijlstra 已提交
2649 2650 2651 2652
	 * 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.
2653 2654
	 */
	if (vma)
2655
		mm->numa_scan_offset = start;
2656 2657 2658
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669

	/*
	 * 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;
	}
2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694
}

/*
 * 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;

2695
	if (now > curr->node_stamp + period) {
2696
		if (!curr->node_stamp)
2697
			curr->numa_scan_period = task_scan_start(curr);
2698
		curr->node_stamp += period;
2699 2700 2701 2702 2703 2704 2705

		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);
		}
	}
}
2706

2707 2708 2709 2710 2711
static void update_scan_period(struct task_struct *p, int new_cpu)
{
	int src_nid = cpu_to_node(task_cpu(p));
	int dst_nid = cpu_to_node(new_cpu);

2712 2713 2714
	if (!static_branch_likely(&sched_numa_balancing))
		return;

2715 2716 2717
	if (!p->mm || !p->numa_faults || (p->flags & PF_EXITING))
		return;

2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737
	if (src_nid == dst_nid)
		return;

	/*
	 * Allow resets if faults have been trapped before one scan
	 * has completed. This is most likely due to a new task that
	 * is pulled cross-node due to wakeups or load balancing.
	 */
	if (p->numa_scan_seq) {
		/*
		 * Avoid scan adjustments if moving to the preferred
		 * node or if the task was not previously running on
		 * the preferred node.
		 */
		if (dst_nid == p->numa_preferred_nid ||
		    (p->numa_preferred_nid != -1 && src_nid != p->numa_preferred_nid))
			return;
	}

	p->numa_scan_period = task_scan_start(p);
2738 2739
}

2740 2741 2742 2743
#else
static void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
}
2744 2745 2746 2747 2748 2749 2750 2751

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)
{
}
2752

2753 2754 2755 2756
static inline void update_scan_period(struct task_struct *p, int new_cpu)
{
}

2757 2758
#endif /* CONFIG_NUMA_BALANCING */

2759 2760 2761 2762
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2763
	if (!parent_entity(se))
2764
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2765
#ifdef CONFIG_SMP
2766 2767 2768 2769 2770 2771
	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);
	}
2772
#endif
2773 2774 2775 2776 2777 2778 2779
	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);
2780
	if (!parent_entity(se))
2781
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2782
#ifdef CONFIG_SMP
2783 2784
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2785
		list_del_init(&se->group_node);
2786
	}
2787
#endif
2788 2789 2790
	cfs_rq->nr_running--;
}

2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831
/*
 * 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)
{
2832 2833 2834 2835
	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;
2836 2837 2838 2839 2840
}

static inline void
dequeue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2841 2842 2843 2844 2845
	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);
2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871
}

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

2872
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
2873
			    unsigned long weight, unsigned long runnable)
2874 2875 2876 2877 2878 2879 2880 2881 2882 2883
{
	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);

2884
	se->runnable_weight = runnable;
2885 2886 2887
	update_load_set(&se->load, weight);

#ifdef CONFIG_SMP
2888 2889 2890 2891 2892 2893 2894
	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);
2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910
#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]);

2911
	reweight_entity(cfs_rq, se, weight, weight);
2912 2913 2914
	load->inv_weight = sched_prio_to_wmult[prio];
}

2915
#ifdef CONFIG_FAIR_GROUP_SCHED
2916
#ifdef CONFIG_SMP
2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954
/*
 * 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
2955
 *   ge->load.weight = ----------------------------- = tg->weight   (4)
2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968
 *			    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
 *
2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980
 * 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)
2981 2982 2983 2984 2985 2986 2987 2988 2989
 *
 * 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!
 */
2990
static long calc_group_shares(struct cfs_rq *cfs_rq)
2991
{
2992 2993 2994 2995
	long tg_weight, tg_shares, load, shares;
	struct task_group *tg = cfs_rq->tg;

	tg_shares = READ_ONCE(tg->shares);
2996

2997
	load = max(scale_load_down(cfs_rq->load.weight), cfs_rq->avg.load_avg);
2998

2999
	tg_weight = atomic_long_read(&tg->load_avg);
3000

3001 3002 3003
	/* Ensure tg_weight >= load */
	tg_weight -= cfs_rq->tg_load_avg_contrib;
	tg_weight += load;
3004

3005
	shares = (tg_shares * load);
3006 3007
	if (tg_weight)
		shares /= tg_weight;
3008

3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020
	/*
	 * 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.
	 */
3021
	return clamp_t(long, shares, MIN_SHARES, tg_shares);
3022
}
3023 3024

/*
3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049
 * 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).
3050 3051 3052
 */
static long calc_group_runnable(struct cfs_rq *cfs_rq, long shares)
{
3053 3054 3055 3056 3057 3058 3059
	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));
3060 3061 3062 3063

	runnable *= shares;
	if (load_avg)
		runnable /= load_avg;
3064

3065 3066
	return clamp_t(long, runnable, MIN_SHARES, shares);
}
3067
#endif /* CONFIG_SMP */
3068

3069 3070
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

3071 3072 3073 3074 3075
/*
 * 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
3076
{
3077 3078
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);
	long shares, runnable;
P
Peter Zijlstra 已提交
3079

3080
	if (!gcfs_rq)
3081 3082
		return;

3083
	if (throttled_hierarchy(gcfs_rq))
P
Peter Zijlstra 已提交
3084
		return;
3085

3086
#ifndef CONFIG_SMP
3087
	runnable = shares = READ_ONCE(gcfs_rq->tg->shares);
3088 3089

	if (likely(se->load.weight == shares))
3090
		return;
3091
#else
3092 3093
	shares   = calc_group_shares(gcfs_rq);
	runnable = calc_group_runnable(gcfs_rq, shares);
3094
#endif
P
Peter Zijlstra 已提交
3095

3096
	reweight_entity(cfs_rq_of(se), se, shares, runnable);
P
Peter Zijlstra 已提交
3097
}
3098

P
Peter Zijlstra 已提交
3099
#else /* CONFIG_FAIR_GROUP_SCHED */
3100
static inline void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
3101 3102 3103 3104
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

3105
static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq, int flags)
3106
{
3107 3108
	struct rq *rq = rq_of(cfs_rq);

3109
	if (&rq->cfs == cfs_rq || (flags & SCHED_CPUFREQ_MIGRATION)) {
3110 3111 3112
		/*
		 * There are a few boundary cases this might miss but it should
		 * get called often enough that that should (hopefully) not be
3113
		 * a real problem.
3114 3115 3116 3117 3118 3119 3120 3121 3122 3123
		 *
		 * 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().
		 */
3124
		cpufreq_update_util(rq, flags);
3125 3126 3127
	}
}

3128
#ifdef CONFIG_SMP
3129
#ifdef CONFIG_FAIR_GROUP_SCHED
3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142
/**
 * 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'.
 *
3143
 * Updating tg's load_avg is necessary before update_cfs_share().
3144
 */
3145
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
3146
{
3147
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
3148

3149 3150 3151 3152 3153 3154
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

3155 3156 3157
	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;
3158
	}
3159
}
3160

3161
/*
3162
 * Called within set_task_rq() right before setting a task's CPU. The
3163 3164 3165 3166 3167 3168
 * 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)
{
3169 3170 3171
	u64 p_last_update_time;
	u64 n_last_update_time;

3172 3173 3174 3175 3176 3177 3178 3179 3180 3181
	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.
	 */
3182 3183
	if (!(se->avg.last_update_time && prev))
		return;
3184 3185

#ifndef CONFIG_64BIT
3186
	{
3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200
		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);
3201
	}
3202
#else
3203 3204
	p_last_update_time = prev->avg.last_update_time;
	n_last_update_time = next->avg.last_update_time;
3205
#endif
3206 3207
	__update_load_avg_blocked_se(p_last_update_time, cpu_of(rq_of(prev)), se);
	se->avg.last_update_time = n_last_update_time;
3208
}
3209

3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220

/*
 * 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.
 *
3221 3222 3223
 * 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).
3224 3225 3226 3227 3228 3229 3230 3231
 *
 * 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:
 *
3232
 *   grq->avg.load_avg = grq->load.weight * grq->avg.runnable_avg	(3)
3233 3234 3235
 *
 * And per (1) we have:
 *
3236
 *   ge->avg.runnable_avg == grq->avg.runnable_avg
3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254
 *
 * 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).
 *
3255 3256 3257 3258 3259 3260
 * 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.
3261
 *
3262
 * So we'll have to approximate.. :/
3263
 *
3264
 * Given the constraint:
3265
 *
3266
 *   ge->avg.running_sum <= ge->avg.runnable_sum <= LOAD_AVG_MAX
3267
 *
3268 3269
 * We can construct a rule that adds runnable to a rq by assuming minimal
 * overlap.
3270
 *
3271
 * On removal, we'll assume each task is equally runnable; which yields:
3272
 *
3273
 *   grq->avg.runnable_sum = grq->avg.load_sum / grq->load.weight
3274
 *
3275
 * XXX: only do this for the part of runnable > running ?
3276 3277 3278
 *
 */

3279
static inline void
3280
update_tg_cfs_util(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3281 3282 3283 3284 3285 3286 3287
{
	long delta = gcfs_rq->avg.util_avg - se->avg.util_avg;

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

3288 3289 3290 3291 3292 3293 3294 3295
	/*
	 * 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.
	 */

3296 3297 3298 3299 3300 3301 3302 3303 3304 3305
	/* 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
3306
update_tg_cfs_runnable(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3307
{
3308 3309 3310 3311
	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;
3312

3313 3314
	if (!runnable_sum)
		return;
3315

3316
	gcfs_rq->prop_runnable_sum = 0;
3317

3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340
	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
3341
	 * As running sum is scale with CPU capacity wehreas the runnable sum
3342 3343 3344 3345 3346 3347
	 * 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);

3348 3349
	load_sum = (s64)se_weight(se) * runnable_sum;
	load_avg = div_s64(load_sum, LOAD_AVG_MAX);
3350

3351 3352
	delta_sum = load_sum - (s64)se_weight(se) * se->avg.load_sum;
	delta_avg = load_avg - se->avg.load_avg;
3353

3354 3355 3356 3357
	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);
3358

3359 3360
	runnable_load_sum = (s64)se_runnable(se) * runnable_sum;
	runnable_load_avg = div_s64(runnable_load_sum, LOAD_AVG_MAX);
3361 3362
	delta_sum = runnable_load_sum - se_weight(se) * se->avg.runnable_load_sum;
	delta_avg = runnable_load_avg - se->avg.runnable_load_avg;
3363

3364 3365
	se->avg.runnable_load_sum = runnable_sum;
	se->avg.runnable_load_avg = runnable_load_avg;
3366

3367
	if (se->on_rq) {
3368 3369
		add_positive(&cfs_rq->avg.runnable_load_avg, delta_avg);
		add_positive(&cfs_rq->avg.runnable_load_sum, delta_sum);
3370 3371 3372
	}
}

3373
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum)
3374
{
3375 3376
	cfs_rq->propagate = 1;
	cfs_rq->prop_runnable_sum += runnable_sum;
3377 3378 3379 3380 3381
}

/* Update task and its cfs_rq load average */
static inline int propagate_entity_load_avg(struct sched_entity *se)
{
3382
	struct cfs_rq *cfs_rq, *gcfs_rq;
3383 3384 3385 3386

	if (entity_is_task(se))
		return 0;

3387 3388
	gcfs_rq = group_cfs_rq(se);
	if (!gcfs_rq->propagate)
3389 3390
		return 0;

3391 3392
	gcfs_rq->propagate = 0;

3393 3394
	cfs_rq = cfs_rq_of(se);

3395
	add_tg_cfs_propagate(cfs_rq, gcfs_rq->prop_runnable_sum);
3396

3397 3398
	update_tg_cfs_util(cfs_rq, se, gcfs_rq);
	update_tg_cfs_runnable(cfs_rq, se, gcfs_rq);
3399 3400 3401 3402

	return 1;
}

3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421
/*
 * 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:
	 */
3422
	if (gcfs_rq->propagate)
3423 3424 3425 3426 3427 3428 3429 3430 3431 3432
		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;
}

3433
#else /* CONFIG_FAIR_GROUP_SCHED */
3434

3435
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
3436 3437 3438 3439 3440 3441

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

3442
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum) {}
3443

3444
#endif /* CONFIG_FAIR_GROUP_SCHED */
3445

3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456
/**
 * 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.
 *
3457 3458 3459 3460
 * 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.
3461
 */
3462
static inline int
3463
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
3464
{
3465
	unsigned long removed_load = 0, removed_util = 0, removed_runnable_sum = 0;
3466
	struct sched_avg *sa = &cfs_rq->avg;
3467
	int decayed = 0;
3468

3469 3470
	if (cfs_rq->removed.nr) {
		unsigned long r;
3471
		u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;
3472 3473 3474 3475

		raw_spin_lock(&cfs_rq->removed.lock);
		swap(cfs_rq->removed.util_avg, removed_util);
		swap(cfs_rq->removed.load_avg, removed_load);
3476
		swap(cfs_rq->removed.runnable_sum, removed_runnable_sum);
3477 3478 3479 3480
		cfs_rq->removed.nr = 0;
		raw_spin_unlock(&cfs_rq->removed.lock);

		r = removed_load;
3481
		sub_positive(&sa->load_avg, r);
3482
		sub_positive(&sa->load_sum, r * divider);
3483

3484
		r = removed_util;
3485
		sub_positive(&sa->util_avg, r);
3486
		sub_positive(&sa->util_sum, r * divider);
3487

3488
		add_tg_cfs_propagate(cfs_rq, -(long)removed_runnable_sum);
3489 3490

		decayed = 1;
3491
	}
3492

3493
	decayed |= __update_load_avg_cfs_rq(now, cpu_of(rq_of(cfs_rq)), cfs_rq);
3494

3495 3496 3497 3498
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
3499

3500
	if (decayed)
3501
		cfs_rq_util_change(cfs_rq, 0);
3502

3503
	return decayed;
3504 3505
}

3506 3507 3508 3509
/**
 * 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
3510
 * @flags: migration hints
3511 3512 3513 3514
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3515
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3516
{
3517 3518 3519 3520 3521 3522 3523 3524 3525
	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
	 */
3526
	se->avg.last_update_time = cfs_rq->avg.last_update_time;
3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544
	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;

3545
	enqueue_load_avg(cfs_rq, se);
3546 3547
	cfs_rq->avg.util_avg += se->avg.util_avg;
	cfs_rq->avg.util_sum += se->avg.util_sum;
3548 3549

	add_tg_cfs_propagate(cfs_rq, se->avg.load_sum);
3550

3551
	cfs_rq_util_change(cfs_rq, flags);
3552 3553
}

3554 3555 3556 3557 3558 3559 3560 3561
/**
 * 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.
 */
3562 3563
static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3564
	dequeue_load_avg(cfs_rq, se);
3565 3566
	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
3567 3568

	add_tg_cfs_propagate(cfs_rq, -se->avg.load_sum);
3569

3570
	cfs_rq_util_change(cfs_rq, 0);
3571 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 3597 3598 3599
/*
 * 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)) {

3600 3601 3602 3603 3604 3605 3606 3607
		/*
		 * 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);
3608 3609 3610 3611 3612 3613
		update_tg_load_avg(cfs_rq, 0);

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

3614
#ifndef CONFIG_64BIT
3615 3616
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
3617
	u64 last_update_time_copy;
3618
	u64 last_update_time;
3619

3620 3621 3622 3623 3624
	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);
3625 3626 3627

	return last_update_time;
}
3628
#else
3629 3630 3631 3632
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
3633 3634
#endif

3635 3636 3637 3638 3639 3640 3641 3642 3643 3644
/*
 * 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);
3645
	__update_load_avg_blocked_se(last_update_time, cpu_of(rq_of(cfs_rq)), se);
3646 3647
}

3648 3649 3650 3651 3652 3653 3654
/*
 * 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);
3655
	unsigned long flags;
3656 3657

	/*
3658 3659 3660 3661 3662 3663 3664
	 * 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.
3665 3666
	 */

3667
	sync_entity_load_avg(se);
3668 3669 3670 3671 3672

	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;
3673
	cfs_rq->removed.runnable_sum	+= se->avg.load_sum; /* == runnable_sum */
3674
	raw_spin_unlock_irqrestore(&cfs_rq->removed.lock, flags);
3675
}
3676

3677 3678
static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq)
{
3679
	return cfs_rq->avg.runnable_load_avg;
3680 3681 3682 3683 3684 3685 3686
}

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

3687
static int idle_balance(struct rq *this_rq, struct rq_flags *rf);
3688

3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715
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;
3716
	enqueued += (_task_util_est(p) | UTIL_AVG_UNCHANGED);
3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741
	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;

3742 3743 3744 3745
	/* 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));
3746 3747 3748 3749 3750 3751 3752 3753 3754
	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;

3755 3756 3757 3758 3759 3760 3761 3762
	/*
	 * 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;

3763 3764 3765 3766
	/*
	 * Skip update of task's estimated utilization when its EWMA is
	 * already ~1% close to its last activation value.
	 */
3767
	ue.enqueued = (task_util(p) | UTIL_AVG_UNCHANGED);
3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794
	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);
}

3795 3796
#else /* CONFIG_SMP */

3797 3798
#define UPDATE_TG	0x0
#define SKIP_AGE_LOAD	0x0
3799
#define DO_ATTACH	0x0
3800

3801
static inline void update_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int not_used1)
3802
{
3803
	cfs_rq_util_change(cfs_rq, 0);
3804 3805
}

3806
static inline void remove_entity_load_avg(struct sched_entity *se) {}
3807

3808
static inline void
3809
attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) {}
3810 3811 3812
static inline void
detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}

3813
static inline int idle_balance(struct rq *rq, struct rq_flags *rf)
3814 3815 3816 3817
{
	return 0;
}

3818 3819 3820 3821 3822 3823 3824
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) {}

3825
#endif /* CONFIG_SMP */
3826

P
Peter Zijlstra 已提交
3827 3828 3829 3830 3831 3832 3833 3834 3835
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)
3836
		schedstat_inc(cfs_rq->nr_spread_over);
P
Peter Zijlstra 已提交
3837 3838 3839
#endif
}

3840 3841 3842
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3843
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3844

3845 3846 3847 3848 3849 3850
	/*
	 * 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
3851
	if (initial && sched_feat(START_DEBIT))
3852
		vruntime += sched_vslice(cfs_rq, se);
3853

3854
	/* sleeps up to a single latency don't count. */
3855
	if (!initial) {
3856
		unsigned long thresh = sysctl_sched_latency;
3857

3858 3859 3860 3861 3862 3863
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3864

3865
		vruntime -= thresh;
3866 3867
	}

3868
	/* ensure we never gain time by being placed backwards. */
3869
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3870 3871
}

3872 3873
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885
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())  {
3886
		printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
3887
			     "stat_blocked and stat_runtime require the "
3888
			     "kernel parameter schedstats=enable or "
3889 3890 3891 3892 3893
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912

/*
 * 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)
 *
3913
 *	->migrate_task_rq_fair() (p->state == TASK_WAKING)
3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924
 *	  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.
 */

3925
static void
3926
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3927
{
3928 3929 3930
	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
	bool curr = cfs_rq->curr == se;

3931
	/*
3932 3933
	 * If we're the current task, we must renormalise before calling
	 * update_curr().
3934
	 */
3935
	if (renorm && curr)
3936 3937
		se->vruntime += cfs_rq->min_vruntime;

3938 3939
	update_curr(cfs_rq);

3940
	/*
3941 3942 3943 3944
	 * 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.
3945
	 */
3946 3947 3948
	if (renorm && !curr)
		se->vruntime += cfs_rq->min_vruntime;

3949 3950 3951 3952 3953 3954 3955 3956
	/*
	 * 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
	 */
3957
	update_load_avg(cfs_rq, se, UPDATE_TG | DO_ATTACH);
3958
	update_cfs_group(se);
3959
	enqueue_runnable_load_avg(cfs_rq, se);
3960
	account_entity_enqueue(cfs_rq, se);
3961

3962
	if (flags & ENQUEUE_WAKEUP)
3963
		place_entity(cfs_rq, se, 0);
3964

3965
	check_schedstat_required();
3966 3967
	update_stats_enqueue(cfs_rq, se, flags);
	check_spread(cfs_rq, se);
3968
	if (!curr)
3969
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3970
	se->on_rq = 1;
3971

3972
	if (cfs_rq->nr_running == 1) {
3973
		list_add_leaf_cfs_rq(cfs_rq);
3974 3975
		check_enqueue_throttle(cfs_rq);
	}
3976 3977
}

3978
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3979
{
3980 3981
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3982
		if (cfs_rq->last != se)
3983
			break;
3984 3985

		cfs_rq->last = NULL;
3986 3987
	}
}
P
Peter Zijlstra 已提交
3988

3989 3990 3991 3992
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3993
		if (cfs_rq->next != se)
3994
			break;
3995 3996

		cfs_rq->next = NULL;
3997
	}
P
Peter Zijlstra 已提交
3998 3999
}

4000 4001 4002 4003
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
4004
		if (cfs_rq->skip != se)
4005
			break;
4006 4007

		cfs_rq->skip = NULL;
4008 4009 4010
	}
}

P
Peter Zijlstra 已提交
4011 4012
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
4013 4014 4015 4016 4017
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
4018 4019 4020

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
4021 4022
}

4023
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4024

4025
static void
4026
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
4027
{
4028 4029 4030 4031
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
4032 4033 4034 4035 4036 4037 4038 4039 4040

	/*
	 * 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.
	 */
4041
	update_load_avg(cfs_rq, se, UPDATE_TG);
4042
	dequeue_runnable_load_avg(cfs_rq, se);
4043

4044
	update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
4045

P
Peter Zijlstra 已提交
4046
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4047

4048
	if (se != cfs_rq->curr)
4049
		__dequeue_entity(cfs_rq, se);
4050
	se->on_rq = 0;
4051
	account_entity_dequeue(cfs_rq, se);
4052 4053

	/*
4054 4055 4056 4057
	 * 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.
4058
	 */
4059
	if (!(flags & DEQUEUE_SLEEP))
4060
		se->vruntime -= cfs_rq->min_vruntime;
4061

4062 4063 4064
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

4065
	update_cfs_group(se);
4066 4067 4068 4069 4070 4071 4072

	/*
	 * 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.
	 */
4073
	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) != DEQUEUE_SAVE)
4074
		update_min_vruntime(cfs_rq);
4075 4076 4077 4078 4079
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
4080
static void
I
Ingo Molnar 已提交
4081
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4082
{
4083
	unsigned long ideal_runtime, delta_exec;
4084 4085
	struct sched_entity *se;
	s64 delta;
4086

P
Peter Zijlstra 已提交
4087
	ideal_runtime = sched_slice(cfs_rq, curr);
4088
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
4089
	if (delta_exec > ideal_runtime) {
4090
		resched_curr(rq_of(cfs_rq));
4091 4092 4093 4094 4095
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106
		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;

4107 4108
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
4109

4110 4111
	if (delta < 0)
		return;
4112

4113
	if (delta > ideal_runtime)
4114
		resched_curr(rq_of(cfs_rq));
4115 4116
}

4117
static void
4118
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
4119
{
4120 4121 4122 4123 4124 4125 4126
	/* '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.
		 */
4127
		update_stats_wait_end(cfs_rq, se);
4128
		__dequeue_entity(cfs_rq, se);
4129
		update_load_avg(cfs_rq, se, UPDATE_TG);
4130 4131
	}

4132
	update_stats_curr_start(cfs_rq, se);
4133
	cfs_rq->curr = se;
4134

I
Ingo Molnar 已提交
4135 4136 4137 4138 4139
	/*
	 * 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):
	 */
4140
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
4141 4142 4143
		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
4144
	}
4145

4146
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
4147 4148
}

4149 4150 4151
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

4152 4153 4154 4155 4156 4157 4158
/*
 * 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
 */
4159 4160
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4161
{
4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172
	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 */
4173

4174 4175 4176 4177 4178
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
4179 4180 4181 4182 4183 4184 4185 4186 4187 4188
		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;
		}

4189 4190 4191
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
4192

4193 4194 4195 4196 4197 4198
	/*
	 * 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;

4199 4200 4201 4202 4203 4204
	/*
	 * 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;

4205
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4206 4207

	return se;
4208 4209
}

4210
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4211

4212
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
4213 4214 4215 4216 4217 4218
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
4219
		update_curr(cfs_rq);
4220

4221 4222 4223
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

4224
	check_spread(cfs_rq, prev);
4225

4226
	if (prev->on_rq) {
4227
		update_stats_wait_start(cfs_rq, prev);
4228 4229
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
4230
		/* in !on_rq case, update occurred at dequeue */
4231
		update_load_avg(cfs_rq, prev, 0);
4232
	}
4233
	cfs_rq->curr = NULL;
4234 4235
}

P
Peter Zijlstra 已提交
4236 4237
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
4238 4239
{
	/*
4240
	 * Update run-time statistics of the 'current'.
4241
	 */
4242
	update_curr(cfs_rq);
4243

4244 4245 4246
	/*
	 * Ensure that runnable average is periodically updated.
	 */
4247
	update_load_avg(cfs_rq, curr, UPDATE_TG);
4248
	update_cfs_group(curr);
4249

P
Peter Zijlstra 已提交
4250 4251 4252 4253 4254
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
4255
	if (queued) {
4256
		resched_curr(rq_of(cfs_rq));
4257 4258
		return;
	}
P
Peter Zijlstra 已提交
4259 4260 4261 4262 4263 4264 4265 4266
	/*
	 * 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
4267
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
4268
		check_preempt_tick(cfs_rq, curr);
4269 4270
}

4271 4272 4273 4274 4275 4276

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

#ifdef CONFIG_CFS_BANDWIDTH
4277

4278
#ifdef CONFIG_JUMP_LABEL
4279
static struct static_key __cfs_bandwidth_used;
4280 4281 4282

static inline bool cfs_bandwidth_used(void)
{
4283
	return static_key_false(&__cfs_bandwidth_used);
4284 4285
}

4286
void cfs_bandwidth_usage_inc(void)
4287
{
4288
	static_key_slow_inc_cpuslocked(&__cfs_bandwidth_used);
4289 4290 4291 4292
}

void cfs_bandwidth_usage_dec(void)
{
4293
	static_key_slow_dec_cpuslocked(&__cfs_bandwidth_used);
4294
}
4295
#else /* CONFIG_JUMP_LABEL */
4296 4297 4298 4299 4300
static bool cfs_bandwidth_used(void)
{
	return true;
}

4301 4302
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
4303
#endif /* CONFIG_JUMP_LABEL */
4304

4305 4306 4307 4308 4309 4310 4311 4312
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
4313 4314 4315 4316 4317 4318

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

P
Paul Turner 已提交
4319
/*
4320 4321 4322
 * Replenish runtime according to assigned quota. We use sched_clock_cpu
 * directly instead of rq->clock to avoid adding additional synchronization
 * around rq->lock.
P
Paul Turner 已提交
4323 4324 4325
 *
 * requires cfs_b->lock
 */
4326
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
4327
{
4328 4329
	if (cfs_b->quota != RUNTIME_INF)
		cfs_b->runtime = cfs_b->quota;
P
Paul Turner 已提交
4330 4331
}

4332 4333 4334 4335 4336
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

4337 4338 4339 4340
/* 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))
4341
		return cfs_rq->throttled_clock_task - cfs_rq->throttled_clock_task_time;
4342

4343
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
4344 4345
}

4346 4347
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4348 4349 4350
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
4351
	u64 amount = 0, min_amount;
4352 4353 4354 4355 4356 4357 4358

	/* 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;
4359
	else {
P
Peter Zijlstra 已提交
4360
		start_cfs_bandwidth(cfs_b);
4361 4362 4363 4364 4365 4366

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
4367 4368 4369 4370
	}
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
4371 4372

	return cfs_rq->runtime_remaining > 0;
4373 4374
}

4375
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
4376 4377
{
	/* dock delta_exec before expiring quota (as it could span periods) */
4378
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
4379 4380

	if (likely(cfs_rq->runtime_remaining > 0))
4381 4382
		return;

4383 4384
	if (cfs_rq->throttled)
		return;
4385 4386 4387 4388 4389
	/*
	 * 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))
4390
		resched_curr(rq_of(cfs_rq));
4391 4392
}

4393
static __always_inline
4394
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
4395
{
4396
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
4397 4398 4399 4400 4401
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

4402 4403
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
4404
	return cfs_bandwidth_used() && cfs_rq->throttled;
4405 4406
}

4407 4408 4409
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
4410
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436
}

/*
 * 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) {
4437
		/* adjust cfs_rq_clock_task() */
4438
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
4439
					     cfs_rq->throttled_clock_task;
4440 4441 4442 4443 4444 4445 4446 4447 4448 4449
	}

	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)];

4450 4451
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
4452
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
4453 4454 4455 4456 4457
	cfs_rq->throttle_count++;

	return 0;
}

4458
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
4459 4460 4461 4462 4463
{
	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
4464
	bool empty;
4465 4466 4467

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

4468
	/* freeze hierarchy runnable averages while throttled */
4469 4470 4471
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
4472 4473 4474 4475 4476 4477 4478

	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;
4479 4480 4481
		if (dequeue) {
			if (se->my_q != cfs_rq)
				cgroup_idle_start(se);
4482
			dequeue_entity(qcfs_rq, se, DEQUEUE_SLEEP);
4483
		}
4484 4485 4486 4487 4488 4489 4490
		qcfs_rq->h_nr_running -= task_delta;

		if (qcfs_rq->load.weight)
			dequeue = 0;
	}

	if (!se)
4491
		sub_nr_running(rq, task_delta);
4492 4493

	cfs_rq->throttled = 1;
4494
	cfs_rq->throttled_clock = rq_clock(rq);
4495
	raw_spin_lock(&cfs_b->lock);
4496
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4497

4498 4499
	/*
	 * Add to the _head_ of the list, so that an already-started
4500 4501
	 * distribute_cfs_runtime will not see us. If disribute_cfs_runtime is
	 * not running add to the tail so that later runqueues don't get starved.
4502
	 */
4503 4504 4505 4506
	if (cfs_b->distribute_running)
		list_add_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
	else
		list_add_tail_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4507 4508 4509 4510 4511 4512 4513 4514

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

4515 4516 4517
	raw_spin_unlock(&cfs_b->lock);
}

4518
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
4519
{
4520
	struct cfs_rq *bottom_cfs_rq = cfs_rq;
4521 4522 4523 4524 4525 4526
	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;

4527
	se = cfs_rq->tg->se[cpu_of(rq)];
4528 4529

	cfs_rq->throttled = 0;
4530 4531 4532

	update_rq_clock(rq);

4533
	raw_spin_lock(&cfs_b->lock);
4534
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
4535 4536 4537
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

4538 4539 4540
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

4541 4542 4543 4544 4545 4546 4547 4548 4549
	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);
4550 4551 4552
		if (enqueue) {
			if (se->my_q != bottom_cfs_rq)
				cgroup_idle_end(se);
4553
			enqueue_entity(cfs_rq, se, ENQUEUE_WAKEUP);
4554
		}
4555 4556 4557 4558 4559 4560 4561
		cfs_rq->h_nr_running += task_delta;

		if (cfs_rq_throttled(cfs_rq))
			break;
	}

	if (!se)
4562
		add_nr_running(rq, task_delta);
4563

4564
	/* Determine whether we need to wake up potentially idle CPU: */
4565
	if (rq->curr == rq->idle && rq->cfs.nr_running)
4566
		resched_curr(rq);
4567 4568
}

4569
static void distribute_cfs_runtime(struct cfs_bandwidth *cfs_b)
4570 4571
{
	struct cfs_rq *cfs_rq;
4572
	u64 runtime, remaining = 1;
4573 4574 4575 4576 4577

	rcu_read_lock();
	list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
				throttled_list) {
		struct rq *rq = rq_of(cfs_rq);
4578
		struct rq_flags rf;
4579

4580
		rq_lock(rq, &rf);
4581 4582 4583
		if (!cfs_rq_throttled(cfs_rq))
			goto next;

4584 4585 4586
		/* By the above check, this should never be true */
		SCHED_WARN_ON(cfs_rq->runtime_remaining > 0);

4587
		raw_spin_lock(&cfs_b->lock);
4588
		runtime = -cfs_rq->runtime_remaining + 1;
4589 4590 4591 4592 4593
		if (runtime > cfs_b->runtime)
			runtime = cfs_b->runtime;
		cfs_b->runtime -= runtime;
		remaining = cfs_b->runtime;
		raw_spin_unlock(&cfs_b->lock);
4594 4595 4596 4597 4598 4599 4600 4601

		cfs_rq->runtime_remaining += runtime;

		/* we check whether we're throttled above */
		if (cfs_rq->runtime_remaining > 0)
			unthrottle_cfs_rq(cfs_rq);

next:
4602
		rq_unlock(rq, &rf);
4603 4604 4605 4606 4607 4608 4609

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

4610 4611 4612 4613 4614 4615 4616 4617
/*
 * 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)
{
4618
	int throttled;
4619 4620 4621

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
4622
		goto out_deactivate;
4623

4624
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4625
	cfs_b->nr_periods += overrun;
4626

4627 4628 4629 4630 4631 4632
	/*
	 * 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
4633 4634 4635

	__refill_cfs_bandwidth_runtime(cfs_b);

4636 4637 4638
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
4639
		return 0;
4640 4641
	}

4642 4643 4644
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

4645
	/*
4646
	 * This check is repeated as we release cfs_b->lock while we unthrottle.
4647
	 */
4648 4649
	while (throttled && cfs_b->runtime > 0 && !cfs_b->distribute_running) {
		cfs_b->distribute_running = 1;
4650 4651
		raw_spin_unlock(&cfs_b->lock);
		/* we can't nest cfs_b->lock while distributing bandwidth */
4652
		distribute_cfs_runtime(cfs_b);
4653 4654
		raw_spin_lock(&cfs_b->lock);

4655
		cfs_b->distribute_running = 0;
4656 4657
		throttled = !list_empty(&cfs_b->throttled_cfs_rq);
	}
4658

4659 4660 4661 4662 4663 4664 4665
	/*
	 * 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;
4666

4667 4668 4669 4670
	return 0;

out_deactivate:
	return 1;
4671
}
4672

4673 4674 4675 4676 4677 4678 4679
/* 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;

4680 4681 4682 4683
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
4684
 * hrtimer base being cleared by hrtimer_start. In the case of
4685 4686
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711
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;

4712 4713 4714 4715 4716
	/* don't push forwards an existing deferred unthrottle */
	if (cfs_b->slack_started)
		return;
	cfs_b->slack_started = true;

P
Peter Zijlstra 已提交
4717 4718 4719
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731
}

/* 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);
4732
	if (cfs_b->quota != RUNTIME_INF) {
4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747
		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)
{
4748 4749 4750
	if (!cfs_bandwidth_used())
		return;

4751
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765
		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();

	/* confirm we're still not at a refresh boundary */
4766
	raw_spin_lock(&cfs_b->lock);
4767
	cfs_b->slack_started = false;
4768 4769 4770 4771 4772
	if (cfs_b->distribute_running) {
		raw_spin_unlock(&cfs_b->lock);
		return;
	}

4773 4774
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
4775
		return;
4776
	}
4777

4778
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
4779
		runtime = cfs_b->runtime;
4780

4781 4782 4783
	if (runtime)
		cfs_b->distribute_running = 1;

4784 4785 4786 4787 4788
	raw_spin_unlock(&cfs_b->lock);

	if (!runtime)
		return;

4789
	distribute_cfs_runtime(cfs_b);
4790 4791

	raw_spin_lock(&cfs_b->lock);
4792
	cfs_b->distribute_running = 0;
4793 4794 4795
	raw_spin_unlock(&cfs_b->lock);
}

4796 4797 4798 4799 4800 4801 4802
/*
 * 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)
{
4803 4804 4805
	if (!cfs_bandwidth_used())
		return;

4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819
	/* 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);
}

4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833
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;
4834
	cfs_rq->throttled_clock_task = rq_clock_task(cpu_rq(cpu));
4835 4836
}

4837
/* conditionally throttle active cfs_rq's from put_prev_entity() */
4838
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4839
{
4840
	if (!cfs_bandwidth_used())
4841
		return false;
4842

4843
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4844
		return false;
4845 4846 4847 4848 4849 4850

	/*
	 * 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))
4851
		return true;
4852 4853

	throttle_cfs_rq(cfs_rq);
4854
	return true;
4855
}
4856 4857 4858 4859 4860

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
4861

4862 4863 4864 4865 4866
	do_sched_cfs_slack_timer(cfs_b);

	return HRTIMER_NORESTART;
}

4867 4868
extern const u64 max_cfs_quota_period;

4869 4870 4871 4872 4873 4874
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;
4875
	int count = 0;
4876

4877
	raw_spin_lock(&cfs_b->lock);
4878
	for (;;) {
P
Peter Zijlstra 已提交
4879
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4880 4881 4882
		if (!overrun)
			break;

4883 4884 4885
		if (++count > 3) {
			u64 new, old = ktime_to_ns(cfs_b->period);

4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907
			/*
			 * Grow period by a factor of 2 to avoid losing precision.
			 * Precision loss in the quota/period ratio can cause __cfs_schedulable
			 * to fail.
			 */
			new = old * 2;
			if (new < max_cfs_quota_period) {
				cfs_b->period = ns_to_ktime(new);
				cfs_b->quota *= 2;

				pr_warn_ratelimited(
	"cfs_period_timer[cpu%d]: period too short, scaling up (new cfs_period_us = %lld, cfs_quota_us = %lld)\n",
					smp_processor_id(),
					div_u64(new, NSEC_PER_USEC),
					div_u64(cfs_b->quota, NSEC_PER_USEC));
			} else {
				pr_warn_ratelimited(
	"cfs_period_timer[cpu%d]: period too short, but cannot scale up without losing precision (cfs_period_us = %lld, cfs_quota_us = %lld)\n",
					smp_processor_id(),
					div_u64(old, NSEC_PER_USEC),
					div_u64(cfs_b->quota, NSEC_PER_USEC));
			}
4908 4909 4910 4911 4912

			/* reset count so we don't come right back in here */
			count = 0;
		}

4913 4914
		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4915 4916
	if (idle)
		cfs_b->period_active = 0;
4917
	raw_spin_unlock(&cfs_b->lock);
4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929

	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
4930
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4931 4932 4933
	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;
4934
	cfs_b->distribute_running = 0;
4935
	cfs_b->slack_started = false;
4936 4937 4938 4939 4940 4941 4942 4943
}

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
4944
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4945
{
P
Peter Zijlstra 已提交
4946
	lockdep_assert_held(&cfs_b->lock);
4947

4948 4949 4950 4951
	if (cfs_b->period_active)
		return;

	cfs_b->period_active = 1;
4952
	hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
4953
	hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
4954 4955 4956 4957
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4958 4959 4960 4961
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4962 4963 4964 4965
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4966
/*
4967
 * Both these CPU hotplug callbacks race against unregister_fair_sched_group()
4968 4969 4970 4971 4972 4973
 *
 * The race is harmless, since modifying bandwidth settings of unhooked group
 * bits doesn't do much.
 */

/* cpu online calback */
4974 4975
static void __maybe_unused update_runtime_enabled(struct rq *rq)
{
4976
	struct task_group *tg;
4977

4978 4979 4980 4981 4982 4983
	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)];
4984 4985 4986 4987 4988

		raw_spin_lock(&cfs_b->lock);
		cfs_rq->runtime_enabled = cfs_b->quota != RUNTIME_INF;
		raw_spin_unlock(&cfs_b->lock);
	}
4989
	rcu_read_unlock();
4990 4991
}

4992
/* cpu offline callback */
4993
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4994
{
4995 4996 4997 4998 4999 5000 5001
	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)];
5002 5003 5004 5005 5006 5007 5008 5009

		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
5010
		cfs_rq->runtime_remaining = 1;
5011
		/*
5012
		 * Offline rq is schedulable till CPU is completely disabled
5013 5014 5015 5016
		 * in take_cpu_down(), so we prevent new cfs throttling here.
		 */
		cfs_rq->runtime_enabled = 0;

5017 5018 5019
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
5020
	rcu_read_unlock();
5021 5022 5023
}

#else /* CONFIG_CFS_BANDWIDTH */
5024 5025
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
5026
	return rq_clock_task(rq_of(cfs_rq));
5027 5028
}

5029
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
5030
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
5031
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
5032
static inline void sync_throttle(struct task_group *tg, int cpu) {}
5033
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
5034 5035 5036 5037 5038

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049

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;
}
5050 5051 5052 5053 5054

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) {}
5055 5056
#endif

5057 5058 5059 5060 5061
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) {}
5062
static inline void update_runtime_enabled(struct rq *rq) {}
5063
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
5064 5065 5066

#endif /* CONFIG_CFS_BANDWIDTH */

5067 5068 5069 5070
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
5071 5072 5073 5074 5075 5076
#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);

5077
	SCHED_WARN_ON(task_rq(p) != rq);
P
Peter Zijlstra 已提交
5078

5079
	if (rq->cfs.h_nr_running > 1) {
P
Peter Zijlstra 已提交
5080 5081 5082 5083 5084 5085
		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)
5086
				resched_curr(rq);
P
Peter Zijlstra 已提交
5087 5088
			return;
		}
5089
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
5090 5091
	}
}
5092 5093 5094 5095 5096 5097 5098 5099 5100 5101

/*
 * 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;

5102
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
5103 5104 5105 5106 5107
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
5108
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
5109 5110 5111 5112
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
5113 5114 5115 5116

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
5117 5118
#endif

5119 5120 5121 5122 5123
/*
 * 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:
 */
5124
static void
5125
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5126 5127
{
	struct cfs_rq *cfs_rq;
5128
	struct sched_entity *se = &p->se;
5129

5130 5131 5132 5133 5134 5135 5136 5137
	/*
	 * 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);

5138 5139 5140 5141 5142 5143
	/*
	 * 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)
5144
		cpufreq_update_util(rq, SCHED_CPUFREQ_IOWAIT);
5145

5146
	for_each_sched_entity(se) {
5147
		if (se->on_rq)
5148 5149
			break;
		cfs_rq = cfs_rq_of(se);
5150
		enqueue_entity(cfs_rq, se, flags);
5151

5152 5153 5154
		if (!entity_is_task(se))
			cgroup_idle_end(se);

5155 5156 5157 5158 5159
		/*
		 * 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.
5160
		 */
5161 5162 5163 5164 5165
		if (cfs_rq_throttled(cfs_rq)) {
#ifdef CONFIG_FAIR_GROUP_SCHED
			if (cfs_rq->nr_running == 1)
				cgroup_idle_end(se->parent);
#endif
5166
			break;
5167
		}
5168
		cfs_rq->h_nr_running++;
5169

5170
		flags = ENQUEUE_WAKEUP;
5171
	}
P
Peter Zijlstra 已提交
5172

P
Peter Zijlstra 已提交
5173
	for_each_sched_entity(se) {
5174
		cfs_rq = cfs_rq_of(se);
5175
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
5176

5177 5178 5179
		if (cfs_rq_throttled(cfs_rq))
			break;

5180
		update_load_avg(cfs_rq, se, UPDATE_TG);
5181
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5182 5183
	}

Y
Yuyang Du 已提交
5184
	if (!se)
5185
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
5186

5187
	hrtick_update(rq);
5188 5189
}

5190 5191
static void set_next_buddy(struct sched_entity *se);

5192 5193 5194 5195 5196
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
5197
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5198 5199
{
	struct cfs_rq *cfs_rq;
5200
	struct sched_entity *se = &p->se;
5201
	int task_sleep = flags & DEQUEUE_SLEEP;
5202 5203 5204

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5205
		dequeue_entity(cfs_rq, se, flags);
5206

5207 5208 5209
		if (!entity_is_task(se))
			cgroup_idle_start(se);

5210 5211 5212 5213 5214 5215
		/*
		 * 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.
		*/
5216 5217 5218 5219 5220
		if (cfs_rq_throttled(cfs_rq)) {
#ifdef CONFIG_FAIR_GROUP_SCHED
			if (!cfs_rq->nr_running)
				cgroup_idle_start(se->parent);
#endif
5221
			break;
5222
		}
5223
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5224

5225
		/* Don't dequeue parent if it has other entities besides us */
5226
		if (cfs_rq->load.weight) {
5227 5228
			/* Avoid re-evaluating load for this entity: */
			se = parent_entity(se);
5229 5230 5231 5232
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
5233 5234
			if (task_sleep && se && !throttled_hierarchy(cfs_rq))
				set_next_buddy(se);
5235
			break;
5236
		}
5237
		flags |= DEQUEUE_SLEEP;
5238
	}
P
Peter Zijlstra 已提交
5239

P
Peter Zijlstra 已提交
5240
	for_each_sched_entity(se) {
5241
		cfs_rq = cfs_rq_of(se);
5242
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5243

5244 5245 5246
		if (cfs_rq_throttled(cfs_rq))
			break;

5247
		update_load_avg(cfs_rq, se, UPDATE_TG);
5248
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5249 5250
	}

Y
Yuyang Du 已提交
5251
	if (!se)
5252
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
5253

5254
	util_est_dequeue(&rq->cfs, p, task_sleep);
5255
	hrtick_update(rq);
5256 5257
}

5258
#ifdef CONFIG_SMP
5259 5260 5261 5262 5263

/* 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);

5264
#ifdef CONFIG_NO_HZ_COMMON
5265 5266 5267 5268 5269
/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
5270
 * The exact cpuload calculated at every tick would be:
5271
 *
5272 5273
 *   load' = (1 - 1/2^i) * load + (1/2^i) * cur_load
 *
5274 5275
 * 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:
5276 5277 5278
 *
 *   load_n   = (1 - 1/2^i)^n * load_0
 *   load_n+1 = (1 - 1/2^i)   * load_n + (1/2^i) * cur_load
5279 5280 5281
 *
 * decay_load_missed() below does efficient calculation of
 *
5282 5283 5284 5285 5286 5287
 *   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())
5288
 *
5289
 * The calculation is approximated on a 128 point scale.
5290 5291
 */
#define DEGRADE_SHIFT		7
5292 5293 5294 5295 5296 5297 5298 5299 5300

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 }
};
5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329

/*
 * 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;
}
5330 5331 5332 5333

static struct {
	cpumask_var_t idle_cpus_mask;
	atomic_t nr_cpus;
5334
	int has_blocked;		/* Idle CPUS has blocked load */
5335
	unsigned long next_balance;     /* in jiffy units */
5336
	unsigned long next_blocked;	/* Next update of blocked load in jiffies */
5337 5338
} nohz ____cacheline_aligned;

5339
#endif /* CONFIG_NO_HZ_COMMON */
5340

5341
/**
5342
 * __cpu_load_update - update the rq->cpu_load[] statistics
5343 5344 5345 5346
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 *
5347
 * Update rq->cpu_load[] statistics. This function is usually called every
5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373
 * 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
5374
 * term.
5375
 */
5376 5377
static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
			    unsigned long pending_updates)
5378
{
5379
	unsigned long __maybe_unused tickless_load = this_rq->cpu_load[0];
5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390
	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 */

5391
		old_load = this_rq->cpu_load[i];
5392
#ifdef CONFIG_NO_HZ_COMMON
5393
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
5394 5395 5396 5397 5398 5399 5400 5401 5402
		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;
		}
5403
#endif
5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416
		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;
	}
}

5417
/* Used instead of source_load when we know the type == 0 */
5418
static unsigned long weighted_cpuload(struct rq *rq)
5419
{
5420
	return cfs_rq_runnable_load_avg(&rq->cfs);
5421 5422
}

5423
#ifdef CONFIG_NO_HZ_COMMON
5424 5425
/*
 * There is no sane way to deal with nohz on smp when using jiffies because the
5426
 * CPU doing the jiffies update might drift wrt the CPU doing the jiffy reading
5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440
 * 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)
5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451
{
	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.
		 */
5452
		cpu_load_update(this_rq, load, pending_updates);
5453 5454 5455
	}
}

5456 5457 5458 5459
/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
5460
static void cpu_load_update_idle(struct rq *this_rq)
5461 5462 5463 5464
{
	/*
	 * bail if there's load or we're actually up-to-date.
	 */
5465
	if (weighted_cpuload(this_rq))
5466 5467
		return;

5468
	cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), 0);
5469 5470 5471
}

/*
5472 5473 5474 5475
 * 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.
5476
 */
5477
void cpu_load_update_nohz_start(void)
5478 5479
{
	struct rq *this_rq = this_rq();
5480 5481 5482 5483 5484 5485

	/*
	 * 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.
	 */
5486
	this_rq->cpu_load[0] = weighted_cpuload(this_rq);
5487 5488 5489 5490 5491 5492 5493
}

/*
 * Account the tickless load in the end of a nohz frame.
 */
void cpu_load_update_nohz_stop(void)
{
5494
	unsigned long curr_jiffies = READ_ONCE(jiffies);
5495 5496
	struct rq *this_rq = this_rq();
	unsigned long load;
5497
	struct rq_flags rf;
5498 5499 5500 5501

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

5502
	load = weighted_cpuload(this_rq);
5503
	rq_lock(this_rq, &rf);
5504
	update_rq_clock(this_rq);
5505
	cpu_load_update_nohz(this_rq, curr_jiffies, load);
5506
	rq_unlock(this_rq, &rf);
5507
}
5508 5509 5510 5511 5512 5513 5514 5515
#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)
{
5516
#ifdef CONFIG_NO_HZ_COMMON
5517 5518
	/* See the mess around cpu_load_update_nohz(). */
	this_rq->last_load_update_tick = READ_ONCE(jiffies);
5519
#endif
5520 5521
	cpu_load_update(this_rq, load, 1);
}
5522 5523 5524 5525

/*
 * Called from scheduler_tick()
 */
5526
void cpu_load_update_active(struct rq *this_rq)
5527
{
5528
	unsigned long load = weighted_cpuload(this_rq);
5529 5530 5531 5532 5533

	if (tick_nohz_tick_stopped())
		cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), load);
	else
		cpu_load_update_periodic(this_rq, load);
5534 5535
}

5536
/*
5537
 * Return a low guess at the load of a migration-source CPU weighted
5538 5539 5540 5541 5542 5543 5544 5545
 * 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);
5546
	unsigned long total = weighted_cpuload(rq);
5547 5548 5549 5550 5551 5552 5553 5554

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

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

/*
5555
 * Return a high guess at the load of a migration-target CPU weighted
5556 5557 5558 5559 5560
 * according to the scheduling class and "nice" value.
 */
static unsigned long target_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
5561
	unsigned long total = weighted_cpuload(rq);
5562 5563 5564 5565 5566 5567 5568

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

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

5569
static unsigned long capacity_of(int cpu)
5570
{
5571
	return cpu_rq(cpu)->cpu_capacity;
5572 5573
}

5574 5575 5576 5577 5578
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

5579 5580 5581
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5582
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
5583
	unsigned long load_avg = weighted_cpuload(rq);
5584 5585

	if (nr_running)
5586
		return load_avg / nr_running;
5587 5588 5589 5590

	return 0;
}

P
Peter Zijlstra 已提交
5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607
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
5608 5609
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
P
Peter Zijlstra 已提交
5610
 *
M
Mike Galbraith 已提交
5611
 * A waker of many should wake a different task than the one last awakened
P
Peter Zijlstra 已提交
5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623
 * 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
5624
 */
5625 5626
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
5627 5628
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
5629
	int factor = this_cpu_read(sd_llc_size);
5630

M
Mike Galbraith 已提交
5631 5632 5633 5634 5635
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
5636 5637
}

5638
/*
5639 5640 5641
 * 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.
5642
 *
5643 5644
 * wake_affine_idle() - only considers 'now', it check if the waking CPU is
 *			cache-affine and is (or	will be) idle.
5645 5646 5647 5648
 *
 * wake_affine_weight() - considers the weight to reflect the average
 *			  scheduling latency of the CPUs. This seems to work
 *			  for the overloaded case.
5649
 */
5650
static int
5651
wake_affine_idle(int this_cpu, int prev_cpu, int sync)
5652
{
5653 5654 5655 5656 5657
	/*
	 * 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.
5658 5659 5660 5661 5662 5663
	 *
	 * 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.
5664
	 */
5665 5666
	if (available_idle_cpu(this_cpu) && cpus_share_cache(this_cpu, prev_cpu))
		return available_idle_cpu(prev_cpu) ? prev_cpu : this_cpu;
5667

5668
	if (sync && cpu_rq(this_cpu)->nr_running == 1)
5669
		return this_cpu;
5670

5671
	return nr_cpumask_bits;
5672 5673
}

5674
static int
5675 5676
wake_affine_weight(struct sched_domain *sd, struct task_struct *p,
		   int this_cpu, int prev_cpu, int sync)
5677 5678 5679 5680
{
	s64 this_eff_load, prev_eff_load;
	unsigned long task_load;

5681 5682 5683 5684 5685
	if (sched_feat(WA_STATIC_WEIGHT))
		this_eff_load =
			scale_load_down(cpu_rq(this_cpu)->cfs.load.weight);
	else
		this_eff_load = target_load(this_cpu, sd->wake_idx);
5686 5687

	if (sync) {
5688 5689 5690 5691 5692 5693
		unsigned long current_load;

		if (sched_feat(WA_STATIC_WEIGHT))
			current_load = task_h_load_static(current);
		else
			current_load = task_h_load(current);
5694

5695
		if (current_load > this_eff_load)
5696
			return this_cpu;
5697

5698
		this_eff_load -= current_load;
5699 5700
	}

5701 5702 5703 5704
	if (sched_feat(WA_STATIC_WEIGHT))
		task_load = task_h_load_static(p);
	else
		task_load = task_h_load(p);
5705

5706 5707 5708 5709
	this_eff_load += task_load;
	if (sched_feat(WA_BIAS))
		this_eff_load *= 100;
	this_eff_load *= capacity_of(prev_cpu);
5710

5711 5712 5713 5714 5715
	if (sched_feat(WA_STATIC_WEIGHT))
		prev_eff_load =
			scale_load_down(cpu_rq(prev_cpu)->cfs.load.weight);
	else
		prev_eff_load = source_load(prev_cpu, sd->wake_idx);
5716 5717 5718 5719
	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);
5720

5721 5722 5723 5724 5725 5726 5727 5728 5729 5730
	/*
	 * 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;
5731 5732
}

5733
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
5734
		       int this_cpu, int prev_cpu, int sync)
5735
{
5736
	int target = nr_cpumask_bits;
5737

5738
	if (sched_feat(WA_IDLE))
5739
		target = wake_affine_idle(this_cpu, prev_cpu, sync);
5740

5741 5742
	if (sched_feat(WA_WEIGHT) && target == nr_cpumask_bits)
		target = wake_affine_weight(sd, p, this_cpu, prev_cpu, sync);
5743

5744
	schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts);
5745 5746
	if (target == nr_cpumask_bits)
		return prev_cpu;
5747

5748 5749 5750
	schedstat_inc(sd->ttwu_move_affine);
	schedstat_inc(p->se.statistics.nr_wakeups_affine);
	return target;
5751 5752
}

5753
static unsigned long cpu_util_without(int cpu, struct task_struct *p);
5754

5755
static unsigned long capacity_spare_without(int cpu, struct task_struct *p)
5756
{
5757
	return max_t(long, capacity_of(cpu) - cpu_util_without(cpu, p), 0);
5758 5759
}

5760 5761 5762
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
5763 5764
 *
 * Assumes p is allowed on at least one CPU in sd.
5765 5766
 */
static struct sched_group *
P
Peter Zijlstra 已提交
5767
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
5768
		  int this_cpu, int sd_flag)
5769
{
5770
	struct sched_group *idlest = NULL, *group = sd->groups;
5771
	struct sched_group *most_spare_sg = NULL;
5772 5773 5774
	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;
5775
	unsigned long most_spare = 0, this_spare = 0;
5776
	int load_idx = sd->forkexec_idx;
5777 5778 5779
	int imbalance_scale = 100 + (sd->imbalance_pct-100)/2;
	unsigned long imbalance = scale_load_down(NICE_0_LOAD) *
				(sd->imbalance_pct-100) / 100;
5780

5781 5782 5783
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

5784
	do {
5785 5786
		unsigned long load, avg_load, runnable_load;
		unsigned long spare_cap, max_spare_cap;
5787 5788
		int local_group;
		int i;
5789

5790
		/* Skip over this group if it has no CPUs allowed */
5791
		if (!cpumask_intersects(sched_group_span(group),
5792
					&p->cpus_allowed))
5793 5794 5795
			continue;

		local_group = cpumask_test_cpu(this_cpu,
5796
					       sched_group_span(group));
5797

5798 5799 5800 5801
		/*
		 * Tally up the load of all CPUs in the group and find
		 * the group containing the CPU with most spare capacity.
		 */
5802
		avg_load = 0;
5803
		runnable_load = 0;
5804
		max_spare_cap = 0;
5805

5806
		for_each_cpu(i, sched_group_span(group)) {
5807
			/* Bias balancing toward CPUs of our domain */
5808 5809 5810 5811 5812
			if (local_group)
				load = source_load(i, load_idx);
			else
				load = target_load(i, load_idx);

5813 5814 5815
			runnable_load += load;

			avg_load += cfs_rq_load_avg(&cpu_rq(i)->cfs);
5816

5817
			spare_cap = capacity_spare_without(i, p);
5818 5819 5820

			if (spare_cap > max_spare_cap)
				max_spare_cap = spare_cap;
5821 5822
		}

5823
		/* Adjust by relative CPU capacity of the group */
5824 5825 5826 5827
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
		runnable_load = (runnable_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
5828 5829

		if (local_group) {
5830 5831
			this_runnable_load = runnable_load;
			this_avg_load = avg_load;
5832 5833
			this_spare = max_spare_cap;
		} else {
5834 5835 5836
			if (min_runnable_load > (runnable_load + imbalance)) {
				/*
				 * The runnable load is significantly smaller
5837
				 * so we can pick this new CPU:
5838 5839 5840 5841 5842 5843 5844 5845
				 */
				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
5846
				 * blocked load into account through avg_load:
5847 5848
				 */
				min_avg_load = avg_load;
5849 5850 5851 5852 5853 5854 5855
				idlest = group;
			}

			if (most_spare < max_spare_cap) {
				most_spare = max_spare_cap;
				most_spare_sg = group;
			}
5856 5857 5858
		}
	} while (group = group->next, group != sd->groups);

5859 5860 5861 5862 5863 5864
	/*
	 * 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.
5865 5866 5867 5868
	 *
	 * 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.
5869
	 */
5870 5871 5872
	if (sd_flag & SD_BALANCE_FORK)
		goto skip_spare;

5873
	if (this_spare > task_util(p) / 2 &&
5874
	    imbalance_scale*this_spare > 100*most_spare)
5875
		return NULL;
5876 5877

	if (most_spare > task_util(p) / 2)
5878 5879
		return most_spare_sg;

5880
skip_spare:
5881 5882 5883
	if (!idlest)
		return NULL;

5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895
	/*
	 * 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;

5896
	if (min_runnable_load > (this_runnable_load + imbalance))
5897
		return NULL;
5898 5899 5900 5901 5902

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

5903 5904 5905 5906
	return idlest;
}

/*
5907
 * find_idlest_group_cpu - find the idlest CPU among the CPUs in the group.
5908 5909
 */
static int
5910
find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
5911 5912
{
	unsigned long load, min_load = ULONG_MAX;
5913 5914 5915 5916
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
5917 5918
	int i;

5919 5920
	/* Check if we have any choice: */
	if (group->group_weight == 1)
5921
		return cpumask_first(sched_group_span(group));
5922

5923
	/* Traverse only the allowed CPUs */
5924
	for_each_cpu_and(i, sched_group_span(group), &p->cpus_allowed) {
5925
		if (available_idle_cpu(i)) {
5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946
			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;
			}
5947
		} else if (shallowest_idle_cpu == -1) {
5948
			load = weighted_cpuload(cpu_rq(i));
5949
			if (load < min_load) {
5950 5951 5952
				min_load = load;
				least_loaded_cpu = i;
			}
5953 5954 5955
		}
	}

5956
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
5957
}
5958

5959 5960 5961
static inline int find_idlest_cpu(struct sched_domain *sd, struct task_struct *p,
				  int cpu, int prev_cpu, int sd_flag)
{
5962
	int new_cpu = cpu;
5963

5964 5965 5966
	if (!cpumask_intersects(sched_domain_span(sd), &p->cpus_allowed))
		return prev_cpu;

5967
	/*
5968 5969
	 * We need task's util for capacity_spare_without, sync it up to
	 * prev_cpu's last_update_time.
5970 5971 5972 5973
	 */
	if (!(sd_flag & SD_BALANCE_FORK))
		sync_entity_load_avg(&p->se);

5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990
	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);
5991
		if (new_cpu == cpu) {
5992
			/* Now try balancing at a lower domain level of 'cpu': */
5993 5994 5995 5996
			sd = sd->child;
			continue;
		}

5997
		/* Now try balancing at a lower domain level of 'new_cpu': */
5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011
		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;
}

6012
#ifdef CONFIG_SCHED_SMT
6013
DEFINE_STATIC_KEY_FALSE(sched_smt_present);
6014
EXPORT_SYMBOL_GPL(sched_smt_present);
6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042

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
6043
void __update_idle_core(struct rq *rq)
6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055
{
	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;

6056
		if (!available_idle_cpu(cpu))
6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072
			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);
6073
	int core, cpu;
6074

P
Peter Zijlstra 已提交
6075 6076 6077
	if (!static_branch_likely(&sched_smt_present))
		return -1;

6078 6079 6080
	if (!test_idle_cores(target, false))
		return -1;

6081
	cpumask_and(cpus, sched_domain_span(sd), &p->cpus_allowed);
6082

6083
	for_each_cpu_wrap(core, cpus, target) {
6084 6085 6086 6087
		bool idle = true;

		for_each_cpu(cpu, cpu_smt_mask(core)) {
			cpumask_clear_cpu(cpu, cpus);
6088
			if (!available_idle_cpu(cpu))
6089 6090 6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110
				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
6111 6112 6113
	if (!static_branch_likely(&sched_smt_present))
		return -1;

6114
	for_each_cpu(cpu, cpu_smt_mask(target)) {
6115
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6116
			continue;
6117
		if (available_idle_cpu(cpu))
6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141
			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).
6142
 */
6143 6144
static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int target)
{
6145
	struct sched_domain *this_sd;
6146
	u64 avg_cost, avg_idle;
6147 6148
	u64 time, cost;
	s64 delta;
6149
	int cpu, nr = INT_MAX;
6150

6151 6152 6153 6154
	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
	if (!this_sd)
		return -1;

6155 6156 6157 6158
	/*
	 * Due to large variance we need a large fuzz factor; hackbench in
	 * particularly is sensitive here.
	 */
6159 6160 6161 6162
	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)
6163 6164
		return -1;

6165 6166 6167 6168 6169 6170 6171 6172
	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;
	}

6173 6174
	time = local_clock();

6175
	for_each_cpu_wrap(cpu, sched_domain_span(sd), target) {
6176 6177
		if (!--nr)
			return -1;
6178
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6179
			continue;
6180
		if (available_idle_cpu(cpu))
6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193
			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.
6194
 */
6195
static int select_idle_sibling(struct task_struct *p, int prev, int target)
6196
{
6197
	struct sched_domain *sd;
6198
	int i, recent_used_cpu;
6199

6200
	if (available_idle_cpu(target))
6201
		return target;
6202 6203

	/*
6204
	 * If the previous CPU is cache affine and idle, don't be stupid:
6205
	 */
6206
	if (prev != target && cpus_share_cache(prev, target) && available_idle_cpu(prev))
6207
		return prev;
6208

6209
	/* Check a recently used CPU as a potential idle candidate: */
6210 6211 6212 6213
	recent_used_cpu = p->recent_used_cpu;
	if (recent_used_cpu != prev &&
	    recent_used_cpu != target &&
	    cpus_share_cache(recent_used_cpu, target) &&
6214
	    available_idle_cpu(recent_used_cpu) &&
6215 6216 6217
	    cpumask_test_cpu(p->recent_used_cpu, &p->cpus_allowed)) {
		/*
		 * Replace recent_used_cpu with prev as it is a potential
6218
		 * candidate for the next wake:
6219 6220 6221 6222 6223
		 */
		p->recent_used_cpu = prev;
		return recent_used_cpu;
	}

6224
	sd = rcu_dereference(per_cpu(sd_llc, target));
6225 6226
	if (!sd)
		return target;
6227

6228 6229 6230
	i = select_idle_core(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
6231

6232 6233 6234 6235 6236 6237 6238
	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;
6239

6240 6241
	return target;
}
6242

6243 6244 6245 6246 6247 6248 6249
/**
 * 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).
6250 6251 6252 6253 6254 6255 6256 6257 6258 6259
 *
 * 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.
 *
6260 6261 6262 6263 6264 6265 6266 6267
 * 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.
 *
6268 6269 6270 6271 6272 6273 6274 6275 6276 6277
 * 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).
6278 6279
 *
 * Return: the (estimated) utilization for the specified CPU
6280
 */
6281
static inline unsigned long cpu_util(int cpu)
6282
{
6283 6284 6285 6286 6287 6288 6289 6290
	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));
6291

6292
	return min_t(unsigned long, util, capacity_orig_of(cpu));
6293
}
6294

6295
/*
6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306
 * cpu_util_without: compute cpu utilization without any contributions from *p
 * @cpu: the CPU which utilization is requested
 * @p: the task which utilization should be discounted
 *
 * The utilization of a CPU is defined by the utilization of tasks currently
 * enqueued on that CPU as well as tasks which are currently sleeping after an
 * execution on that CPU.
 *
 * This method returns the utilization of the specified CPU by discounting the
 * utilization of the specified task, whenever the task is currently
 * contributing to the CPU utilization.
6307
 */
6308
static unsigned long cpu_util_without(int cpu, struct task_struct *p)
6309
{
6310 6311
	struct cfs_rq *cfs_rq;
	unsigned int util;
6312 6313

	/* Task has no contribution or is new */
6314
	if (cpu != task_cpu(p) || !READ_ONCE(p->se.avg.last_update_time))
6315 6316
		return cpu_util(cpu);

6317 6318 6319
	cfs_rq = &cpu_rq(cpu)->cfs;
	util = READ_ONCE(cfs_rq->avg.util_avg);

6320
	/* Discount task's util from CPU's util */
6321
	util -= min_t(unsigned int, util, task_util(p));
6322

6323 6324 6325 6326 6327 6328
	/*
	 * 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:
6329
	 *      cpu_util_without = (cpu_util - task_util) = 0
6330 6331 6332 6333 6334 6335
	 *
	 * 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:
6336
	 *      cpu_util_without = (cpu_util - task_util) >= 0
6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348
	 *
	 * 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.
	 */
6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375
	if (sched_feat(UTIL_EST)) {
		unsigned int estimated =
			READ_ONCE(cfs_rq->avg.util_est.enqueued);

		/*
		 * Despite the following checks we still have a small window
		 * for a possible race, when an execl's select_task_rq_fair()
		 * races with LB's detach_task():
		 *
		 *   detach_task()
		 *     p->on_rq = TASK_ON_RQ_MIGRATING;
		 *     ---------------------------------- A
		 *     deactivate_task()                   \
		 *       dequeue_task()                     + RaceTime
		 *         util_est_dequeue()              /
		 *     ---------------------------------- B
		 *
		 * The additional check on "current == p" it's required to
		 * properly fix the execl regression and it helps in further
		 * reducing the chances for the above race.
		 */
		if (unlikely(task_on_rq_queued(p) || current == p)) {
			estimated -= min_t(unsigned int, estimated,
					   (_task_util_est(p) | UTIL_AVG_UNCHANGED));
		}
		util = max(util, estimated);
	}
6376 6377 6378 6379 6380 6381 6382

	/*
	 * 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));
6383 6384
}

6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402
/*
 * 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;

6403 6404 6405
	/* Bring task utilization in sync with prev_cpu */
	sync_entity_load_avg(&p->se);

6406 6407 6408
	return min_cap * 1024 < task_util(p) * capacity_margin;
}

6409
/*
6410 6411 6412
 * 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.
6413
 *
6414 6415
 * 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.
6416
 *
6417
 * Returns the target CPU number.
6418 6419 6420
 *
 * preempt must be disabled.
 */
6421
static int
6422
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
6423
{
6424
	struct sched_domain *tmp, *sd = NULL;
6425
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
6426
	int new_cpu = prev_cpu;
6427
	int want_affine = 0;
6428
	int sync = (wake_flags & WF_SYNC) && !(current->flags & PF_EXITING);
6429

P
Peter Zijlstra 已提交
6430 6431
	if (sd_flag & SD_BALANCE_WAKE) {
		record_wakee(p);
6432
		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu)
6433
			      && cpumask_test_cpu(cpu, &p->cpus_allowed);
P
Peter Zijlstra 已提交
6434
	}
6435

6436
	rcu_read_lock();
6437
	for_each_domain(cpu, tmp) {
6438
		if (!(tmp->flags & SD_LOAD_BALANCE))
M
Mike Galbraith 已提交
6439
			break;
6440

6441
		/*
6442
		 * If both 'cpu' and 'prev_cpu' are part of this domain,
6443
		 * cpu is a valid SD_WAKE_AFFINE target.
6444
		 */
6445 6446
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
6447 6448 6449 6450
			if (cpu != prev_cpu)
				new_cpu = wake_affine(tmp, p, cpu, prev_cpu, sync);

			sd = NULL; /* Prefer wake_affine over balance flags */
6451
			break;
6452
		}
6453

6454
		if (tmp->flags & sd_flag)
6455
			sd = tmp;
M
Mike Galbraith 已提交
6456 6457
		else if (!want_affine)
			break;
6458 6459
	}

6460 6461
	if (unlikely(sd)) {
		/* Slow path */
6462
		new_cpu = find_idlest_cpu(sd, p, cpu, prev_cpu, sd_flag);
6463 6464 6465 6466 6467 6468 6469
	} 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;
6470
	}
6471
	rcu_read_unlock();
6472

6473
	return new_cpu;
6474
}
6475

6476 6477
static void detach_entity_cfs_rq(struct sched_entity *se);

6478
/*
6479
 * Called immediately before a task is migrated to a new CPU; task_cpu(p) and
6480
 * cfs_rq_of(p) references at time of call are still valid and identify the
6481
 * previous CPU. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
6482
 */
6483
static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
6484
{
6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510
	/*
	 * 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;
	}

6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529
	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);
	}
6530 6531 6532

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
6533 6534

	/* We have migrated, no longer consider this task hot */
6535
	p->se.exec_start = 0;
6536 6537

	update_scan_period(p, new_cpu);
6538
}
6539 6540 6541 6542 6543

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
6544 6545
#endif /* CONFIG_SMP */

6546
static unsigned long wakeup_gran(struct sched_entity *se)
6547 6548 6549 6550
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
6551 6552
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
6553 6554 6555 6556 6557 6558 6559 6560 6561
	 *
	 * 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.
6562
	 */
6563
	return calc_delta_fair(gran, se);
6564 6565
}

6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587
/*
 * 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;

6588
	gran = wakeup_gran(se);
6589 6590 6591 6592 6593 6594
	if (vdiff > gran)
		return 1;

	return 0;
}

6595 6596
static void set_last_buddy(struct sched_entity *se)
{
6597 6598 6599
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6600 6601 6602
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6603
		cfs_rq_of(se)->last = se;
6604
	}
6605 6606 6607 6608
}

static void set_next_buddy(struct sched_entity *se)
{
6609 6610 6611
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6612 6613 6614
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6615
		cfs_rq_of(se)->next = se;
6616
	}
6617 6618
}

6619 6620
static void set_skip_buddy(struct sched_entity *se)
{
6621 6622
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
6623 6624
}

6625 6626 6627
/*
 * Preempt the current task with a newly woken task if needed:
 */
6628
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
6629 6630
{
	struct task_struct *curr = rq->curr;
6631
	struct sched_entity *se = &curr->se, *pse = &p->se;
6632
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
6633
	int scale = cfs_rq->nr_running >= sched_nr_latency;
6634
	int next_buddy_marked = 0;
6635

I
Ingo Molnar 已提交
6636 6637 6638
	if (unlikely(se == pse))
		return;

6639
	/*
6640
	 * This is possible from callers such as attach_tasks(), in which we
6641 6642 6643 6644 6645 6646 6647
	 * 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;

6648
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
6649
		set_next_buddy(pse);
6650 6651
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
6652

6653 6654 6655
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
6656 6657 6658 6659 6660 6661
	 *
	 * 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.
6662 6663 6664 6665
	 */
	if (test_tsk_need_resched(curr))
		return;

6666 6667 6668 6669 6670
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

6671
	/*
6672 6673
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
6674
	 */
6675
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
6676
		return;
6677

6678
	find_matching_se(&se, &pse);
6679
	update_curr(cfs_rq_of(se));
6680
	BUG_ON(!pse);
6681 6682 6683 6684 6685 6686 6687
	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);
6688
		goto preempt;
6689
	}
6690

6691
	return;
6692

6693
preempt:
6694
	resched_curr(rq);
6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708
	/*
	 * 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);
6709 6710
}

6711
static struct task_struct *
6712
pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
6713 6714 6715
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
6716
	struct task_struct *p;
6717
	int new_tasks;
6718

6719
again:
6720
	if (!cfs_rq->nr_running)
6721
		goto idle;
6722

6723
#ifdef CONFIG_FAIR_GROUP_SCHED
6724
	if (prev->sched_class != &fair_sched_class)
6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743
		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.
		 */
6744 6745 6746 6747 6748
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
6749

6750 6751 6752
			/*
			 * This call to check_cfs_rq_runtime() will do the
			 * throttle and dequeue its entity in the parent(s).
6753
			 * Therefore the nr_running test will indeed
6754 6755
			 * be correct.
			 */
6756 6757 6758 6759 6760 6761
			if (unlikely(check_cfs_rq_runtime(cfs_rq))) {
				cfs_rq = &rq->cfs;

				if (!cfs_rq->nr_running)
					goto idle;

6762
				goto simple;
6763
			}
6764
		}
6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797

		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);
	}

6798
	goto done;
6799 6800
simple:
#endif
6801

6802
	put_prev_task(rq, prev);
6803

6804
	do {
6805
		se = pick_next_entity(cfs_rq, NULL);
6806
		set_next_entity(cfs_rq, se);
6807 6808 6809
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
6810
	p = task_of(se);
6811

6812
done: __maybe_unused;
6813 6814 6815 6816 6817 6818 6819 6820 6821
#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

6822 6823
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
6824 6825

	return p;
6826 6827

idle:
6828 6829
	new_tasks = idle_balance(rq, rf);

6830 6831 6832 6833 6834
	/*
	 * 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.
	 */
6835
	if (new_tasks < 0)
6836 6837
		return RETRY_TASK;

6838
	if (new_tasks > 0)
6839 6840 6841
		goto again;

	return NULL;
6842 6843 6844 6845 6846
}

/*
 * Account for a descheduled task:
 */
6847
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
6848 6849 6850 6851 6852 6853
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
6854
		put_prev_entity(cfs_rq, se);
6855 6856 6857
	}
}

6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882
/*
 * 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);
6883 6884 6885 6886 6887
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
6888
		rq_clock_skip_update(rq);
6889 6890 6891 6892 6893
	}

	set_skip_buddy(se);
}

6894 6895 6896 6897
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

6898 6899
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
6900 6901 6902 6903 6904 6905 6906 6907 6908 6909
		return false;

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

	yield_task_fair(rq);

	return true;
}

6910
#ifdef CONFIG_SMP
6911
/**************************************************
P
Peter Zijlstra 已提交
6912 6913 6914 6915 6916
 * Fair scheduling class load-balancing methods.
 *
 * BASICS
 *
 * The purpose of load-balancing is to achieve the same basic fairness the
6917
 * per-CPU scheduler provides, namely provide a proportional amount of compute
P
Peter Zijlstra 已提交
6918 6919 6920 6921
 * 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)
 *
6922
 * Where W_i,n is the n-th weight average for CPU i. The instantaneous weight
P
Peter Zijlstra 已提交
6923 6924 6925 6926
 * W_i,0 is defined as:
 *
 *   W_i,0 = \Sum_j w_i,j                                             (2)
 *
6927
 * Where w_i,j is the weight of the j-th runnable task on CPU i. This weight
6928
 * is derived from the nice value as per sched_prio_to_weight[].
P
Peter Zijlstra 已提交
6929 6930 6931 6932 6933 6934
 *
 * 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)
 *
6935
 * C_i is the compute capacity of CPU i, typically it is the
P
Peter Zijlstra 已提交
6936 6937 6938 6939 6940 6941
 * 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):
 *
6942
 *   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
6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955
 *
 * 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)
6956
 * for all i,j solution, we create a tree of CPUs that follows the hardware
P
Peter Zijlstra 已提交
6957
 * topology where each level pairs two lower groups (or better). This results
6958
 * in O(log n) layers. Furthermore we reduce the number of CPUs going up the
P
Peter Zijlstra 已提交
6959
 * tree to only the first of the previous level and we decrease the frequency
6960
 * of load-balance at each level inv. proportional to the number of CPUs in
P
Peter Zijlstra 已提交
6961 6962 6963 6964 6965 6966 6967 6968
 * 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
6969
 *         |         |     `- number of CPUs doing load-balance
P
Peter Zijlstra 已提交
6970 6971 6972 6973 6974 6975 6976
 *         |         `- 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
6977
 * to every other CPU in at most O(log n) steps:
P
Peter Zijlstra 已提交
6978 6979 6980
 *
 * The adjacency matrix of the resulting graph is given by:
 *
6981
 *             log_2 n
P
Peter Zijlstra 已提交
6982 6983 6984 6985 6986 6987 6988
 *   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)
 *
6989
 * Showing there's indeed a path between every CPU in at most O(log n) steps.
P
Peter Zijlstra 已提交
6990 6991 6992 6993 6994 6995 6996 6997 6998
 * 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
6999
 * balancing is more aggressive and has the newly idle CPU iterate up the domain
P
Peter Zijlstra 已提交
7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019
 * 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)
 *
7020
 * 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
7021 7022 7023 7024 7025 7026
 *
 * 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.]
7027
 */
7028

7029 7030
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

7031 7032
enum fbq_type { regular, remote, all };

7033
#define LBF_ALL_PINNED	0x01
7034
#define LBF_NEED_BREAK	0x02
7035 7036
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
7037
#define LBF_NOHZ_STATS	0x10
7038
#define LBF_NOHZ_AGAIN	0x20
7039 7040 7041 7042 7043

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
7044
	int			src_cpu;
7045 7046 7047 7048

	int			dst_cpu;
	struct rq		*dst_rq;

7049 7050
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
7051
	enum cpu_idle_type	idle;
7052
	long			imbalance;
7053 7054 7055
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

7056
	unsigned int		flags;
7057 7058 7059 7060

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
7061 7062

	enum fbq_type		fbq_type;
7063
	struct list_head	tasks;
7064 7065
};

7066 7067 7068
/*
 * Is this task likely cache-hot:
 */
7069
static int task_hot(struct task_struct *p, struct lb_env *env)
7070 7071 7072
{
	s64 delta;

7073 7074
	lockdep_assert_held(&env->src_rq->lock);

7075 7076 7077 7078 7079 7080 7081 7082 7083
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
7084
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
7085 7086 7087 7088 7089 7090 7091 7092 7093
			(&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;

7094
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
7095 7096 7097 7098

	return delta < (s64)sysctl_sched_migration_cost;
}

7099
#ifdef CONFIG_NUMA_BALANCING
7100
/*
7101 7102 7103
 * 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.
7104
 */
7105
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
7106
{
7107
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
7108 7109
	unsigned long src_weight, dst_weight;
	int src_nid, dst_nid, dist;
7110

7111
	if (!static_branch_likely(&sched_numa_balancing))
7112 7113
		return -1;

7114
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
7115
		return -1;
7116 7117 7118 7119

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

7120
	if (src_nid == dst_nid)
7121
		return -1;
7122

7123 7124 7125 7126 7127 7128 7129
	/* 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;
	}
7130

7131 7132
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
7133
		return 0;
7134

7135
	/* Leaving a core idle is often worse than degrading locality. */
7136
	if (env->idle == CPU_IDLE)
7137 7138
		return -1;

7139
	dist = node_distance(src_nid, dst_nid);
7140
	if (numa_group) {
7141 7142
		src_weight = group_weight(p, src_nid, dist);
		dst_weight = group_weight(p, dst_nid, dist);
7143
	} else {
7144 7145
		src_weight = task_weight(p, src_nid, dist);
		dst_weight = task_weight(p, dst_nid, dist);
7146 7147
	}

7148
	return dst_weight < src_weight;
7149 7150
}

7151
#else
7152
static inline int migrate_degrades_locality(struct task_struct *p,
7153 7154
					     struct lb_env *env)
{
7155
	return -1;
7156
}
7157 7158
#endif

7159 7160 7161 7162
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
7163
int can_migrate_task(struct task_struct *p, struct lb_env *env)
7164
{
7165
	int tsk_cache_hot;
7166 7167 7168

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

7169 7170
	/*
	 * We do not migrate tasks that are:
7171
	 * 1) throttled_lb_pair, or
7172
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
7173 7174
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
7175
	 */
7176 7177 7178
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

7179
	if (!cpumask_test_cpu(env->dst_cpu, &p->cpus_allowed)) {
7180
		int cpu;
7181

7182
		schedstat_inc(p->se.statistics.nr_failed_migrations_affine);
7183

7184 7185
		env->flags |= LBF_SOME_PINNED;

7186
		/*
7187
		 * Remember if this task can be migrated to any other CPU in
7188 7189 7190
		 * our sched_group. We may want to revisit it if we couldn't
		 * meet load balance goals by pulling other tasks on src_cpu.
		 *
7191 7192
		 * Avoid computing new_dst_cpu for NEWLY_IDLE or if we have
		 * already computed one in current iteration.
7193
		 */
7194
		if (env->idle == CPU_NEWLY_IDLE || (env->flags & LBF_DST_PINNED))
7195 7196
			return 0;

7197
		/* Prevent to re-select dst_cpu via env's CPUs: */
7198
		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
7199
			if (cpumask_test_cpu(cpu, &p->cpus_allowed)) {
7200
				env->flags |= LBF_DST_PINNED;
7201 7202 7203
				env->new_dst_cpu = cpu;
				break;
			}
7204
		}
7205

7206 7207
		return 0;
	}
7208 7209

	/* Record that we found atleast one task that could run on dst_cpu */
7210
	env->flags &= ~LBF_ALL_PINNED;
7211

7212
	if (task_running(env->src_rq, p)) {
7213
		schedstat_inc(p->se.statistics.nr_failed_migrations_running);
7214 7215 7216 7217 7218
		return 0;
	}

	/*
	 * Aggressive migration if:
7219 7220 7221
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
7222
	 */
7223 7224 7225
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
7226

7227
	if (tsk_cache_hot <= 0 ||
7228
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
7229
		if (tsk_cache_hot == 1) {
7230 7231
			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
			schedstat_inc(p->se.statistics.nr_forced_migrations);
7232
		}
7233 7234 7235
		return 1;
	}

7236
	schedstat_inc(p->se.statistics.nr_failed_migrations_hot);
Z
Zhang Hang 已提交
7237
	return 0;
7238 7239
}

7240
/*
7241 7242 7243 7244 7245 7246 7247
 * 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;
7248
	deactivate_task(env->src_rq, p, DEQUEUE_NOCLOCK);
7249 7250 7251
	set_task_cpu(p, env->dst_cpu);
}

7252
/*
7253
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
7254 7255
 * part of active balancing operations within "domain".
 *
7256
 * Returns a task if successful and NULL otherwise.
7257
 */
7258
static struct task_struct *detach_one_task(struct lb_env *env)
7259
{
7260
	struct task_struct *p;
7261

7262 7263
	lockdep_assert_held(&env->src_rq->lock);

7264 7265
	list_for_each_entry_reverse(p,
			&env->src_rq->cfs_tasks, se.group_node) {
7266 7267
		if (!can_migrate_task(p, env))
			continue;
7268

7269
		detach_task(p, env);
7270

7271
		/*
7272
		 * Right now, this is only the second place where
7273
		 * lb_gained[env->idle] is updated (other is detach_tasks)
7274
		 * so we can safely collect stats here rather than
7275
		 * inside detach_tasks().
7276
		 */
7277
		schedstat_inc(env->sd->lb_gained[env->idle]);
7278
		return p;
7279
	}
7280
	return NULL;
7281 7282
}

7283 7284
static const unsigned int sched_nr_migrate_break = 32;

7285
/*
7286 7287
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
7288
 *
7289
 * Returns number of detached tasks if successful and 0 otherwise.
7290
 */
7291
static int detach_tasks(struct lb_env *env)
7292
{
7293 7294
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
7295
	unsigned long load;
7296 7297 7298
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
7299

7300
	if (env->imbalance <= 0)
7301
		return 0;
7302

7303
	while (!list_empty(tasks)) {
7304 7305 7306 7307 7308 7309 7310
		/*
		 * 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;

7311
		p = list_last_entry(tasks, struct task_struct, se.group_node);
7312

7313 7314
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
7315
		if (env->loop > env->loop_max)
7316
			break;
7317 7318

		/* take a breather every nr_migrate tasks */
7319
		if (env->loop > env->loop_break) {
7320
			env->loop_break += sched_nr_migrate_break;
7321
			env->flags |= LBF_NEED_BREAK;
7322
			break;
7323
		}
7324

7325
		if (!can_migrate_task(p, env))
7326 7327 7328
			goto next;

		load = task_h_load(p);
7329

7330
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
7331 7332
			goto next;

7333
		if ((load / 2) > env->imbalance)
7334
			goto next;
7335

7336 7337 7338 7339
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
7340
		env->imbalance -= load;
7341 7342

#ifdef CONFIG_PREEMPT
7343 7344
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
7345
		 * kernels will stop after the first task is detached to minimize
7346 7347
		 * the critical section.
		 */
7348
		if (env->idle == CPU_NEWLY_IDLE)
7349
			break;
7350 7351
#endif

7352 7353 7354 7355
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
7356
		if (env->imbalance <= 0)
7357
			break;
7358 7359 7360

		continue;
next:
7361
		list_move(&p->se.group_node, tasks);
7362
	}
7363

7364
	/*
7365 7366 7367
	 * 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().
7368
	 */
7369
	schedstat_add(env->sd->lb_gained[env->idle], detached);
7370

7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381
	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);
7382
	activate_task(rq, p, ENQUEUE_NOCLOCK);
7383
	p->on_rq = TASK_ON_RQ_QUEUED;
7384 7385 7386 7387 7388 7389 7390 7391 7392
	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)
{
7393 7394 7395
	struct rq_flags rf;

	rq_lock(rq, &rf);
7396
	update_rq_clock(rq);
7397
	attach_task(rq, p);
7398
	rq_unlock(rq, &rf);
7399 7400 7401 7402 7403 7404 7405 7406 7407 7408
}

/*
 * 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;
7409
	struct rq_flags rf;
7410

7411
	rq_lock(env->dst_rq, &rf);
7412
	update_rq_clock(env->dst_rq);
7413 7414 7415 7416

	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
		list_del_init(&p->se.group_node);
7417

7418 7419 7420
		attach_task(env->dst_rq, p);
	}

7421
	rq_unlock(env->dst_rq, &rf);
7422 7423
}

7424 7425 7426 7427 7428 7429 7430 7431 7432 7433 7434
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;
}

7435
static inline bool others_have_blocked(struct rq *rq)
7436 7437 7438 7439
{
	if (READ_ONCE(rq->avg_rt.util_avg))
		return true;

7440 7441 7442
	if (READ_ONCE(rq->avg_dl.util_avg))
		return true;

7443
#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
7444 7445 7446 7447
	if (READ_ONCE(rq->avg_irq.util_avg))
		return true;
#endif

7448 7449 7450
	return false;
}

7451 7452
#ifdef CONFIG_FAIR_GROUP_SCHED

7453
static void update_blocked_averages(int cpu)
7454 7455
{
	struct rq *rq = cpu_rq(cpu);
7456
	struct cfs_rq *cfs_rq;
7457
	const struct sched_class *curr_class;
7458
	struct rq_flags rf;
7459
	bool done = true;
7460

7461
	rq_lock_irqsave(rq, &rf);
7462
	update_rq_clock(rq);
7463

7464 7465 7466 7467
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
7468
	for_each_leaf_cfs_rq(rq, cfs_rq) {
7469 7470
		struct sched_entity *se;

7471 7472 7473
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
7474

7475
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq))
7476
			update_tg_load_avg(cfs_rq, 0);
7477

7478 7479 7480
		/* Propagate pending load changes to the parent, if any: */
		se = cfs_rq->tg->se[cpu];
		if (se && !skip_blocked_update(se))
7481
			update_load_avg(cfs_rq_of(se), se, 0);
7482

7483 7484
		/* Don't need periodic decay once load/util_avg are null */
		if (cfs_rq_has_blocked(cfs_rq))
7485
			done = false;
7486
	}
7487 7488 7489 7490

	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);
7491
	update_irq_load_avg(rq, 0);
7492
	/* Don't need periodic decay once load/util_avg are null */
7493
	if (others_have_blocked(rq))
7494
		done = false;
7495 7496 7497

#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7498 7499
	if (done)
		rq->has_blocked_load = 0;
7500
#endif
7501
	rq_unlock_irqrestore(rq, &rf);
7502 7503
}

7504
/*
7505
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
7506 7507 7508
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
7509
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
7510
{
7511 7512
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
7513
	unsigned long now = jiffies;
7514
	unsigned long load;
7515

7516
	if (cfs_rq->last_h_load_update == now)
7517 7518
		return;

7519
	WRITE_ONCE(cfs_rq->h_load_next, NULL);
7520 7521
	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
7522
		WRITE_ONCE(cfs_rq->h_load_next, se);
7523 7524 7525
		if (cfs_rq->last_h_load_update == now)
			break;
	}
7526

7527
	if (!se) {
7528
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
7529 7530 7531
		cfs_rq->last_h_load_update = now;
	}

7532
	while ((se = READ_ONCE(cfs_rq->h_load_next)) != NULL) {
7533
		load = cfs_rq->h_load;
7534 7535
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
7536 7537 7538 7539
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
7540 7541
}

7542
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
7543
{
7544
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
7545

7546
	update_cfs_rq_h_load(cfs_rq);
7547
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
7548
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
7549
}
7550 7551 7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566 7567 7568 7569 7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580 7581 7582 7583 7584 7585 7586 7587 7588 7589 7590 7591

static void update_cfs_rq_h_load_static(struct cfs_rq *cfs_rq)
{
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
	unsigned long now = jiffies;
	unsigned long load;

	if (cfs_rq->last_h_load_update == now)
		return;

	WRITE_ONCE(cfs_rq->h_load_next, NULL);
	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
		WRITE_ONCE(cfs_rq->h_load_next, se);
		if (cfs_rq->last_h_load_update == now)
			break;
	}

	if (!se) {
		cfs_rq->h_load = scale_load_down(cfs_rq->load.weight);
		cfs_rq->last_h_load_update = now;
	}

	while ((se = READ_ONCE(cfs_rq->h_load_next)) != NULL) {
		load = cfs_rq->h_load;
		load = div64_ul(load * se->load.weight,
			cfs_rq->load.weight + 1);
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
}

static unsigned long task_h_load_static(struct task_struct *p)
{
	struct cfs_rq *cfs_rq = task_cfs_rq(p);

	update_cfs_rq_h_load_static(cfs_rq);
	return div64_ul(p->se.load.weight * cfs_rq->h_load,
			cfs_rq->load.weight + 1);
}
P
Peter Zijlstra 已提交
7592
#else
7593
static inline void update_blocked_averages(int cpu)
7594
{
7595 7596
	struct rq *rq = cpu_rq(cpu);
	struct cfs_rq *cfs_rq = &rq->cfs;
7597
	const struct sched_class *curr_class;
7598
	struct rq_flags rf;
7599

7600
	rq_lock_irqsave(rq, &rf);
7601
	update_rq_clock(rq);
7602
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq);
7603 7604 7605 7606

	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);
7607
	update_irq_load_avg(rq, 0);
7608 7609
#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7610
	if (!cfs_rq_has_blocked(cfs_rq) && !others_have_blocked(rq))
7611
		rq->has_blocked_load = 0;
7612
#endif
7613
	rq_unlock_irqrestore(rq, &rf);
7614 7615
}

7616
static unsigned long task_h_load(struct task_struct *p)
7617
{
7618
	return p->se.avg.load_avg;
7619
}
7620 7621 7622 7623 7624

static unsigned long task_h_load_static(struct task_struct *p)
{
	return scale_load_down(p->se.load.weight);
}
P
Peter Zijlstra 已提交
7625
#endif
7626 7627

/********** Helpers for find_busiest_group ************************/
7628 7629 7630 7631 7632 7633 7634

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

7635 7636 7637 7638 7639 7640 7641
/*
 * 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
7642
	unsigned long load_per_task;
7643
	unsigned long group_capacity;
7644
	unsigned long group_util; /* Total utilization of the group */
7645 7646 7647
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
7648
	enum group_type group_type;
7649
	int group_no_capacity;
7650 7651 7652 7653
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
7654 7655
};

J
Joonsoo Kim 已提交
7656 7657 7658 7659 7660 7661 7662
/*
 * 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 */
7663
	unsigned long total_running;
J
Joonsoo Kim 已提交
7664
	unsigned long total_load;	/* Total load of all groups in sd */
7665
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
7666 7667 7668
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
7669
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
7670 7671
};

7672 7673 7674 7675 7676 7677 7678 7679 7680 7681 7682
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,
7683
		.total_running = 0UL,
7684
		.total_load = 0UL,
7685
		.total_capacity = 0UL,
7686 7687
		.busiest_stat = {
			.avg_load = 0UL,
7688 7689
			.sum_nr_running = 0,
			.group_type = group_other,
7690 7691 7692 7693
		},
	};
}

7694 7695 7696
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
7697
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
7698 7699
 *
 * Return: The load index.
7700 7701 7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718 7719 7720 7721
 */
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;
}

7722
static unsigned long scale_rt_capacity(struct sched_domain *sd, int cpu)
7723 7724
{
	struct rq *rq = cpu_rq(cpu);
7725
	unsigned long max = arch_scale_cpu_capacity(sd, cpu);
7726 7727
	unsigned long used, free;
	unsigned long irq;
7728

7729
	irq = cpu_util_irq(rq);
7730

7731 7732
	if (unlikely(irq >= max))
		return 1;
7733

7734 7735
	used = READ_ONCE(rq->avg_rt.util_avg);
	used += READ_ONCE(rq->avg_dl.util_avg);
7736

7737 7738
	if (unlikely(used >= max))
		return 1;
7739

7740
	free = max - used;
7741 7742

	return scale_irq_capacity(free, irq, max);
7743 7744
}

7745
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
7746
{
7747
	unsigned long capacity = scale_rt_capacity(sd, cpu);
7748 7749
	struct sched_group *sdg = sd->groups;

7750
	cpu_rq(cpu)->cpu_capacity_orig = arch_scale_cpu_capacity(sd, cpu);
7751

7752 7753
	if (!capacity)
		capacity = 1;
7754

7755 7756
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
7757
	sdg->sgc->min_capacity = capacity;
7758 7759
}

7760
void update_group_capacity(struct sched_domain *sd, int cpu)
7761 7762 7763
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
7764
	unsigned long capacity, min_capacity;
7765 7766 7767 7768
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
7769
	sdg->sgc->next_update = jiffies + interval;
7770 7771

	if (!child) {
7772
		update_cpu_capacity(sd, cpu);
7773 7774 7775
		return;
	}

7776
	capacity = 0;
7777
	min_capacity = ULONG_MAX;
7778

P
Peter Zijlstra 已提交
7779 7780 7781 7782 7783 7784
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

7785
		for_each_cpu(cpu, sched_group_span(sdg)) {
7786
			struct sched_group_capacity *sgc;
7787
			struct rq *rq = cpu_rq(cpu);
7788

7789
			/*
7790
			 * build_sched_domains() -> init_sched_groups_capacity()
7791 7792 7793
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
7794 7795
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
7796
			 *
7797
			 * This avoids capacity from being 0 and
7798 7799 7800
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
7801
				capacity += capacity_of(cpu);
7802 7803 7804
			} else {
				sgc = rq->sd->groups->sgc;
				capacity += sgc->capacity;
7805
			}
7806

7807
			min_capacity = min(capacity, min_capacity);
7808
		}
P
Peter Zijlstra 已提交
7809 7810 7811 7812
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
7813
		 */
P
Peter Zijlstra 已提交
7814 7815 7816

		group = child->groups;
		do {
7817 7818 7819 7820
			struct sched_group_capacity *sgc = group->sgc;

			capacity += sgc->capacity;
			min_capacity = min(sgc->min_capacity, min_capacity);
P
Peter Zijlstra 已提交
7821 7822 7823
			group = group->next;
		} while (group != child->groups);
	}
7824

7825
	sdg->sgc->capacity = capacity;
7826
	sdg->sgc->min_capacity = min_capacity;
7827 7828
}

7829
/*
7830 7831 7832
 * 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
7833 7834
 */
static inline int
7835
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
7836
{
7837 7838
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
7839 7840
}

7841 7842
/*
 * Group imbalance indicates (and tries to solve) the problem where balancing
7843
 * groups is inadequate due to ->cpus_allowed constraints.
7844
 *
7845 7846
 * 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.
7847 7848
 * Something like:
 *
7849 7850
 *	{ 0 1 2 3 } { 4 5 6 7 }
 *	        *     * * *
7851 7852 7853
 *
 * 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
7854
 * cpu 3 and leave one of the CPUs in the second group unused.
7855 7856
 *
 * The current solution to this issue is detecting the skew in the first group
7857 7858
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
7859 7860
 *
 * When this is so detected; this group becomes a candidate for busiest; see
7861
 * update_sd_pick_busiest(). And calculate_imbalance() and
7862
 * find_busiest_group() avoid some of the usual balance conditions to allow it
7863 7864 7865 7866 7867 7868 7869
 * 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.
 */

7870
static inline int sg_imbalanced(struct sched_group *group)
7871
{
7872
	return group->sgc->imbalance;
7873 7874
}

7875
/*
7876 7877 7878
 * 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
7879 7880
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
7881 7882 7883 7884 7885
 * 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.
7886
 */
7887 7888
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
7889
{
7890 7891
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
7892

7893
	if ((sgs->group_capacity * 100) >
7894
			(sgs->group_util * env->sd->imbalance_pct))
7895
		return true;
7896

7897 7898 7899 7900 7901 7902 7903 7904 7905 7906 7907 7908 7909 7910 7911 7912
	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;
7913

7914
	if ((sgs->group_capacity * 100) <
7915
			(sgs->group_util * env->sd->imbalance_pct))
7916
		return true;
7917

7918
	return false;
7919 7920
}

7921 7922 7923 7924 7925 7926 7927 7928 7929 7930 7931
/*
 * 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;
}

7932 7933 7934
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
7935
{
7936
	if (sgs->group_no_capacity)
7937 7938 7939 7940 7941 7942 7943 7944
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

7945
static bool update_nohz_stats(struct rq *rq, bool force)
7946 7947 7948 7949
{
#ifdef CONFIG_NO_HZ_COMMON
	unsigned int cpu = rq->cpu;

7950 7951 7952
	if (!rq->has_blocked_load)
		return false;

7953
	if (!cpumask_test_cpu(cpu, nohz.idle_cpus_mask))
7954
		return false;
7955

7956
	if (!force && !time_after(jiffies, rq->last_blocked_load_update_tick))
7957
		return true;
7958 7959

	update_blocked_averages(cpu);
7960 7961 7962 7963

	return rq->has_blocked_load;
#else
	return false;
7964 7965 7966
#endif
}

7967 7968
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
7969
 * @env: The load balancing environment.
7970 7971 7972 7973
 * @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.
7974
 * @overload: Indicate more than one runnable task for any CPU.
7975
 */
7976 7977
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
7978 7979
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
7980
{
7981
	unsigned long load;
7982
	int i, nr_running;
7983

7984 7985
	memset(sgs, 0, sizeof(*sgs));

7986
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
7987 7988
		struct rq *rq = cpu_rq(i);

7989
		if ((env->flags & LBF_NOHZ_STATS) && update_nohz_stats(rq, false))
7990
			env->flags |= LBF_NOHZ_AGAIN;
7991

7992
		/* Bias balancing toward CPUs of our domain: */
7993
		if (local_group)
7994
			load = target_load(i, load_idx);
7995
		else
7996 7997 7998
			load = source_load(i, load_idx);

		sgs->group_load += load;
7999
		sgs->group_util += cpu_util(i);
8000
		sgs->sum_nr_running += rq->cfs.h_nr_running;
8001

8002 8003
		nr_running = rq->nr_running;
		if (nr_running > 1)
8004 8005
			*overload = true;

8006 8007 8008 8009
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
8010
		sgs->sum_weighted_load += weighted_cpuload(rq);
8011 8012 8013 8014
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
8015
			sgs->idle_cpus++;
8016 8017
	}

8018 8019
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
8020
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
8021

8022
	if (sgs->sum_nr_running)
8023
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
8024

8025
	sgs->group_weight = group->group_weight;
8026

8027
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
8028
	sgs->group_type = group_classify(group, sgs);
8029 8030
}

8031 8032
/**
 * update_sd_pick_busiest - return 1 on busiest group
8033
 * @env: The load balancing environment.
8034 8035
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
8036
 * @sgs: sched_group statistics
8037 8038 8039
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
8040 8041 8042
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
8043
 */
8044
static bool update_sd_pick_busiest(struct lb_env *env,
8045 8046
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
8047
				   struct sg_lb_stats *sgs)
8048
{
8049
	struct sg_lb_stats *busiest = &sds->busiest_stat;
8050

8051
	if (sgs->group_type > busiest->group_type)
8052 8053
		return true;

8054 8055 8056 8057 8058 8059
	if (sgs->group_type < busiest->group_type)
		return false;

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

8060 8061 8062 8063 8064 8065 8066 8067 8068 8069 8070 8071 8072 8073
	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:
8074 8075
	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
8076 8077
		return true;

8078
	/* No ASYM_PACKING if target CPU is already busy */
8079 8080
	if (env->idle == CPU_NOT_IDLE)
		return true;
8081
	/*
T
Tim Chen 已提交
8082 8083 8084
	 * 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.
8085
	 */
T
Tim Chen 已提交
8086 8087
	if (sgs->sum_nr_running &&
	    sched_asym_prefer(env->dst_cpu, sg->asym_prefer_cpu)) {
8088 8089 8090
		if (!sds->busiest)
			return true;

8091
		/* Prefer to move from lowest priority CPU's work */
T
Tim Chen 已提交
8092 8093
		if (sched_asym_prefer(sds->busiest->asym_prefer_cpu,
				      sg->asym_prefer_cpu))
8094 8095 8096 8097 8098 8099
			return true;
	}

	return false;
}

8100 8101 8102 8103 8104 8105 8106 8107 8108 8109 8110 8111 8112 8113 8114 8115 8116 8117 8118 8119 8120 8121 8122 8123 8124 8125 8126 8127 8128 8129
#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 */

8130
/**
8131
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
8132
 * @env: The load balancing environment.
8133 8134
 * @sds: variable to hold the statistics for this sched_domain.
 */
8135
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
8136
{
8137 8138
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
8139
	struct sg_lb_stats *local = &sds->local_stat;
J
Joonsoo Kim 已提交
8140
	struct sg_lb_stats tmp_sgs;
8141
	int load_idx, prefer_sibling = 0;
8142
	bool overload = false;
8143 8144 8145 8146

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

8147
#ifdef CONFIG_NO_HZ_COMMON
8148
	if (env->idle == CPU_NEWLY_IDLE && READ_ONCE(nohz.has_blocked))
8149 8150 8151
		env->flags |= LBF_NOHZ_STATS;
#endif

8152
	load_idx = get_sd_load_idx(env->sd, env->idle);
8153 8154

	do {
J
Joonsoo Kim 已提交
8155
		struct sg_lb_stats *sgs = &tmp_sgs;
8156 8157
		int local_group;

8158
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_span(sg));
J
Joonsoo Kim 已提交
8159 8160
		if (local_group) {
			sds->local = sg;
8161
			sgs = local;
8162 8163

			if (env->idle != CPU_NEWLY_IDLE ||
8164 8165
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
8166
		}
8167

8168 8169
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
8170

8171 8172 8173
		if (local_group)
			goto next_group;

8174 8175
		/*
		 * In case the child domain prefers tasks go to siblings
8176
		 * first, lower the sg capacity so that we'll try
8177 8178
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
8179 8180 8181 8182
		 * 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).
8183
		 */
8184
		if (prefer_sibling && sds->local &&
8185 8186
		    group_has_capacity(env, local) &&
		    (sgs->sum_nr_running > local->sum_nr_running + 1)) {
8187
			sgs->group_no_capacity = 1;
8188
			sgs->group_type = group_classify(sg, sgs);
8189
		}
8190

8191
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
8192
			sds->busiest = sg;
J
Joonsoo Kim 已提交
8193
			sds->busiest_stat = *sgs;
8194 8195
		}

8196 8197
next_group:
		/* Now, start updating sd_lb_stats */
8198
		sds->total_running += sgs->sum_nr_running;
8199
		sds->total_load += sgs->group_load;
8200
		sds->total_capacity += sgs->group_capacity;
8201

8202
		sg = sg->next;
8203
	} while (sg != env->sd->groups);
8204

8205 8206 8207 8208 8209 8210 8211 8212 8213
#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

8214 8215
	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
8216 8217 8218 8219 8220 8221

	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;
	}
8222 8223 8224 8225
}

/**
 * check_asym_packing - Check to see if the group is packed into the
8226
 *			sched domain.
8227 8228 8229 8230 8231 8232 8233 8234 8235 8236 8237 8238 8239 8240
 *
 * 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.
 *
8241
 * Return: 1 when packing is required and a task should be moved to
8242
 * this CPU.  The amount of the imbalance is returned in env->imbalance.
8243
 *
8244
 * @env: The load balancing environment.
8245 8246
 * @sds: Statistics of the sched_domain which is to be packed
 */
8247
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
8248 8249 8250
{
	int busiest_cpu;

8251
	if (!(env->sd->flags & SD_ASYM_PACKING))
8252 8253
		return 0;

8254 8255 8256
	if (env->idle == CPU_NOT_IDLE)
		return 0;

8257 8258 8259
	if (!sds->busiest)
		return 0;

T
Tim Chen 已提交
8260 8261
	busiest_cpu = sds->busiest->asym_prefer_cpu;
	if (sched_asym_prefer(busiest_cpu, env->dst_cpu))
8262 8263
		return 0;

8264
	env->imbalance = DIV_ROUND_CLOSEST(
8265
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
8266
		SCHED_CAPACITY_SCALE);
8267

8268
	return 1;
8269 8270 8271 8272 8273 8274
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
8275
 * @env: The load balancing environment.
8276 8277
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
8278 8279
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8280
{
8281
	unsigned long tmp, capa_now = 0, capa_move = 0;
8282
	unsigned int imbn = 2;
8283
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
8284
	struct sg_lb_stats *local, *busiest;
8285

J
Joonsoo Kim 已提交
8286 8287
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8288

J
Joonsoo Kim 已提交
8289 8290 8291 8292
	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;
8293

J
Joonsoo Kim 已提交
8294
	scaled_busy_load_per_task =
8295
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8296
		busiest->group_capacity;
J
Joonsoo Kim 已提交
8297

8298 8299
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
8300
		env->imbalance = busiest->load_per_task;
8301 8302 8303 8304 8305
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
8306
	 * however we may be able to increase total CPU capacity used by
8307 8308 8309
	 * moving them.
	 */

8310
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
8311
			min(busiest->load_per_task, busiest->avg_load);
8312
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
8313
			min(local->load_per_task, local->avg_load);
8314
	capa_now /= SCHED_CAPACITY_SCALE;
8315 8316

	/* Amount of load we'd subtract */
8317
	if (busiest->avg_load > scaled_busy_load_per_task) {
8318
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
8319
			    min(busiest->load_per_task,
8320
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
8321
	}
8322 8323

	/* Amount of load we'd add */
8324
	if (busiest->avg_load * busiest->group_capacity <
8325
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
8326 8327
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
8328
	} else {
8329
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8330
		      local->group_capacity;
J
Joonsoo Kim 已提交
8331
	}
8332
	capa_move += local->group_capacity *
8333
		    min(local->load_per_task, local->avg_load + tmp);
8334
	capa_move /= SCHED_CAPACITY_SCALE;
8335 8336

	/* Move if we gain throughput */
8337
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
8338
		env->imbalance = busiest->load_per_task;
8339 8340 8341 8342 8343
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
8344
 * @env: load balance environment
8345 8346
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
8347
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8348
{
8349
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
8350 8351 8352 8353
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8354

8355
	if (busiest->group_type == group_imbalanced) {
8356 8357
		/*
		 * In the group_imb case we cannot rely on group-wide averages
8358
		 * to ensure CPU-load equilibrium, look at wider averages. XXX
8359
		 */
J
Joonsoo Kim 已提交
8360 8361
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
8362 8363
	}

8364
	/*
8365 8366 8367 8368
	 * 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:
8369
	 */
8370 8371
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
8372 8373
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
8374 8375
	}

8376
	/*
8377
	 * If there aren't any idle CPUs, avoid creating some.
8378 8379 8380
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
8381
		load_above_capacity = busiest->sum_nr_running * SCHED_CAPACITY_SCALE;
8382
		if (load_above_capacity > busiest->group_capacity) {
8383
			load_above_capacity -= busiest->group_capacity;
8384
			load_above_capacity *= scale_load_down(NICE_0_LOAD);
8385 8386
			load_above_capacity /= busiest->group_capacity;
		} else
8387
			load_above_capacity = ~0UL;
8388 8389 8390
	}

	/*
8391
	 * We're trying to get all the CPUs to the average_load, so we don't
8392
	 * want to push ourselves above the average load, nor do we wish to
8393
	 * reduce the max loaded CPU below the average load. At the same time,
8394 8395
	 * we also don't want to reduce the group load below the group
	 * capacity. Thus we look for the minimum possible imbalance.
8396
	 */
8397
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
8398 8399

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
8400
	env->imbalance = min(
8401 8402
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
8403
	) / SCHED_CAPACITY_SCALE;
8404 8405 8406

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
8407
	 * there is no guarantee that any tasks will be moved so we'll have
8408 8409 8410
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
8411
	if (env->imbalance < busiest->load_per_task)
8412
		return fix_small_imbalance(env, sds);
8413
}
8414

8415 8416 8417 8418
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
8419
 * if there is an imbalance.
8420 8421 8422 8423
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
8424
 * @env: The load balancing environment.
8425
 *
8426
 * Return:	- The busiest group if imbalance exists.
8427
 */
J
Joonsoo Kim 已提交
8428
static struct sched_group *find_busiest_group(struct lb_env *env)
8429
{
J
Joonsoo Kim 已提交
8430
	struct sg_lb_stats *local, *busiest;
8431 8432
	struct sd_lb_stats sds;

8433
	init_sd_lb_stats(&sds);
8434 8435 8436 8437 8438

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
8439
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
8440 8441
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
8442

8443
	/* ASYM feature bypasses nice load balance check */
8444
	if (check_asym_packing(env, &sds))
8445 8446
		return sds.busiest;

8447
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
8448
	if (!sds.busiest || busiest->sum_nr_running == 0)
8449 8450
		goto out_balanced;

8451
	/* XXX broken for overlapping NUMA groups */
8452 8453
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
8454

P
Peter Zijlstra 已提交
8455 8456
	/*
	 * If the busiest group is imbalanced the below checks don't
8457
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
8458 8459
	 * isn't true due to cpus_allowed constraints and the like.
	 */
8460
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
8461 8462
		goto force_balance;

8463 8464 8465 8466 8467
	/*
	 * 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) &&
8468
	    busiest->group_no_capacity)
8469 8470
		goto force_balance;

8471
	/*
8472
	 * If the local group is busier than the selected busiest group
8473 8474
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
8475
	if (local->avg_load >= busiest->avg_load)
8476 8477
		goto out_balanced;

8478 8479 8480 8481
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
8482
	if (local->avg_load >= sds.avg_load)
8483 8484
		goto out_balanced;

8485
	if (env->idle == CPU_IDLE) {
8486
		/*
8487
		 * This CPU is idle. If the busiest group is not overloaded
8488
		 * and there is no imbalance between this and busiest group
8489
		 * wrt idle CPUs, it is balanced. The imbalance becomes
8490 8491
		 * significant if the diff is greater than 1 otherwise we
		 * might end up to just move the imbalance on another group
8492
		 */
8493 8494
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
8495
			goto out_balanced;
8496 8497 8498 8499 8500
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
8501 8502
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
8503
			goto out_balanced;
8504
	}
8505

8506
force_balance:
8507
	/* Looks like there is an imbalance. Compute it */
8508
	calculate_imbalance(env, &sds);
8509
	return env->imbalance ? sds.busiest : NULL;
8510 8511

out_balanced:
8512
	env->imbalance = 0;
8513 8514 8515 8516
	return NULL;
}

/*
8517
 * find_busiest_queue - find the busiest runqueue among the CPUs in the group.
8518
 */
8519
static struct rq *find_busiest_queue(struct lb_env *env,
8520
				     struct sched_group *group)
8521 8522
{
	struct rq *busiest = NULL, *rq;
8523
	unsigned long busiest_load = 0, busiest_capacity = 1;
8524 8525
	int i;

8526
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
8527
		unsigned long capacity, wl;
8528 8529 8530 8531
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
8532

8533 8534 8535 8536 8537 8538 8539 8540 8541 8542 8543 8544 8545 8546 8547 8548 8549 8550 8551 8552 8553 8554
		/*
		 * 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;

8555
		capacity = capacity_of(i);
8556

8557
		wl = weighted_cpuload(rq);
8558

8559 8560
		/*
		 * When comparing with imbalance, use weighted_cpuload()
8561
		 * which is not scaled with the CPU capacity.
8562
		 */
8563 8564 8565

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
8566 8567
			continue;

8568
		/*
8569 8570 8571
		 * 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
8572
		 * potentially running at a lower capacity.
8573
		 *
8574
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
8575
		 * multiplication to rid ourselves of the division works out
8576 8577
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
8578
		 */
8579
		if (wl * busiest_capacity > busiest_load * capacity) {
8580
			busiest_load = wl;
8581
			busiest_capacity = capacity;
8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592 8593 8594
			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

8595
static int need_active_balance(struct lb_env *env)
8596
{
8597 8598 8599
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
8600 8601 8602

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
T
Tim Chen 已提交
8603 8604
		 * lower priority CPUs in order to pack all tasks in the
		 * highest priority CPUs.
8605
		 */
T
Tim Chen 已提交
8606 8607
		if ((sd->flags & SD_ASYM_PACKING) &&
		    sched_asym_prefer(env->dst_cpu, env->src_cpu))
8608
			return 1;
8609 8610
	}

8611 8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623
	/*
	 * 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;
	}

8624 8625 8626
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

8627 8628
static int active_load_balance_cpu_stop(void *data);

8629 8630 8631 8632 8633
static int should_we_balance(struct lb_env *env)
{
	struct sched_group *sg = env->sd->groups;
	int cpu, balance_cpu = -1;

8634 8635 8636 8637 8638 8639 8640
	/*
	 * 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;

8641
	/*
8642
	 * In the newly idle case, we will allow all the CPUs
8643 8644 8645 8646 8647
	 * to do the newly idle load balance.
	 */
	if (env->idle == CPU_NEWLY_IDLE)
		return 1;

8648
	/* Try to find first idle CPU */
8649
	for_each_cpu_and(cpu, group_balance_mask(sg), env->cpus) {
8650
		if (!idle_cpu(cpu))
8651 8652 8653 8654 8655 8656 8657 8658 8659 8660
			continue;

		balance_cpu = cpu;
		break;
	}

	if (balance_cpu == -1)
		balance_cpu = group_balance_cpu(sg);

	/*
8661
	 * First idle CPU or the first CPU(busiest) in this sched group
8662 8663
	 * is eligible for doing load balancing at this and above domains.
	 */
8664
	return balance_cpu == env->dst_cpu;
8665 8666
}

8667 8668 8669 8670 8671 8672
/*
 * 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,
8673
			int *continue_balancing)
8674
{
8675
	int ld_moved, cur_ld_moved, active_balance = 0;
8676
	struct sched_domain *sd_parent = sd->parent;
8677 8678
	struct sched_group *group;
	struct rq *busiest;
8679
	struct rq_flags rf;
8680
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
8681

8682 8683
	struct lb_env env = {
		.sd		= sd,
8684 8685
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
8686
		.dst_grpmask    = sched_group_span(sd->groups),
8687
		.idle		= idle,
8688
		.loop_break	= sched_nr_migrate_break,
8689
		.cpus		= cpus,
8690
		.fbq_type	= all,
8691
		.tasks		= LIST_HEAD_INIT(env.tasks),
8692 8693
	};

8694
	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
8695

8696
	schedstat_inc(sd->lb_count[idle]);
8697 8698

redo:
8699 8700
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
8701
		goto out_balanced;
8702
	}
8703

8704
	group = find_busiest_group(&env);
8705
	if (!group) {
8706
		schedstat_inc(sd->lb_nobusyg[idle]);
8707 8708 8709
		goto out_balanced;
	}

8710
	busiest = find_busiest_queue(&env, group);
8711
	if (!busiest) {
8712
		schedstat_inc(sd->lb_nobusyq[idle]);
8713 8714 8715
		goto out_balanced;
	}

8716
	BUG_ON(busiest == env.dst_rq);
8717

8718
	schedstat_add(sd->lb_imbalance[idle], env.imbalance);
8719

8720 8721 8722
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

8723 8724 8725 8726 8727 8728 8729 8730
	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.
		 */
8731
		env.flags |= LBF_ALL_PINNED;
8732
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
8733

8734
more_balance:
8735
		rq_lock_irqsave(busiest, &rf);
8736
		update_rq_clock(busiest);
8737 8738 8739 8740 8741

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
8742
		cur_ld_moved = detach_tasks(&env);
8743 8744

		/*
8745 8746 8747 8748 8749
		 * 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.
8750
		 */
8751

8752
		rq_unlock(busiest, &rf);
8753 8754 8755 8756 8757 8758

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

8759
		local_irq_restore(rf.flags);
8760

8761 8762 8763 8764 8765
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

8766 8767 8768 8769
		/*
		 * 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
8770
		 * iterate on same src_cpu is dependent on number of CPUs in our
8771 8772 8773 8774 8775 8776 8777 8778 8779 8780 8781 8782 8783 8784
		 * 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.
		 */
8785
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
8786

8787
			/* Prevent to re-select dst_cpu via env's CPUs */
8788 8789
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

8790
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
8791
			env.dst_cpu	 = env.new_dst_cpu;
8792
			env.flags	&= ~LBF_DST_PINNED;
8793 8794
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
8795

8796 8797 8798 8799 8800 8801
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
8802

8803 8804 8805 8806
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
8807
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
8808

8809
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
8810 8811 8812
				*group_imbalance = 1;
		}

8813
		/* All tasks on this runqueue were pinned by CPU affinity */
8814
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
8815
			cpumask_clear_cpu(cpu_of(busiest), cpus);
8816 8817 8818 8819 8820 8821 8822 8823 8824
			/*
			 * 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)) {
8825 8826
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
8827
				goto redo;
8828
			}
8829
			goto out_all_pinned;
8830 8831 8832 8833
		}
	}

	if (!ld_moved) {
8834
		schedstat_inc(sd->lb_failed[idle]);
8835 8836 8837 8838 8839 8840 8841 8842
		/*
		 * 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++;
8843

8844
		if (need_active_balance(&env)) {
8845 8846
			unsigned long flags;

8847 8848
			raw_spin_lock_irqsave(&busiest->lock, flags);

8849 8850 8851 8852
			/*
			 * Don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest CPU can't be
			 * moved to this_cpu:
8853
			 */
8854
			if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) {
8855 8856
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
8857
				env.flags |= LBF_ALL_PINNED;
8858 8859 8860
				goto out_one_pinned;
			}

8861 8862 8863 8864 8865
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
8866 8867 8868 8869 8870 8871
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
8872

8873
			if (active_balance) {
8874 8875 8876
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
8877
			}
8878

8879
			/* We've kicked active balancing, force task migration. */
8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890 8891 8892
			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
8893
		 * detach_tasks).
8894 8895 8896 8897 8898 8899 8900 8901
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
8902 8903
	/*
	 * We reach balance although we may have faced some affinity
8904 8905
	 * constraints. Clear the imbalance flag only if other tasks got
	 * a chance to move and fix the imbalance.
8906
	 */
8907
	if (sd_parent && !(env.flags & LBF_ALL_PINNED)) {
8908 8909 8910 8911 8912 8913 8914 8915 8916 8917 8918 8919
		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.
	 */
8920
	schedstat_inc(sd->lb_balanced[idle]);
8921 8922 8923 8924

	sd->nr_balance_failed = 0;

out_one_pinned:
8925 8926 8927 8928 8929 8930 8931 8932 8933 8934 8935
	ld_moved = 0;

	/*
	 * idle_balance() disregards balance intervals, so we could repeatedly
	 * reach this code, which would lead to balance_interval skyrocketting
	 * in a short amount of time. Skip the balance_interval increase logic
	 * to avoid that.
	 */
	if (env.idle == CPU_NEWLY_IDLE)
		goto out;

8936
	/* tune up the balancing interval */
8937
	if (((env.flags & LBF_ALL_PINNED) &&
8938
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
8939 8940 8941 8942 8943 8944
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;
out:
	return ld_moved;
}

8945 8946 8947 8948 8949 8950 8951 8952 8953 8954 8955 8956 8957 8958 8959 8960
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
8961
update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
8962 8963 8964
{
	unsigned long interval, next;

8965 8966
	/* used by idle balance, so cpu_busy = 0 */
	interval = get_sd_balance_interval(sd, 0);
8967 8968 8969 8970 8971 8972
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

8973
/*
8974
 * active_load_balance_cpu_stop is run by the CPU stopper. It pushes
8975 8976 8977
 * 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.
8978
 */
8979
static int active_load_balance_cpu_stop(void *data)
8980
{
8981 8982
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
8983
	int target_cpu = busiest_rq->push_cpu;
8984
	struct rq *target_rq = cpu_rq(target_cpu);
8985
	struct sched_domain *sd;
8986
	struct task_struct *p = NULL;
8987
	struct rq_flags rf;
8988

8989
	rq_lock_irq(busiest_rq, &rf);
8990 8991 8992 8993 8994 8995 8996
	/*
	 * 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;
8997

8998
	/* Make sure the requested CPU hasn't gone down in the meantime: */
8999 9000 9001
	if (unlikely(busiest_cpu != smp_processor_id() ||
		     !busiest_rq->active_balance))
		goto out_unlock;
9002 9003 9004

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
9005
		goto out_unlock;
9006 9007 9008 9009

	/*
	 * This condition is "impossible", if it occurs
	 * we need to fix it. Originally reported by
9010
	 * Bjorn Helgaas on a 128-CPU setup.
9011 9012 9013 9014
	 */
	BUG_ON(busiest_rq == target_rq);

	/* Search for an sd spanning us and the target CPU. */
9015
	rcu_read_lock();
9016 9017 9018 9019 9020 9021 9022
	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)) {
9023 9024
		struct lb_env env = {
			.sd		= sd,
9025 9026 9027 9028
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
9029
			.idle		= CPU_IDLE,
9030 9031 9032 9033 9034 9035 9036
			/*
			 * 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,
9037 9038
		};

9039
		schedstat_inc(sd->alb_count);
9040
		update_rq_clock(busiest_rq);
9041

9042
		p = detach_one_task(&env);
9043
		if (p) {
9044
			schedstat_inc(sd->alb_pushed);
9045 9046 9047
			/* Active balancing done, reset the failure counter. */
			sd->nr_balance_failed = 0;
		} else {
9048
			schedstat_inc(sd->alb_failed);
9049
		}
9050
	}
9051
	rcu_read_unlock();
9052 9053
out_unlock:
	busiest_rq->active_balance = 0;
9054
	rq_unlock(busiest_rq, &rf);
9055 9056 9057 9058 9059 9060

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

9061
	return 0;
9062 9063
}

9064 9065 9066 9067 9068 9069 9070 9071 9072 9073 9074 9075 9076 9077 9078 9079 9080 9081 9082 9083 9084 9085 9086 9087 9088 9089 9090 9091 9092 9093 9094 9095 9096 9097 9098 9099 9100 9101 9102 9103 9104 9105 9106 9107 9108 9109 9110 9111 9112 9113 9114 9115 9116 9117 9118 9119 9120 9121 9122 9123 9124 9125 9126 9127 9128 9129 9130 9131 9132 9133 9134 9135 9136 9137 9138 9139 9140 9141 9142 9143 9144 9145 9146 9147 9148 9149 9150 9151 9152 9153 9154 9155 9156 9157 9158 9159 9160 9161 9162 9163 9164 9165 9166 9167 9168 9169 9170 9171 9172 9173 9174 9175 9176 9177 9178 9179 9180 9181
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
	}
}

9182 9183 9184 9185 9186
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

9187
#ifdef CONFIG_NO_HZ_COMMON
9188 9189 9190 9191 9192
/*
 * 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.
9193 9194
 * - HK_FLAG_MISC CPUs are used for this task, because HK_FLAG_SCHED not set
 *   anywhere yet.
9195
 */
9196

9197
static inline int find_new_ilb(void)
9198
{
9199
	int ilb;
9200

9201 9202 9203 9204 9205
	for_each_cpu_and(ilb, nohz.idle_cpus_mask,
			      housekeeping_cpumask(HK_FLAG_MISC)) {
		if (idle_cpu(ilb))
			return ilb;
	}
9206 9207

	return nr_cpu_ids;
9208 9209
}

9210
/*
9211 9212
 * Kick a CPU to do the nohz balancing, if it is time for it. We pick any
 * idle CPU in the HK_FLAG_MISC housekeeping set (if there is one).
9213
 */
9214
static void kick_ilb(unsigned int flags)
9215 9216 9217 9218 9219
{
	int ilb_cpu;

	nohz.next_balance++;

9220
	ilb_cpu = find_new_ilb();
9221

9222 9223
	if (ilb_cpu >= nr_cpu_ids)
		return;
9224

9225
	flags = atomic_fetch_or(flags, nohz_flags(ilb_cpu));
P
Peter Zijlstra 已提交
9226
	if (flags & NOHZ_KICK_MASK)
9227
		return;
9228

9229 9230
	/*
	 * Use smp_send_reschedule() instead of resched_cpu().
9231
	 * This way we generate a sched IPI on the target CPU which
9232 9233 9234 9235
	 * is idle. And the softirq performing nohz idle load balance
	 * will be run before returning from the IPI.
	 */
	smp_send_reschedule(ilb_cpu);
9236 9237 9238 9239 9240 9241 9242 9243 9244 9245 9246 9247 9248 9249 9250 9251 9252 9253 9254
}

/*
 * 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;
9255
	unsigned int flags = 0;
9256 9257 9258 9259 9260 9261 9262 9263

	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.
	 */
9264
	nohz_balance_exit_idle(rq);
9265 9266 9267 9268 9269 9270 9271 9272

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

9273 9274
	if (READ_ONCE(nohz.has_blocked) &&
	    time_after(now, READ_ONCE(nohz.next_blocked)))
9275 9276
		flags = NOHZ_STATS_KICK;

9277
	if (time_before(now, nohz.next_balance))
9278
		goto out;
9279 9280

	if (rq->nr_running >= 2) {
9281
		flags = NOHZ_KICK_MASK;
9282 9283 9284 9285 9286 9287 9288 9289 9290 9291 9292 9293
		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) {
9294
			flags = NOHZ_KICK_MASK;
9295 9296 9297 9298 9299 9300 9301 9302 9303
			goto unlock;
		}

	}

	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
9304
			flags = NOHZ_KICK_MASK;
9305 9306 9307 9308 9309 9310 9311 9312 9313 9314 9315 9316
			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)) {
9317
				flags = NOHZ_KICK_MASK;
9318 9319 9320 9321 9322 9323 9324
				goto unlock;
			}
		}
	}
unlock:
	rcu_read_unlock();
out:
9325 9326
	if (flags)
		kick_ilb(flags);
9327 9328
}

9329
static void set_cpu_sd_state_busy(int cpu)
9330
{
9331
	struct sched_domain *sd;
9332

9333 9334
	rcu_read_lock();
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
9335

9336 9337 9338 9339 9340 9341 9342
	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

	atomic_inc(&sd->shared->nr_busy_cpus);
unlock:
	rcu_read_unlock();
9343 9344
}

9345 9346 9347 9348 9349 9350 9351 9352 9353 9354 9355 9356 9357 9358 9359
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)
9360 9361 9362 9363
{
	struct sched_domain *sd;

	rcu_read_lock();
9364
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
9365 9366 9367 9368 9369

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

9370
	atomic_dec(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
9371
unlock:
9372 9373 9374
	rcu_read_unlock();
}

9375
/*
9376
 * This routine will record that the CPU is going idle with tick stopped.
9377
 * This info will be used in performing idle load balancing in the future.
9378
 */
9379
void nohz_balance_enter_idle(int cpu)
9380
{
9381 9382 9383 9384
	struct rq *rq = cpu_rq(cpu);

	SCHED_WARN_ON(cpu != smp_processor_id());

9385
	/* If this CPU is going down, then nothing needs to be done: */
9386 9387 9388
	if (!cpu_active(cpu))
		return;

9389
	/* Spare idle load balancing on CPUs that don't want to be disturbed: */
9390
	if (!housekeeping_cpu(cpu, HK_FLAG_SCHED))
9391 9392
		return;

9393 9394 9395 9396 9397 9398 9399 9400 9401 9402 9403 9404 9405
	/*
	 * 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
	 */
9406
	if (rq->nohz_tick_stopped)
9407
		goto out;
9408

9409
	/* If we're a completely isolated CPU, we don't play: */
9410
	if (on_null_domain(rq))
9411 9412
		return;

9413 9414
	rq->nohz_tick_stopped = 1;

9415 9416
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
9417

9418 9419 9420 9421 9422 9423 9424
	/*
	 * 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();

9425
	set_cpu_sd_state_idle(cpu);
9426 9427 9428 9429 9430 9431 9432

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);
9433 9434 9435
}

/*
9436 9437 9438 9439 9440
 * 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.
9441
 */
9442 9443
static bool _nohz_idle_balance(struct rq *this_rq, unsigned int flags,
			       enum cpu_idle_type idle)
9444
{
9445
	/* Earliest time when we have to do rebalance again */
9446 9447
	unsigned long now = jiffies;
	unsigned long next_balance = now + 60*HZ;
9448
	bool has_blocked_load = false;
9449
	int update_next_balance = 0;
P
Peter Zijlstra 已提交
9450 9451
	int this_cpu = this_rq->cpu;
	int balance_cpu;
9452
	int ret = false;
P
Peter Zijlstra 已提交
9453
	struct rq *rq;
9454

P
Peter Zijlstra 已提交
9455
	SCHED_WARN_ON((flags & NOHZ_KICK_MASK) == NOHZ_BALANCE_KICK);
9456

9457 9458 9459 9460 9461 9462 9463 9464 9465 9466 9467 9468 9469 9470 9471 9472
	/*
	 * 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();

9473
	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
9474
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
9475 9476 9477
			continue;

		/*
9478 9479
		 * If this CPU gets work to do, stop the load balancing
		 * work being done for other CPUs. Next load
9480 9481
		 * balancing owner will pick it up.
		 */
9482 9483 9484 9485
		if (need_resched()) {
			has_blocked_load = true;
			goto abort;
		}
9486

V
Vincent Guittot 已提交
9487 9488
		rq = cpu_rq(balance_cpu);

9489
		has_blocked_load |= update_nohz_stats(rq, true);
9490

9491 9492 9493 9494 9495
		/*
		 * If time for next balance is due,
		 * do the balance.
		 */
		if (time_after_eq(jiffies, rq->next_balance)) {
9496 9497
			struct rq_flags rf;

9498
			rq_lock_irqsave(rq, &rf);
9499
			update_rq_clock(rq);
9500
			cpu_load_update_idle(rq);
9501
			rq_unlock_irqrestore(rq, &rf);
9502

P
Peter Zijlstra 已提交
9503 9504
			if (flags & NOHZ_BALANCE_KICK)
				rebalance_domains(rq, CPU_IDLE);
9505
		}
9506

9507 9508 9509 9510
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
9511
	}
9512

9513 9514 9515 9516 9517 9518
	/* 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
9519 9520 9521
	if (flags & NOHZ_BALANCE_KICK)
		rebalance_domains(this_rq, CPU_IDLE);

9522 9523 9524
	WRITE_ONCE(nohz.next_blocked,
		now + msecs_to_jiffies(LOAD_AVG_PERIOD));

9525 9526 9527
	/* The full idle balance loop has been done */
	ret = true;

9528 9529 9530 9531
abort:
	/* There is still blocked load, enable periodic update */
	if (has_blocked_load)
		WRITE_ONCE(nohz.has_blocked, 1);
9532

9533 9534 9535 9536 9537 9538 9539
	/*
	 * 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
9540

9541 9542 9543 9544 9545 9546 9547 9548 9549 9550 9551 9552 9553 9554 9555 9556 9557 9558 9559 9560 9561 9562 9563 9564 9565 9566 9567 9568 9569
	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
9570
	return true;
9571
}
9572 9573 9574 9575 9576 9577 9578 9579 9580 9581 9582 9583 9584 9585 9586 9587 9588 9589 9590 9591 9592 9593 9594 9595 9596 9597 9598 9599 9600 9601 9602 9603 9604

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);
}

9605 9606 9607
#else /* !CONFIG_NO_HZ_COMMON */
static inline void nohz_balancer_kick(struct rq *rq) { }

9608
static inline bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
P
Peter Zijlstra 已提交
9609 9610 9611
{
	return false;
}
9612 9613

static inline void nohz_newidle_balance(struct rq *this_rq) { }
9614
#endif /* CONFIG_NO_HZ_COMMON */
9615

P
Peter Zijlstra 已提交
9616 9617 9618 9619 9620 9621 9622 9623 9624 9625 9626 9627 9628 9629 9630 9631 9632 9633 9634 9635 9636 9637 9638 9639 9640 9641 9642 9643 9644 9645 9646 9647 9648 9649
/*
 * 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) {
9650

P
Peter Zijlstra 已提交
9651 9652 9653 9654 9655 9656
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, &next_balance);
		rcu_read_unlock();

9657 9658
		nohz_newidle_balance(this_rq);

P
Peter Zijlstra 已提交
9659 9660 9661 9662 9663 9664 9665 9666 9667 9668 9669 9670 9671 9672 9673 9674 9675 9676 9677 9678 9679 9680 9681 9682 9683 9684 9685 9686 9687 9688 9689 9690 9691 9692 9693 9694 9695 9696 9697 9698 9699 9700 9701 9702 9703 9704 9705 9706 9707
		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;

9708
out:
P
Peter Zijlstra 已提交
9709 9710 9711 9712 9713 9714 9715 9716 9717 9718 9719 9720 9721 9722 9723 9724 9725 9726 9727 9728 9729 9730 9731 9732
	/*
	 * 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;
}

9733 9734 9735 9736
/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
9737
static __latent_entropy void run_rebalance_domains(struct softirq_action *h)
9738
{
9739
	struct rq *this_rq = this_rq();
9740
	enum cpu_idle_type idle = this_rq->idle_balance ?
9741 9742 9743
						CPU_IDLE : CPU_NOT_IDLE;

	/*
9744 9745
	 * If this CPU has a pending nohz_balance_kick, then do the
	 * balancing on behalf of the other idle CPUs whose ticks are
9746
	 * stopped. Do nohz_idle_balance *before* rebalance_domains to
9747
	 * give the idle CPUs a chance to load balance. Else we may
9748 9749
	 * load balance only within the local sched_domain hierarchy
	 * and abort nohz_idle_balance altogether if we pull some load.
9750
	 */
P
Peter Zijlstra 已提交
9751 9752 9753 9754 9755
	if (nohz_idle_balance(this_rq, idle))
		return;

	/* normal load balance */
	update_blocked_averages(this_rq->cpu);
9756
	rebalance_domains(this_rq, idle);
9757 9758 9759 9760 9761
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
9762
void trigger_load_balance(struct rq *rq)
9763 9764
{
	/* Don't need to rebalance while attached to NULL domain */
9765 9766 9767 9768
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
9769
		raise_softirq(SCHED_SOFTIRQ);
9770 9771

	nohz_balancer_kick(rq);
9772 9773
}

9774 9775 9776
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
9777 9778

	update_runtime_enabled(rq);
9779 9780 9781 9782 9783
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
9784 9785 9786

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
9787 9788
}

9789
#endif /* CONFIG_SMP */
9790

9791
/*
9792 9793 9794 9795 9796 9797
 * 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.
9798
 */
P
Peter Zijlstra 已提交
9799
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
9800 9801 9802 9803 9804 9805
{
	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
9806
		entity_tick(cfs_rq, se, queued);
9807
	}
9808

9809
	if (static_branch_unlikely(&sched_numa_balancing))
9810
		task_tick_numa(rq, curr);
9811 9812 9813
}

/*
P
Peter Zijlstra 已提交
9814 9815 9816
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
9817
 */
P
Peter Zijlstra 已提交
9818
static void task_fork_fair(struct task_struct *p)
9819
{
9820 9821
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
P
Peter Zijlstra 已提交
9822
	struct rq *rq = this_rq();
9823
	struct rq_flags rf;
9824

9825
	rq_lock(rq, &rf);
9826 9827
	update_rq_clock(rq);

9828 9829
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;
9830 9831
	if (curr) {
		update_curr(cfs_rq);
9832
		se->vruntime = curr->vruntime;
9833
	}
9834
	place_entity(cfs_rq, se, 1);
9835

P
Peter Zijlstra 已提交
9836
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
9837
		/*
9838 9839 9840
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
9841
		swap(curr->vruntime, se->vruntime);
9842
		resched_curr(rq);
9843
	}
9844

9845
	se->vruntime -= cfs_rq->min_vruntime;
9846
	rq_unlock(rq, &rf);
9847 9848
}

9849 9850 9851 9852
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
9853 9854
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
9855
{
9856
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
9857 9858
		return;

9859 9860 9861 9862 9863
	/*
	 * 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
9864
	if (rq->curr == p) {
9865
		if (p->prio > oldprio)
9866
			resched_curr(rq);
9867
	} else
9868
		check_preempt_curr(rq, p, 0);
9869 9870
}

9871
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
9872 9873 9874 9875
{
	struct sched_entity *se = &p->se;

	/*
9876 9877 9878 9879 9880 9881 9882 9883 9884 9885
	 * 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
9886
	 *
9887 9888 9889 9890
	 * - 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
9891
	 */
9892 9893
	if (!se->sum_exec_runtime ||
	    (p->state == TASK_WAKING && p->sched_remote_wakeup))
9894 9895 9896 9897 9898
		return true;

	return false;
}

9899 9900 9901 9902 9903 9904 9905 9906 9907 9908 9909 9910 9911 9912 9913 9914 9915 9916
#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;

9917
		update_load_avg(cfs_rq, se, UPDATE_TG);
9918 9919 9920 9921 9922 9923
	}
}
#else
static void propagate_entity_cfs_rq(struct sched_entity *se) { }
#endif

9924
static void detach_entity_cfs_rq(struct sched_entity *se)
9925 9926 9927
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

9928
	/* Catch up with the cfs_rq and remove our load when we leave */
9929
	update_load_avg(cfs_rq, se, 0);
9930
	detach_entity_load_avg(cfs_rq, se);
9931
	update_tg_load_avg(cfs_rq, false);
9932
	propagate_entity_cfs_rq(se);
P
Peter Zijlstra 已提交
9933 9934
}

9935
static void attach_entity_cfs_rq(struct sched_entity *se)
9936
{
9937
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
9938 9939

#ifdef CONFIG_FAIR_GROUP_SCHED
9940 9941 9942 9943 9944 9945
	/*
	 * 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
9946

9947
	/* Synchronize entity with its cfs_rq */
9948
	update_load_avg(cfs_rq, se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
9949
	attach_entity_load_avg(cfs_rq, se, 0);
9950
	update_tg_load_avg(cfs_rq, false);
9951
	propagate_entity_cfs_rq(se);
9952 9953 9954 9955 9956 9957 9958 9959 9960 9961 9962 9963 9964 9965 9966 9967 9968 9969 9970 9971 9972 9973 9974 9975 9976
}

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);
9977 9978 9979 9980

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
9981

9982 9983 9984 9985 9986 9987 9988 9989
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);
9990

9991
	if (task_on_rq_queued(p)) {
9992
		/*
9993 9994 9995
		 * 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.
9996
		 */
9997 9998 9999 10000
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
10001
	}
10002 10003
}

10004 10005 10006 10007 10008 10009 10010 10011 10012
/* 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;

10013 10014 10015 10016 10017 10018 10019
	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);
	}
10020 10021
}

10022 10023
void init_cfs_rq(struct cfs_rq *cfs_rq)
{
10024
	cfs_rq->tasks_timeline = RB_ROOT_CACHED;
10025 10026 10027 10028
	cfs_rq->min_vruntime = (u64)(-(1LL << 20));
#ifndef CONFIG_64BIT
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
10029
#ifdef CONFIG_SMP
10030
	raw_spin_lock_init(&cfs_rq->removed.lock);
10031
#endif
10032 10033
}

P
Peter Zijlstra 已提交
10034
#ifdef CONFIG_FAIR_GROUP_SCHED
10035 10036 10037 10038 10039 10040 10041 10042 10043 10044 10045 10046 10047 10048 10049 10050 10051 10052 10053 10054 10055 10056 10057 10058 10059 10060 10061 10062 10063 10064 10065 10066 10067 10068 10069 10070 10071 10072
#ifdef CONFIG_SCHED_SLI
static void update_nr_iowait_fair(struct task_struct *p, long inc)
{
	unsigned long flags;
	struct sched_entity *se = p->se.parent;
	u64 clock;

	if (!schedstat_enabled())
		return;

	clock = __rq_clock_broken(cpu_rq(p->cpu));

	for_each_sched_entity(se) {
		/*
		 * Avoid locking rq->lock from try_to_wakeup hot path, in
		 * the price of poor consistency among cgroup hierarchy,
		 * which we can tolerate.
		 * While accessing se->on_rq does need to hold rq->lock. We
		 * already do, because when inc==1, the caller is __schedule
		 * and task_move_group_fair
		 */
		spin_lock_irqsave(&se->iowait_lock, flags);
		if (!se->on_rq && !schedstat_val(se->cg_nr_iowait) && inc > 0)
			__schedstat_set(se->cg_iowait_start, clock);
		if (schedstat_val(se->cg_iowait_start) > 0 &&
			schedstat_val(se->cg_nr_iowait) + inc == 0) {
			__schedstat_add(se->cg_iowait_sum, clock -
				schedstat_val(se->cg_iowait_start));
			__schedstat_set(se->cg_iowait_start, 0);
		}
		__schedstat_add(se->cg_nr_iowait, inc);
		spin_unlock_irqrestore(&se->iowait_lock, flags);
	}
}
#else
static void update_nr_iowait_fair(struct task_struct *p, long inc) {}
#endif

10073 10074 10075 10076 10077 10078 10079 10080
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;
}

10081
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
10082
{
10083 10084
	if (p->in_iowait)
		update_nr_iowait_fair(p, -1);
10085
	detach_task_cfs_rq(p);
10086
	set_task_rq(p, task_cpu(p));
10087 10088 10089 10090 10091

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
10092
	attach_task_cfs_rq(p);
10093 10094
	if (p->in_iowait)
		update_nr_iowait_fair(p, 1);
P
Peter Zijlstra 已提交
10095
}
10096

10097 10098 10099 10100 10101 10102 10103 10104 10105 10106 10107 10108 10109
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;
	}
}

10110 10111 10112 10113 10114 10115 10116 10117 10118
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]);
10119
		if (tg->se)
10120 10121 10122 10123 10124 10125 10126 10127 10128 10129
			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;
10130
	struct cfs_rq *cfs_rq;
10131 10132
	int i;

K
Kees Cook 已提交
10133
	tg->cfs_rq = kcalloc(nr_cpu_ids, sizeof(cfs_rq), GFP_KERNEL);
10134 10135
	if (!tg->cfs_rq)
		goto err;
K
Kees Cook 已提交
10136
	tg->se = kcalloc(nr_cpu_ids, sizeof(se), GFP_KERNEL);
10137 10138 10139 10140 10141 10142 10143 10144 10145 10146 10147 10148 10149 10150 10151 10152 10153 10154 10155 10156
	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]);
10157
		init_entity_runnable_average(se);
10158 10159 10160 10161 10162 10163 10164 10165 10166 10167
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

10168 10169 10170
void online_fair_sched_group(struct task_group *tg)
{
	struct sched_entity *se;
10171
	struct rq_flags rf;
10172 10173 10174 10175 10176 10177
	struct rq *rq;
	int i;

	for_each_possible_cpu(i) {
		rq = cpu_rq(i);
		se = tg->se[i];
10178
		rq_lock_irq(rq, &rf);
10179
		update_rq_clock(rq);
10180
		attach_entity_cfs_rq(se);
10181
		sync_throttle(tg, i);
10182
		rq_unlock_irq(rq, &rf);
10183 10184 10185
	}
}

10186
void unregister_fair_sched_group(struct task_group *tg)
10187 10188
{
	unsigned long flags;
10189 10190
	struct rq *rq;
	int cpu;
10191

10192 10193 10194
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
10195

10196 10197 10198 10199 10200 10201 10202 10203 10204 10205 10206 10207 10208
		/*
		 * 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);
	}
10209 10210 10211 10212 10213 10214 10215 10216 10217 10218 10219 10220 10221 10222 10223 10224 10225 10226 10227
}

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
10228
	if (!parent) {
10229
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
10230 10231
		se->depth = 0;
	} else {
10232
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
10233 10234
		se->depth = parent->depth + 1;
	}
10235 10236

	se->my_q = cfs_rq;
10237 10238
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
10239
	se->parent = parent;
10240
	seqcount_init(&se->idle_seqcount);
10241
	spin_lock_init(&se->iowait_lock);
10242
	se->cg_idle_start = se->cg_init_time = cpu_clock(cpu);
10243 10244 10245 10246 10247 10248 10249 10250 10251 10252 10253 10254 10255 10256 10257 10258 10259 10260 10261 10262 10263 10264 10265
}

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);
10266 10267
		struct sched_entity *se = tg->se[i];
		struct rq_flags rf;
10268 10269

		/* Propagate contribution to hierarchy */
10270
		rq_lock_irqsave(rq, &rf);
10271
		update_rq_clock(rq);
10272
		for_each_sched_entity(se) {
10273
			update_load_avg(cfs_rq_of(se), se, UPDATE_TG);
10274
			update_cfs_group(se);
10275
		}
10276
		rq_unlock_irqrestore(rq, &rf);
10277 10278 10279 10280 10281 10282 10283 10284 10285 10286 10287 10288 10289 10290 10291
	}

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;
}

10292 10293
void online_fair_sched_group(struct task_group *tg) { }

10294
void unregister_fair_sched_group(struct task_group *tg) { }
10295 10296 10297

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
10298

10299
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
10300 10301 10302 10303 10304 10305 10306 10307 10308
{
	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)
10309
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
10310 10311 10312 10313

	return rr_interval;
}

10314 10315 10316 10317 10318 10319 10320 10321 10322 10323
#ifdef CONFIG_SCHED_SLI
static void update_nr_uninterruptible_fair(struct task_struct *p, long inc)
{
	struct sched_entity *se = &p->se;

	for_each_sched_entity(se)
		cfs_rq_of(se)->nr_uninterruptible += inc;
}
#endif

10324 10325 10326
/*
 * All the scheduling class methods:
 */
10327
const struct sched_class fair_sched_class = {
10328
	.next			= &idle_sched_class,
10329 10330 10331
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
10332
	.yield_to_task		= yield_to_task_fair,
10333

I
Ingo Molnar 已提交
10334
	.check_preempt_curr	= check_preempt_wakeup,
10335 10336 10337 10338

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

10339
#ifdef CONFIG_SMP
L
Li Zefan 已提交
10340
	.select_task_rq		= select_task_rq_fair,
10341
	.migrate_task_rq	= migrate_task_rq_fair,
10342

10343 10344
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
10345

10346
	.task_dead		= task_dead_fair,
10347
	.set_cpus_allowed	= set_cpus_allowed_common,
10348
#endif
10349

10350
	.set_curr_task          = set_curr_task_fair,
10351
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
10352
	.task_fork		= task_fork_fair,
10353 10354

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
10355
	.switched_from		= switched_from_fair,
10356
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
10357

10358 10359
	.get_rr_interval	= get_rr_interval_fair,

10360 10361
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
10362
#ifdef CONFIG_FAIR_GROUP_SCHED
10363
	.task_change_group	= task_change_group_fair,
P
Peter Zijlstra 已提交
10364
#endif
10365 10366 10367

#ifdef CONFIG_SCHED_SLI
	.update_nr_uninterruptible = update_nr_uninterruptible_fair,
10368
	.update_nr_iowait	= update_nr_iowait_fair,
10369
#endif
10370 10371 10372
};

#ifdef CONFIG_SCHED_DEBUG
10373
void print_cfs_stats(struct seq_file *m, int cpu)
10374
{
10375
	struct cfs_rq *cfs_rq;
10376

10377
	rcu_read_lock();
10378
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
10379
		print_cfs_rq(m, cpu, cfs_rq);
10380
	rcu_read_unlock();
10381
}
10382 10383 10384 10385 10386 10387

#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;
10388
	struct numa_group *ng;
10389

10390 10391
	rcu_read_lock();
	ng = rcu_dereference(p->numa_group);
10392 10393 10394 10395 10396
	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)];
		}
10397 10398 10399
		if (ng) {
			gsf = ng->faults[task_faults_idx(NUMA_MEM, node, 0)],
			gpf = ng->faults[task_faults_idx(NUMA_MEM, node, 1)];
10400 10401 10402
		}
		print_numa_stats(m, node, tsf, tpf, gsf, gpf);
	}
10403
	rcu_read_unlock();
10404 10405 10406
}
#endif /* CONFIG_NUMA_BALANCING */
#endif /* CONFIG_SCHED_DEBUG */
10407 10408 10409 10410 10411 10412

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

10413
#ifdef CONFIG_NO_HZ_COMMON
10414
	nohz.next_balance = jiffies;
10415
	nohz.next_blocked = jiffies;
10416 10417 10418 10419 10420
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
#endif
#endif /* SMP */

}